Image identification system
Abstract
This record has no abstract on file.
Term
No projected expiry on record.
- Priority
- Filed
- Published
- Today
9 claims: 2 independent, 7 dependent
- 1CLAIMS :1. A method for efficiently and accurately comparing a first image to a plurality of other images, the method comprising: 5 deriving a first image data set based on the first image and a plurality of other image data sets based on the plurality of other images, wherein each data set includes a plurality of data elements and is a model representation of a corresponding image;10 storing the first and each of the other image data sets in memory;and comparing at least one data element in the first image data set with data elements in at least one of the plurality of other image data sets;15 characterised in that comparing comprises: creating a directory that progressively lists, based on at least one measured characteristic, a plurality of a first type of data elements that appear in the plurality of other image data sets, wherein each data element in the 20 directory is listed with an identifier that represents association with a particular image data set within which the data element appears;storing said directory in memory as a searchable database;25 accessing said database and comparing data elements of the first type taken from the first image data set to the data elements that are progressively listed/ in the directory;and providing an output according to whether or not the 30 first image matches one of the plurality of other images.
- 2The method of Claim 1, wherein comparing data elements of the first type taken from the first image data set to the data elements that are progressively listed in the directory comprises:creating an array having a two-entry cell for each of a plurality of potential configurations for the first type of data element;recording in a first entry of at least one two-entry cell, a quantity value representing a number of consecutive data elements in the directory that demonstrate characteristics that are approximately similar to characteristics of the one of the plurality of potential configurations for the first type of data element that is associated with the two-entry cell within which the first entry is being recorded;and recording in a second entry of at least one two-entry cell, an index value corresponding to an initial data element that begins the number of consecutive data elements listed in the directory.
- 3The method of Claim 2, wherein comparing data elements of the first type taken from the first image data set to data elements that. are progressively listed in the directory comprises:identifying a two-entry cell in the array that is associated with a data element configuration having characteristics approximately identical to a target data element taken from the first image data set;and comparing the target data element to a group of consecutive data elements listed in the directory, as indicated by the two-entry cell associated with the target data element;and repeating the identifying and comparing steps for at least one additional target data element.
- 4The method of Claim 3, further comprising generating a count for each of the plurality of other image data sets, wherein generating the count comprises:utilizing the identifiers associated with the data elements listed in the directory to calculate and note a quantity of data elements in each of the plurality of other image data sets that approximately match data elements taken from the first image data set.
- 5The method of Claim 4, further comprising:selecting from the plurality of other image data sets a predetermined number of image data sets having the most data elements that approximately match data elements taken from the first image data set;and performing a more thorough comparison of the predetermined number of image data sets to the first image data set.
- 6The method of Claim 4, further comprising:repeating the previous steps substituting at least one other type of data elements in place of the first type of data elements;selecting from the plurality of other image data sets a predetermined number of image data sets having the most data elements that approximately match data elements taken from the first image data set;and performing a more thorough comparison of the predetermined number of image data sets to the first image data set.
- 7A method for efficiently and accurately comparing a first image data set to a plurality of other image data sets, characterised in that the first and plurality of other image data sets are individually derived from fingerprint images and include a plurality of data elements of a plurality of types, the plurality of types comprising at least one of bifurcation representations, end of ridge segment representations, vector segments representing part of a ridge line, microminutia points such as an area devoid of ridge lines, and combinations thereof, the method comprising:storing the first and each of the other image data sets in memory;creating a directory that progressively lists, based on at least one measured characteristic, a plurality of a first type of data elements that appear in the plurality of other image data sets, wherein each data element in the directory includes an identifier that represents association with a particular image data set within which the data element appears;storing said directory in memory as a׳ searchable database;accessing said database and creating an array having a two-entry cell for each of a plurality of potential configurations for the first type of data element;recording in a first entry of at least one two-entry cell, a quantity value representing a number of consecutive data elements in the directory that demonstrate characteristics that are approximately similar to characteristics of the one of the plurality of potential configurations for the first type of data element that is associated with the two-entry cell within which the first entry is being recorded;recording in a second entry of at least one two-entry cell, an index value corresponding to an initial data element that begins the number of consecutive data elements listed- in the directory;identifying a two-entry cell in the array that is associated with a data element configuration having characteristics approximately identical to a target data element of the first type taken from the first image data set;comparing the target data element to a group of consecutive data elements listed in the directory, as indicated by the two-entry cell associated with the target data element;and providing an output according to whether or not the first image matches one of the plurality of other images.
- 8The method of Claim 7, further comprising:repeating the identifying and comparing steps for at least one additional target data element;utilizing the identifiers associated with the data elements listed in the directory to calculate and note a quantity of data elements in each of the plurality of other image data sets that approximately match data elements taken from the first image data set;selecting from the plurality of other image data sets a predetermined number of image data sets having the most data elements that approximately match data elements taken from the first image data set;and performing a more thorough comparison of the predetermined number of image data sets to the first image data set.
- 9The method of Claim 7, further comprising:repeating the identifying and comparing steps for at least one additional target data element;utilizing the identifiers associated with the data elements listed in the directory to calculate and note a quantity of data elements in each of the plurality of other image data sets that approximately match data elements taken from the first image data set;repeating the previous steps substituting at least one other type of data elements in place of the first type of 5 data elements;selecting from the plurality of other image data sets a predetermined number of image data sets having the most data elements that approximately match data elements taken from the first image data sets;and 10 performing a more thorough comparison of the predetermined number of image data sets to the first image data set.
Independent claims9
885 paragraphs in 20 sections, as filed
BACKGROUND OF THE INVENTION
The present invention relates generally to image identification systems. More specifically, the present invention relates to methods and procedures for improving the performance and reliability of image identification systems.
Image identification systems have been used in the past, one application being biometric image identification systems. One type of biometric image identification system is a fingerprint identification system. In a fingerprint identification system, a user places the tip of a finger on a scanning surface of a fingerprint image reader device. Each ridge of the epidermis (outer skin) is dotted with sweat glands that produce moisture that, in combination with oily secretions and other substances naturally present on the tip of a finger, enable an image of a fingerprint to be scanned. (The present invention can also be successfully applied to images generated from readers that do not rely on the moisture content of the skin to capture an image.). The fingerprint image reader device creates an image scan by capturing a picture of fingerprint ridge 'characteristics present on the tip of a finger. In many systems, the image is then compared to a database of other stored fingerprint images or fingerprint image models for verification, authentication, or some other form of analysis.
Security systems that implement . fingerprint identification technology have the potential of being reliable and easy to use. These benefits arise from the fact that the technology does not require a system user to־ retain any piece of knowledge, such as a password, personal identification number, combination or any other code. Neither must a user possess a card, key or any other physical device to gain access to a secured environment. A fingerprint security authentication key, as opposed to a knowledge or possession based security authentication key is nearly impossible to lose, steal, or be forgotten.
Development of practical . security system applications that incorporate fingerprint image identification technology has been hindered by. a general non-repeatability of data from one image scan to another. In particular, physical variations present in the environment of a fingerprint reader device can cause substantial incongruities from one image scan of a fingerprint as compared to a subsequently taken image scan of the same fingerprint. Differences in the temperature, amount of pressure applied to the scanning surface,. moisture content of the finger, as well as the effects of medications and differences' in blood pressure can all contribute, to substantial incongruities from one image scan to another. These incongruous results hinder the development of most fingerprint identification technology applications because inconsistent data leads to an unacceptably high number of . false acceptances (multiple identifications, which include matching to wrong people) and false rejections (not recognizing an enrolled user) for applications that 5 might require instantaneous and unsupervised comparisons to be made between a scanned fingerprint image and a database of. fingerprint images or fingerprint models. Another problem associated with many .image identification systems is the small 10 amount of ,data gleaned by the typical system from each image. For instance, most' fingerprint identification systems are minutiae-based, typically meaning that only rods, islands, and bifurcations are cataloged and made available for analysis. An ideal irmjge scan 15 performed by a minutiae-based system will typically glean a maximum of approximately 50 useful data' points, and this count may be further compromised by data points that' might not appear in the scanned image due to previously discussed interference in the image 20 reader environment. The discrimination capability of the typical minutiae-based identification system is not adequate for applications that require . instantaneous and accurate comparisons to be made between a real-time scanned image and a database of 25 potential matching images or models. In addition, systems that glean only a small number of useful data points are more susceptible to fraudulently produced' . fingerprint forgeries.
Yet another problem associated with the average image identification system is that they prove to be an inefficient model for making comparisons between a real-time scanned image and a database of potential matching images or models. Most systems compare the real-time scanned image or model derived from that scan with each of the images or models contained within a database of images or models on a one-to-one basis until a matching pair is located. Depending on the size of the database, the time required to locate a matching pair can be substantial.
Due to these classical limitations on image identification technology, image identification applications have typically been limited to use in low security and/or supervised environments within which quick processing is not a priority. For instance, many law enforcement agencies that currently utilize fingerprint identification systems operate within the confines of minutiae-based matching. A minutiae-based system may be adequate in such an environment where a fingerprint expert may be available to take the time necessary to supervise the system and act as the arbiter in cases of multiple matches to an online database.
Minutiae-based systems, and other traditional fingerprint identification systems, are not adequate for unsupervised mass market applications, such as an automatic teller machine (ATM) that incorporates a fingerprint identification system and requires the
I user to submit a valid fingerprint scan when using an ATM card to make a money transaction. Neither are traditional systems appropriate for authentication systems designed to selectively and instantaneously 5 provide access to places and devices such as computers, computer networks, facilities, automobiles and appliances based on the receipt of an authorized image. Efficient and effective functionality of these types of applications depend on a level of rapid and 10 accurate analysis that cannot be consistently achieved by the traditional fingerprint image identification system.
Another benefit associated with an authentication . system that incorporates image identification is that 15 such a system is tunable, meaning the discrimination level or the match requirements during image comparison can be adjusted based on the nature of the environment to be secured and the desired level of security associated therewith. Due to burdens of non20 repeatability of data, false match acceptances, and false match rejections, the range and number of levels within which a traditional image identification system can be tuned is narrowly limited. Such a system may not be tunable at all. Even the highest level of 25 discrimination in a traditional system provides a substantially limited amount of discrimination.
SUMMARY OF THE INVENTION
Methods . and procedures for improving the performance and reliability of image analysis within 5 an image identification system include a series of image qualification functions designed to quickly process a fraction of available scanned image data and to provide feedback to a system User pertaining to image quality and authenticity. In one embodiment, if 10 image qualification leads to the conclusion that the scanned image is fraudulent or of insufficient quality, then processing of £he image is interrupted.
Also included in ' the׳ present invention are functions designed .to produce image models based on 15 original image data and to catalogue such image models into a searchable database. In accordance with one embodiment, the, creation of an image model involves analyzing and manipulating image data received from an image reader, and new data sets originating therefrom. 20 Image models enrolled within a searchable database, in accordance with one embodiment of the present invention, can be derived either from a single scan of. an object or from two or more scans of the same obj ect.
The present invention also includes, functions for comparing one image, model to another. In accordance with one embodiment, a series of shift and rotate algorithms are applied to at least one of the image models until a position at which the two models best
-דcompare is identified. A score that represents a relationship or percentage of data elements that are common between the two image models is computed. In accordance with one embodiment, the level of similarity required in order for two image models to be considered matching is tunable.
Finally, the present invention provides functions for making a quick determination as to which, if any, of a potential thousands (or more, i.e., millions to hundreds of millions) of image models within a searchable database exhibit a desired level of similarity, as compared to a target image model. in accordance with one embodiment, rather than' comparing image models specifically, a set of database index keys that describe different image model characteristics are defined and enable general, rather than specific comparisons to be made. In accordance with one embodiment, discrimination levels can be tuned.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a block diagram of a fingerprint imaging system.
FIG. 2 is a flow diagram illustrating operations to be carried out. within the fingerprint imaging system according to the present invention.
FIG. 3 is a pictorial representation of' an' example set of image scan parameters.
FIG. 4 is a block diagram illustrating a set of procedural components corresponding to an image qualification portion of the operations shown in FIG.
2.
FIG. 5 is an illustration of a raw scan image.
FIG. 6 is an illustration of an intermediate image produced in accordance with a preprocessing portion of the operations shown in FIG.4.
FIG. 7 is an illustration of a monochrome image produced in accordance with the preprocessing portion of the operations shown in FIG. 4;
FIG. 8A is an illustration of a monochrome image derived from a Mylar film source using an LED light source within an image reader.
FIG. 8B is an illustration of a monochrome image derived from .a paper source using an LED light source within an image reader.
FIG. 8C is an illustration of a monochrome image derived from a live'finger source using an LED light source within an image reader.
FIG. 9A is an illustration of a monochrome image derived from a Mylar film source using an infrared light source within an image reader.
FIG. 9B is an illustration of a monochrome image derived from a paper source using an infrared light source within ah image reader.
FIG. 9C is an illustration of a monochrome image derived from a live finger source using an infrared light source within an image reader.
FIG. 10 is an illustration of a monochrome image after a contour trace has been completed in accordance with a slope table generation portion of the operations shown in FIG. 4.
FIG. 11 is an illustration of a monochrome image with a slope overlay based on a slope table completed in accordance with the slope table generation portion of the operations shown in FIG. 4.
FIG. 12 is an illustration of a histogram completed in accordance with a histogram generation portion of the operations shown in FIG. 4.
FIG. 13 is an illustration of the histogram overlaying a raw scan image from which the histogram was derived.
FIG. 14 is an illustration of a histogram cell.
FIG. 15 is a block diagram illustrating a set of procedural components corresponding to a model creation portion of the operations shown in FIG. 2.
FIG. 16 is a block diagram illustrating a set of procedural components corresponding to a preprocessing portion of the operations shown in FIG. 15.
FIG. 17 is an illustration of a corrected raw scan image.
FIG. 18 is ' an illustration of an intermediate image produced in accordance with a preprocessing portion of the operations shown in FIG. 15.
FIG. 19 is an illustration of an enhanced image.
FIG. 20 is an illustration of a monochrome image produced in accordance with a preprocessing portion of the operations shown in FIG. 15.
FIG. 21 is an illustration of a monochrome image after irregularities in the image have been located and filled.
FIG. 22 is an illustration of a filled monochrome image.
FIG. 23 is an illustration of a smoothed and filled monochrome image.
FIG: ;24 is a pictorial representation of an alternate set of example image scan parameters.
FIG. 25 is an illustration .of a monochrome image after a contour trace has been completed in accordance with a slope table׳ generation portion of the operations shown in FIG. .15.
FIG. 26 is an illustration of a monochrome image with a slope overlay based on a slope table completed in accordance with the slope table generation portion of the operations shown in FIG. 15.
FIG. 27 is a block diagram illustrating a set of procedural components corresponding to a wire-frame generation portion of the operations shown in FIG. 15.
FIG. 28 is. an illustration of a monochrome image after a first removal of pixels from image ridge lines.
FIG. 29 is an illustration of a monochrome image with a comprehensive representation of pixel removal passes made during the thinning of the monochrome image to the center-most ridge line pixels.
FIG. 30 is an illustration of FIG. 29 further including an overlay of a thinned monochrome image having raw wire-frame lines.
FIG. 31 is an illustration of a thinned monochrome image with raw wire-frame lines.
FIG. 32 is an illustration of a thinned monochrome image after excess pixels have been removed from the raw wire-frame lines.
FIG. 33 is an illustration . of the relationship between a thinned monochrome image, after excess pixels have been removed, and a corresponding monochrome image.
FIG. 34 is an illustration of a thinned monochrome image, with excess pixels removed, that includes a representation of data from an end-point table.
FIG. 35 is an illustration of a thinned monochrome image, with excess pixels removed, that includes a representation of data from a. center-point table.
FIG. 36 is an illustration of a refined set of wire-frame lines.
FIG. 37 is an illustration demonstrating the relationship between the refined set of wire-frame lines and a corresponding monochrome image.
FIG. 38 is an illustration demonstrating the relationship between a further refined set of wire frame lines, Including, fixed end-points, and a corresponding monochrome image.
FIG. 39 is an illustration demonstrating the relationship between the further refined set of wireframe lines, including fixed and joined end-points, and a corresponding monochrome image.
FIG. 40 is a graphical representation of a fingerprint bifurcation image element.
FIG. 41 is a graphical representation of a fingerprint rod image element.
FIG. 42 is an illustration of .a wire-frame fingerprint image within which, qualified bifurcations and rods have been circled.
FIG. 43 is an illustration of the wire-frame fingerprint image within which qualified bifurcations and rods have been circled and vector segments, have been traced.
FIG. 44 is a block diagram illustrating a set of procedural components associated with a one-to-one image comparison process.
FIG. 45 is a block diagram illustrating a set'of procedural components associated with a one-to-many database image comparison process.
FIG. 46 is a block diagram illustrating a set of procedural components associated with another one-tomany database image comparison process.
DETAILED DESCRIPTION OF ILLUSTRATIVE EMBODIMENTS
The present invention relates to methods and procedures for improving . the. performance and reliability of image identification systems generally. The inventive concepts could be applied within systems designed to operate in conjunction with a broad range of image types, including but not limited to' license plate images, graphic' images and text based images. In addition, the present .invention provides methods and procedures that are particularly suitable for improving the performance and reliability of fingerprint image identification systems .specifically. While the remainder of the detailed description will discuss the present invention in relation to fingerprint image identification systems, it is to be understood that the concepts of the present invention could just as easily be applied within other types of image identification systems.
FIG. 1 is a ' block diagram of a fingerprint imaging system 10 within which the methods and procedures of the present invention could be applied. Imaging system 10 includes a reader portion 12, image analyzer/processor 14 and searchable database 16, which further includes an output 15. Reader portion 12 could be any of a number of known systems capable of scanning an image of a fingerprint and transferring data pertaining to the image to an image analyzer, such as image analyzer/processor 14.
In many cases, reader portion 12 will include an optical device that includes a reflecting face designed to receive the finger to be imaged. Light is input into the optical device by a light emitter and an optical image of the finger is reflected out of the optical device to an imager which receives׳the image and produces an analog image signal indicative of the optical signal received. In many׳ systems, the analog 5 signal is then transferred to a conventional analog/digital converter, which produces a digital representation of the analog signal. The digital signal is reformatted into a digitized image which can be stored and, in accordance with an embodiment of the 10 present invention, manipulated. Finally, the digitized image is transferred out of the reader portion to an image analyzer/processor 14. Image analyzer/processor 14 varies with application, but generally analyzes the image data received for a wide 15 variety of purposes and applications.
In an embodiment of the present invention, as will be discussed in more detail below, image analyzer/processor 14 creates an image model based on the particular features and characteristics of each 20 image received from reader portion 12. These image models are more than facsimiles of their associated fingerprint images and include a unique range of data elements that provide analytical opportunities that are a part of the present invention.
In one embodiment of the present invention, image analyzer/processor 14 compares data elements of one image model to data elements of at least one other image model stored within searchable database 16. The image models contained in database 16 correspond to . -15previously obtained scanned images, while the image model being compared typically corresponds to a contemporaneously scanned image. Fingerprint imaging system 10, through the incorporation of this process, 5 is able to quickly and efficiently make a determination as to whether the image model corresponding to the contemporaneously scanned fingerprint is substantially similar to any of the image models included within the searchable' database 10 16. As will be discussed more fully below, system 10 requires a particular level of similarity for a match to be indicated. In accordance with one embodiment, the level of required similarity is adjustable and can be tuned based on the nature' of the environment for 15 which system 10 is designed to provide security. In this manner, fingerprint imaging system 10 provides .an efficient and accurate fingerprint image identification system that can be used, for instance, as a security measure to determine whether the person 20 who places a finger on the reader portion 12 should be authorized to enter a room, to access a bank account or to take any other variety of actions.
As is shown in FIG. 1, searchable database 16 includes an output'15. The precise nature of output 25 15 depends on the context within which imaging system is to be applied. For instance, output 15 could be an identification indicator of an image contained in searchable database 16 that substantially matches the image scanned by reader portion 12. This is but one example of the many potential forms of output 15.
FIG. 2 is a flow diagram illustrating operations to be carried out within system 10, specifically within analyzer/processor 14, in accordance with an embodiment of the present invention. The process begins when image analyzer/processor 14 receives image data from reader portion 12. After receiving image data, image analyzer/processor 14 first performs, as is indicated by block 18 in FIG. 2, a series of image qualification functions.
Details pertaining to image qualification 18. will be discussed in greater detail with respect to FIG. 4. Briefly,; image qualification 18 involves quickly processing a fraction of the available image data to ensure that the received image is a scan of a real fingerprint (as opposed to a fraudulent fingerprint) and of sufficient quality to proceed with processing. In one embodiment, if the image qualification process leads to the conclusion that the scanned image is fraudulent or of insufficient quality, then processing of the image is stopped or interrupted. In such a case, the system user is provided with feedback pertaining to identified inadequacies and is allowed to continue processing only when the inadequacies have been corrected. Only a fraction of available image data is processed during image qualification 18 in order to expedite processing and to enable feedback to be provided to a system user on a substantially real time basis.
Once the image has been qualified, the next step, as is indicated by block 20 in FIG. 2, is the creation of an image model. Model creation 20 will be described in greater detail with respect to FIG. 15. 'Briefly, model creation 20 involves analyzing and manipulating image data received from reader portion 12, and new data sets originating therefrom, until an image model is produced. Due to an increased need for accuracy, the image data processed during model creation 20 is a complete set of image data, as opposed to the fractional set processed during image qualification 18. While the procedure for creating and the composition of an image model will be described in greater detail below, it should be emphasized that an image model is a collection of data based on the original print image and is not a facsimile of the original print image.
After an image model has been created, in accordance with an embodiment of ׳ the present invention, the image model is utilized for one of two purposes. First, as is indicated in FIG. 2, is model enrollment 22. Model enrollment 22 is the process with which image models are entered into and catalogued within searchable database 16. Image models enrolled within database 16, in accordance with one embodiment of the present invention, can be derived either from a single scan of a* fingerprint image or from two or more scans of the same fingerprint image. When two or more scans are used to create an image model, consistent model elements that show up from scan to scan are noted in the ־ image model. Inconsistent model elements, for example, discrepancies in the' image data that are the result of previously mentioned variations in the reader environment, are eliminated.
In one embodiment of the present invention, when two or more scans are being utilized during model enrollment 22, the finger is removed from the reader portion 12 after each scan and then is subsequently replaced before the next scan is taken. In accordance with another embodiment, a significant amount of time may pass between scans. Because environmental factors such as finger pressure, finger moisture and finger positioning can vary from ,scan to scan, removing the finger from reader portion 12 between scans increases the likelihood that environmental inconsistencies will be eliminated when they do not show up in each individual scan.
As is indicated by block 24 in FIG. 2, and in accordance with another . embodiment of the present invention, the other purpose for which an image model can be utilized is model comparison 24. Model comparison 24 will be described in greater detail below. Briefly, model comparison 24 is a process that can be utilized to compare one image model to another. Model comparison 24 is accomplished by applying a series of shift and rotate algorithms to at least one of the image models until a position at which the two models best compare is identified. Then, a score that represents, a relationship or percentage of data elements that are common between the two image models is computed.
As is indicated by block 26 in FIG. 2, and in accordance with an illustrative embodiment of the present, invention, . database search 26 could be performed in place of or in combination with model comparison 24. Database search 26 will be described in greater detail below. Briefly, database search 26 involves a quick and efficient determination as to which, if any, of. a potential thousands, or even millions, of image models within database 16 exhibit a desired level of similarity, as compared to a target image model. In accordance with one embodiment,' the target image model is an image model associated with a contemporaneously scanned image. Rather than comparing image models specifically, a set of database keys that describe different image model characteristics are defined and enable general, rather than specific comparisons to be made during. the database search 26 process. The desired level of similarity is adjustable and could be selected based on a desired processing speed, a desired level of security, and other characteristics indicative of the environment for which system 10 is designed to provide security.
It should be emphasized that nearly all, with an anti-spoofing procedure to be discussed later in this application being a primary exception, of the methods and procedures of the present invention are not dependent upon the' inclusion of a particular reader portion 12 and can .be retrofitted to work with any reader technology. For the purpose of illustrating embodiments of the present invention, however,. an example set of image scan parameters that correspond to an example reader portion 12 will be adopted. In particular, the example parameters will correspond to a SACcat™ fingerprint reader device offered and marketed by Secured Access Control Technologies (doing business as ΒΙΟ-key International), of Eagan, Minnesota.
FIG. 3 is a pictorial representation of details pertaining to the example set of image scan parameters. The example parameters are generally indicated by reference numeral 28 and are not critical to the present invention. The. example reader portion 12, which produces the example parameters 28, illustratively includes a camera that has an aspect ratio of 4 to 3 and provides 64 levels of gray-scale, where neither value is critical to the present invention. As is illustrated by FIG. 3, image scan parameters 28 .include a scan area 30 which is larger than a processing area 32. Processing area 32 is part of scan area 30 and is the only portion of scan area 30 that provides data that is actually captured for analysis. Within scan area 30, there are 510 Lines and 483 pixels per line. For the purpose of simplifying explanation of the present invention, it is to be assumed that the reader.portion 12 produces 5 no linear distortion due to optics (a flat image is assumed).
FIG. 4 is a block diagram illustrating a set of procedural components' corresponding to the image qualification 18 portion of the operations illustrated 10 in FIG. 2. It should be emphasized that the primary purpose of image qualification 18 is to ensure that the image data received by image analyzer/processor 14 from reader portion 12 is a scan of a non-fraudulent fingerprint and of suitable quality for subsequent 15 image processing.
In accordance with an embodiment of the present invention, as was previously alluded to, all of the functions within image qualification 18 are carried out utilizing a fraction of the image data potentially 20 available for analysis. In one embodiment, analyzer/processor 14 receives a complete set of image data from reader portion 12 but utilizes only every other line and every other pixel of information for analysis during image qualification 18. In other 25 words, only one quarter of the data within processing area 32 is analyzed during image . qualification 18. The purpose of processing only a fraction of available data is to expedite processing, thereby enabling feedback pertaining to image quality and authenticity to be provided to a system user in a substantially real time manner. Upon receiving real time feedback, a system user is then allowed to adjust .variables (change pressure applied to scanning surface, produce a non-fraudulent image source, wipe excessive moisture from finger, etc.) until all negative feedback is remedied and the image scan is of sufficient quality to continue with the processing of the image.
In more detail, image qualification 18 begins with preprocessing 34 (see FIG. 4). When analyzer/processor 14 receives image data from reader portion 12, it is in a raw scan, also known as grayscale, format. The general purpose of preprocessing 34 is to reduce data. More particularly, the purpose is to convert the raw scan image into a monochrome image or binary image, which is desirable for subsequent image qualification 18 processing. In accordance with one embodiment, preprocessing 34 is utilized to convert the raw scan image into an image that incorporates single bits that are black or white.
During preprocessing 34, a raw scan image, similar .to raw scan image 46 in FIG. 5, is received from reader portion 12 and first transformed into an intermediate image similar to intermediate image 48 in FIG. 6. As the Figures illustrate, intermediate image 48 is similar to raw scan image 46 but includes enhancements of primary features. To accomplish the image transformation, in accordance with an embodiment of the present invention, each pixel in intermediate י -23image 48 is created by averaging an n x n pixel (where n is greater than 1) array taken from the raw scan image 46. In accordance with one embodiment, 3x3 pixel arrays are utilized. The pixel (new pixel 5 value) at row y and column x in intermediate image 48 is given by:
Equation 1
Set new pixel value to zero.
For xl values of x-1 to x+1 do
For yl values of y-1 to y+1 do
Add to new pixel value the value of the pixel in Raw. Scan at xl and yl.
Divide new pixel value by 9.
Store new pixel value in an intermediate image buffer at row y and column x.
The next step in preprocessing 34, in accordance with one embodiment, is to convert intermediate image <sub>v </sub>20 48 (FIG. 6) into a monochrome image similar to monochrome image 50 in FIG. 7. In accordance with an embodiment of the present invention, the transformation from intermediate image 48 (FIG. 6) to monochrome image 50 (FIG. 7) is accomplished . as 25 follows: Each pixel in the monochrome image is created by comparing the 5x5 average value of a pixel taken from intermediate image 4 8 with a 3 x 3 . average for t!he. same pixel location. It should be pointed out that different sizes of pixel arrays could 30 be utilized without departing from the scope of the present invention. The pixel (new pixel value) at row y and column x in the monochrome image is given by:
Equation 2.
Set average_l value to zero. For xl values of x-2 to x+2 do For yl values of y-2 to y+2 do Add to average_l value the value of the pixel in enhanced image at xl and yl
Divide average_l value by 25 (5 multiplied by 5).
Set average_2 value to zero. For xl values of x-1 to x+1 do For yl values of y-1 to y+1 do Add to average_2 value the value of the pixel in enhanced image at xl and yl Divide average_2 value by 9 (3 multiplied by 3). If average_2 value is greater than average_l value Then set pixel value to zero Else, set pixel value to 255.
Store pixel value in monochrome image at row y and column x.
With reference to FIG. 4, another process incorporated within image qualification 18 is antispoofing 36. In order for fingerprint imaging system 10 to incorporate anti-spoofing techniques as described herein, a reader portion 12 that includes both an infrared light source and an LED light source for shining light into an optical device must be
ע incorporated in the system. Of course, with other readers, other anti-spoofing technologies can be implemented. Anti-spoofing 36 is a method for detecting a non-live finger, such as a fake finger, a drawing qn paper or a photo-plot on Mylar film. Antispoofing also provides protection against the prerecorded/playback of live scans. Anti-spoofing, in accordance with one embodiment of the present invention, involves the capture and comparison of two consecutive images, where the .first image is side-lit by an infra-red light source and the second image is 5 back-lit by a visible LED light source.
The anti-spoafing process starts by' insuring that the back-lit LED light source is turned off. Next, the side-lit infra-red light source is turned on. In accordance with one embodiment, this switching of 10 light sources is performed on a random basis to defeat prerecorded/playback spoofing' attack scenarios. The infra-red lit image is captured and, in one ' embodiment, is preprocessed in accordance with previously described preprocessing 34 to produce a '15 first monochrome image. Monochrome images 58, 59., and 60, respectively depicted in FIGS. 9A, 9B and 9C, illustrate images derived using an infra-red light source to scan images contained on a Mylar . film source, a paper source and a live finger source, also 20 respectively.
The next step in the anti-spoofing process is to turn off the infra-red light source and to turn on.the back-lit LED lightsource in order to capture a second image, which in accordance with one embodiment, is 25 preprocessed and transformed into a second monochrome image. Monochrome images 52, 54, and 56, respectively depicted in FIGS. 8A, 8B and 8C, illustrate images derived using a back-lit LED light source to scan images from a Mylar film source, a paper source and a live finger source, also respectively.
In the final step of anti-spoofing, the infra-red originated monochrome images are compared to the LED originated monochrome images and matching pixel values are noted. Generally, live finger image scans will produce a very high correlation of like values as compared to images based on fraudulent image sources. Illustratively, images 56 and 62 are substantially the same, whereas the primary features in images . 52 and 58, and images 54 and 60 include pixels having values substantially opposite to one another (i.e. a feature that is black in one image is not black in the corresponding comparison image). In one embodiment of the present invention, fingerprint imaging system 10, when confronted with results that indicate a non-live image has been presented, will terminate further processing until a live finger is presented for scanning.
It should be noted that while the anti-spoofing method has. been described, in relation to the comparison of monochrome scan images, the antispoofing process could just as easily be applied to raw scan or other image configurations. Because monochrome images, however, are comprised of a limited range of pixel values, they provide a smooth comparative model that typically produces a clear and accurate result., ־27Referring to FIG.
4, another component within the process ׳ is slope table The purpose of the slope table, once not to create information directly provide feedback to the system statistical tool that is used image qualification 1θ the slope table could user, but to a3 an aid in processing.
be used could be to used the
34, image qua11!נ generation 38. it is generated, is used to create a subsequent Specifically, supplement histogram generation 40 and during print center determination 42.
. !ס begin slope table generation 3B, monochrome image created during preprocessing illustratively monochrome image SO (FI<j. 1 .(י<sup>s f1rst </sup>divided into an array ot n x ״ pixel (where n is greater th1 ״־).' In embodiment, ah array of 3 x 3 pixel grids is utilized. In accordance with this embodiment, and in accordance with example image ־can parameters 23 (FIG. 3), an array of 3 x ־ ^ids yields 27 grids in the x direction and 29 grids ih the y direction.
To aid in the creation of the slope table, a raw slope table i־ ־tested. The raw slope data table is illustratively, in accordance with example parameters 23, a two dimensional array 27 x 29 where each entry in the table contains three entries:.
1. A count of the changes in the x coordinate.
2. ft count of the changes in the y. coordinate.
3. A count of the pixels tested.
The raw slope data table is created by doing a contour trace of the features within'each pixel grid of the array of pixel grids into which monochrome image 50 has-been divided. As the trace migrates 5 through the pixel grids, the three elements included in the raw slope data table are incremented. Below is a diagram showing the values to be added to the raw slope data table for the eight possible next pixel combinations (P is the. current pixel, N is the next 10 pixel, * represents an ordinary pixel and serves as a filler for display purposes):
Equation 3
<td colspan="3"> ***</td><td rowspan="4"> *p* **N x = y =</td><td colspan="2" rowspan="4"> + 1 + 1</td><td colspan="3"> * * *</td><td colspan="2" rowspan="2"> ★ ★ ★ *p* N**.</td>
<td rowspan="3"> 15</td><td colspan="2" rowspan="2"> *PN * .fr ★־ X = + 1</td><td rowspan="3"> *P* *N* x = y = +.</td><td colspan="2" rowspan="3"> 0 1</td>
<td rowspan="2"> X = y = +</td><td rowspan="2"> 1 1</td>
<td> y =</td><td> 0</td>
<td> 20</td><td> ** *</td><td></td><td> N**</td><td></td><td></td><td> *N*</td><td></td><td></td><td> **N</td><td></td>
<td></td><td> NP*</td><td></td><td> *p*</td><td></td><td></td><td> *p*</td><td></td><td></td><td> *p*</td><td></td>
<td></td><td> ** .*</td><td></td><td> *.**</td><td></td><td></td><td> ***</td><td></td><td></td><td> * ★ #</td><td></td>
<td> 25</td><td> x =</td><td></td><td> X =</td><td></td><td></td><td> X =</td><td> 0</td><td></td><td> x = +</td><td> 1 ן־</td>
<td></td><td> y =</td><td> 0</td><td> y =</td><td></td><td></td><td> y =־ -</td><td> 1</td><td></td><td> y = -</td><td> !</td>
<td></td><td></td><td> Image</td><td> 64 in</td><td> FIG.</td><td> 10 :</td><td> is an</td><td colspan="3"> illustration</td><td> of a</td>
<td></td><td colspan="2"> monochrome</td><td> image</td><td> after</td><td> the</td><td colspan="2"> contour</td><td> trace</td><td> has</td><td> been</td>
<td> 30</td><td colspan="2"> completed.</td><td></td><td></td><td></td><td></td><td></td><td></td><td></td><td> t</td>
<td></td><td></td><td> When</td><td colspan="2"> the contour</td><td colspan="3"> trace has ,</td><td> been</td><td colspan="2"> completed</td>
<td></td><td colspan="2"> throughout</td><td> every</td><td> pixel</td><td> grid</td><td> and</td><td> the</td><td> raw</td><td> slope</td><td> data</td>
table is complete, the slope table is ready to be generated. The Slope table is a two-dimensional array and, in accordance with example parameters 28 (FIG. 3), is 27 x 29. Each entry in the slope table consists . of a single entry, namely the slope of a ridge or ridges flowing through each particular pixel grid. Initially, all entries in the slope table are set to a -one (invalid slope) . The slope for each pixel grid is calculated utilizing information from the raw slope data table and is specifically computed as follows:
Equation 4
Set x coordinate count to zero.
Set y coordinate count to zero.
Set pixel count value to zero.
For xl values of x-1 to.x+1 do
For yl values of y-1 to y+1 do from raw slope table.at coordinates xl and yl do
Add to pixel count the count .of. pixels tested.
Add to x coordinate count the changes in the x . coordinate.
Add to y coordinate count the changes in the y coordinate.
from raw slope table at coordinates x and y do
Add to pixel count the count of pixels tested then divide by 2. ,
Add to x coordinate count the changes in the x coordinate then divide by 2.
Add to y coordinate count the changes in the y coordinate then divide by 2.
If the pixel count is greater than 10
Then compute the slope using the trig function arcsine.
Find angle function ־30Input: delta y and delta x (computed previously above)
Set quadrant to 0
If delta y is less than 0
Then add 2 to quadrant
If delta x is less than 0
Then add 1 to quadrant
Hypotenuse == square root of ( (delta x times delta x) + (delta y times delta y))
Angle = arcsine (delta y divided by hypotenuse) times degrees per radian.
If quadrant is 1
Then angle = 180-angle
Else if quadrant is 2
Then angle = 360-angle
Else if quadrant is 3
Then angle = 180+angle
Since slopes have values between 0 and 180, the angle is converted to a slope as follows:
If angle is equal to or greater than 180 . Then slope is angle minus 180
Else slope is the angle Increment number of pixels<processed by one.
Image 66 in FIG. 11 is an illustration- of a monochrome image with a slope overlay ' based on a completed slope table.
Referring to FIG. 4, another component of image qualification 18 is histogram generation 40. A completed histogram is used within imaging system 10 to determine the quality of scanned fingerprint image data and the adequacy of the image data for subsequent processing.
A completed histogram is a multiple dimensioned n x n array (where n is greater than .1), illustratively two dimensional, and in accordance with example parameters 28 (FIG. 3), a 6 x 6 array. Each cell within the array corresponds to a portion of the image data under analysis. Image 68 in FIG. 12 is an illustration of a completed histogram that includes cell 67, in addition to other unlabeled cells. Image 70 in FIG. 13 is an illustration of the same completed histogram overlaying the raw scan image from which the histogram was illustratively derived. Assigning different portions of an image to different cells of the histogram array enables multiple individual quality determinations to be made for limited quantities of image data corresponding to each, of the different histogram cells, rather than a single quality determination being made for the entire set of image data. In one embodiment of the present invention, these multiple quality determinations can be utilized to selectively exclude portions of the image data corresponding to cells that demonstrate low quality characteristics. After low quality cells have been excluded, a positive or negative system determination can be made as .to whether enough data cells of acceptable quality are available for subsequent processing.
In accordance with an embodiment of the present invention, each cell of a histogram includes a histogram list. The histogram list, in accordance with the above described example reader portion 12, is an array of 64 entries (zero to 63) . Each entry is assigned a pixel value (example reader portion 12 has 64 potential pixel values) and includes a count of the number of image data pixels having the assigned pixel value. Each histogram cell. also illustratively includes a count of the number of pixels within the cell that are processed and classified in the histogram list.
It is to be understood that some reader technologies may require histograms with different configurations in order for accurate quality determinations to be made. For instance, some reader portion 12 technologies my include a broader or narrower range of pixel values. It is to be understood that histograms tailored to accommodate other reader portion 12 technologies are still'within the׳ scope of the present invention.
A more detailed description of the functions performed during the generation of an illustrative two dimensional, 6 x 6 histogram array during the histogram generation 40 portion of image qualification 18 is as follows:
Equation 5.
Box width is defined as the number of pixels per line divided by 6 (every other pixel included).
Box height is defined as the number of lines divided by 6 (every other line included).
For x values of zero to line length do׳
For y values of zero to number of lines do Pixel value is contents of raw scan image at coordinates x and y.
Slope table x coordinate is at x divided by (illustrative slope table grid size).
Slope table y coordinate is at y divided by (illustrative slope .table grid size).
If the contents of the slope table is not 1 (Recall -1 represents an area that the slope could not be computed).
Then
Histogramtable x coordinate is at x divided by Box width.
Histogram table y coordinate is at y divided by Box height.
Increment histogram list, at index pixel value, by one.
In one embodiment of histogram generation 40, 20 image quality is divided into four classifications:
1. Excellent.
2. Good.
3. Fair.
4. Poor.
In addition, those areas that are considered to have fair or poor quality may have two additional attributes: too dark or too light.
The precise details as to the types of data elements recorded in a completed histogram, and how 30 those data elements are interpreted to make image quality classifications differ depending on the type of data desired and the reader portion 12 that is being used within fingerprint imaging system 10. In other words, quality classification can be tuned in accordance with the type of image quality data desired and in accordance with a particular reader portion 12.
In one embodiment of quality classification, the data recorded in each histogram cell includes seven particular data elements. In the interest of simplifying description, the seven data elements shall be given labels A-G. Histogram cell 72 in FIG. 14 includes data elements A-G, which illustratively correspond to the following information.
Equation 6
A. Represents the number of pixels in the histogram list corresponding to the most white 25% of the listed pixel values.
B. Represents the number of pixels in the histogram list corresponding to the most white 35% of the listed pixel values.
C. Maximum height between points B and F. (not used in quality determination)
D. Average pixel value, (not used in quality determination)
E. Minimum height between points B and F. (not used in quality determination)
F. Represents the number of pixels in the histogram list corresponding to the most black 35% of the listed pixel values.
G. Represents the number of pixels in the histogram list corresponding to the most black 25% of the . listed pixel' values.
In accordance with one embodiment of quality determination, image data quality is determined by comparing the columns associated with points A, B, F <sub>a</sub>nd G. Specifically, image data quality is illustratively determined as follows:
־3-5Equation 7
<td rowspan="3"></td><td> Excellent: A - B <- 2 and</td>
<td> and ־. 59 < B</td>
<td> F - G <= 2 and</td>
<td> 5 10</td><td> F <5 Good: A - B <= 2 and B >55 and F - G <= 2 and F < 9 Fair light: B >59 and , F > 10 Fair dark: B <59 and G <10</td>
<td> 15</td><td> Poor light: B >59 and F >30 Poor dark: B <59 and G <30</td>
<td> 20</td><td> As will be discussed in more detail below, feedback relating to the ascertained image quality is provided to a system user in accordance with feedback interaction 44 . Referring again to FIG. 4, print center</td>
<td> 25</td><td> determination 42 is another component that could be included within image qualification 18. Print center determination 42 is performed by analyzing image data in order to find the center of the .associated print image. One way that print center determination 42</td>
<td> 30</td><td> could be accomplished is through the application of a set of filter rules to data contained in the slope table generated during, slope table generation 38. After the print center has been determined, a further determination is made as to whether a new scan should</td>
<td> 35</td><td> be taken with the system user's finger repositioned on</td>
an imaging surface of reader portion 12. Feedback relating to this further determination is provided to a system user in accordance with feedback interaction 44.
Referring once again to FIG. 4,. feedback interaction 44 is another potential component of image qualification 18. As was previously mentioned, reader portion 12 of fingerprint imaging 10 (.FIG. 1) is capable of capturing a live scan of a fingerprint. In 10 accordance with feedback interaction 44, as image analyzer/processor 14 receives . fingerprint image data from reader portion 12 and performs the functions of image qualification 18 on a fraction of that data, • substantially real time feedback and instructions are 15 provided to the user of system 10 as to inadequate characteristics of the scanned image data that might be improved. Feedback and instructions for the correction of inadequacies of image data might pertain to the proper positioning of the user's finger on 20 reader portion 12 (print center determination 42) .
Alternatively, they may pertain to the detection of a live finger (anti-spoofing 36) or to image data quality characteristics (histogram generation 40) . In one embodiment of the .present invention, feedback 25 pertaining to the moisture content ’of the system • user's finger may also be provided.
Once image qualification 18 has been completed, the next step, as is indicated by block 20 in FIG. 2 is the creation of an image model. FIG. 15 is a block diagram illustrating a set of procedural components that, in accordance with an embodiment of the present invention, make up model creation 20. To enhance the accuracy of model creation 20, substantially all available image data, in one embodiment, all the image data included within example processing area 32 (FIG. 3), is made available to image analyzer/processor 14 for model creation 20 processing. This stands in contrast to the fraction of data processed during image qualification, 18 for speed and efficiency purposes. While some of the components of model creation 20 ־ are similar . to components of imaging qualification .18, none of. the data sets generated during image qualification 18 are utilized during model creation 20. Model creation 20, like image qualification 18, starts with a set of raw scan image data and proceeds from that point.
Model creation 20, in accordance with FIG. 15, begins with anti-spoofing 74. Anti-spoofing 74 is an optional step and is performed in substantially the same manner and for the same reasons described above in relation to anti-spoofing. 36, a procedural component of image qualification 18. One key difference' between anti-spoofing 74 and anti-spoofing 36, however, is that anti-spoofing 74 is performed utilizing a complete data set, whereas anti-spoofing 36 is performed utilizing only a fraction of available data. The purpose of anti-spoofing 74 is to provide further . insurance that the. source of raw scan image data is not a fraudulent one. In accordance with one embodiment of the present invention, when antispoofing 74 leads to the indication that the source of raw scan data is fraudulent, then subsequent processing is terminated until a valid image source is submitted to system 10.
Anti-spoofing 7 4 could be performed utilizing raw scan image, data or an alternate image data format produced during model creation 20. For instance, anti-spoofing 74 could be preformed utilizing monochrome images that, as will be discussed below, are the product of preprocessing 76. In other.words, while anti-spoofing 74 has been illustrated in FIG. 15 as the first step in model creation 20, it could be performed later in the model creation 20 process, or, because anti-spoofing 74 is optional, the step could be eliminated altogether.
An early step in the model creation 20 process, as is indicated by block 76 in FIG. 15, is preprocessing 76. The purpose of preprocessing 76 is to produce a monochrome image with an adjusted aspect ratio and with 'smooth and distinct features suitable for subsequent processing.׳ Preprocessing 76 . is different than the preprocessing step described above in relation to image qualification 18. In particular, preprocessing 76 is performed utilizing a complete, rather than fractional, set of available image data. In addition, preprocessing 76 includes some additional steps intended to . eliminate irregularities and inconsistencies in the resulting monochrome image. These steps., while unnecessary for image qualification 18, prove to be beneficial to subsequent processing during model creation 20 processing.
FIG. 16 is a block diagram illustrating the primary procedural components of preprocessing 76, in accordance with an embodiment of the present invention. As is illustrated by block 90, an early step in the process is to generate a set of image data similar to the raw scan image data but with a modified aspect ratio. In accordance with one embodiment, the aspect ratio is adjusted to a 1 to 1 configuration. The correction of the aspect ratio is necessary for subsequent processing that involves rotation of the image and its corresponding data. Conceivably, model creation 20 could be carried out without adjusting the image aspect ratio, but the adjustment is beneficial to procedures carried out after model creation 20, such as model comparison 24.
In accordance with an embodiment of the present invention, the aspect ratio of a raw scan image is modified by copying the raw scan image line by line and replicating lines at appropriate times and places so as to produce a corrected raw scan image with the desired aspect ratio scale. Image 98 in FIG. 17 is an illustration of a corrected raw scan image, wherein the aspect ratio of a raw scan image has been adjusted to 1 to 1.
Another component of preprocessing 76, in accordance with block 92 in FIG.' 16, is the conversion of a corrected raw scan (image 98 in FIG. 16) into a monochrome image. Because a monochrome image produced accordingly will be based on all available image data associated with an image having a modified aspect ratio, it is unlikely that this monochrome image will be identical to the one generated during image qualification 18. In addition, characteristics within the model' creation 20 monochrome image, as will be .described below, are eventually modified and manipulated to emphasize particular image characteristics. This emphasizing of image characteristics is beneficial to model creation 20 but is unnecessary for image qualification 18.
In accordance with an embodiment of the present invention, the first 'step in the conversion of a corrected raw scanned image to a monochrome image is the creation of an intermediate image. The purpose of creating an intermediate image is to average features within the corrected raw scan image that are predominantly too light or too dark, possibly due to the moisture content of a system user's finger or lighting characteristics. Averaging of these features creates a resultant image that provides for more complete and consistent wire frame generation which follows in subsequent processing steps.' In accordance with one embodiment, to create the intermediate image, each pixel is selected by averaging a 5 x 5 pixel array taken from a corrected raw scan of an image (corrected meaning that the image aspect ratio has been adjusted). The pixel (new pixel value) at row .y in column x in the intermediate image is given by:
Equation 8
Set new pixel value to zero.
. For xl values of. x-2 to x+2 do
For yl values of y-2 to y+2 do
Add to new pixel value the value of the pixel in Corrected Raw Scan at xl and yl
Divide new pixel value by 25 and round to the nearest integer value.
Store new pixel value in intermediate image at row y and column x.
Image 100 in FIG. 18 is an illustration of an intermediate image. . It is to be understood that, without departing from the spirit of the present invention, other sized pixel arrays could be utilized during the transformation to intermediate image, format.
In. accordance with one embodiment of the
ר . ’ ׳ ' ’ transformation from a corrected raw scan image format to a monochrome image format, after an intermediate image has been obtained, an edge detect algorithm is applied to the intermediate image in order to produce an enhanced image. In accordance with one embodiment, the edge detect algorithm is applied as follows:
Equation 9
Set new pixel value to six times the value of the pixel in intermediate image at'row y and column x.
Subtract from new pixel value the value of the pixel in intermediate image at row y-1 and column xSubtract from new pixel value the value of the pixel in intermediate image at row y-1 and column x+1.
Subtract from new pixel value the value of the pixel in intermediate image at row y+1 and column xSubtract from new pixel value the value of the pixel in intermediate image at row y+1 and column. x+1.
If new pixel, value is less than zero,. set new pixel value to zero.
Store new pixel value in enhanced image at row y and column x.
Image 102. in ,FIG. 19 is an illustration of an enhanced image after the edged detect algorithm has been applied.
The final step of the block 92 portion of preprocessing 7 6 (FIG. 16) is to transform the • enhanced image into a monochrome image format. In one illustrative embodiment, each pixel in the monochrome image is created by determining an average pixel value of a large grid area and comparing this average to the average pixel value of a smaller grid area for the same pixel location. A threshold separation variance between the average pixel values of the large and small pixel grids is utilized ,for the determination of setting the corresponding pixel at that location to .a white or black level (i.e., monochrome image result). Pixel grid sizes and threshold values can be chosen to accommodate the' characteristics of the image reader being utilized. In accordance With one embodiment, ־-43the pixel (new pixel value) at row y in column x in the monochrome image is given by:
Equation 10
Set average_l value to zero.
For xl values of x-6 to x+6 do
For yl values of y-6 to y+6 do
Add to average_l value the value of the pixel in edge detect image at xl and yl
Divide average_l value by 169 (13 multiplied by 13) .
Set average_2 value to zero.
For xl values of x-1 to x+1 do
For yl values of y-1 to y+1 do
Add to average_2 value the value of the pixel in edge detect image at xl and yl
Divide average_2 value by 9 (3 multiplied by 3).
If average_2 value is greater than average_l . value plus 4
Then set pixel value to zero . Else, set pixel value to 255.
Store pixel value in monochrome image at row y and column x.
Image 104 in FIG. 20 is an illustration of a monochrome image produced accordingly.
Another component of preprocessing 76, as is indicated by block 94 in FIG. 16, is to locate and fill irregularities, in particular small holes, within the monochrome image. Locating and filling small holes in the monochrome image is necessary so that a wire-frame image that is subsequently derived from the monochrome image during model creation . 20 will not have bubbles or voids in it. The irregularity location process is performed by scanning a monochrome image, such as image 10.4 in FIG. 20, until an unprocessed pixel value of zero is detected. At that point, the detected pixel is marked as being processed and a recursive descent routine is called. Each zero value pixel is. marked as processed and corresponding x and y coordinates stored in a table. When no further, zero value pixels can be located, the . size of zero value pixel areas is calculated (number of. x and y coordinate entries in the table). Illustratively, if the area.size is 35 pixels or less, the 35 pixel value being ' selected based on the image characteristics outlined in relation to FIG. 3, and provided the shape of the area is roughly a circle, the. area is filled using a pixel value of 255. The number of pixels required in order for an area to be considered for filling can be adjusted to accommodate a particular reader portion 12 without departing from the spirit of the present invention. When the entire monochrome image has been checked, the block 94 process is complete.
Image 106 in FIG. 21 is an illustration of a monochrome image, such as image 104 in FIG. 20, after irregularities have been located and, for illustrative purposes, filled with pixels having a substantially white value. ' In accordance with one embodiment of the present invention, the center coordinates of the white filled areas within image 106 in FIG. 21, and the size of these areas are stored in a table and classified as data element points. These data element, points, illustratively called micro-minutia, classified by their location and associated slope value of the ridge they reside on, are image data points that are unique to a particular system user and, in combination with other data element .points, can be catalogued and utilized in the comparison of one set of image scan data to another. The micro-minutia points are small, (as small as one thousandth of an inch in diameter), and likely represent the locations of a system user s sweat glands. It should be noted that the number of micro-minutiae points identified in an image scan is substantially dependent upon the resolution capabilities of a particular reader portion 12 and upon the condition of the fingertip (wet, dry, damaged, etc.). The higher the resolution capability of the reader portion 12, the more micro-minutiae points available for identification.
The final component of preprocessing 76, in accordance with block 95 in FIG. 16, is to smooth and fill image elements within the monochrome image. In the case of fingerprint image data, the image elements within the monochrome image are typically fingerprint ridge elements. The smooth and fill process is designed to add. and remove pixels on the border of fingerprint ridge elements. Smoothing the boundaries along the ridge elements optimizes the quality of subsequently produced wire-frame images, which are derived from the completed, monochrome, image later m the model creation 20 process.
The input to the smooth and fill 96 process, in one embodiment, is the monochrome image after it has been filled in accordance with block 94. Image 108 in FIG. 22 is an illustration of a filled monochrome image, wherein filled pixels no longer includes a substantially white value, as was the case in FIG. 21. The output from the smooth and fill 96 process. is a smooth monochrome image similar to image 110 in FIG. 23. To make the transformation, each pixel in the filled monochrome image is used as the center of a 3 x 3<sub>ג</sub> array. Each of the surrounding eight pixels are used to , form an index into a table. The content of each table entry contains two flags:
1. . Do nothing
2. Set the center pixel
Below are the indexes into the table for those values that contain the set flag. All other table entries contain the do nothing flag.
Equation 11 | 6 I 7 [where the index value is equal to ---ן---1--- the binary sum of pixel locations ן p 0 ן that are present and is represented ---ן — ן as a hexadecimal value (i.e.; pixel 3|2|1 location 4 if present = 2<sup>4th</sup> which equals decimal value 16); and where Count equals the number of pixels present in a 3x3 array surrounding the center P pixel location.]
-W-
<td></td><td> index</td><td> value</td><td> OxlF:</td><td> FILL</td><td colspan="2"> - Count =</td><td> ή</td>
<td></td><td> index</td><td> value</td><td> 0x7C:</td><td> Jr Jr Jr</td><td> **</td><td></td><td> kk</td>
<td></td><td> index</td><td> value</td><td> OxFl:</td><td> *X*</td><td> *X</td><td> *X*</td><td> X*</td>
<td> 5</td><td> index</td><td> value</td><td> 0xC7:</td><td></td><td> **</td><td> * * *</td><td> kk</td>
<td></td><td> index</td><td> value</td><td> 0x3F:</td><td> FILL</td><td colspan="2"> - Count =</td><td> 6</td>
<td></td><td> index</td><td> value</td><td> OxFC:</td><td> **λ</td><td> kk</td><td> *</td><td> kkk</td>
<td></td><td> index</td><td> value</td><td> 0xF3:</td><td> *X*</td><td> *X</td><td> *X*</td><td> X*</td>
<td> 10</td><td> index</td><td> value</td><td> OxCF:.</td><td> * .</td><td> k k k</td><td> ***</td><td> kk</td>
<td></td><td> index</td><td> value</td><td> 0x3E: '</td><td> FILL</td><td colspan="2"> - Count =</td><td> 5</td>
<td></td><td> index</td><td> value</td><td> OxF8:</td><td> ***</td><td> *</td><td> *</td><td> kkk</td>
<td></td><td> index</td><td> value</td><td> 0xF3:</td><td> *X</td><td> *X</td><td> X*</td><td> X*</td>
<td> 15</td><td> index</td><td> I value</td><td> 0x8F:</td><td> k</td><td> kkk</td><td> kkk</td><td> *</td>
<td></td><td> index</td><td> value</td><td> OxBF:</td><td> FILL</td><td colspan="2"> - Count =</td><td> 7</td>
<td></td><td> index</td><td> value</td><td> OxFE:</td><td> trick</td><td> kkk</td><td> k k</td><td> * * *</td>
<td></td><td> index</td><td> value</td><td> OxFB:</td><td> *X*</td><td> *X</td><td> *X*</td><td> X*</td>
<td> 20</td><td> index</td><td> value</td><td> OxEF:</td><td> * *</td><td> kkk</td><td> kkk</td><td> Jr Jr J־</td>
<td></td><td> index</td><td> value</td><td> 0x7E:</td><td> FILL</td><td colspan="2"> - Count =</td><td> 6</td>
<td></td><td> index</td><td> value</td><td> 0xF9:</td><td> ***</td><td> *</td><td> Jr Jr</td><td> kkk</td>
<td></td><td> index</td><td> value</td><td> 0xE7:</td><td> *X</td><td> *X*</td><td> X*</td><td> *X*</td>
<td> 25</td><td> index</td><td> value</td><td> 0x9F:</td><td> Jr Jr</td><td> *Jr*</td><td> ***</td><td> k</td>
<td></td><td> index</td><td> value</td><td> 0x3D:</td><td> FILL׳</td><td colspan="2"> - Count =</td><td> 5</td>
<td></td><td> index</td><td> value</td><td> 0xF4:</td><td> Jr Jr</td><td> *</td><td> ★</td><td> kkk</td>
<td></td><td> index</td><td> value</td><td> 0xD3:</td><td> '*X*</td><td> *X</td><td> *X* .</td><td> X*</td>
<td></td><td> index value</td><td> 0x4F:</td><td> *</td><td> *Jrjr Ή</td><td> *</td>
<td></td><td> index value</td><td> 0x10:'</td><td> FILL</td><td> - Count =</td><td> 4</td>
<td></td><td> index value</td><td> 0x74:</td><td> **</td><td> *</td><td> ★★</td>
<td> 5</td><td> index value</td><td> OxDl:</td><td> *X*</td><td> *X *X*</td><td> X*</td>
<td></td><td> index value</td><td> 0x47:</td><td></td><td> ** **</td><td> *</td>
<td></td><td> index value</td><td> 0x7B:</td><td> FILL</td><td> - Count =</td><td> 6</td>
<td></td><td> index value</td><td> OxED:</td><td> Je .k</td><td> ** **</td><td> ***</td>
<td> 10</td><td> index value</td><td> OxB7:</td><td> *X*.</td><td> x* *x*</td><td> *X</td>
<td></td><td> index value</td><td> OxDE:</td><td> **</td><td> Jr Jr .k Jr Jr</td><td> **</td>
<td></td><td> index value</td><td> 0x37 :</td><td> FILL</td><td> - Count =</td><td> 5</td>
<td></td><td> index value</td><td> 0-xDC:</td><td></td><td> * * *</td><td> **</td>
<td> 15</td><td> index value</td><td> 0x73 :</td><td> *X*</td><td> *X *X*</td><td> X*</td>
<td></td><td> index value</td><td> OxCD:</td><td> *</td><td> ★ * ★ Je</td><td> **</td>
<td></td><td> index value</td><td> 0x5E:</td><td> FILL</td><td> - Count =</td><td> 5</td>
<td></td><td> index value</td><td> 0x79:</td><td></td><td> * *</td><td> **</td>
<td> 20</td><td> index value</td><td> 0xE5:</td><td> *X</td><td> *x* x*</td><td> *X*</td>
<td></td><td> index value</td><td> 0x97:</td><td> ★</td><td> ** ***</td><td> *</td>
<td></td><td> index value</td><td> 0x16:</td><td> FILL</td><td> - Count =</td><td> 3</td>
<td></td><td> index value</td><td> 0x58 :</td><td> ★ *</td><td> *</td><td> A</td>
<td> 25</td><td> index value</td><td> 0x61:</td><td> *X</td><td> *x x*</td><td> X*</td>
<td></td><td> index value</td><td> 0x85:</td><td></td><td> * **</td><td> *</td>
<td></td><td> index value</td><td> 0x67 :</td><td> FILL</td><td> - Count.=</td><td> ; 5</td>
<td></td><td> index value</td><td> 0x9D:</td><td> **</td><td> ** **</td><td> * </td>
<td></td><td> index</td><td> value</td><td> 0x76:</td><td> X*</td><td> *X*</td><td> *X</td><td> *X*</td>
<td></td><td> index</td><td> value</td><td> 0xD9:</td><td> Jr'*</td><td> Jlr</td><td> **</td><td></td>
<td></td><td> index</td><td> value</td><td> 0x6F:</td><td> FILL</td><td colspan="2"> - Count +</td><td> 6</td>
<td> 5</td><td> index</td><td> value</td><td> OxBD:</td><td> ***</td><td> * ★</td><td> **</td><td> * *</td>
<td></td><td> index</td><td> value</td><td> 0xF6:</td><td> X*</td><td> *X*</td><td> *X</td><td> *X*</td>
<td></td><td> index</td><td> value</td><td> OxDB:</td><td> fc ־k</td><td> * *</td><td> jr</td><td> **</td>
<td></td><td> index</td><td> value</td><td> 0x17:</td><td> FILL</td><td colspan="2"> - Count =</td><td> 4</td>
<td> 10</td><td> index</td><td> value</td><td> 0x5C:</td><td> ★ *</td><td> **</td><td></td><td> it</td>
<td></td><td> index</td><td> value</td><td> 0x71:</td><td> *x* .</td><td> *X</td><td> *X*</td><td> X* </td>
<td></td><td> index</td><td> value</td><td> 0xC5:</td><td></td><td> *</td><td></td><td> k ־*</td>
<td></td><td> index</td><td> value</td><td> OxFF:</td><td> FILL</td><td colspan="2"> - Count =</td><td> 8</td>
<td> 15</td><td></td><td></td><td></td><td> ***</td><td> ★ ★</td><td> ***</td><td></td>
<td></td><td></td><td></td><td></td><td> *X*</td><td> *X*</td><td> *X*</td><td> *X*</td>
<td></td><td></td><td></td><td></td><td> ***</td><td> ***</td><td> ***</td><td> ***</td>
Referring to FIG. 16, after image smooth and fill 20 96 has been completed, preprocessing 76, in accordance with an embodiment of the present invention, is also completed.
Another procedural component of model creation 20, in accordance with block 78 in FIG. 15, is slope 25 table generation 78. Slope table generation 78 is substantially similar to slope table generation 38 described above in relation to FIG. 4 and image qualification 18. The primary difference between slope table generation 78 and slope table generation is that during slope table generation 78, all available image data is processed rather than a fraction of available image data. In addition, slope table generation 78 involves the processing of a 5 unique monochrome image formed in accordance with the procedures of preprocessing 76, rather than the limited monochrome image formed in accordance with the procedures of preprocessing 34. .
While slope table generation 38, in accordance 10 with example image scan parameters 28, defined in relation to FIG. 3, involved the processing of an illustrative array of 8 x 8 pixel grids with 27 grids in the x direction and 29 grids in the y direction, slope table generation 78 involves the processing of a 15 more complete data set and a correspondingly different grid configuration. In addition, during model creation 20, example parameters 28 may vary in accordance with aspect ratio adjustments made during preprocessing 76. For. example, FIG. 24 is an 20 illustration of an alternate set of example image scan parameters ' 112 that include a scan area 114 and a processing area 116. Alternate example image scan parameters 112 are similar to example scan parameters 28, but processing area 116 reflects an example change 25 in size configuration that may occur when the aspect ratio of' the image scan is adjusted during preprocessing 76.
Therefore, illustratively, in accordance with example image scan parameters 112, slope table generation 78 is accomplished by dividing the image corresponding to processing area 116 into an illustrative array of 10 . x 10 pixel grids. Considering that every pixel and every line is to be analyzed, this yields 44 grids in the x direction and 60 grids in .the y direction. It should be emphasized that the precise values incorporated into the slope table generation process depend on the characteristics of the particular reader portion 12 being utilized. Analysis can be tailored to accommodate any reader portion 12.
As was explained above in relation to slope table, generation 38, there are two tables created during the slope table generation process: the raw slope data table and the slope table. In accordance with example scan parameters 112, the raw slope data table is a two dimensional י array consisting of 44 x 60 cells, where each cell in the raw slope data table consists of three individual entries :
1. A count of the changes in the x coordinate.
2. A count of the changes in the y coordinate.
3. A count of the pixels tested.
The raw slope data table is created by doing a contour trace of the monochrome image produced during preprocessing 76. As the trace migrates through the pixel grids, the three elements included in the raw slope data table are incremented. Below is a diagram showing the .values to be added for the eight next pixel combinations (P is the current pixel, N is the next pixel, * represents an ordinary pixel and is a filler for display purposes):
<td> Equation 12</td><td></td><td></td><td></td>
<td> ★ ★ ★</td><td> Ά־ ★</td><td> Ar</td><td> ***</td>
<td> *PN</td><td> *p*</td><td> *p*</td><td> *p*</td>
<td> * * A:</td><td> **N</td><td> *N*</td><td> N**</td>
<td> x=+l</td><td> x=+l</td><td> x=0</td><td> x=-l</td>
<td> y+ 0</td><td> y=+l</td><td> y=+l</td><td> y=+l</td>
<td> Ar</td><td> N**</td><td> *N*</td><td> **N</td>
<td> NP*</td><td> *p*</td><td> *p*</td><td> *p*</td>
<td> Ar ★</td><td> * + Λ</td><td> m</td><td> ***</td>
<td> x=-l</td><td> x»-l</td><td> x=0</td><td> x־+l</td>
<td> y= 0</td><td> y=-l</td><td> y=-l</td><td> y=-l</td>
<td> Image 118</td><td> 1 in</td><td colspan="2"> FIG. 25 is an illustration of the</td>
<td colspan="2"> monochrome image,</td><td> produced</td><td> in accordance with</td>
<td> preprocessing</td><td> 76,</td><td colspan="2"> after a contour trace has been</td>
<td colspan="4"> completed. When the entire image has been traced and</td>
<td> the raw slope</td><td> data</td><td colspan="2"> table has been completed, the slope</td>
<td colspan="2"> table is generated</td><td colspan="2"> Illustratively, the slope table</td>
<td> is also a two-</td><td colspan="2"> dimensional array,</td><td> consisting of 44 x 60</td>
<td> cells. Each</td><td> entry</td><td colspan="2"> in .the. slope table consists of a</td>
<td> single entry,</td><td colspan="2"> namely, the slope</td><td> of the ridge or ridges</td>
<td> going through</td><td> the ׳</td><td> corresponding</td><td> grid. Initially, all</td>
<td colspan="3"> entries in the slope table are</td><td> set׳ to a -1 (invalid</td>
<td> slope) ., The</td><td> slope</td><td colspan="2"> for each pixel grid is calculated</td>
utilizing information from the raw slope data table and is computed as follows:
Equation 13
Set x coordinate count .to zero.
Set y coordinate count to zero.
Set pixel count value to zero.
For xl values of x-1 to x+1 do
For yl values of y-1 to y+1 do . from raw slope table at coordinates xl and yl.do ־
Add to pixel count the count of pixels tested.
Add to x coordinate count the changes in the x coordinate.
Add to y coordinate count the changes in the y coordinate.
from raw slope table at coordinates x and y do ־
Add to pixel count the count of pixels tested then divide by 2.
Add to x coordinate count the changes in the x coordinate then divide by 2.
Add to y coordinatecount the changes in the y coordinate then divide by 2.
If the pixel count is greater than 20 Then compute the slope using the trig function arcsine.
Find angle function
Input: delta y and delta x (computed previously above)
Set quadrant to 0
If delta y is less' than 0
Then add 2 to quadrant
If delta x is less than 0
Then add 1 to quadrant
Hypotenuse = square root of ((delta x times delta x) + (deltay times delta y))
Angle = arcsine (delta y divided by hypotenuse) times degrees per radian.
If quadrant is 1
Then angle .= 180-angle
Else if quadrant is 2
Then angle = 360-angle
Else if quadrant is 3
Then angle = 180+angle
Since slopes have values between 0 and 180, the angle is converted to a slope as follows:
If angle is equal to or greater than 180
Then slope is angle minus 180 ]_5 Else slope is the angle
Increment number of pixels processed by one.
Image 120 in FIG. 26 is an illustration of a monochrome image produced in accordance with preprocessing 76 and with a slope overlay consistent with a completed slope table. The completed slope, table is used in subsequent processing. Specifically, it is used as an aid during wire-frame generation 82 in the extension of wire-frame lines and in the 25׳ removal of unwanted wire-frame lines.
In accordance with block 80 in FIG. 15, and in accordance with an embodiment of the present ‘ invention, a histogram could be generated during model creation 20. A histogram resulting from histogram 30 generation 80 could be utilized to make' a determination as to image quality within different portions of the image data under analysis. If a determination is made that a portion of the grid is of insufficient quality to proceed with subsequent processing, then that' portion of the grid could he independently excluded .from the remainder of the׳ model creation 20 process. It should be noted that, in accordance with the present invention, histogram' generation 80 could be performed at any point in the model creation 20 process.
Histogram generation 80 is accomplished in substantially the same manner as histogram generation 40, described above in relation to FIG. 4 and image qualification 18. The primary differences between histogram generation 80 and histogram generation 40 are that all, as opposed to a fraction of, available image data is processed during histogram generation 80 and that slope table values utilized during histogram generation 80 are the values generated during slope, table generation 78 and not during slope table generation 38.
An important component to model creation 20 is indicated by block 82 in FIG. 15, namely wire-frame generation 82. Wire-frame generation 82 is the process of thinning to a special set of thin lines, illustratively. referred to. as wire-frame lines, the features, specifically, the fingerprint ridge line features, included within the monochrome. image produced in accordance with preprocessing 76. FIG. 27 is a block diagram providing a detailed illustration of the procedural components of wire-frame generation 82 in accordance with an embodiment of the present invention.
.1
One of the components of wire-frame generation 82, as is indicated by block 122 in FIG. 2.7, is the location of pixels within the monochrome image that are the approximate center of a plurality of edges, in the current case, the approximate center of fingerprint ridge lines. Once the center pixels have been located, a thinned version of the monochrome image is created by thinning. the ridge lines to the center pixels, so as to create a set of wire-frame lines. An effective method to locate pixels that are the approximate center of ridge lines and, at the same time, to thin the monochrome image to a set of wireframe lines is as follows (of course, other methods could be used as well):
Equation 14
Set test pixel value 255.
Set replace pixel value to 254.
Repeat until no pixels are replaced
COMMENT process the image in the horizontal
For y coordinate values of zero to number of lines do.
For x coordinate values of zero to number of pixels per lines do.
If pixel at image coordinates x and y is test pixel value
If pixel at image coordinates x+1 and y is 255
Setpixel at image coordinates x+1 and y to replace pixel value.
For x coordinate values of number of pixels per lines down to zero do.
If pixel at image coordinates x and y is test pixel value
If pixel at image coordinates x-1 and y is 255
Set pixel at image coordinates x1 and y to replace pixel value.
COMMENT process the image in the vertical
For x coordinate values of zero to number of pixels per lines do.
For y coordinate values of zero to number of lines do. .׳
If pixel at image coordinates x and y is test pixel value
If pixel at image coordinates x and y+1 is 255
Set pixel at image coordinates x and y+1 to replace pixel value.
For y coordinate values of number of lines down to zero do.
If pixel at image coordinates x and y is test pixel value
If pixel at image coordinates x and y-1 is 255
Set pixel at image coordinates x and y-1 to replace pixel value.
Decrement test pixel value by one.
Decrement replace pixel value by one
At this point, the center pixels will have the lowest values. The process described below locates these pixels.
Clear wire-frame image.
COMMENT process the image in the horizontal
For y coordinate values of zero to number of lines do.
For x coordinate values of zero to number ot pixels per lines do.
Set test pixel to value in image at x and y If test pixel is not zero
While pixel at image at x+1 and y is less than test pixel .
Set test pixel to value in image at x+1 and y
Increment x by one.
Set pixel in wire-frame image at x-1 and y to 255.
If pixel at x and y in image equals test pixel ' <sup>58</sup>־~
Set pixel in wire-frame image at x־l and y to 255.
Image 134 in FIG. 28 is an illustration of a monochrome image׳ after a first removal of pixels from image ridge lines.' Image 136 in FIG. 29 is an illustration of a monochrome image with a comprehensive representation of pixel removal passes made during the thinning of the monochrome image to center image line pixels. Image 138 in FIG. 30 is an illustration of FIG. 29 further including an overlay of a thinned monochrome image having raw wire-frame lines. . Finally, image 140 in FIG. 31 is an illustration of a thinned monochrome image with raw wire-frame lines.
The block 122 (FIG. 27) thinning process produces a thinned version of the monochrome image that includes wire-frame lines that may be, in some places, more than one pixel thick. In addition, the thinned image as a whole may contain pixels that do not belong to any line. Accordingly, as is indicated by block 124 in FIG. 27, another component of wire-frame generation 82 is the removal of excess pixels 124. In accordance with one embodiment, component 124 proceeds as follows: the thinned version of the monochrome image is scanned until a non-zero valued pixel is located. Then, the surrounding eight pixels are utilized to form an index into a table. The entry in the table contains flags that define the potential operations to be performed, namely:
1. Remove center pixel.
2. Remove center pixel and set a new pixel (M).
3. Set a new pixel (N).
Below are the indexes into the table:
Equation 15 | 6 | 7 [where the index value is equal to — ן — ן—. the binary sum of pixel locations ן p | 0 that are present and is represented
---1---ן--- as a hexadecimal value (i.e.‘; pixel | 2 | 1 location 4 if present = 2<sup>4th</sup> which equals decimal value 16); and where Count equals the number of pixels present in a 3x3 array surrounding the center P pixel location.]
<td> index</td><td> value</td><td> 0x00:</td><td> Remove</td><td> Point</td><td> - Count</td><td> = 0</td>
<td> index</td><td> value</td><td> 0x00:</td><td></td><td></td><td></td><td></td>
<td> index</td><td> value</td><td> 0x00:</td><td> R</td><td> R</td><td> R</td><td> R</td>
<td> index</td><td> value</td><td> 0x00:</td><td> 1</td><td></td><td></td><td></td>
<td> index</td><td> value</td><td> 0x05:</td><td> Remove</td><td> Point -</td><td> Count =</td><td> = 2</td>
<td> index</td><td> value</td><td> 0x14:</td><td> *</td><td> k</td><td></td><td></td>
<td> index</td><td> value</td><td> 0x50:</td><td> R*</td><td> *R</td><td> *R</td><td> R*</td>
<td> index</td><td> value</td><td> 0x41:</td><td></td><td></td><td> *</td><td> *</td>
<td> index</td><td> value</td><td> 0x07:</td><td> Remove</td><td> Point</td><td> - Count</td><td></td>
<td> index</td><td> value</td><td> OxlC:</td><td> **</td><td> **</td><td></td><td></td>
<td> index</td><td> value</td><td> 0x70:</td><td> R* *R</td><td> *R</td><td> R*</td>
<td> index</td><td> value</td><td> OxCl:</td><td></td><td> .k ie</td><td> i׳ ★</td>
<td> index</td><td> value</td><td> OxOD:</td><td> Remove Point -</td><td> Count =</td><td> 3</td>
<td> index</td><td> value</td><td> 0x34:</td><td> ir ir</td><td></td><td> ir</td>
<td> index</td><td> value</td><td> OxDO:</td><td> R* *R</td><td> *R</td><td> R* '</td>
<td> index</td><td> value</td><td> 0x4 3:</td><td> *</td><td> Hr</td><td> *</td>
<td> index</td><td> value</td><td> OxOE:</td><td> Remove Point</td><td> - Count</td><td> = 3</td>
<td> index</td><td> value</td><td> 0x38:</td><td> ***. k</td><td></td><td> *</td>
<td> index</td><td> value</td><td> OxEO:</td><td> R *R</td><td> R</td><td> R*</td>
<td> index</td><td> value</td><td> 0x83:</td><td> ★</td><td> ir ir ir</td><td> *</td>
<td> index</td><td> value</td><td> OxOF:</td><td> Remove Point</td><td> - Count</td><td> = 4</td>
<td> index</td><td> value</td><td> 0x3C:</td><td> *** **</td><td></td><td> *</td>
<td> index</td><td> value</td><td> OxFO:</td><td> R* *R</td><td> *R</td><td> R*</td>
<td> index</td><td> value</td><td> 0xC3:</td><td> *</td><td> ir irir</td><td> ir ir</td>
<td> index</td><td> value</td><td> 0x16:</td><td> Remove Point</td><td> - Count</td><td> = 3</td>
<td> index</td><td> value</td><td> 0x58:</td><td> if ir ir</td><td></td><td> *</td>
<td> index</td><td> value</td><td> 0x61:</td><td> *R *R</td><td> R*</td><td> R*</td>
<td> index</td><td> value</td><td> 0x85:</td><td> *</td><td> **</td><td> *</td>
<td> index</td><td> value</td><td> OxlE:</td><td> Remove Point</td><td> - Count</td><td> = 4</td>
<td> index</td><td> value</td><td> 0x78:</td><td> if ir if if</td><td></td><td> * *</td>
<td> index</td><td> value</td><td> OxEl: .</td><td> *R *R</td><td> R*</td><td> R*</td>
<td> index</td><td> value</td><td> 0x87:</td><td> **</td><td> ir ir ir</td><td> *</td>
<td> index</td><td> value</td><td> 0x36:</td><td> Remove Point</td><td> - Count</td><td> = 4</td>
<td> index</td><td> value</td><td> 0xD8:</td><td> k k</td><td> *</td><td> *</td><td> ★ ־Ar</td>
<td> index</td><td> value</td><td> 0x63:</td><td> *R</td><td> *R</td><td> R*</td><td> R*</td>
<td> index</td><td> value</td><td> 0x8D:</td><td> Λ</td><td> **</td><td> **</td><td> k</td>
<td> index</td><td> value</td><td> 0x3D:</td><td> Remove</td><td> 1 Point</td><td> - Count</td><td> 5 ־</td>
<td> index</td><td> value</td><td> 0xF4:</td><td> ★ k</td><td> *</td><td> *</td><td> ***</td>
<td> index</td><td> value</td><td> 0xD3:</td><td> *R*</td><td> *R</td><td> *R*</td><td> R*</td>
<td> index</td><td> value</td><td> 0x4F:</td><td> *</td><td> ***</td><td> **</td><td> *</td>
<td> index</td><td> value</td><td> 0x3E:</td><td> Remove</td><td> ! Point</td><td> - Count</td><td> = 5</td>
<td> index</td><td> value</td><td> 0xF8 :</td><td> ***</td><td> *</td><td> ★</td><td> ***</td>
<td> index</td><td> value</td><td> 0xE3:</td><td> *R</td><td> *R</td><td> R*</td><td> R*</td>
<td> index</td><td> value</td><td> 0x8 F:</td><td> k</td><td> ***</td><td> ***</td><td> *</td>
<td> index</td><td> value</td><td> 0x3F:</td><td> Remove</td><td> Point</td><td> - Count</td><td> = 6</td>
<td> index</td><td> value</td><td> OxFC:</td><td> # ★ ★</td><td> + A־</td><td> *</td><td> ***</td>
<td> index</td><td> value</td><td> 0xF3:</td><td> *R*</td><td> *R</td><td> *R*</td><td> R*</td>
<td> index</td><td> value</td><td> OxCF:</td><td> *</td><td> ***</td><td> k kk</td><td> k k</td>
<td> index</td><td> value</td><td> OxBF:</td><td> Remove</td><td> Point</td><td> - Count</td><td> = 7</td>
<td> index</td><td> value</td><td> OxFE:</td><td> *</td><td> ***</td><td> * *</td><td> kkk</td>
<td> index</td><td> value</td><td> OxFB:</td><td> *R*</td><td> *R</td><td> *R*</td><td> R*</td>
<td> index</td><td> value</td><td> OxEF:</td><td> * *</td><td> ***</td><td> kkk</td><td> ***</td>
***********************
<td> index</td><td> value OxOA:</td><td> Move</td><td> Point - Count = 2</td>
<td> index</td><td> value 0x28:</td><td></td><td> * *</td>
<td> index</td><td> value OxAO:</td><td> X</td><td> MX X XM</td>
--62index value 0x82:
*M* ***********************
<td> index</td><td> value</td><td> 0x29:</td><td> New</td><td> Point -</td><td> Count =</td><td> 3</td>
<td> index</td><td> value</td><td> 0xA4:</td><td> *</td><td> ★</td><td> *</td><td> *N*</td>
<td> index</td><td> value</td><td> 0x92:</td><td> NX*</td><td> X</td><td> *XN</td><td> X</td>
<td> index</td><td> value</td><td> 0x4A:</td><td> *</td><td> *N*</td><td> *</td><td> *</td>
<td> index</td><td> value</td><td> 0x2B:</td><td> New</td><td> Point -</td><td> Count =</td><td> 4</td>
<td> index</td><td> value</td><td> OxAC:</td><td> *N*</td><td> **</td><td> *</td><td> ★ .k</td>
<td> index</td><td> value</td><td> 0xB2:</td><td> X*</td><td> NX</td><td> *X</td><td> XN</td>
<td> index</td><td> value</td><td> OxCA:</td><td> *</td><td> * *</td><td> *N*</td><td> **</td>
<td> index</td><td> value</td><td> 0x6A:</td><td> New</td><td> Point ־</td><td> Count -</td><td> 4</td>
<td> index</td><td> value</td><td> 0xA9:</td><td> *N*</td><td> *</td><td> **</td><td> * *</td>
<td> index</td><td> value</td><td> ΟχΑβ:</td><td> X</td><td> NX*</td><td> X</td><td> *XN</td>
<td> index</td><td> value</td><td> 0x9A:</td><td> **</td><td> * ־Ar .</td><td> *N*</td><td> *</td>
<td> index</td><td> value</td><td> OxAB:</td><td> New</td><td> Point -</td><td> Count =</td><td> 5</td>
<td> index</td><td> value</td><td> OxAE:</td><td> * *</td><td> * *</td><td> * *</td><td> *N*</td>
<td> index</td><td> value</td><td> OxBA:</td><td> NX*</td><td> X</td><td> *XN</td><td> • X</td>
<td> index</td><td> value</td><td> OxEA:</td><td> * *</td><td> *N*</td><td> ★ Λ</td><td> Jr ★ *</td>
Image 142 in FIG. 32 is an illustration of a thinned version of a monochrome image after excess pixels have been removed. Image 144 in FIG. 33 is an illustration demonstrating the relationship between a thinned version of a monochrome without excess pixels and a corresponding monochrome image (monochrome image ridge lines in white).
In accordance with an embodiment of the present invention, after the thinned version of the monochrome image has been created and excess pixels have been removed from the wire-frame lines, as is illustrated by block 126 in FIG. 27, end-point and center-point tables are constructed. To create these tables, the wire-frame lines are scanned and all end-points (those pixels on the wire-frame lines that touch only one other pixel) are catalogued in the end-point table. Image 146 in FIG. 34 is an illustration of a thinned monochrome image (with excess pixels removed) that includes a representation of data from an end-point table (white dots represent end-points). As the wireframe lines are scanned, the center-points (those pixels that touch more than two other pixels) are catalogued in the center-point table. Image 148 in FIG. 35 is an illustration of a thinned monochrome image that includes a representation of data from a center-point table (white dots represent centerpoints) .
The points within end-point and center-point tables are used in subsequent processing. In one embodiment, the points are utilized to identify unique data elements, such as spikes, mouse bites and antiridges. These data elements can be identified and catalogued through data describing the precise location of end-points or center-points, and slope values associated with lines attached to the points. The orientation of these data elements are unique to an individual system user and can be utilized to authenticate or match one set of image data to one or more other sets of image data.
After excess pixels have been removed from the wire-frame lines contained in the thinned monochrome image, the next step in wire-frame generation 82, iri accordance with block 128 .in FIG. 27, is to create a refined set of wire-frame lines by pruning excess branches. As can be seen in image 142 in FIG. 32, the. wire-frame lines within the thinned monochrome image include many small branches protruding from the main lines. The removal of a selection of these branches<sup>1 </sup>leaves a refined set of wire-frame lines that are relatively smooth for subsequent minutiae and vector segment extraction.
In accordance with one embodiment, the block.128 branch removal process relies upon data taken from end-point and slope tables, both generated previously<sup>1 </sup>in the model creation 20 process. Each entry in the end-point table is used to locate a line segment that includes an end-point. The segment is traced back from the end-point along the corresponding line segment for seven pixels or until a center-point is. located. If the segment length is less than five pixels, it is unconditionally removed (the pixels are!
cleared from the image). If the segment terminates on a center-point, the slope of the segment is compared to the slope of image elements in the same proximity. Slope data is derived from the slope table. If the difference between the two slopes is greater than 25 degrees, the segment is removed. Image 150 in FIG. 36 is an illustration of a refined set of wire-frame lines that result from the removal of excess wireframe branches. Image 152 in FIG. 37 is an illustration demonstrating the relationship between the refined set of wire-frame lines and a corresponding monochrome image (image ridge lines are in white).
After the refined set of wire-frame lines have been created, in accordance with one embodiment, excess pixels are once again removed from the image. In one embodiment, excess pixels are located and removed in the same manner as described above in relation to Equation 15. In one embodiment, the endpoint table and center-point table are recomputed either before, but illustratively after excess pixels have once again been removed.
As is demonstrated by block 130 in FIG. 27, another procedural component of wire-frame generation 82 is the fixing of end points. As can be seen in image 150 in FIG. 36, segments at the end of a line within the refined set of wire-frame lines may curl or may demonstrate a slope angle that is inconsistent with slope table entries in the same proximity.
In accordance with one embodiment and with block 130, in order to correct these deficiencies, each entry in the end-point table is utilized to assist in the creation of a further refined set of wire-frame lines. Accordingly, each entry in the end-point table is used to trace a corresponding line segment back seven pixels or until a center-point is encountered. If a center-point is encountered, the segment is restored. If no center-point is encountered, the line is removed from the wire—frame image. After the line is removed, the slope table entry for the line segment termination point is retrieved from the slope table.' This slope value is utilized to create a new. line. segment, using a line draw algorithm, from the termination point to the end of the monochrome image. Image 154 in FIG. 38 is an illustration demonstrating the relationship between a further refined set of wire-frame lines, including fixed end points, and a corresponding monochrome image (image ridge lines are in white).
In accordance with one embodiment, the end-point and center-point tables are recomputed after endpoints have been fixed in accordance with block 130 and the creation of a further refined set of wireframe lines.
As is demonstrated by block 132 in FIG., 27, another procedural component of wire-frame generation 82 is the joining of end-points. Fingerprint ridge data elements may be broken due to paper cuts, blisters, burns, skin wrinkles, wet/dry conditions, or due to other image scanning problems, such as those caused by environmental influences in the image reader 12 environment. In accordance with end-point joining 132, an attempt is made to join end-points in a manner that bridges certain ridge gaps.
In accordance with one embodiment of end-point joining 132, each entry in the end-point table is compared to all other entries in the end-point table. If any two points are within six pixels of each other and the slope of the corresponding lines are within 25 degrees of each other,' the segments are joined. Image 156 in FIG. 39 is an illustration demonstrating the. relationship between a further refined set of wireframe lines, including fixed and joined end-points, and a corresponding monochrome image (image ridge lines are in . white) . In accordance with one embodiment, the end'-point and center-point tables are recomputed after end-points have been joined. After the end-points have been joined within the further refined set of wire-frame lines, a complete wire-frame image, based on a corresponding monochrome image, in accordance with wire-frame generation 82, will have been completed.
After a monochrome image has been transformed into a completed wire-frame image, in accordance with an embodiment of the present invention, the next component of model creation 20, in accordance with block 84 in FIG. 15, is to analyze the completed wire ־68frame image in order to locate fingerprint bifurcations and rods to be catalogued and included within an image model, along with other data elements. In accordance with one embodiment, relative to portion 12 image resolution and with reference .to FIG. 40, general data elements related to a bifurcation are as follows:
Equation 16 • Leg segments 162, 164 and 168. These segments are of equal length and each has at least one originating point at center-point 158. In accordance with one embodiment, each segment is set to 17 pixels.
• The coordinates of center-point 158. This point is used to define a bifurcation. The upper left corner of the image is assumed to have coordinates of 0,0. Positive x coordinates are right and positive y coordinates are down.
• First separation angle 160. This is the angle ; between leg segments 162 and 164, which is used to define a bifurcation. In accordance with one embodiment, first separation angle 160, by definition, cannot exceed 120 degrees.
• Direction angle 166. This angle is the direction of a bifurcation and is used to define a bifurcation. In accordance with one embodiment, direction angle 166 can have values between 0 and 359.
• It should be noted that the angle between leg segments 162 and 168 is the largest angle of all the angles between leg segments.
• A count of the number of 20 (illustratively 20) pixel segments tracing from point 158 along the wire-frame, line connected to leg segment 162 is used to define data points associated with a bifurcation. It is assumed that this is the first leg segment array.
• A list (of length count) of the x and y coordinates of the 20 (illustratively 20) pixel segment end-points is constructed in order to catalogue data points within a first leg segment array. Leg segment array data points can also be called vector segment data points and, in accordance with one embodiment, and in accordance with the present invention, can be utilized to compare one image model to another.
Bifurcations are used as one origin to catalogue vector segment data points.
• A count of the ‘number of 20 (illustratively 20) pixel segments tracing from point 158. along the wire-frame line connected to leg segment 164 is .15 constructed in order to catalogue data points within a second leg segment array.
• A list (of length count) of the x and y coordinates of the 20 (illustratively 20) pixel segment end-points is constructed in order to catalogue second leg segment array data points (vector segment data points).
• A count of the number of 20 (illustratively 20) pixel segment end-points tracing from point 158 along a wire-frame line connected to leg segment
168. י is used to define data points associated with a bifurcation. It is assumed that this is the third leg segment array.
• A list (of length count) of the x and y coordinates of the 20 (illustratively 20) pixel segment end-points is constructed in order to catalogue third leg segment array data points (vector segment data points).
• In accordance with one embodiment, the maximum number of leg segments with each array of data points is 20 (illustratively 20).
It should be emphasized that the precise values, in particular values corresponding to pixel counts and segment counts, could be modified without departing from the current invention. Different reader portion 12 (FIG. 1) technologies may require such modifications. The particular values provided in the present description in the context of illustrative embodiments, are to be considered illustrative values only.
In accordance with one embodiment of the present invention, location of a possible bifurcation starts with the center-point table created during wire frame generation 82. Each entry in the center-point table is considered a potential bifurcation. Starting with a center-point entry in the center-point table, each segment extending therefrom is traced. When a length of 17 pixels is reached, the corresponding x and y coordinates are placed in a list and a leg segment count is incremented. Tracing, however, will terminate upon one of the following conditions.
1. A count of 20 (illustratively 20) leg segments is reached.
2. An end-point is detected (from EP table).
3. A center-point from center-point table is detected.
When the line tracing has been completed, three angles can be computed for potential bifurcations:
1. The angle between leg segments 162 and 164.
2. The angle between leg segments 164 and 168.
3, The angle between leg segments 162 and 168.
These angles are sorted in ascending order. The smallest angle, between segments 162 and 164, is saved as the separation angle. Next, angle 166 is computed using point 167 as coordinates 0,0. After leg
ך segments have been identified and corresponding angles have been computed, a list of bifurcations will have been constructed. The bifurcations are illustratively defined through the center-point table.
A rod is considered a special case bifurcation. With reference to FIG. 41, wherein elements common to FIGS. 40 and 41 include identical labels, illustrative general data elements related to a rod, and assumptions based thereon, are as follows:
Equation 17 • The end-point 172 of rod 165; This point is the center point 158 of the special case bifurcation.
• The upper left corner of the image, again, is the coordinate 0,0. Positive values of x are to the right and positive values of y are down.
• Using as the direction the direction of the rod 165 segment extending from point 170 to endpoint 172, the rod end is extended to points 162 and 164.
• Points 162 and 164 coincide.
• Direction angle 174 is the direction of the rod. In accordance with one embodiment, it is assumed that the direction angle can have.values between 0 and 359.
• The angle between the segment extending between points 172 and 162, and the segment extending between point 172 and .164 (first separation angle 160) is set to zero.
• The segment extending from end-point 172 to points 162/164 is the same length as the segment extending from point 170 to end-point 172. In accordance with one embodiment, it is assumed that each segment is set to 17 (this value may vary) pixels.
• A count of the number of 20 (this value may vary) pixel segment end-points tracing from point 158 along rod 165 is used to define data points associated with a rod. It is assumed that this is the first leg segment array list.
• A list (of length count) of the x and y coordinates of the 20 (this value may vary) pixel segment end-points is constructed in order to catalogue the first leg segment array data points. These data points are vector segment data points and can be utilized to compare one image model to another.
• The two remaining segment array lists contain only one point, 162/164. This point is computed by extending the segment associated with the.
first segment array list by 17 (this value may vary) pixels.
• The maximum .number of 20 (this value may Vary) pixel segments in the first leg segment array is illustratively 20. .
In accordance with one embodiment of the present invention, location of a possible rod starts with the end-point table created during wire frame generation 25 82. Each entry in the end-point table is considered a potential rod. Starting with an entry in the endpoint table, the corresponding line segment is traced. When a length of 17 (this value may vary) pixels is reached, the corresponding x and y coordinates are 30 place in a list and a leg . segment count is incremented. Tracing terminates upon one of the following conditions:
1. A count of 20 (this value may vary) segments is reached.
2. An end-point is detected (from EP table).
3. A center-point from center-point table is
-כדdetected .
In accordance with an embodiment of the present invention, for rods to be included within an image model, they must meet certain qualifying standards. Accordingly, after a vector segment has been traced, the segment extending from point 170 to end-point 172 is extended along the same angle to 162/164. If during extension, an image ridgeline is crossed, the. corresponding rod is not saved and is not entered into. an image model. In addition, a line perpendicular to the segment extending from end-point 172 to point 162/164 is extended in both direction for a distance of an illustrative 20 (this value may vary) pixels. If neither of these 20 (this' value may vary) pixel lines intersect an image ridgeline, the rod is not saved. The distances during the perpendicular extension are then compared to see if end-point 172 is approximately the mid-poiht. If it is not, the rod is not saved. The output from this process is a list of qualified rods, which are defined through the endpoint table. Direction angle 174 is used to assist in defining rods and is computed using point 170 as coordinates 0,0.
In addition to being located, in order to be included within an image model, rods and bifurcations must meet certain qualifications. As was previously mentioned, rods are qualified at the same time as they are located. Bifurcations, however, in accordance with block 86 in FIG. 15, in order to be included within an image model, must fulfill several qualifications in addition to those imposed during the bifurcation location process. With reference to FIG. 40, in accordance with one embodiment, the following is a list of filter rules used to further qualify bifurcations for entry into an image model:
Equation 18 • As first separation angle 160 approaches 120 degrees, the risk of the separation angle being defined using the wrong legs increases. Accordingly, the first filter rule is that first separation angle 160 must be less than 115 degrees.
• At least two of the bifurcation legs must have three or more 20 (this value may vary) pixel' segments.
• The slope of leg segments 162, 164 and 168 must be within 30 degrees of the slope table for the same area.
Image 176 in FIG. 42 is an illustration of a wjpp—-frame image within which qualified bifurcations and rods have been circled. End-points and centerpoints are identified with white dots. Image 178 in FIG.43 is an illustration of a wire-frame image within which the same qualified bifurcations and rods have been circled. Within image 178, vector segment data points (the illustrative 20 pixel segments extending from a qualified rod or bifurcation) have been traced and are shown in white. Vector segment data points approximately track wire-frame lines that connect to a qualified bifurcation and rod and can be used to compare one image model to another.
In accordance with block 88 in FIG. 15, the final step in the model creation 20 process is the building of an image model based on a completed wire-frame image and image elements derived therefrom. A completed image model consists of the following elements:
Equation 19 • A count of qualified bifurcations • A count of qualified rods • A bifurcation and rod list which consists of: .
—The center point (center-point 158 for bifurcations or end-point 172 for rods, see FIGS. 40 and 41) identified with x and y coordinates.
—The direction angle (angle 166 for bifurcations or angle 174 for rods) . --The . separation angle (angle 160 for bifurcations or a zero value for rods) —Three leg segment arrays (vector segment arrays), which consist of:
a) Count of 20 (this value may vary) pixel segments extending from a point within corresponding דrod/bifurcation.
b) List of x coordinate end-points corresponding to 20 (this value may vary) pixel segments.
c) List of y coordinate end-points corresponding to 20 (this value may vary) pixel segments.
• Data representations of line segments or vector segments not used by bifurcations or rods.
Bifurcations, within an image model, are defined by the intersection of ridge segments. Referring to FIG. 40, in accordance with one embodiment of the present invention, the information recorded in an image model for a bifurcation is as follows:
Equation 20 • The x and y coordinates of center point 158.
• First separation angle 160.
• Direction angle 166.
• A list of the x and y coordinates that trace the vector/leg segments emanating from the • center point 159. These points are equal distance (illustratively 20 pixels) apart. Tracing and collection of 20 (this value may vary) pixel vector/leg segments . continues until:
—The trace terminates at the center of
-דדa bifurcation .
—The end of the segment is reached.
—A maximum of forty 20 (this value may vary) pixel segments are recorded.
Rods are defined by the ending of a ridge segment. The end-point must be approximately half way between two ridge segments and if extended for a particular amount of pixels, illustratively 10 or 20 (or some other appropriate value), must not touch another ridge segment. In accordance with one embodiment, with reference to FIG.' 41, the information recorded in an image model for a rod is as follows:
Equation 21 • The x and y coordinates of end-point 172.
• First separation angle 160 (by definition, this value is zero). .
• Direction angle 174.
• A list of the x and y coordinates that trace the vector/leg segments emanating from point 172. These points are equal distance (illustratively 20 pixels) apart. Tracing continues until:
—Arrive at center of a bifurcation.
—The end of the wire-frame segment is reached.
—Maximum of forty 20 (this value may vary) pixel segments are recorded.
I
In accordance with an embodiment of the present invention, ridge segments not used by bifurcations or rods are recorded in an image model as follows:
Equation 22 • A list of x and ־y coordinates that trace the ridge segments emanating from end-points. These points are equal distance (illustratively 20 pixels) apart. Tracing continues until:
-The trace terminates at the center of ׳a bifurcation.
-The end of the segment is reached.
-A maximum of forty 20 (this value may vary) pixel segments are recorded.
In accordance with one embodiment, information within an image model can be stored in accordance with the following data storage format: Equation 23 . typedef struct tag_LEG (
int seperation_angle; // angle to next leg int count; // number of entries in row and col int row[ 40 ]; // y coordinate list int col[ 40 ]; // x coordinate list }LEG, * LEG_PTR;
typedef struct tag_D_POINT // bifurcation or rod information (
int row; // bifurcation or rod center y דint col;
int leg_orientation;
LEG leg [ 3 ];
} D_POINT, * D_POINT_PTR;
typedef struct tag_SEG {
int count;
int row[ 4 0 ];
int col[ 40 ];
' } SEG, * SEG_PTR;
coordinate .
// bifurcation or rod center y coordinate // orientation angle // leg trace info // for extra segment information // number of entries in row and col // y coordinate list // x coordinate list typedef struct tag_PRINT // model structure {
int min_point_center_col;
int min_point_center_row;
int no_of_extra_segs; // entries in segment_list int no_of_points; // # of bifurcations in point_list int .no_of_islands; // number of rods in point_list
D_POINT point_list[ 100 ]; // bifurcations and rods list
SEG segment_list[ 100 ]; H extra segment list } PRINT, * PRINT_PTR;
[DATA LIMITS CAN BE EXTENDED FOR DIFFERENT IMAGES]
It should again be emphasized that the precise values, in particular values corresponding to pixel counts and segments counts, could be modified without departing from the current invention. Different reader portion 12 (FIG. 1) technologies my require such modifications. The values provided, whether explicitly indicated or not, are illustrative values only.
As is indicated by block 24 in FIG. 2, and in accordance with an embodiment of the present invention, a completed image model can be utilized as the basis for a model comparison. Model comparison 24 is a process for comparing one ׳image model to another.
Model comparison can be accomplished by using a series of shift and rotate algorithms to adjust at least one image model and then counting the number of matching data elements. Data elements might include, but are not limited to bifurcation representations', rod representations, vector segments associated with bifurcation/rod representations, micro-minutia points and vector segments not incorporated into or associated with bifurcation/rod representations, and combinations thereof. When shifting and rotating causes the data element count to reach or approach a maximum value, the two models are at or approaching their maximum comparison point. Using the counts of the matching data elements at or near the maximum comparison point, a score, which represents the relationship or percentage of match, can be calculated.
The theory behind model comparison 24 is that as one model is rotated and shifted, the number of data element points that. match will increase or decrease. If the number of points that match for each angle is ־81plotted, a. bell shaped curve will result. The high point for the curve will represent the point at which the two models best compare.
In accordance with one embodiment of model comparison 24, the process starts by loosely comparing each bifurcation and rod in a model (model A) to all bifurcations and rods in another model׳ (model B). The loose comparison enables a nonmatching pair to be discovered without a significant. investment of processing time. A more detailed׳ comparison could illustratively be substituted for the loose comparison. In one embodiment, requirements utilized during the loose comparison are defined as:
Equation 24 • The center points must be within 90 (this value may vary) pixels in the x direction.
• The center points must be within 120 (this value may vary) pixels in the y direction.
• The difference between separation angles is within plus or minus 8 degrees.
• The difference between direction angles is within plus or minus 40 degrees.
In accordance with one embodiment, a possible match table is generated based on the loose comparison. The possible match table is indexed by the bifurcation or rod index into model A. Each entry in the possible match table consists of:
Equation 25 • A count of possible matches • A list of the bifurcation or rod indexes in model B. This list defines all those points in model B that loosely match to the point in model
A.
The general comparison process is best described by a simple coding example. Specifically, in accordance with one embodiment, the coding is defined as follows :
Equation 26
Generate possible match table.
Set maximum match count to zero.
Set rotate angle to zero.
Clear master x and y shift counts.
TOP OF LOOP
If rotate angle is not zero
Rotate model B by rotate angle. This is a standard trig, function.
Shift process on model B (this is described below).
Count the bifurcations, rods and segments that match between model A and model B.
• The center points (both x and y) must be within 10 (this value may vary) pixels.
• ' The leg segment points, (both x and y) must be within 10 (this value may vary) pixels.
• The difference for separation angle is within plus or minus 6 (this value may vary) degrees.
• The difference for direction angle is within plus or minus 10 (this value may vary) degrees.
If match count is less than maximum match count minus two
Exit the loop.
If match count is greater than maximum match count
Set maximum match count to maximum count.
Save rotate angle.
Save x and y shift counts. This is an output of the above shift process.
Increment rotate angle by one.
If rotate angle is less than 30
Goto TOPOF LOOP
The above loop is repeated for angles between -1 and -30.
Copy working possible match table to possible match table.,This will be used in the enrollment process.
Using the saved variables rotate angle and x and y shift counts, shift and rotate the entire model B.
Count the bifurcations, rods and segments that match between model A and model B.
• The center points (both x and y) must be within 10 (this value may vary) pixels.
• The leg segment points (both x and y) must be within 10 (this value may vary) pixels.
• The difference for separation,angle is within plus or minus 6 -(this value may vary) degrees.
• The difference for direction angle is within plus, or minus 10 (this value may vary) degrees.
SHIFT PROCESS DESCRIPTION
Set max test x to 90 (this value may vary) pix.
Set max text y to 120 (this value may vary) pix. Clear master x and y shift counts.
Make a working copy of the possible match table.
TOP OF LOOP 1
Clear saved points count.
Clear delta x and delta y count.
Set model A index to zero.
TOP OF LOOP 2
Set possible index to zero.
TOP OF LOOP 3 ־84Extract model B index from possible match table at model A index and possible index.
Compute the distance between the bifurcation or rod in model A and model B.
If the distance is less than max test x and max test y
Increment saved points count.
Add distance in x direction to delta x count.
Add distance in y direction to delta y count.
Else
Remove this point from the working copy of the possible match table.
Increment possible index by one.
If possible index is less than count value in possible match tables at model A index.
Goto TOP OF LOOP 3
Increment model A index by one.
If model A index is less than the bifurcation and rod count in model A
Goto TOP OF LOOP 2.
If saved points count is zero
Exit the routine and return a value of zero (no compare).
Add delta x count divided by saved points count • to master x count.
Add delta y count divided by saved points count 30 to master y count.
Adjust all x and y coordinates in model B using master x and y counts.
If max test x is greater than 20 (this value may vary) pixels
Multiply max.test x by 0.75.
Multiply max test y by 0.75.
Goto TOP OF LOOP 1
Exit the routine and return the value of. saved points count.(Those labeled and other specifically 40 noted values may be application-dependent.) // END CODE EXAMPLE
The result of the final shift and rotate process is several counts:
Equation 27 • The number of bifurcations that match.
• The number of rods that match.
• The number of segments that match.
• The number of segments that don't match.
For the discussion below, the following definitions are used:
Equation 28 • Total points: the sum of the number of bifurcations that match and the number of rods that match.
• Segments per leg: total segments matched divided by number of legs compared.
• Percent of points: total points (multiplied by . 100) divided by the minimum of total bifurcations and rods in model A or model B.
• Raw percentage: the number of segments that match (multiplied by 100) divided by the sum of the segments that match and the segments that don't match.
Using the above counts, a percentage of match can be calculated (below is illustrative example based one sample reader device..values may change depending on the particular reader technology).
Equation 29 • If the number of segments that match is zero, the percentage of match is zero.
• If segments per leg is less than four, the percentage of match is. zero.
• If total points is less than three, the percentage of match is zero.
• If total points is three or four, the percentage of match is M of the percent of points.
• If total points is five and segments per leg is greater than four, the percentage of match is raw percentage multiplied by percent of points divided by 800.
• If total points is six through ten and • Segments per leg is four, the percentage of match is raw percentage multiplied by percent of points divided by 700.
• Segments per leg is five through seven, the percentage of match is raw percentage multiplied by percent of points divided by 500.
• Segments per leg is greater than eight, the percentage of match is raw percentage multiplied by percent of points divided by 300.
• If total points is greater than 10, the percentage of match is raw percentage multiplied by percent of points divided by 100.
Alternatively, using the above counts, a relationship of match can be calculated.
'Alternate Equation 29 • If the number of total points is equal to zero then the relationship is equal to zero.
• If the rotation angle to align the images is more than the maximum allowed then the relationship is equal to zero. This value is tunable to work with different image sizes and aspect ratios.
• The relationship between ' two images can be u defined as the relationship between scoring elements. These elements can include but are not limited to:
• The number of bifurcation representations that match • The number of bifurcation representations that do not match • The number of rods that match • The number of rods that do not match • The number of segment representations that match • The number of segment representations that do not match • The number of segment representations per leg ♦ A series of calculations are then done to determine the scoring element values, which includes:
• To determine the scoring element for the number of bifurcations that match, we first use the number of bifurcations that match to decide the weighting factor to use in the calculation. The larger the number of birfurcations that match, the more you can weight the final score.
• To determine the scoring element for the number of bifurcations that do not match, we first use 'the number of bifurcatoins that match to decide the weighting factor to use in the calculation. The larger the number of bifurcations that match, the less the weighting factor for the bifurcations that do not match.
• To determine the scoring element for the number of rods that match, we first use the number of bifurcations that match to decide the weighting factor to use in the calculation. The number of rods that match are then used to further obtain the weighting factor for this element. The larger the number of rods that match and the number of bifurcations that match, the more you can weight the final score.
• To determine the scoring element for the number of rods that do not match, we first use the number of bifurcations that match to decide the weighting factor to use in the calculation. The number of rods that match are then used to further obtain the weighting factor for this element. The larger the number of rods that match and the number of bifurcations that match, the less the weighting factor for the rods that do not match.
• To determine the scoring element for the number of segments that match, the total segments that match are multiplied by a weighting factor and then divided by the total segments that are within the fingerprint model's overlap area.
• To determine the scoring element for the number, of segments that do not match, the lets that do not match are multiplied by a weighting factor and divided by the number of legs that match.
• To determine the scoring element for the segments per leg, we first use the number of bifurcations that match to decide the weighting factor to use in the calculation. The number of rods that match are then used to further obtain the weighting factor for this element.
• These calculations are tunable to accommodate different image sizes and aspect ratios. These values are then added . together to form a probability of relationship between the images. This value is then adjusted by how much of the two images actually overlap. The smaller the overlap area, the greater the correction to the probability of relationship value.
In accordance with one embodiment of the present invention, the level of similarity required, within an image identification based security system, before two image models are declared matching is adjustable. Similarity levels can be tuned based on the nature of the 'environment being secured. A strictly tuned system would provide maximum security by requiring a high match relationship or percentage, but may be more prone to accidentally rejecting a matching pair of image models. Conversely, a loosely tuned system would require a lower match relationship or percentage and, accordingly, would provide a lower level of security. A loosely tuned system, however, would be more prone to mistaken match determinations and less prone to match rejections.
FIG. 44 is a block diagram illustrating a set of procedural components associated with a one-to-one image comparison process, in accordance with an embodiment of the present invention. Block 180 indicates a first illustrative step in the process, specifically, deriving a first image data set based on a first image and a second image data set based on the second image. The first and second image data sets include a plurality of data elements, which are illustratively, but not necessarily, associated with fingerprint images. The two data sets are model representations of the first and second images and are illustratively created as described above in relation to other Figures incorporated into the description of the present invention.
Block 182 indicates an optional performance of a loose comparison process. Illustratively, the first and second data element sets include a plurality of data elements types. Performance of the loose comparison, as described above, economizes the comparison process by eliminating those data elements in one of the two sets that do not fit within a predetermined qualifying range of deviation from a data element of a same type in the other of the two sets.
Block 184 indicates a comparison of data. elements from the first image data set with data elements from the second image data sets. Illustratively, as described above, data elements include defining characteristics that are numeric or mathematical in nature and facilitate comparison of data elements. For example, in accordance with one embodiment, during the comparison process, a qualifying range of deviation is selected or predetermined. The qualifying range of deviation represents a tolerated differential between a defining characteristic of at least one data element, of the first type, from the first image data set, and a corresponding at least one defining characteristic, of at least one data element, of the first type, from the second׳ image data set. In accordance with one embodiment, multiple defining characteristics, each with a qualifying range of deviation, can be used to define a single data element type. In accordance with one embodiment, a count is generated of data elements from the first and second image data sets wherein the amount of deviation between the defining characteristics (or characteristic) is within the qualifying range of deviation. In essence,.’ this count is a count of data elements in the first image data set that approximately match data elements in the second image data set.
Block 186 indicates an examination of the count of matching data elements to evaluate whether the count is a maximum value. If only one comparison has been performed, the maximum value will not have yet been found. In situations where the maximum value has not been obtained, as is indicated by block 188, one of the first and second data image sets, and its incorporated data elements, is re-positioned (e.g., rotated, shifted, or both). Then, the process returns to block 184, where a new comparison is made and a new count is generated.
The compare, count, re-position steps are illustratively repeated to produce a plurality of additional counts of the data elements in the first image data set that approximately match data elements in the second image data set. When, as described above, the count approaches a maximum value, the process proceeds.
Block 190 indicates comparison of the count value with a predetermined target value. In accordance with one embodiment, the predetermined target value represents a selected level of similarity required for the first, and second images to be considered matching. The .target value can be adjusted based on a desired level and comprehensiveness of an associated security application. As is indicated by block 192, if the count is greater than the predetermined target value, a positive match is indicated. As is indicated by block 194, if the count is less than the predetermined target value, a negative match is indicated. In accordance with one embodiment, the count comparison could be performed on a relationship or percentage basis and relationships or percentages could be compared rather than counts.
As is indicated by block 26 in FIG. 2, and in accordance with an illustrative embodiment of the present invention, database search 26 could be performed in place of or in combination with model comparison 24. Database search 26 involves a quick and efficient determination as to which, if any, of potentially thousands (or more, i.e., millions) of image models included within a database exhibit a desired level of similarity, as compared to a supplied, i.e., a target image model.
In accordance with an embodiment of the present invention, in order to quickly identify a fingerprint from a database containing thousands (or millions or more) of prints, a set of database keys is defined. Due to many factors (finger too dry, finger too wet, position of the finger when placed on the scanner, distortions due to pressure, etc.), keys are general (approximate), rather than specific in nature. In accordance with the embodiment, these general keys form the basic foundation for high-speed general indexing.
In accordance with one embodiment, a set of keys is generated using some of the characteristics associated with bifurcation and rod image model elements (defined above in relation to FIGS. 40 and 41) . The information used in key generation for a bifurcation, with reference to FIG. 40, consists of the following characteristics :
Equation 30 • The coordinates-of center point 158.- The upper left corner of the image is assumed to have coordinates of 0,0. Positive x coordinates are right and positive y. coordinates are down.
• A first separation angle between leg segments 162 and 164 (separation angle 160).
• A second separation angle between segments 164 and 168.
• The direction of the bifurcation (direction angle 166).
A rod is considered to be a special case of a bifurcation with the following assumptions made with reference to FIG. 41:.
Equation 31.
• The end point of the rod (end-point 172) becomes the center point of the bifurcation (centerpoint 158).
• Using as the direction the direction of the segment extending from point 170 to end-point 158/172, the rod end is extended to points 162/164.
• Points 162 and 164 coincide.
• The angle between the segments extending from point 158/172 to points 162 and 164 (first .separation angle, previously referred to as separation angle 160) is zero.
• The angle between the segments extending from point 158/172 to points 164 and 170 (second separation angle, previously referred to as direction angle 174) becomes the direction angle.
The information used in key generation for a rod is.
Equation 32 • The coordinates of center-point/endpoint 158/172.
• The first separation angle (separation angle 160) Using the above assumption, this angle is zero.
. . The second separation angle (direction angle 174).
As print models are stored in the database, two tables are updated. The first table, KEY_1, is a two dimensional array. The first index is the first separation angle the second index is the second separation angle. It should be noted that the first separation angle cannot exceed 120 degrees, and, in accordance with one embodiment, could be limited to 115 degrees. In addition, second separation angle cannot exceed 180 degrees. The direction angle can be any value .between zero and 359. Therefore the array size of KEY_1 is 116 (0 to 115) by 180 (0 to 179). The second table, KEY_2, contains the key information. KEY 2 is a single dimension array that is open-ended.
Each entry in KEY_1 contains two elements, a count and an index into the KEY_2 table. The element count defines the number of bifurcations or rods that have identical separation and direction angles. The index element defines the starting point for the keys contained in table KEY_2. An entry in the KEY_2 table consist of the following elements:
Equation 33 • The x coordinate (X_CENTER) of the bifurcation center or rod end-point. The upper left corner of the image is assumed to have coordinates of 0,0.
• The y coordinate (Y_CENTER) of the bifurcation center or rod end-point. The upper left corner of the image is assumed to have coordinates of 0,0.
• The index to the bifurcation or rod in the model.
» The direction angle.
• The model identification number. As model records are stored in the database, they are assigned unique numbers. These numbers start at a predetermined value, usually 10,000,000, and increment, by one, with each record added.
• The finger identification number. Each finger is assigned a numerical value starting with zero. The left pinkie is assigned zero, the left ring assigned one, the last assigned, right pinkie, is nine (left to right).
• Control flags. This element contains information about the finger model:
Is the record marked as deleted?
In accordance with an embodiment of the present invention, updating of the tables, KEY_1 and KEY_2, is best described by the following coding example:
Equation 34
Set model index to zero. // loop through all bifurcations and rods in the model
TOP OF LOOP 1 .
Extract the direction angle (DA) from the model. Extract the first separation angle (SAI) from the model.
Extract the second separation angle(SA2) from the model.
Build the new entry for the KEY_2 table.
• The x coordinate. Extracted from the model entry.
• The y coordinate. Extracted from the model entry.
• The direction angle.
• The index. Set to model index.
• The model identification number. Passed into this routine.
• The finger identification number. Passed into this routine.
• Control flags. Set to zero.
Increment the count field in table KEY_1[ SAI][ SA2 ] by one.
// Increment all remaining key 2 index fields in the KEY_1 table.
// At this point the table KEY_1 can be thought of as a single dimension array with 41,760 (116 times 360) entries.
Set update index to the product of SAI times SA2.
TOP OF LOOP 2
Increment update index by one.
If update index is less than 41,760 increment KEY_1[update index].key_2_start_index. Increment update index.
Go to TOP OF LOOP 2 // make room in the KEY_2 table by sliding all entries, starting at // KEY_1[ SAI ][ SA2 ] ].key_2_start_index, up one.
Set update index to the number of entries contained in the KEY_2 table minus one.
Set stop index to the value stored in KEY_l.[ SAI ][ SA2 ].key_2_start_index.
TOP OF LOOP 3
If update index is greater than or equal tio stop index
Copy the entry at update index to update index plus one.
Decrement update index by one.
Go to TOP OF LOOP 3 i
Store the new key, generated above, in the'. KEY_2 table at stop index.
Increment the number of entries contained ,in the KEY_2 table by one.
// END CODE EXAMPLE .
Tables KEY_1 and KEY_2 provide a means of locating all bifurcations and rods that have identical angle characteristics. Table KEY_1 provides a count and a starting index. Table KEY_2 provides the key data. Also, it should be noted that the data stored in table KEY_2 is grouped such that all bifurcations and rods that have identical angle characteristics are contiguous.
. After the tables have been constructed, :he next step is to identify matching model images. Those prints that have a large number of bifurcations and rods that loosely match to an entry in the 'database are most likely. to compare with ,a high score. Using the above tables and some filter rules, a possible match table is created. Each entry in the possible match table contains the following information:
Equation 35. ׳ • A count of the number of bifurcations and (rods that loosely match.
• The distance, in both the x and y, between the two center points (stored as an ordered pair).
Where loosely matched is defined to be: j • The center points must be within 90 (this !and other specifically named values may vary) !pixels in the x direction.
• The center points must be within 120 (may ‘vary) pixels in the y direction.
• The angle difference for first separation !angle is within plus or minus 8 degrees.
• The angle difference for second separation angle is within plus or minus 8 (may vary) degrees.
• The direction angle is within plus or minus 20 (this value may vary) degrees. ;
This table is indexed using the model identification number and finger identification number. In accordance with one embodiment, the table is created as follows:
Equation 36
Clear the possible match table. !
Set model index to zero.|
TOP OF LOOP 1
Extract a model element.
// PROCESSING FOR BIFURCATIONS
I
If the element is a bifurcation |
Extract first separation angle from njodel. Extract second separation angle from ;model. Extract direction angle from model. Extract center point coordinate from :model. Set start index 1 to first separation angle minus the deviation (8 degrees).
' If start index 1 is less than one Set start index 1 to 1.
Set stop index 1 to first separation Jangle plus the deviation (8 degrees). j
If stop index 1 is greater than 116 Set stop index 1 to 116.
Set start index 2 to second separation angle minus the deviation (8 degrees).
If start index 2 is less than one Set start index 2 to 1.
Set stop index 2 to second separation plus the deviation (8 degrees).
Set index 1 to start index 1.
angle
TOP OF LOOP 2 I
Set index 2 to start index 2.
TOP OF LOOP 3
Set loop count to the value of the count element in table KEY1 [ index 1 ][ indsx 2 ] If loop count is not zero
Set key 2 index to index value intable KEY1 [ index 1 ][ index 2 ]
TOP OF LOOP 4 matched rules defined above.
If the loosely are met ,
Use model identification ana finger model numbers to form an Ind into the possible match table.
Increment the match count and store the distance .difference.
30'
Increment key 2 index by one. :
Decrement loop count
If loop count is greater than zepo
Goto TOP OF LOOP 4ן
Increment index 2
If index 2 is greater than stop index 2
Goto TOP OF LOOP 3
Increment index 1;
If index 1 is greater than stop index! 1 .
Goto TOP OF LOOP 2.
// PROCESSING FOR RODSI
If the element׳ is a rod
Extract direction angle from model. )
Extract center point coordinate from model.
Set start index 1 to direction minus(the deviation (20 degrees).
If start index 1 is less than zero Set start index 1 to zero.
deviation
Set stop index (20 degrees).
to direction plus the
If stop index 1 is greater than 359 !
Set stop index 1 to 359. i
Set index 1 to start index 1. j
TOP OF LOOP 5 ,
Set loop count to the value of the ci element in table KEY1[ 0 ][ index 2 :
table KEY1 [ 0
TOP OF LOOP 6
If loop count is not Set key 2 index ][ index 2 ] zero to index value in matched rules defined above,
If the loosely are met
Use model identification aAd finger model numbers to form an into the possible match table. . Increment the match count and the distance difference. Increment key 2 index by one. Decrement loop count If loop count is
Goto TOP OF greater than
LOOP 6 index st.ore zero
Increment index 1
If index 1 is greater
Goto TOP OF LOOP than stop index 1 5
Increment model index by 1. ί
If model index is less than total model elements Goto TOP OF LOOP 1 // END OF CODE EXAMPLE
At this point, the possible matc'h table contains a list of all models in the database that are candidates for comparison. The count field contains a count of.the bifurcations and rods that loosely match. The distance list contains an ordered list, count value entries, of the distances from the center point of model to those of entries in the database. Some of these entries are invalid. The filter rule below attempts to remove some invalid values. To remove invalid entries the distance list is first sorted, using the x distance, in ascending order (see example below). This, list is then scanned looking for the longest sequence of numbers whose ' distance is less than 40. The start and end index of this list is then used to sort the y component of the ordered pairs. The list is then scanned looking for the longest sequence of numbers whose distance is less than 40. The start and end index of this list provides a new count to replace the old count.
Equation 37 x -bl -3« -32 -23 -1a -113־ 7־ v 8b bM -31 רו b 11 י11. ח־b » « * « y ab bi -י<sup>11</sup>־ י<sup>5</sup>־ יגג־ 125־ 31 רו
S 11 IM IM IM 2M 2b26
11M -125 7 -lb 32 -12 -7b 125-87 « x « «» b 7 32 UM 11b -12 -7b 12587־
..
In the above example the original count is 17.
Following the first sort and index location, the start index is 4 and the end index is 12 (represented by asterisk). The second sort and index location yields a count of 3.
When all match counts have been adjusted, the table is sorted, in descending order, by count. Those entries at the top of the list (the highest count values) represent those models (from the database) that should be compared to the supplied live scan.
Qualification parameters can be adjusted at several levels to accommodate the reader 12 characteristics and overall security level as it relates to false matches versus false rejections. As an example, the match table distance parameter could be expanded to consider entries whose distance is greater than 60 versus 40, which would have the net effect of lowering the security level thereby reducing false rejections at the expense of increasing false accepts.
FIG. 45 is a block diagram illustrating a set of procedural components associated with a one-to-many, or database search, image comparison process, in accordance with an embodiment of the present invention. Block 196 indicates a first illustrative step in the process, specifically, deriving a first image data set based on a first image and a plurality of other image data sets based on a plurality of other images. The first and other image data sets include, a plurality of data elements, which are illustratively, but not necessarily, associated with fingerprint images. Different data element types associated with fingerprint images are described above in relation to one-to-one comparisons, as well as in other locations within the present description.
The image data sets derived for the first and other images are illustratively, model representations of corresponding first and other images. The image data sets are illustratively generated as described above in relation to other Figures incorporated into the description of the present invention.
’ The next step, indicated by block 198, is the creation of a directory that progressively lists, 5' based on at least one measured characteristic (e.g., a particular angle, length, etc.), a substantial number of a particular type, of data elements (e.g., one of bifurcation representations, rod representations, vectors associated and not . associated therewith, 10 microminutia points, etc.) that appear in the plurality of other image data sets. Each data element in the directory is illustratively listed with an identifier that represents association with a particular image data set within which' the data 15 element appears.
Block 200 indicates the next step in the process, wherein an array is created that includes a two-entry cell for each of a range (or substantially all) of potential configurations for the same one type of data 20 elements that are listed in the directory created in step 198.
Block 202 indicates a recording of process information within array cells. In accordance with one embodiment, in one entry of each two-entry cell is 25 recorded a quantity value representing a number of consecutive data elements in the directory that demonstrate characteristics that are approximately similar to characteristics of the one of the range of potential configurations that corresponds to the data element configuration associated with that two-entry cell. Illustratively, in the other entry of each twoentry cell is recorded an index value corresponding to an initial data element that begins the number of 5. consecutive data elements listed in the directory. In other words, the cells, are utilized to record locations within the progressive directory of data elements that potentially match the data element associated with a given cell location.
Block 204 indicates the next step, wherein there is an identification of a two-entry cell in the array that is associated with a data element configuration having characteristics approximately identical to a target data element, a data element taken from the 15 first image data set.
Block 206 indicates the next step, wherein the target data element is compared to a group of consecutive data elements listed in the directory, as indicated by the two-entry cell associated with that 20 target data element. As is indicated by block 208, this identification and comparison of target data elements to associated consecutive data elements in the directory is repeated for additional target data elements.
Block 210 indicates the next step. The step comprises calculating and noting a quantity of data elements in each of the plurality of other image data sets that approximately match data elements, illustratively target data elements, taken from the first image data set. As target data elements are found to match data elements in the directory, the above-mentioned identifiers assist in knowing which image data set to which an increase in count should be made.
As is indicated by block 212, similar directories, arrays, and counts can be made for numerous data element types. Illustratively, steps 198 through 210 are performed utilizing a variety of data element types and a single running count is kept for each of the other image data sets.
Block 214 indicates the next process step. The step comprises selecting from the plurality of other image .data sets a predetermined number of image data sets that include the highest count, or the most data elements that approximately match target data elements taken from the first image data set.
Finally, block 216 indicates performing a more thorough comparison of the predetermined number of image data sets to the first image data set. The more thorough comparison process is illustratively utilized to distinguish between the image data sets, and consequently images, that should or should not be considered matching. In accordance with one embodiment, the one-to-one comparison process described in relation to FIG. 4 4 is utilized as the more thorough comparison process.
FIG. 4 6 is a block diagram illustrating a set of procedural components associated with another one-to many, dr database search, image comparison process, in accordance with another embodiment of the present invention. Block 218 indicates a first illustrative step in the process, specifically, deriving a first image data set based on a first image and a plurality of other image data sets based on a plurality of other images. The first and other image data sets include a plurality of data elements, which are illustratively, but not necessarily, associated with fingerprint images. Different data element types associated with fingerprint images are described above in relation to one-to-one comparisons, as well as in other locations within the present description.
The image data sets derived for the first and other images are illustratively model representations of corresponding first and other images. The image data sets are illustratively generated as described above in relation to other Figures incorporated into the description of the present invention.
Block 220 indicates a step in the process comprising the creation of a B~tree data file for each of a plurality of data element types (e.g., bifurcation representations, rod representations, vectors associated with or independent of vector/rod representations, microminutia points, etc.). It should be pointed out that a B-tree data file is a data storage structure that is well known in the fields of data storage and data analysis. Illustratively, each B-tree data file is associated with a different data element type and includes substantially all of the data elements included in the other image data sets (not the first/target image data set) . The data elements from the other image data sets are illustratively, listed and categorized in the B-tree data files based on data normalization rules. Each data element listed in each B-tree data file is stored with an identifier that represents association with a particular image data set in which that data element appears. In accordance with one embodiment, each B-tree data file reflects normalization rules for a corresponding data element type, including variations of relative position and/or rotation. In accordance with another embodiment, each B-tree data file reflects substantially all potential relative associations between data elements stored in the data file and the plurality of data element types.
Block 222 indicates a comparing of at least one target data element (data element from first image data set) to data elements in B-tree data files having data elements of a same type as the target data element being compared. In other words, a target data element is taken from the first image data set and progressively compared through a corresponding (of the same data element type) B-tree. The purpose is to look for data elements in the B-tree data files that approximately match the target data element.
The process steps indicated by blocks 224, 226 and 228 .are substantially and respectively similar to ־108blocks 210, 214 and 216 described in relation to FIG. 45. For example, as is indicated by block 224, as target data elements are found to match data elements stored in a B-tree data file, for each' of the plurality of other image data sets, a count is generated of included data elements that approximately match a target data element.
Block 226 indicates, the next process step. The step comprises selecting from the plurality of other image data sets a predetermined number of. image data sets that include the highest count, or the most data elements that approximately match target data elements taken from the first image data set.' Finally, block 228 indicates performing a more thorough comparison of the predetermined number of image data sets to the first image data set. The more thorough comparison process is illustratively utilized to distinguish between the image data sets, and consequently images, that should or should not be considered matching. In accordance with one embodiment, the one-to-one comparison process described in relation to FIG. 44 is utilized as the more thorough comparison process.
Although the present invention has been described with reference to illustrative embodiments, workers skilled in the art will recognize that changes may be made in form and detail without departing from the spirit and scope of the invention.
Contents20
44 sheets
Sheet 1 Sheet 2 Sheet 3 Sheet 4 Sheet 5 Sheet 6 Sheet 7 Sheet 8 Sheet 9 Sheet 10 Sheet 11 Sheet 12 Sheet 13 Sheet 14 Sheet 15 Sheet 16 Sheet 17 Sheet 18 Sheet 19 Sheet 20 Sheet 21 Sheet 22 Sheet 23 Sheet 24 Sheet 25 Sheet 26 Sheet 27 Sheet 28 Sheet 29 Sheet 30 Sheet 31 Sheet 32 Sheet 33 Sheet 34 Sheet 35 Sheet 36 Sheet 37 Sheet 38 Sheet 39 Sheet 40 Sheet 41 Sheet 42 Sheet 43 Sheet 44
27 members in 10 offices
Priority claims8
| Document | Office | Kind | Date |
|---|---|---|---|
| 99158901 | United States of America | A | |
| 99158901 | United States of America | A | |
| 0236934 | United States of America | W | |
| 0236934 | United States of America | W | |
| 09991589 | – | – | – |
| PCTUS2002036934 | – | – | – |
| US20010991589 | – | – | – |
| WO2002US36934 | – | – | – |
Members27
| Document | Office | Kind | |
|---|---|---|---|
| CA2467529A1 | Canada | A1 | |
| CA2817686A1 | Canada | A1 | |
| WO03044725A2 | World Intellectual Property Organization (WIPO) | A2 | |
| AU2002348288A1 | Australia | A1 | |
| US2003118218A1 | United States of America | A1 | |
| WO03044725A3 | World Intellectual Property Organization (WIPO) | A3 | |
| EP1446762A2 | European Patent Office (EPO) | A2 | |
| EP1467308A1 | European Patent Office (EPO) | A1 | |
| EP1501040A1 | European Patent Office (EPO) | A1 | |
| US6895104B2 | United States of America | B2 | |
| ZA200403899B | South Africa | B | |
| US2005201597A1 | United States of America | A1 | |
| RU2004118064A | Russian Federation | A | |
| IL162003A0 | Israel | A0 | |
| IL162003D0 | Israel | D0 | |
| EP1467308B1 | European Patent Office (EPO) | B1 | |
| AT322052T | Austria | T | |
| ATE322052T1 | Austria | T1 | |
| DE60210348D1 | Germany | D1 | |
| DE60210348T2 | Germany | T2 | |
| RU2302656C2 | Russian Federation | C2 | |
| US7359553B1 | United States of America | B1 | |
| AU2002348288B2 | Australia | B2 | |
| US7539331B2 | United States of America | B2 | |
| IL162003AThis record | Israel | A | |
| CA2467529C | Canada | C | |
| CA2817686C | Canada | C |
2 legal events, as the office reported them to INPADOC
Over the term
Point at a mark for the eventEvents
| Event | Code | |
|---|---|---|
| Patent not in force due to non-payment of renewal feesMM9K | MM9K | |
| Patent grantedGrantedFF | FF |
Numbers
- Publication, DOCDB
- 162003
- Publication, EPODOC
- IL162003
- Application
- 162003
- Application, DOCDB
- 16200302
- Application, EPODOC
- IL20020162003
Titles
- English
- IMAGE IDENTIFICATION SYSTEM
Classification
- CPC, 3
- G06V40/1359
- G06V40/1365
- G06V10/993
- IPC, 1
- G06K9 00