Method and apparatus for the electronic capture of rolled fingerprints

0001The invention relates to a method and an arrangement for the electronic recording of a moving object, in particular for receiving a finger which has been unrolled on a support surface. The invention finds primarily application for the production of electronic fingerprints, but can also be used to advantage in object tracking for the intelligent control of a variable in position and / or size sampling window.

0002The background for the use of fingerprints in forensics is the uniqueness of the dermal scar images, which are not inheritable and which are immutable from the fourth embryonic month to dissolution after death. The image template can be the finger itself, a fingerprint applied to paper by ink or a trace photogram. The latter two techniques were and are mainly used by the police. With the help of pattern recognition techniques capable of extracting the characteristics of a fingerprint, it is possible to automate the identification and verification of fingerprints.

0003In recent years, the above-mentioned techniques have increasingly been integrated into electronic systems that allow the finger to be picked up directly. This shortens the recording and evaluation time, which at the same time improves the quality of the images. If a finger has not been detected correctly, it is possible to immediately repeat the process of recording that one finger. The electronic image capture of fingerprints is usually done with matrix or line sensors based on CCD or CMOS technology. Here, the fingerprint is converted as an image template by special optics and sensors in an electronic image and then digitized, creating a raster image with a fixed spatial and gray scale resolution. However, capacitive, thermal, ultrasound-based or pressure-sensitive sensors can also be used to record electronic fingerprints.

0004In addition to the recording of mere fingerprints of the fingers, the police also create images of unrolled fingers. The procedure for producing unrolled fingerprints with ink and paper is simple enough, as ink is applied to the finger and the finger is then unrolled onto paper. Rotation and deformationsNegreases at the ends of the finger are accepted and are permissible. In electronic systems, such a procedure is somewhat more complicated. Here, depending on how the finger is detected electronically (with line or matrix), an overall picture is composed of many individual pictures. There is the problem that compared to the unrolling with ink on paper, not a complete shot, but a variety of discrete-time scans arises, which entail a loss of information in itself. The causes for the loss of information are due not only to the deformation of the finger and the resulting changes at two different points in time, especially in too little temporal scanning of the finger by the image acquisition unit. That is, to ensure that a finger is picked up correctly, the image acquisition unit must have a minimum frame rate, depending on the method of composition used, so that disturbing operations, such as Twisting or slipping during the rolling process of the finger, can be detected.

0005Methods of how to get from the unavoidable individual images to an error-minimized overall picture have been described several times, with the production of the image of a rolled-out finger always starting from a successively recorded sequence of electronic individual pictures. The overall picture is composed of these individual pictures by means of various methods. Here are - the individual images decomposed into strips. In the pamphlets <patcit id="pcit0001" dnum="US4933976A"><text>US 4,933,976</text></patcit>, <patcit id="pcit0002" dnum="US5230025A"><text>US 5,230,025</text></patcit>, <patcit id="pcit0003" dnum="US6483932B"><text>US 6,483,932</text></patcit>, <patcit id="pcit0004" dnum="WO9741528A"><text>WO 97/41528</text></patcit> and <patcit id="pcit0005" dnum="DE19851544C1"><text>DE 198 51 544 C1</text></patcit> (<patcit id="pcit0006" dnum="WO0028470A"><text>WO 00/28470</text></patcit>) a strip is taken from the individual images and the overall picture is composed of these strips. The above-mentioned patents differ from each other in their different algorithms with which the strips are determined and put together in order, if possible, to assemble only the parts of the individual images which correspond to the actual contact surface of the finger.

0006In all of the references listed above, the frames coming from the imaging unit have a fixed size and are based on a fixed time regime (dictated by a clock generator or an event trigger). As a result, the unwinding speed (and possibly its change) as well as the size of the finger contact surface (and its real change in the unrolling process) can not be taken into account in the readout regime of the image pickup unit. In order to allow a correct calculation of the result image, two neighboring fingerprint strips must have a sufficiently large intersection for the respective method. This can only be achieved if a fast time sampling occurs when rolling the finger. This is currently being realized by sensors with a high image read rate (frame rate) of more than 25 frames / second (B / s). When increasing the spatial resolution (eg from 500 dpi to 1000 dpi), it hits the technical limits. Either the frame rate is reduced (eg when changing from 500 dpi to 1000 dpi reduction of frame rate)¹/ 4) or the price of the sensor with its electronics becomes unacceptably high.

0007The invention is therefore based on the object to find a new way to electronically record unrolled fingerprints, the high-resolution frames and a tight sequence of images for seamless composition of the individual images to an overall picture without resorting to expensive image sensors with high frame rate (read speed) got to.

0008According to the invention, the object is achieved in a method for electronically recording a moving object, in particular for recording a finger unrolled on a recording surface, wherein the object is detected with its structures during movement by means of a spatially resolving image recording unit in a series of individual images and processed as a two-dimensional electronic image , solved by in that a position enclosing the object is determined from the position and size of the object in at least one of the individual images read from the image recording unit, that a rectangle which is enlarged by tolerance additions and aligned parallel to the row and column direction of the sensor is determined from the enclosing figure . and that the enlarged rectangle for predicting and adjusting the size and location of an active pixel area of ​​the image pickup unit is used for at least one subsequently readable image so that the read active pixel area of ​​the image pickup unit is always kept small due to its matching and a higher frame rate in the sensor readout or data transfer is realized.

0009Advantageously, the enclosing figure of the object image of the currently scanned image is determined by a respective gradient image formed in the column and in the row direction, a difference of adjacent gray values ​​in the row or in the column direction being formed for the generation of the gradient image for each pixel. In this case, it is expedient to form the difference of adjacent gray values ​​from the gray values ​​of the predecessor and successor pixels of the currently examined pixel of the row or column in order to produce each pixel of the gradient image. In order to be able to extract edges of the fingerprint with particular reliability, the pixels of each gradient image are preferably calculated from the difference between the gray values ​​of the preceding and succeeding pixel of the currently examined pixel of the lines or Formed columns. For a current image, start and end values ​​of the enclosing figure of the object image are advantageously determined from at least one gradient image of the rows or columns aligned in the direction of the object movement, a first and a last significant difference of adjacent gray values ​​per row or column Calculation of the start and end values ​​of the figure enclosing the object is determined. The significant differences are expediently determined by exceeding threshold values. It proves advantageous to calculate an average value or a median value of the start and end values ​​determined from the gradient image in at least one dimension which corresponds to the main movement direction of the object, the mean or median values ​​then being linear boundaries of the object image being rectangular form enclosing figure.

0010In a preferred embodiment, the start and end values ​​are entered into a location histogram at least from the gradient image that corresponds to the main movement direction of the object, the locations at which the integral frequency distribution reaches a significant threshold form the boundaries of a figure enclosing the object image in a rectangular manner , This threshold value can make sense 5% of the integral frequency distribution. But it is also possible to choose the maximum value of the frequency distribution To save on computational and storage capacity, it is advantageous to use only the start and end values ​​of selected rows or columns of the gradient image for determining the boundaries of the rectangular enclosing figure.

0011In order to avoid a trimming of the object image, a tolerance addition for generating the enlarged rectangle is appropriately determined differently depending on a detected movement sequence of the object, wherein different evaluation algorithms are applied from at least two previously read-out images after the determination of the boundaries of the respectively enclosing figure on the basis of the temporal change of the boundaries of the enclosing figure in the successive images.

0012Preferably, a finger rolling on a receiving surface is detected as the object, a fingerprint being recorded as the instantaneous contact surface of the finger as the object image in each image read. The following steps are advantageously applied to each of the read-out images for setting the boundaries of the individual images:<ul id="ul0001" list-style="dash" compact="compact"><li>a pad detection routine that, upon detection of an insufficient number of fingerprint differential values, inhibits calculation of the tolerance allowance and subsequent predictions and adjustments of the active pixel area of ​​the imaging unit;</li><li>a motion-tracking routine, which, when the movement of the surrounding figure of a fingerprint is not clearly directed, determines an all-round uniform tolerance addition around the boundaries of the surrounding figure, which is chosen so large that, with any movement of the object in object-typical dimensions, the enlarged rectangle will move the object within does not curtail the time of reading at least one next picture; and</li><li>a tracing-following routine is performed when, in a defined number of successive pictures, the boundaries of the enclosing figure have clearly shifted in an excellent direction, and for the limit of the enclosing figure moving in the direction of movement of the object, from a calculated velocity of the moving boundary Tolerance allowance is determined the enlarged rectangle does not clip the current object image (fingerprint) within the time of at least one subsequently read-out image.</li></ul>

0013Since the roll-over trace routine is the centerpiece of the image capture of a rolled-up finger to provide a complete and temporally high-resolution series of individual images for the composition of the complete footprint of a rolled-up finger, roll-off tracking is conveniently accomplished by roll-off detection, which terminates the motion tracking in the roll-over tracking when there is a certain set number of progressions in the same direction of the boundaries of the surrounding figure, and terminated by a roll-off detection, which terminates the roll-tracking upon reversing the direction of movement of the bounding character boundaries.

0014When rolling a finger along a recording surface, an intermediate storage of all images used for the start detection is advantageously carried out within the Abrollstarterkennung to keep the already read images for the subsequent complete composition of the unrolled finger available.

0015Furthermore, the object of the invention in an arrangement for electronically recording a moving object, in particular for recording a finger unrolled on a recording surface, with an image recording unit for recording a sequence of two-dimensional individual images of a moving object, wherein the composition of an overall image of the moving object Individual images have local overlaps, according to the invention solved by in that the image acquisition unit is followed by a logic unit for continuously calculating a restricted pixel region which closely encloses the object image on the basis of at least one image previously read by the image acquisition unit, in which the logic unit is assigned a program memory, a data memory, a processor and an interface the processor is provided for controlling the data flows, and that the interface for data transfer of the images, which are controlled by the logic unit in terms of size and location of your pixel area defined, is present on an external computer unit, wherein the computer unit contains the image processing for assembling the temporally successively recorded images into an overall image.

0016The image acquisition unit preferably has an optoelectronic sensor, but may also be equipped with a capacitive, inductive, thermal, ultrasonic or other type of contact sensor. Suitable optoelectronic image sensors are CCD arrays or CMOS arrays (each in the form of a matrix or line array).

0017Advantageously, the image acquisition unit can also contain a sensor with a low frame rate (&lt;25 B / s), in which the low frame rate is based only on the reading of all pixels of the sensor, but the reading of images from an arbitrarily programmable active pixel area with a much larger frame rate is feasible. In an expanded variant of the invention, the image acquisition unit preferably has a large-area sensor in which a complete footprint or handprint with high resolution but low frame rate can be realized by a single sensor reading, in which a defined partial area of ​​the receiving surface is defined for receiving individual unrolled fingers with the sensor in a correspondingly limited active pixel area, which is assigned to the defined subregion of the receiving surface for the finger to be unrolled, can be controlled to read out pixel-reduced images, so that the image rate for recording unrolled fingerprints is considerably increased compared to footprints or handprints.

0018In a modified variant of the invention, the image acquisition unit may also be a high-frame-rate sensor (≥ 25 B / s) in which the interface is the limiting element of data transmission, so that only a small portion of the image data can be transmitted in "real time" , In this case, the readout of data-reduced images according to the invention takes place from arbitrarily accessible pixel areas of the data memory in order subsequently to increase the data rate of the transmission via the interface.

0019The method according to the invention for determining the boundaries of the active pixel area advantageously takes place in a logic unit (hardware) which precedes (the arithmetic unit for assembling the images of the object) and which is expediently an FPGA (freely programmable gate array). The logic unit can also be a PLD (programmable logic element) or an ASIC (application-specific integrated circuit). The processor used to control the processes and data streams of the logic unit may advantageously be a microprocessor (MPU), a digital signal processor (DSP) or a microcontroller (MCU). The processor can also be integrated directly into the logic unit, the program memory or the external computer unit.

0020The invention is based on the basic idea of ​​a gray scale image of variable surface area and / or structure that can be represented as function G (x, y), in which x and y are the coordinates of a respective pixel of the image and G records its gray values ​​electronically. A moving object with such image characteristic is t at different times_n sequentially with t_{n + 1} > t_n with different views as function G_n taken to the overall picture G from pictures G_n(x, y) are formed, wherein the image pickup unit only interesting portions of the object surface are recorded, the size of which vary depending on the rotational speed of the object. The essence of the method according to the invention is to define an adapted limited pixel section from each currently read image for the next image respectively to predict, which with the necessary security completely contains the fingerprint to be detected and is set as a read window of the image pickup unit to increase the read rate (frame rate) of the image pickup unit. In this way, it is possible to use inexpensive sensors that do not meet the otherwise required high frame rate of ≥ 25 B / s for the reading of complete individual images, but have a so-called WOI (Window of Interest) adjustability, and the have full local resolution. In CMOS technology, this application is called "Window of Interest" or "Region of Interest", "Active Window" and "Windowing" depending on the manufacturer. In CCD technology, the name "Fast Dump" for skipping lines and "overclocking" for overclocking the unneeded columns has been established.

0021With the invention, it is possible to realize an electronic recording of unrolled fingerprints, which generates the high resolution images adapted to the object size and movement in close sequence and allows their reliable composition of the images to the overall image without the need for expensive image sensors or interfaces with high Frame rate must be used. The invention allows to choose the frame rate much higher than is possible with the currently used high-resolution sensors or image processing interfaces with complete reading of the image area. In addition, due to the electronic adjustment of the sensor readout window, applications such as taking a complete hand and rolling a finger with one and the same device are possible.

0022The invention will be explained below with reference to exemplary embodiments. The drawings show:<dl id="dl0001" compact="compact"><dt>Fig. 1:</dt><dd>a schematic representation of the method according to the invention,</dd><dt>Fig. 2:</dt><dd>a representation of the prior art for the composition of individual images of unrolled fingers due to insufficient sampling density of the individual images,</dd><dt>3:</dt><dd>a representation of the problems of the prior art in the composition of papillary lines of unrolled fingers due to lack of overlap of the individual images,</dd><dt>4:</dt><dd>a schematic representation of an optoelectronic device for obtaining fingerprints of a rolling on a receiving surface finger,</dd><dt>Fig. 5:</dt><dd>a basic structure of the control hardware according to the invention for setting the active pixel area for the reading of adapted, the current fingerprint sure enclosing images,</dd><dt>Fig. 6</dt><dd>a flow chart for a run detection routine,</dd><dt>Fig. 7:</dt><dd>a flowchart for a motion tracking routine for determining the figure enclosing the moving finger by means of spatial histograms,</dd><dt>Fig. 8:</dt><dd>a schematic representation of the definition of the starting position for the rolling of a finger and division of the receiving surface,</dd><dt>Fig. 9:</dt><dd>a schematic representation of the data buffering for storing read-out images that were used for the roll-off starter recognition, for the subsequent composition of the overall image,</dd><dt>Fig. 10:</dt><dd>a schematic representation of the rolling of a finger to illustrate the different sizes of the fingerprint, the enclosing figure and the enlarged rectangle,</dd><dt>Fig. 11:</dt><dd>schematic representation of the rolling process of a finger for explaining the determination of the speeds and accelerations of the boundaries of the surrounding figure of the fingerprint,</dd><dt>Fig. 12:</dt><dd>Representation of the functional progressions of path, velocity and acceleration of the boundaries of the enclosing figure, plotted over the number of the swept sensor pixel,</dd><dt>Fig. 13:</dt><dd>schematic representation of the monitoring of tolerance addition for the position of the border of the rectangle as active pixel area for the next image reading to avoid the fingerprint cropping in the following image.</dd></dl>

0023In <figref idref="f0001">Fig. 1</figref> is the basic principle of the method for electronically recording an imprint image (hereinafter: fingerprint 11) from the surface of a finger 1 shown. The method is based on the fact that a finger 1 is recorded at different times sequentially in different, overlapping views, to subsequently - not in accordance with the invention - to put together an overall image of the unrolled finger 1. Due to the above-mentioned "roll-off rule" for the finger 1, the finger surface can be picked up by the image pickup unit 2 only piecemeal as a successive series of frames. The size and position of the recorded images 31 in this case vary depending on the current impression surface and the rolling speed of the finger. 1 For this purpose, a complex control method is proposed, depending on the current impression surface and the unwinding speed of the finger 1 to set and monitor an adapted limited section in the reading of the image pickup unit 2, so that high frame rates can be achieved without having to resort to expensive image sensors.

0024In contrast to rolling a finger 1 with ink on paper when rolling on an optoelectronic image pickup device, as exemplified in <figref idref="f0004">Fig. 4</figref> is specified, no continuous recording of the rolling process instead, but there are discrete-time scans made. To illustrate the invention, a comparison is made with the prior art, the in <figref idref="f0002">Fig. 2</figref> and <figref idref="f0003">Fig. 3</figref> is shown schematically. According to the conventional procedure, an image pickup unit 2, which can be read out, for example, at 25 B / s, takes 75 pictures during a rolling process of 3 s duration. Each of these 75 images is a full image containing the fingerprint 11 at a defined time in a particular location (size and position). In order to generate an overall image 3 of the unrolled finger 1 from these individual images, specific image strips 12 are cut out of these individual frames, which represent the information-relevant parts of the read-out image for the composition. For a correct calculation of the overall image 3 to take place, two adjacent image strips 12 must have a sufficiently large intersection of the fingerprint 11. One common method of composing two image strips 12 is the cross-correlation and overlapping "cementing" of the two image strips 12 at the location of closest coincidence of the edge regions. However, discrete-time scanning entails a loss of information caused by the deformation and movement of the finger 1 and the resulting changes in the image of the fingerprint 11 at two different times of recording. This information loss increases as the unwinding speed of the finger 1 increases relative to the frame rate of the image pickup unit 2. In other words, to ensure that a finger 1 is picked up correctly, the image acquisition unit 2 has to fulfill a minimum image rate for assembly so that, for example, processes such as twisting or slipping during the rolling process can be detected and corrected. In order to clarify the problems when the read-out speed of the image pickup unit 2 is too low, FIG <figref idref="f0002"><b>Fig. 2</b></figref> in the left partial view, the result of an overall picture 3 of a rolled-out finger 1 is stylized, as it occurs when rolling the finger 1 with ink on paper and how it is to be generated as an overall image 3 by composing a sequence of individual images. The overall image 3 is to be assembled according to the detail images of a finger 1 of three images per time interval shown in the upper right row, wherein according to the prior art, a series of three snapshots of the finger 1 are scanned as a frame of the entire image acquisition unit 2 in a fixed time regime , Thereafter, information-bearing image parts are extracted, whereby the three image strips 12a to 12c are formed. If the time interval between the snapshots is too long (ie, the frame rate of the image pickup unit 2 is too low) compared to the movement of the finger 1, there will be allocation problems if the overall image 3 is assembled even if there is an overlap of the edge areas of the image strips 12a to 12c Finger 1 was additionally laterally shifted or rotated within the rolling motion. In this sample, the two upper papillary lines 13 and 14 have performed a translational movement and the uppermost papillary line 13 has additionally experienced a rotational movement. In the enlarged detail on the top right, it is shown that the first segment 131 of the upper papillary line 13 from the strip 12a is closer to the third segment 143 of the next papillary line 14 than the associated second segment 132 of the upper papillary line 13 from the strip 12b. This can lead to the incorrect assembly of at least the papillary lines 13 and 14 in the overall image 3 and thus to misinterpretations of the structures of the unrolled finger 1.

0025Much better done the assembling in the lower row of the detailed representations of <figref idref="f0002"><b>Fig. 2</b></figref>which symbolize a higher frame rate scan by operating on five frames per time interval. The five image strips 12a to 12e extracted therefrom make it possible to detect the sliding and rotational movements superimposed on the unrolling of the finger 1 and thus to achieve a clear, correct composition of the strips 12a to 12e in the overall image 3, since cracks of papillary lines from an image strip 12 to the next are largely limited and thus remain traceable. However, due to the relatively large receiving surface 21 of the image recording unit 2 required for the unwinding of the finger 1, any increase in the frame rate is limited by the sensor costs that are disproportionately increased by the increased frame rate.

0026In <figref idref="f0003"><b>Fig. 3</b></figref> Another problem of the image reading speed which is too low is represented by the fact that the instantaneous area of ​​the fingerprint 11 is represented by an elliptical area per frame reading. The solid ellipses 15a, 15b and 15c illustrate the contents of the temporally too widely spaced impression images in conventional Auslesung inexpensive image sensors, while an increased frame rate is symbolized by additional intermediate scans as dashed ellipses 16d and 16e, as according to the invention with inexpensive sensors or according to the prior art is achieved only with very expensive sensors. In the upper and lower edge regions of the fingerprint 11 takes place - due to the finger geometry - the largest change from one to the next single image recording. As a result, the intersection of two images on the edge area is the smallest. Are now, as in<figref idref="f0003"><b>Fig. 3</b></figref> stylized drawn characteristic features, such as a papillary line branch 161 or papillary ends 162, 163 and 164, outside the temporally successive as solid ellipses 15a-c drawn fingerprint areas, so are interrupted by papillary lines 16 such that in a result image (right detail representations of <figref idref="f0003">Fig. 3</figref>) the line separation 16a of the first ellipse 15a can no longer be unambiguously associated with the line separations 16b and 16c of the next ellipse 15b of the currently detected fingerprint 11. In the enlarged detail of <figref idref="f0003"><b>Figure 3</b></figref> It can clearly be seen that between the two adjacent solid ellipses 15a and 15b as well as 15b and 15c (as stylized instantaneous fingerprints 11) there is in each case one cut surface 17 which does not extend over the entire width of one <figref idref="f0002"><b>Fig. 2</b></figref> The cut-out image strip 12 described goes so that areas of ridge lines 16 are outside the composite total area of ​​the ellipses 15a, 15b and 15c. With the circular left-hand detail representations, two such areas are picked out and enlarged in the event that the image rate is too low and thus includes only the solid ellipses 15a to 15c. In the upper left detail representation, after optimal correlation of the elliptical surfaces 15a-c, the composite overall image 3 appears incompletely. In this case, the branch 161 is mapped in the composition of the fingerprint surfaces 15a-c instead of a feature as three features (line ends) and thus falsified. Even with a context analysis, it is unclear whether it is a branch, a line extension and a line end, or actually three terminating line ends. The lower left detail of <figref idref="f0003"><b>Fig. 3</b></figref> shows another critical area of ​​the unrolled finger 1 between the ellipses 15b and 15c, in the complete overall picture 3 two papillary line ends 162 and 163 and a lying in the scanning gap papillary end 164 can be seen. This patchy composite result image shows five line separations 16a through 16e that appear as five line ends and that do not allow for any degree of closeness. Here, in context analyzes, only speculation is possible, whether at the line separations 16a to 16e<ul id="ul0002" list-style="dash" compact="compact"><li>there is a simple branch from the line partition 16a to the line dividers 16c and 16d or from the line partition 16b to the line dividers 16d and 16e and one continuous line from 16b to 16e or from 16a to 16c, respectively</li><li>a continuous line from the line partition 16a to 16c or 16d or from 16b to 16d or 16e as well as three line ends are to be expected, or</li><li>two solid lines are provided from the line separations 16a and 16b to 16c, 16d or 16e and one line end at one of the line separations 16c to 16e.</li></ul>The complete and correct overall image 3 can only be obtained by a higher frame rate, symbolized by an intermediate scan of the finger 1 between the impression ellipses 15b and 15c shown as dashed ellipse 15e, ie when the frame rate of the image acquisition unit 2 is approximately doubled. This Abtastsituation the finger 1 is in the lower right detail of<figref idref="f0003"><b>Fig. 3</b></figref> represented and leads based on the prior art again to the demand for immensely expensive image sensors with high resolution and much higher (eg double) frame rate.

0027In <figref idref="f0004"><b>Fig. 4</b></figref> an image recording unit 2 for the optoelectronic image recording of a finger 1 according to the principle of total reflection on a receiving surface 21 is shown. In a typical basic construction, the image acquisition unit 2 consists of a diffuse illumination unit 24, a prism 25 and a camera 26 with imaging optics and an optoelectronic sensor 22. In this principle, the base area of ​​the prism 25 is the actual acquisition area 21. The light is coupled by the illumination unit 24 in the prism 25 so that it is totally reflected without applied finger 1 on the base surface, ie when the critical angle for total reflection (here glass / air) is exceeded. Thus, a brightly illuminated image is generated by means of the imaging optics on the sensor 22 in the camera 26. If a finger 1 is placed on the prism 25, changes at the points where the finger 1 rests, the transition glass / air to glass / skin. Since the skin has a higher refractive index than air, the critical angle of total reflection is greater. The total reflection is thus canceled at these points and the light is decoupled. On the sensor 22, which is arranged in the camera 26, the finger 1 is therefore imaged as an image in black and white transitions. The sensor 22 converts these light intensities into electrical signals which are subsequently digitized, as a result of which a raster image with a defined spatial and grayscale resolution is produced. The optoelectronic image recording of fingerprints 11 takes place - without limiting the generality - with matrix or line sensors based on CCD or CMOS technology.

0028In the above-described arrangement of <figref idref="f0004"><b>Fig. 4</b></figref> For simplicity's sake, and not limited to the image pickup principle described above, only the pickup surface 21 and the image pickup unit 2 will be referred to, the unrolled finger being referred to hereinafter as the base of the prism 25 1 in the image pickup unit 2 in any manner in an electronic image converted and output digitized. It is also possible to take fingerprints 11 instead of an opto-electronic sensor 22 to equivalently use capacitive, pressure-sensitive, ultrasound-based or thermal sensors within a suitably designed image pickup unit.

0029The inventive method, preferably in hardware in an arrangement according to <figref idref="f0005"><b>Fig. 5</b></figref> is realized, takes into account different object surfaces and speeds and thus ensures an optimally adapted scanning of the moving object.

0030For receiving a rolled finger 1, the method as in <figref idref="f0001"><b>Fig. 1</b></figref> represented, are divided into the following process sections:<ul id="ul0003" list-style="dash" compact="compact"><li>Pad detection 61 (<u style="single">P</u>ut-on <u style="single">R</u>ecognition - PR routine 61),</li><li>Motion tracking 62 (<u style="single">M</u>ovement <u style="single">T</u>racking - MT routine 62),</li><li>Roll Starter Detection 63 (<u style="single">R</u>oll <u style="single">S</u>tart <u style="single">R</u>ecognition - RSR routine 63),</li><li>Roll Tracking 64 (<u style="single">R</u>oll <u style="single">T</u>racking - RT routine 64) and</li><li>Rolling termination detection 65 (<u style="single">R</u>oll <u style="single">e</u>nd <u style="single">R</u>ecognition - RER routine 65).</li></ul>

0031Starting point of the procedure - as in <figref idref="f0001"><b>Fig. 1</b></figref> displayed - is an (arbitrary) information that tells you that a rolling process should take place. This information is usually transmitted by an external computer unit 4, which processes the recorded data. The information can also be specified by another external system, with the arrangement according to<figref idref="f0005"><b>Fig. 5</b></figref> communicatively communicates via the interface. By this sent information, the arrangement knows that a finger 1 will soon be launched and initiates corresponding detection steps. With this active circuit, the following detection steps of the method can take place.

1. Run detection 61 (PR routine)

0032The task of the PR routine 61 is to check in each image 31 supplied by the image acquisition unit 2 whether a valid fingerprint 11 is present on the support surface 21 or not. In this case, an x-gradient image 71 and a y-gradient image 72 are generated from each image 31 by forming differences of adjacent gray-scale pixels in each line or column. In <figref idref="f0006"><b>Figure 6</b></figref> a possible program sequence for recognizing a "valid fingerprint" is shown, this program sequence in each case referring to a single image 31 within an image sequence. From each electronically converted image 31, which has a gray value distribution G_{i, j} has a gradient image 71 in the x-direction with the values ​​Dx_{i, j} and a gradient image 72 in the y direction with Dy_{i, j} formed by in each row or each column differences 73 and 74 are formed from the recorded adjacent gray scale pixels. It is possible for each pixel of the gradient image 71 to be generated currently within the lines to form the difference 73 from the direct predecessor pixel and the direct successor pixel of the associated gray value pixel in the image 31 or - to improve the sensitivity - from the predecessor and the successor successor pixels. The same applies to the differences 74 in the column direction. In the gradient images 71 and 72 thus generated, a search is made for difference values ​​73 and 74 which exceed a specific gray-level threshold SwA. The gray value threshold SwA defines a difference value size beyond which a difference value 73 or 74 formed from predecessor and successor gray value pixels counts as "valid". This gray value threshold SwA does not have to be firmly defined, but can also be variable (changing from image to image) and thus better adapt to the image quality of the image 31 received. If the difference is "valid", a variable CountR defined in the difference counter 75 for the lines of the x-gradient image 71 is calculated_j or a variable CountC defined in the difference counter 76 for the columns of the γ gradient image 72_i each counted higher by one. Checking the criterion and incrementing the difference counters 75 and 76 is done until the end of the column or row is reached. Thereafter, it is checked whether the respective count in the difference counters 75 and 76 exceeds a minimum threshold. In this case, the rows or columns whose number of "valid" difference values ​​73 or 74 determined in the difference counter 75 or 76 are each a threshold SwB or SwC exceed and thus "valid" lines or columns represent, in a line counter 77 for all valid rows (RowCount) or in a column counter 78 for all valid columns (ColumnCount) added up. The thresholds SwB and SwC may well be the same size, but they do not have to. During a scrolling operation of the finger 1 in the line direction of the image pickup unit 2, the number of "valid" differences 73 per image 31 in the line direction greatly varies. At the beginning and end of a rolling process, this number is approximately two times smaller than in the middle part of the rolling process. The number of "valid" column direction differences 74, on the other hand, is relatively constant during the rolling process and is of the same order of magnitude as the number of "valid" lines 73 in the row direction during middle part of the rolling process. Therefore, the threshold SwC can be selected to be greater than the threshold SwB. In the present example, both thresholds SwC and SwB were chosen to be the same size as to be more sensitive to the placement of a finger 1.

0033When all the rows and columns of the gradient images 71 and 72 have been examined for the valid rows or columns and summed up in the row counter 77 and column counter 78 respectively, the count of the column counter 77 for the valid columns (ColumnCount) and that of the row counter 78 for the valid rows (RowCount) have exceeded a threshold SwD or SwE. In this comparison, the same sizes are chosen for the thresholds SwD and SwE in order to make the verification more sensitive. Formally, however, in analogy to the thresholds SwB and SwC, the threshold SwE can be selected to be larger than the threshold SwD. The decision as to whether a "valid finger" exists can also be limited to having only one of row counter 77 and column counter 78 exceed the thresholds SwD and SwE, respectively. This is when rolling the finger 1 in the row direction, for example the column counter 78 is best. If the output values ​​of the line counter 77 and / or of the column counter 78 exceed the threshold value SwD or SwE, then a "valid finger" rests on the receiving surface 21 and the motion tracking 62 can be initiated.

Motion tracking 62 (MT routine)

0034The MT routine 62, which is described in U.S. Pat <figref idref="f0007"><b>Fig. 7</b></figref> is shown schematically, is used to track the due to a rolling movement of the finger 1 successively changed position of the current fingerprint 11 (footprint). The starting point for the evaluation by the MT routine 62 may also be predetermined by a switch in the control panel, a foot switch or an external device (eg, arithmetic unit 4). At the first pass of the MT routine 62, the image 31 read out of the image pickup unit 2 has the size of the entire active area of ​​the sensor 22 and is thus one frame. In the subsequent runs, depending on the calculated area size, only a selected active pixel area 23 of the sensor 22 is read out, so that the read-out image 31, relative to the entire sensor area, represents a partial image. The MT routine 62, whose function is based on the <figref idref="f0007"><b>Fig. 7</b></figref> and the schematic representations of <figref idref="f0008"><b>Fig. 8</b></figref> is to be explained, the tracking of the moving finger 1 is used to prepare his defined unwinding after it has been recognized as a "valid finger". In this case, the position and size of an information-relevant section of the image 31 is recalculated for each subsequently adapted image 31 to be read out of the image acquisition unit 2 and set as the pixel sector 23 of the sensor 22 to be read out. In the MT routine 62 - as in <figref idref="f0007"><b>Fig. 7</b></figref> are shown as a flowchart - from each image 31, the gray value distribution G_{i, j} has a gradient image 81 in the x-direction with gray values ​​Dx_{i, j} and a gradient image 82 in the y-direction with Dy_{i, j} calculated or the gradient images 71 and 72 taken from the PR routine 61 directly. The difference values ​​are - as in the PR routine 61 - formed from a pixel environment (eg difference from predecessor predecessor and successor of the successor of the pixel to be determined).

0035In these gradient images 81 and 82, search is made for "valid" difference values ​​83 and 84, which exceed a specific gray value threshold SwA. The gray value threshold SwA is defined as in the PR routine 61. In each row or column, the first and the last valid difference values ​​83 and 84 are determined by these "valid difference values ​​83 and 84 and temporarily stored as StartX or StartY and EndX or EndY. Column reached, the start and end values ​​are used, respectively in a place histogram 85 and 86 at the pixel position of the start value StartX or StartY and in a place histogram 87 or 88 pixel position of the final value EndX or EndY the local value according to the Histogram field of <figref idref="f0007"><b>Fig. 7</b></figref> given formula by one higher.

0036If, for example, the pixel number 312 of the image 31 has been determined as the starting value for the line (from the x-gradient image 81), the value there is counted higher by one in the starting value histogram 87 (HistoStartX) at point 312. When all the histograms (start value histograms 85 and 86 and end value histograms 87 and 88, respectively) are determined, the line end for the x direction determination and the column end for the y direction determination are reached, the histograms 85 to 88 are evaluated , In order to find the valid start and end values ​​from the four histograms 85 to 88, there are several possibilities:<ol id="ol0001" compact="compact"><li>1. The mean values ​​of all values ​​occurring in the respective histogram;</li><li>Second The median values ​​(all values ​​ranked by frequency and then the mean) of all values occurring in the respective histogram ;</li><li>Third Center of gravity, eg centroid (quotient of the sum of all products of frequencies H_i with their positions i and the sum of all frequencies)<maths id="math0001"><math display="block"><mi mathvariant="normal">W</mi><mo mathvariant="normal">=</mo><mfrac><mstyle displaystyle="false"><mstyle displaystyle="true"><munder><mo mathvariant="normal">∑</mo><mi mathvariant="normal">i</mi></munder></mstyle><msub><mi mathvariant="normal">H</mi><mi mathvariant="normal">i</mi></msub><mo mathvariant="normal">⋅</mo><mi mathvariant="normal">i</mi></mstyle><mrow><mstyle displaystyle="true"><munder><mo mathvariant="normal">∑</mo><mi mathvariant="normal">i</mi></munder></mstyle><msub><mi mathvariant="normal">H</mi><mi mathvariant="normal">i</mi></msub></mrow></mfrac><mo mathvariant="normal">;</mo></math><img file="EP1542151B1_D0001.tif" /></maths></li><li>4th Minimum (at start values) or maximum (at end values) position where there is a valid value in the line or column of the histogram;</li><li>5th Minimum (at start values) or maximum (at end values) position in the histogram where a valid value exceeds a threshold;</li><li>6th Position of the most frequently occurring value in the histogram;</li><li>7th Position from which a minimum number of valid values ​​in the histogram has been reached;</li><li>8th. Position in the histogram from which a minimum number of valid values ​​exceeds a threshold;</li><li>9th Position at which, starting from the first occurrence of a valid value, the integral of the frequency distribution curve has exceeded a certain threshold as a percentage.</li></ol>(For the methods 4 to 9, where the positions of the valid start or end values ​​are counted, the starting values ​​are counted from the zero position upwards and from the maximum position downwards at the end values.)

0037The method selected by the above-mentioned method is expediently applied to only a part of the columns or lines of each image 31 so as to reduce the computational time to reduce the computational effort. The start and end values ​​determined therefrom form the boundaries of a preferably rectangular enclosing figure 32 which completely contains the fingerprint 11. But there are also ellipses or similar planar figures as enclosing figure 32 makes sense. The enclosing figure 32, as in rectangular shape in <figref idref="f0010">10a and 10b</figref> can also be based on the definition of a left start limit 35 (StartX) and a right end limit 36 ​​(EndX), as in <figref idref="f0011">Fig. 11a to 11c</figref> be limited indicated if the calculation of the enclosing figure 32 for reasons of almost complete utilization of the receiving surface 21 laterally to the rolling direction of the finger 1 not worthwhile.

0038To be sure that the fingerprint 11 is not cropped in any case, you would have to use the above method # 4. But this would have the disadvantage that small disturbances on the receiving surface 21 lead to a too large rectangle 33 and thus the read speed is limited.

0039In conjunction with an optoelectronic image recording unit 2 according to FIG <figref idref="f0005">Fig. 5</figref> the above method No. 9 has been found to be particularly advantageous. In order to exclude a circumcision of the fingerprint 11 as far as possible, a tolerance range is introduced, which is deducted from the found starting values ​​and added to the final values. The same tolerance range can later, when composing the images 31 (subtracted from the image edges), form a boundary area in which no calculations are made to avoid edge problems of clipping. The tolerance range should be selected so that the rolling finger 1 with the rectangle 33 retained also three images 31 later in the rectangle 33, without being circumcised. If, for example, the sensor 22 of a camera 26 is read out at 20 B / s and a normal rolling process takes 2 s, then 40 images are taken during this time. Suppose the StartX line in<figref idref="f0008">Fig. 8a and 8b</figref> represents the (left) start boundary 35 of the rectangle 33, moves 400 pixels in that time, so, assuming there is a uniform motion (no acceleration) between two successive images 31, it moves 10 pixels. If one wants that three pictures 31 continue, the finger 1 is in this rectangle 33, then the tolerance range must be at least 30 pixels! This tolerance range guarantees that the finger 1 will always be in this rectangle 33, regardless of the direction in which it is unrolled. A second aspect in the calculation of the tolerance range is the change in the readout speed with different size of the rectangle 33. If, in the case of the sensor 22 used (such as, for example, LM9638 from National Semiconductors, Inc., US, with so-called "windowing"), it is sensible to change the tolerance range by defining an active pixel region 23. Such a sensor 22, as mentioned in the above example, the complete readout of its active area about 20 For example, in the case of limiting the size of the active pixel area 23 to half of the active sensor area 40, for example, in which the start boundary 35 (left boundary = StartX line of the active pixel area 23) moves from image to image by 10 pixels Read out images per second and, when limited to one third of the active sensor surface, can realize a frame rate of 66 B / s. Thus, instead of 10 pixels / image, the line moves by 5 or 3.33 pixels / image and the tolerance range can be limited to 15 or 10 pixels. For motion tracking in the MT routine 62, a fixed rectangle size can also be used if the rectangle 33 is so large that always a fingerprint 11 fits in and is secured, that is also at least in the next image 31, predefined by the active pixel area 23 , the case is. The rectangle 33 can be positioned in two ways:<ul id="ul0004" list-style="dash" compact="compact"><li>Around the center of gravity (preferably centroid) of the fingerprint 11 into which the center of the rectangle 33 of fixed size is placed.</li><li>By calculating the start and end values ​​of the enclosing Figure 32 according to one of the nine methods described above, and positioning the fixed rectangle 33 by these limits by setting the center of the fixed rectangle 33 to the calculated average of start and end values.</li></ul>

0040If the boundaries of the rectangle 33 with a tolerance range are determined, the found values ​​are entered into the register of the sensor 22 of the image acquisition unit 2, which then only records and reproduces this range when the next image 31 is read out.

Roll Starter Detection 63 (RSR Routine 63)

0041After the MT routine 62 scores the current image 31 and determines the new size of FIG. 32, the RSR routine 63 compares the positions of FIG. 32 with those of the previous image images 32. In <figref idref="f0008"><b>Fig. 8a</b> and <b>Fig. 8b</b></figref> is each stylized a rolling process to the right and to the left. In the sectional view above, it can be seen that the receiving surface 21 is sensibly divided into three thirds in order to position the finger 1 correctly so that the fingerprint 11 is completely unrolled onto the receiving surface 21. For this purpose, the finger 1 is preferably placed in the second third and rolled into the first or third third. In this case, a rolling movement is already performed, whereby the center of gravity 34 and the boundaries 35 and 36 (StartX and EndX) of the rectangle 33 change. If the goal is to perform a rolling operation, for example, from left to right, the finger 1 is placed in the middle of the support surface 21 on the second third and rolled to the left to a starting point for the complete unwinding. The center of gravity, the (left) start boundary 35 and (right) end boundary 36 of the rectangle 33 are also moved to the left. If the starting point is reached in the first third of the support surface 21, the rolling process can be started by the finger 1 is rolled to the right. As the starting point of a rolling process, the changed direction of movement of the (right-hand) end boundary 36 is preferably detected and the occurrence of this event is communicated to all system components which require this information. The detection of the starting point can also be determined by the center of gravity displacement or as a combination of the reversal of motion from the end boundary 36 and the center of gravity 34. In the case of the latter, indicators serve to change both the direction of displacement of the center of gravity 34 and the direction of displacement of the end boundary 36 (EndX line) of the rectangle 33. This variant is more robust than the first two. In order to increase the robustness of the method, it is also possible to make the decision as to whether or not a rolling process is taking place only after a defined (larger) number of images 31, if their center of gravity 34 and / or end limits 36 (right Rectangle boundaries) have monotonically moved across this number of frames in the same direction as the initial direction. However, this means that a kind of ring buffer memory 41 is required to process the read-out images 31 (see FIG <figref idref="f0009"><b>Fig. 9</b></figref>), in which all images 31, which are stored from the actual starting point of the rolling process up to the image 31, from which the starting point is determined, are buffered. Thus, at least a number p of images 31 must fit into the ring buffer memory 41 as needed to determine the beginning of the unrolling process. If the starting point is established, first all the images 31 are read out of the ring buffer memory 41 and forwarded for processing. In the event that you want to unroll to the left, everything behaves the same, except that not the right, but the start boundary 35 (left border of the rectangle 33) is used for viewing.

0042If the rolling process has started according to one of the above criteria, then this is signaled to the external computing unit 4 (eg a PC) connected to the image recording unit 2 via the processing unit 5. The arithmetic unit 4 is also informed of how many images 31 were used for the determination of the unwinding process, so that they are buffered from a buffer memory in which the images 31 already used for the adjustment of the active pixel area 23 of the image recording unit 2 are put together of the overall picture 3 used. For this purpose, either in the processing unit 5, which is designed as a control hardware, or in the external arithmetic unit 4, a ring buffer memory 41 according to <figref idref="f0009"><b>Fig. 9</b></figref> in which the images 31 required for the RSR routine 63 are buffered. That is, the number of memory locations in the buffer 41 must be one greater than the number of frames 31 used to determine the roll-off. If a starting point of a rolling process has been detected, loop 2 in <figref idref="f0001"><b>Fig. 1</b></figref> finished and passed into the loop 3. This starts again with a support test (see PR routine 61) to end the recording of the rolling process in the unexpected start of the finger 1. If a valid fingerprint 11 is still present, it will switch to roll-over tracking.

Roll Tracking 64 (RT Routine 64)

0043The RT routine 64 is used to track the finger 1 during the defined unwinding. It not only determines the center of gravity 34 and the boundaries 35 and 36 of the fingerprint 11, such as the motion track 62, but also determines the speed of the roll-off operation by adjusting the speed of the center of gravity 34 and / or the speed of the respective boundaries 35 and 36 of the rectangle 33 separately be calculated. In addition, the acceleration of the finger center of gravity 34 and / or the respective rectangle boundaries 35 and 36 is determined to detect short-term changes in the unwinding speed. Thus, since the speeds and accelerations of the rolling operation are known and the rolling direction remains unchanged during the rolling operation, the tolerance between the enclosing figure 32 of the fingerprint 11 and the boundaries 35 and 36 of the enlarged rectangle 33 can be made smaller, in contrast to the MT algorithm 62 ,

0044Should the direction of unrolling change over several images 31 or be lifted off at an image 31 of the fingers 1, this is detected by the RER algorithm 65 or the PR algorithm 61 and interpreted as scroll closure. During a rolling process, the finger 1 is moved in a rolling motion over a contact surface 21, on which the fingerprint 11 formed there, no matter what physical principle, is taken at different times. Due to the shape and elasticity of the finger 1, the width and height of the fingerprint 11 varies. Show this <figref idref="f0010"><b>Fig. 10a</b> and <b>Fig. 10b</b></figref> two fingerprints 11 recorded at different times which are not directly successive, in which it can be seen that the width of the fingerprints 11 is different and thus also the enclosing figure 32 of the width ΔS (n) = E (n) -A (n) ΔS (m) = E (m) - A (m) changes. When the width of FIG. 32 reduces around the fingerprint 11, the frame rate of the image pickup unit 2 can be increased due to its narrower active pixel area 23. For MT routine 62, it has already been explained how to find valid start and end values ​​by subtracting a fixed tolerance allowance from the starting value and adding it to the final value. The size of the tolerance range is oriented in the MT routine 62 at the maximum occurring speed of the rolling process and is thus firmly defined. This has the consequence that the tolerance range must be chosen to be relatively large and thus image readout time is wasted. This is not problematic in the MT routine 62 since the captured fingerprint 11 is used only for visualization. In a rolling process, however, it is necessary to read out as many images 31 per second as possible, since any increase in the readout speed will improve the proper composition of the overall image 3. Therefore, with a priori knowledge about the started rolling process, the readout time can be significantly reduced. First, it is known by the Abrollstarterkennung 63, in which direction is rolled, ie the tolerance range must have only in one direction the maximum value. Secondly, by the steady direction of unwinding, the speed and possibly occurring accelerations can be calculated, thereby making a more accurate prediction of the boundaries of the figure 32 enclosing the fingerprint 11 and thus of the required rectangle 33. The procedure for determining the speed is in <figref idref="f0011"><b>Fig. 11a to 11c</b></figref> shown. When rolling in the x-direction three speeds are primarily interesting, the speed of the center of gravity in the x-direction (eg centroid of the imprint, center of gravity calculated as the difference of the boundaries 35 and 36 of the enclosing figure 32 in the x-direction) and the velocities of the two boundaries in X direction. For the velocities in y-directions, this applies analogously if rolling in the y-direction is implemented. The speeds are calculated in each case from the traveled distances and the cycle time (maximum possible integration time). Physically, the speed is defined as:<maths id="math0002" num="(1)"><math display="block"><mi mathvariant="normal">v</mi><mo mathvariant="normal">=</mo><mfrac><mi>ds</mi><mi>dt</mi></mfrac><mo mathvariant="normal">≈</mo><mfrac><mi mathvariant="normal">.DELTA.s</mi><mi mathvariant="normal">.delta.t</mi></mfrac></math><img file="EP1542151B1_D0002.tif" /></maths> Δs is the distance traveled eg the starting limit of the enclosing figure 32 calculated by:<maths id="math0003" num="(2)"><math display="block"><mi mathvariant="normal">.DELTA.s</mi><mfenced><mi mathvariant="normal">n</mi></mfenced><mo mathvariant="normal">=</mo><mi mathvariant="normal">A</mi><mfenced><mi mathvariant="normal">n</mi></mfenced><mo mathvariant="normal">-</mo><mi mathvariant="normal">A</mi><mo>⁢</mo><mfenced><mi mathvariant="normal">n</mi><mo mathvariant="normal">-</mo><mn mathvariant="normal">1</mn></mfenced></math><img file="EP1542151B1_D0003.tif" /></maths> The index n describes an arbitrary instant of the taking of an image 31 in the value range of 0 ≤ n ≤ N-1, with N as the total number of all captured images 31. Thus n-1 is the time of the predecessor and n + 1 of the successor. When using CCD and CMOS sensors, Δt represents the cycle time, ie the time it takes to read out the active pixel region 23 and thus a limited image 31.

0045Sensors without the ability to control the read-out speed as a function of the image size have a constant cycle time, which refers to the reading out of complete images and one can do without the division by Δt (saving of a division) to simplify the calculation. For sensors such as the LM9638 from National Semiconductors, Inc. (US), where the cycle time can be changed depending on the used size of the active pixel area 23, the cycle time is calculated as follows:<maths id="math0004" num="(3)"><math display="block"><mi mathvariant="normal">.delta.t</mi><mo mathvariant="normal">=</mo><mi mathvariant="normal">T</mi><mfenced><mi mathvariant="normal">n</mi></mfenced><mo mathvariant="normal">=</mo><msub><mi mathvariant="normal">T</mi><mi>integration</mi></msub><mo mathvariant="normal">=</mo><mfrac><mrow><msub><mi mathvariant="normal">N</mi><mi>rows</mi></msub><mo mathvariant="normal">⋅</mo><mfenced><msub><mi mathvariant="normal">N</mi><mi>Column</mi></msub><mo mathvariant="normal">+</mo><msub><mi mathvariant="normal">R</mi><mi>opcycle</mi></msub><mo mathvariant="normal">+</mo><msub><mi mathvariant="normal">R</mi><mi>itime</mi></msub></mfenced></mrow><msub><mi mathvariant="normal">f</mi><mi>pixelclock</mi></msub></mfrac></math><img file="EP1542151B1_D0004.tif" /></maths>

0046Here, f_Pixelclock the pixel clock between 12 to 27 MHz of the sensor 22. With this pixel clock, the individual pixels are read out. N_rows and N_Column are the number of rows and columns needed in the active pixel area 23 of the sensor 22. R_opcycle is a fixed integer value of 140. This is the time required to perform all intermediate steps before and after reading one row of the sensor 22. When reading adapted rectangles 33 as each active pixel area 23 ("Window of Interest"), R_itime a fixed integer of 34, otherwise R_itime Zero. This results, for example, in the case of a pixel frequency of 25 MHz and a complete readout of the sensor 22 with 1280 × 1024 pixels, a frame rate of 16.8 images per second. If the RT routine 64 is about to expire, the width of the image 31 is reduced to half (640 x 1024 pixels), for example, and a frame rate of 30 B / s (frames per second) is achieved. Another physical quantity for calculating the prediction of the next boundary 35 or 36 of the rectangle 33 is the acceleration. It takes into account any speed change in the calculation. The acceleration is calculated as follows:<maths id="math0005" num="(4)"><math display="block"><mi mathvariant="normal">a</mi><mo mathvariant="normal">=</mo><mfrac><mi>dv</mi><mi>dt</mi></mfrac><mo mathvariant="normal">≈</mo><mfrac><mi mathvariant="normal">.DELTA.v</mi><mi mathvariant="normal">.delta.t</mi></mfrac><mo mathvariant="normal">=</mo><mfrac><mrow><msub><mi mathvariant="normal">v</mi><mi mathvariant="normal">A</mi></msub><mo>⁢</mo><mfenced><mi mathvariant="normal">n</mi><mo mathvariant="normal">+</mo><mn mathvariant="normal">1</mn></mfenced><mo mathvariant="normal">-</mo><msub><mi mathvariant="normal">v</mi><mi mathvariant="normal">A</mi></msub><mfenced><mi mathvariant="normal">n</mi></mfenced></mrow><mrow><mi mathvariant="normal">T</mi><mo>⁢</mo><mfenced><mi mathvariant="normal">n</mi><mo mathvariant="normal">+</mo><mn mathvariant="normal">1</mn></mfenced></mrow></mfrac></math><img file="EP1542151B1_D0005.tif" /></maths>

0047The acceleration is determined from the two preceding speeds, whereby the first prediction is to be performed without an acceleration calculation. Empirical investigations on scrolling sequences have shown that the movement of boundaries 35 and 36 of surrounding figure 32 (ie in the unwinding direction: start and end of fingerprint 11 in the x-direction) at constant cycle time is similar to a tanh function and can be approximated by the following equation :<maths id="math0006" num="(5)"><math display="block"><msub><mi mathvariant="normal">s</mi><mi mathvariant="normal">n</mi></msub><mo mathvariant="normal">=</mo><mi mathvariant="italic">F</mi><mo>⁢</mo><mfenced open="[" close="]"><mn mathvariant="normal">1</mn><mo mathvariant="normal">+</mo><mi>tanh</mi><mfenced><mi mathvariant="italic">π</mi><mo mathvariant="normal">⋅</mo><mfrac><mrow><mi mathvariant="normal">n</mi><mo mathvariant="normal">-</mo><mfrac><mi mathvariant="normal">N</mi><mn mathvariant="normal">2</mn></mfrac></mrow><mi mathvariant="normal">N</mi></mfrac></mfenced></mfenced><mo mathvariant="normal">+</mo><mi mathvariant="italic">Z</mi><mspace width="1em" /><mfenced open="[" close="]"><mi>pixel</mi></mfenced></math><img file="EP1542151B1_D0006.tif" /></maths>

0048The parameters F and Z are used to match the function to a real rolling process and can be calculated by considering, for example, the starting value at the time t = 0 and for the starting value at the termination of the rolling process t = ∞ as follows:<maths id="math0007" num="(6)"><math display="block"><mi>F</mi><mo>=</mo><mfrac><mrow><msub><mi>begin</mi><mi>∞</mi></msub><mo>-</mo><msub><mi>begin</mi><mn>0</mn></msub></mrow><mrow><mn>2</mn><mo>⋅</mo><mi>tanh</mi><mfenced><mfrac><mi mathvariant="italic">π</mi><mn>2</mn></mfrac></mfenced></mrow></mfrac><mo>;</mo><mspace width="1em" /><mi mathvariant="italic">Z</mi><mo>=</mo><msub><mi>begin</mi><mn>0</mn></msub><mo>-</mo><mi>F</mi><mo>⋅</mo><mfenced open="[" close="]"><mn>1</mn><mo>+</mo><mi>tanh</mi><mfenced><mo>-</mo><mfrac><mi mathvariant="italic">π</mi><mn>2</mn></mfrac></mfenced></mfenced></math><img file="EP1542151B1_D0007.tif" /></maths>

0049For a certain rolling process, for example, have the values <i>F</i> = 132 and <i>Z</i> = 128 result. This gives you a curve, as in the<figref idref="f0012"><b>Fig. 12a</b></figref> is shown. In<figref idref="f0012"><b>Fig. 12a</b></figref> s (n) represents the "rolled-over pixels" of the finger 1 rolling on the support surface 21 of the image acquisition unit 2. It can be seen that between n = 18 to n = 24 the greatest change takes place. This change is also reflected in <figref idref="f0012"><b>Fig. 12b</b></figref> reflected, in which the speeds of the rolling process are shown. Due to the constant cycle time, the division was dispensed with over time. Thus, v (n) represents the pixel change at each time point (how many pixels are rolled over to the next image 31). At n = 22, the maximum speed is reached (9.2 pixels / image), ie in the middle of unrolling the speed is the highest. Therefore, it is particularly important in this area to read the image pickup unit 2 as fast as possible. In <figref idref="f0012"><b>Fig. 12c</b></figref> is entered the acceleration of the rolling process. Again, the division was omitted by the cycle time, since this was chosen constant. Thus, a (n) represents the deviation of the pixel change between two consecutive frames 31. Since the acceleration a (n) is the derivative of v (n), the value of a (n) is equal at the maximum of v (n = 22) "Zero". The function a (n) also has two extreme values ​​(0.5 or -0.5 pixels / image) at n = 13 and at n = 31. At these two points, the change in the speed between two read-out images 31 is greatest, ie the possibility of a miscalculation is greatest at these points, which means that additional tolerance allowances Δ for extending the rectangle 33 for setting a larger active pixel area 23 can not be dispensed with ,

0050A routine is presented below which realizes a more accurate prediction of the boundaries 35 and 36 of the rectangle 33. The basic requirement is that, after the already presented RT routine 64, the exact boundaries 35 and 36 of the fingerprint 11 or the figure 32 enclosing it have been determined, and two assumptions are made as initial conditions. Since at least two images 31 are needed to determine the velocity, an assumption must be made, since otherwise no prediction can be made for determining the position of the boundaries of the second image 31. Therefore, when a rolling operation by the RSR routine 63 has been judged to be started, it is assumed that the position of the start boundary 35 for the first image 31 is equal to the position of the start boundary 35 for the second image 31. This assumption can therefore be made because the finger 1 moves away from the starting position during rolling and thus ensures that the fingerprint 11 can not be cropped. For the position of the end boundary 36 in the second image 31, a fixed value is added to the position of the end boundary 36 from the first image 31, which ensures that the fingerprint 11 in the second image 31 is still before the end boundary 36. A second assumption is necessary for the acceleration prediction. For this you need at least three pictures 31 to make a statement. Therefore, for the third image 31 to be captured, it is preferably assumed that the speed increases by a factor of 2.

0051The following is an explanation of a procedure that realizes an accurate prediction of the next boundaries 35 and 36 of the rectangle 33. In<figref idref="f0013"><b>Fig. 13</b></figref> the results of the method are shown, where s (n) is the movement of the position of the start limit 35 and e (n) is the movement of the position of the end limit 36. The representations of the<figref idref="f0013"><b>Fig. 13a and 13b</b></figref> represent a rolling movement from an origin at pixel 0 to the maximum value of the respective image capturing unit 2. In <figref idref="f0013"><b>Fig. 13a</b></figref> is the movement of the position of the start boundary 35, where the solid line represents the predicted position 351 of the start boundary 35 of the rectangle 33 and the broken line represents the actual reached position 352 of the imaged fingerprint 11. When the start limit 35 is moved, the position 352 actually reached must not be below the solid line of the calculated position 351, since otherwise the fingerprint 11 will be judged.

0052In the movement of the position of the end limit 36 ​​according to <figref idref="f0013"><b>Fig. 13b</b></figref> conversely, here the calculated position 361 must be above the actually reached position 362. During a rolling movement in the opposite direction, ie from the maximum possible position towards the pixel 0 of the active pixel region 23 of the image recording unit 2, the statements about the movement of the position of the start and end limits 35 and 36 are the same. The position 351 or 361 of the start limit 35 or End limit 36 ​​can in principle be precalculated with the following equation:<maths id="math0008" num="(7)"><math display="block"><mi mathvariant="normal">x '</mi><mo>⁢</mo><mfenced><mi mathvariant="normal">n</mi><mo mathvariant="normal">+</mo><mn mathvariant="normal">1</mn></mfenced><mo mathvariant="normal">=</mo><mfrac><mrow><msub><mi mathvariant="normal">a</mi><mi mathvariant="normal">x</mi></msub><mfenced><mi mathvariant="normal">n</mi></mfenced></mrow><mn mathvariant="normal">2</mn></mfrac><mo mathvariant="normal">⋅</mo><msup><mi mathvariant="normal">.delta.t</mi><mn mathvariant="normal">2</mn></msup><mo mathvariant="normal">+</mo><msub><mi mathvariant="normal">v</mi><mi mathvariant="normal">x</mi></msub><mfenced><mi mathvariant="normal">n</mi></mfenced><mo mathvariant="normal">⋅</mo><mi mathvariant="normal">.delta.t</mi><mo mathvariant="normal">+</mo><mi mathvariant="normal">x</mi><mfenced><mi mathvariant="normal">n</mi></mfenced></math><img file="EP1542151B1_D0008.tif" /></maths>

0053Here, x '(n + 1) is either the position to be calculated of the start limit s' (n + 1) or the end limit e' (n + 1). The apostrophe means that it is a predictable quantity that is not used to calculate the value after the second. The size a_x(n) represents the acceleration and v_x(n) the speed of the starting value (a_s, v_s) or the final value (a_e, v_e) at the current time. The current position is represented by x (n). With the above equation (7), a position of the next value of the boundaries of the enclosing figure 32 is calculated on the assumption that speed and acceleration do not change. However, this assumption does not correspond to reality, but the change in the pixel position during an unrolling process describes approximately a tanh function, as already above<figref idref="f0012"><b>Fig. 12a</b></figref><b>-c</b> has been described. Therefore, to predict the exact position of the location of the boundaries 35 and 36 of the rectangle 33 surrounding the fingerprint 11, some precautions must be taken, which in turn are to define a tolerance that surrounds the boundaries of the fingerprint 11 and ensures that the fingerprint 11 is always located within an enlarged rectangle 33 framing the figure 32. The tolerance to be chosen here is smaller than in the MT routine 62, since the direction of unwinding is known. It proves to be useful to carry out a risk assessment, ie at the position of the starting value (limit 35) becomes a minimum value estimation and at the position of the end value (limit 36) becomes a maximum value estimate between the calculated position 351 or 361 and the actually reached position 352 or 362 realized. This adds extra security. One possible formula for the start value and the end value (limits 35 and 36) for scrolling from the smaller pixel position to the larger one is thus:<maths id="math0009" num="(8)"><math display="block"><mtable columnalign="left"><mtr><mtd><mi>start value</mi><mo mathvariant="normal">:</mo><mi mathvariant="normal">s'</mi><mo>⁢</mo><mfenced><mi mathvariant="normal">n</mi><mo mathvariant="normal">+</mo><mn mathvariant="normal">1</mn></mfenced><mo mathvariant="normal">=</mo><mfrac><mrow><mi mathvariant="normal">a</mi><mfenced><mi mathvariant="normal">n</mi></mfenced></mrow><mn mathvariant="normal">2</mn></mfrac><mo mathvariant="normal">⋅</mo><msup><mi mathvariant="normal">.delta.t</mi><mn mathvariant="normal">2</mn></msup><mo mathvariant="normal">+</mo><mi mathvariant="normal">v</mi><mfenced><mi mathvariant="normal">n</mi></mfenced><mo mathvariant="normal">⋅</mo><mi mathvariant="normal">.delta.t</mi><mo mathvariant="normal">+</mo><mi>MIN</mi><mfenced open="[" close="]"><mi mathvariant="normal">s</mi><mfenced><mi mathvariant="normal">n</mi></mfenced><mo mathvariant="normal">,</mo><mi mathvariant="normal">s'</mi><mfenced><mi mathvariant="normal">n</mi></mfenced></mfenced><mo mathvariant="normal">-</mo><mi>Tol</mi><mo>⁢</mo><mn mathvariant="normal">1</mn></mtd></mtr><mtr><mtd><mi>full scale</mi><mo mathvariant="normal">:</mo><mi mathvariant="normal">e '</mi><mo>⁢</mo><mfenced><mi mathvariant="normal">n</mi><mo mathvariant="normal">+</mo><mn mathvariant="normal">1</mn></mfenced><mo mathvariant="normal">=</mo><mfrac><mrow><mi mathvariant="normal">a</mi><mfenced><mi mathvariant="normal">n</mi></mfenced></mrow><mn mathvariant="normal">2</mn></mfrac><mo mathvariant="normal">⋅</mo><msup><mi mathvariant="normal">.delta.t</mi><mn mathvariant="normal">2</mn></msup><mo mathvariant="normal">+</mo><mi mathvariant="normal">v</mi><mfenced><mi mathvariant="normal">n</mi></mfenced><mo mathvariant="normal">⋅</mo><mi mathvariant="normal">.delta.t</mi><mo mathvariant="normal">+</mo><mi>MAX</mi><mfenced open="[" close="]"><mi mathvariant="normal">e</mi><mfenced><mi mathvariant="normal">n</mi></mfenced><mo mathvariant="normal">,</mo><mi mathvariant="normal">e '</mi><mfenced><mi mathvariant="normal">n</mi></mfenced></mfenced><mo mathvariant="normal">-</mo><mi>Tol</mi><mo>⁢</mo><mn mathvariant="normal">2</mn></mtd></mtr></mtable></math><img file="EP1542151B1_D0009.tif" /></maths>

0054Tol1 is the tolerance of the start value when scrolling from left to right. It may be smaller than the tolerance Tol2 of the final value, since the position 352 of the start limit 35 at time n is definitely smaller than the position 352 of the start limit 35 at time n + 1. Thus, as the predicted position 351 of the start limit 35 at time n + 1, theoretically, the actual position 352 of the start limit 35 may be taken at time n and Tol1 may take the value 0. However, since the detection of the roll-off occurs within the roll-off sequence and the direction change is a feature for the roll-off end, a certain size must be used for the tolerance Tol1.

0055In reverse roll direction, the formulas (8) change to:<maths id="math0010" num="(9)"><math display="block"><mtable columnalign="left"><mtr><mtd><mi>start value</mi><mo mathvariant="normal">:</mo><mi mathvariant="normal">s'</mi><mo>⁢</mo><mfenced><mi mathvariant="normal">n</mi><mo>-</mo><mn mathvariant="normal">1</mn></mfenced><mo mathvariant="normal">=</mo><mfrac><mrow><mo>-</mo><mi mathvariant="normal">a</mi><mfenced><mi mathvariant="normal">n</mi></mfenced></mrow><mn mathvariant="normal">2</mn></mfrac><mo mathvariant="normal">⋅</mo><msup><mi mathvariant="normal">.delta.t</mi><mn mathvariant="normal">2</mn></msup><mo>-</mo><mi mathvariant="normal">v</mi><mfenced><mi mathvariant="normal">n</mi></mfenced><mo mathvariant="normal">⋅</mo><mi mathvariant="normal">.delta.t</mi><mo mathvariant="normal">+</mo><mi>MIN</mi><mfenced open="[" close="]"><mi mathvariant="normal">s</mi><mfenced><mi mathvariant="normal">n</mi></mfenced><mo mathvariant="normal">,</mo><mi mathvariant="normal">s'</mi><mfenced><mi mathvariant="normal">n</mi></mfenced></mfenced><mo mathvariant="normal">-</mo><mi>Tol</mi><mo>⁢</mo><mn mathvariant="normal">2</mn></mtd></mtr><mtr><mtd><mi>full scale</mi><mo mathvariant="normal">:</mo><mi mathvariant="normal">e '</mi><mo>⁢</mo><mfenced><mi mathvariant="normal">n</mi><mo>-</mo><mn mathvariant="normal">1</mn></mfenced><mo mathvariant="normal">=</mo><mfrac><mrow><mo>-</mo><mi mathvariant="normal">a</mi><mfenced><mi mathvariant="normal">n</mi></mfenced></mrow><mn mathvariant="normal">2</mn></mfrac><mo mathvariant="normal">⋅</mo><msup><mi mathvariant="normal">.delta.t</mi><mn mathvariant="normal">2</mn></msup><mo>-</mo><mi mathvariant="normal">v</mi><mfenced><mi mathvariant="normal">n</mi></mfenced><mo mathvariant="normal">⋅</mo><mi mathvariant="normal">.delta.t</mi><mo mathvariant="normal">+</mo><mi>MAX</mi><mfenced open="[" close="]"><mi mathvariant="normal">e</mi><mfenced><mi mathvariant="normal">n</mi></mfenced><mo mathvariant="normal">,</mo><mi mathvariant="normal">e '</mi><mfenced><mi mathvariant="normal">n</mi></mfenced></mfenced><mo mathvariant="normal">-</mo><mi>Tol</mi><mo>⁢</mo><mn mathvariant="normal">1</mn></mtd></mtr></mtable></math><img file="EP1542151B1_D0010.tif" /></maths>

0056In this case, Tol1 may again be smaller than Tol2, since the rolling process is carried out from right to left and the final boundary 36 at time n is definitely greater than the position 362 of the end boundary 36 at time n + 1.

0057To predict the limits 35 and 36 of the rectangle 33, there are still many other possibilities of calculation up to the use of a tanh function.

0058It is also possible to work with a fixed size of the rectangle 33 and to use the here presented algorithm for finding the boundaries 35 and 36 of the fingerprint 11 enclosing figure 32 and the position of the rectangle 33 with the fixed size to the changed position of to adapt Figure 32. That is, based on the imaged in the current image 31 fingerprint 11 is determined which maximum rectangle width is necessary to ensure the usual object movement that the fingerprint 11 is not trimmed by the rectangle 33. Thereafter, as described above, it is derived where the next positions of the boundaries 35 and 36 of the enclosing figure 32 are located. The position of the subsequent rectangle 33 results from these values, and the fixed rectangle size must be placed around the calculated enclosing figure 32 such that the distances of the boundaries of the rectangle 33 from the minimum start value and the maximum end value of the enclosing figure 32 are equal on all sides , The resulting position of the solid rectangle 33 is programmed in the image acquisition unit 2 as an active pixel region 23 for the subsequent image 31. The positioning of the rectangle 33, which is always the same size, can also - as described above - be carried out with the aid of a center of gravity algorithm which, for example, searches for the centroid of the fingerprint 11 and places the solid rectangle 33 around it. This has the consequence that a circumcision of the fingerprint 11 can not be excluded.

0059The excising of an enlarged rectangle 33 containing the fingerprint 11 can also take place outside the image acquisition unit 2, if the data rate when reading the complete sensor 22 is sufficient, but the limiting element for the required data rate is the transmission channel. In this case, the method according to the invention can be applied in the same way, except that the frame as picture 31 is present as the basis for calculation and not already a selected active pixel area 23. It is then no longer necessary to make a prediction where the next start or stop is Final limit 35 or 36 is because you have the image 31 as a frame of the image pickup unit 2 available, so you can do without the calculation of speeds and accelerations. ie by a high frame rate image pickup unit 2, the frame in the processing unit 5 (in accordance with FIG <figref idref="f0005"><b>Fig. 5</b></figref><b>)</b> read in, which determines the boundaries 35 and 36 of the fingerprint 11 (enclosing figure 32) and an enlarged rectangle 33 and passes only the part of the frame corresponding to the content of the currently determined enlarged rectangle 33, via the interface 55 to the external arithmetic unit 4 ,

0060If during the rolling process, despite all safety precautions (tolerance additions), the prediction truncates the fingerprint 11, then, for example, when rolling from left to right in the two possible cases, the procedure is as follows.

1. Cropping of the finger through the start boundary 35 of the rectangle 33

0061In this error, which was caused by the fact that the finger 1 has moved slower than predicted, the start limit 35 is maintained as an input value without recalculation of speed or acceleration, but increases the tolerance.

Second Trimming the finger by the end boundary 36 of the rectangle 33

0062Since the cause of the error is that the finger 1 has moved faster than predicted, when calculating the next position of the enclosing rectangle 33, assuming the same speed, the double tolerance allowance is used.

0063This ensures that the fingerprint 11 in the next image 31 is within the start and end limits 35 and 36 of the rectangle 33.

0064RT routine 64 is as in FIG <figref idref="f0001"><b>Fig. 1</b></figref> seen, surrounded by the support detection 61 and a Abrollschlusserkennung 65 (RER routine). PR routine 61 and RER routine 65 check whether the finger 1 has been lifted during the rolling process or the finger 1 has been moved over a series of images 31 in the opposite direction to the detected rolling direction. If the finger 1 was lifted during the unwinding, this is detected by the PR routine 61 and interpreted as a rolling shut-off. The other way of terminating the RT routine 64 is effected by a roll-off detection 65, which will be explained below.

Rolling termination detection 65 (RER routine)

0065With an RER routine 65, it is determined whether or not a coasting operation has been completed. In this case, the positions determined by the RT routine 64 of the start and end limits 35 and 36 of the rectangle 33 are used. In a rolling process, the user should, if he wants to end the rolling process, be prompted to roll the finger 1 against the original rolling direction. This changes the limits determined by the RT algorithm 35 and 36 (eg over more than three images 31) against the original rolling direction. This can then be interpreted unambiguously as rolling closure. In order to make the RER routine 65 robust, it is expedient to use 5 images to make sure that the unwinding process has ended. Also, the roll-off is evaluated as the lifting of the finger 1, which is detected by the PR routine 61 described above. If the roll-off has been detected, the registers in which the adapted rectangle 33 was stored as the active pixel region 23 of the sensor 22 are deleted in the image acquisition unit 2, so that again the entire sensor surface is switched active.

0066The processes described above can also be used when a large active area of ​​the image recording unit 2 has to be used to take a complete footprint or handprint and only a small portion of the same sensor area is required to hold a finger 1 to be unrolled. With a sensor 22, which has the special possibility of "windowing", then both the large-scale recording with the necessary resolution and by the inventive control of the active pixel area 23 of the sensor 22, the significantly higher frame rate for recording a fast image sequence of rolling finger 1 can be realized.

0067To the in <figref idref="f0001"><b>Fig. 1</b></figref> It is necessary to have a fast control hardware as the processing unit 5 so as not to have all the calculation routines for controlling the active pixel area 23, the computing unit 4 provided for image processing in the sense of composing the successively read-out images 31 , to charge. In<figref idref="f0005"><b>Fig. 5</b></figref> a structure is shown, which subsequently to the image pickup unit 2, a logic unit 51, a processor 54, a program memory 52, a data memory 53 and an interface 55, wherein the logic unit 51 and the processor 54 may be combined as a data processing or processing unit. The program and data memory 67 and 68 can be combined into a memory. As processor 54, the CPU of the external computing unit 4 could also be used. Via the control bus 56, the logic unit 51, the processor 54 and the interface 55 exchange relevant information for unrolling. Furthermore, the logic unit 51 and interface 55 are connected to each other via the image data bus 57 for transmitting the image data. The image pickup unit 2 whose sensor is based on CMOS or CCD technology preferably includes a drive circuit and an analog-to-digital converter. Thus, there is a fingerprint 11 as a two-dimensional digital image 31 at the output of the image acquisition unit 2. This is written to the data memory 53 via the logic unit 51. In the processing unit 5, the above-described algorithms and processing routines are executed. The logic unit 51 can be implemented by a FPGA (Field Programmable Gate Array), PLD (Programmable Logic Device), ASIC (Application Specific Integrated Circuit). As the processor 54, a microcontroller (MCU), a microprocessor (MPU) or a digital signal processor (DSP) may be used. The processor 54 can also be completely dispensed with if the logic unit 51 is able to perform all functions, including communication with the interface 55 itself. Conversely, the logic unit 51 can be dispensed with if the processor 54 is able to realize all the tasks of the logic unit 51. For this purpose, the image data bus 57 would then have to be led by the processor 54 to the interface 55 and a direct coupling of the processor 54 to the data memory 53 would exist. The interface 55 represents the interface to the outside and can be arbitrary (eg USB2.0, IEEE1394, Ethernet). In the case of a powerful interface 55, that is to say an interface with a high data transmission rate, the processing routines 61 to 65 can also be partially or completely also outside, for example with the in<figref idref="f0005">Fig. 5</figref> represented external computing unit 4, executed and the setting of the active pixel area 23 in the register of the image pickup unit 2 are controlled via the interface 55 (and possibly the logic unit 51).

LIST OF REFERENCE NUMBERS

0068<dl id="dl0002" compact="compact"><dt>1</dt><dd>finger</dd><dt>11</dt><dd>fingerprint</dd><dt>12, 12a-e</dt><dd>filmstrips</dd><dt>13, 14, 16</dt><dd>papillary</dd><dt>131-133</dt><dd>(shown) segments</dd><dt>141-143</dt><dd>segments</dd><dt>15a-c</dt><dd>solid ellipse</dd><dt>15d, 15e</dt><dd>dashed ellipse</dd><dt>16a-e</dt><dd>line dividers</dd><dt>161</dt><dd>line branching</dd><dt>162-164</dt><dd>ridge ending</dd><dt>17</dt><dd>section</dd></dl><dl id="dl0003" compact="compact"><dt>2</dt><dd>Imaging unit</dd><dt>21</dt><dd>receiving surface</dd><dt>22</dt><dd>sensor</dd><dt>23</dt><dd>active pixel area</dd><dt>24</dt><dd>lighting unit</dd><dt>25</dt><dd>prism</dd><dt>26</dt><dd>camera</dd></dl><dl id="dl0004" compact="compact"><dt>3</dt><dd>overall picture</dd><dt>31</dt><dd>picture</dd><dt>32</dt><dd>enclosing figure</dd><dt>33</dt><dd>enlarged rectangle</dd><dt>34</dt><dd>main emphasis</dd><dt>35</dt><dd>start limit</dd><dt>36</dt><dd>end limit</dd></dl><dl id="dl0005" compact="compact"><dt>4</dt><dd>(external) arithmetic unit</dd><dt>41</dt><dd>Ring buffer memory</dd><dt>5</dt><dd>processing unit</dd><dt>51</dt><dd>logic unit</dd><dt>52</dt><dd>program memory</dd><dt>53</dt><dd>data storage</dd><dt>54</dt><dd>processor</dd><dt>55</dt><dd>interface</dd><dt>56</dt><dd>control bus</dd><dt>57</dt><dd>image data bus</dd></dl><dl id="dl0006" compact="compact"><dt>61</dt><dd>Circulation detection (PR routine),</dd><dt>62</dt><dd>Motion tracking (MT routine),</dd><dt>63</dt><dd>Roll-off detection (RSR routine),</dd><dt>64</dt><dd>Roll Tracking (RT Routine)</dd><dt>65</dt><dd>Unwinding detection (RER routine)</dd></dl><dl id="dl0007" compact="compact"><dt>71, 72</dt><dd>gradient</dd><dt>73, 74</dt><dd>difference value</dd><dt>75, 76</dt><dd>differences counter</dd><dt>77</dt><dd>line counter</dd><dt>78</dt><dd>column counter</dd></dl><dl id="dl0008" compact="compact"><dt>81, 82</dt><dd>gradient</dd><dt>83, 84</dt><dd>difference value</dd><dt>85, 86, 87, 88</dt><dd>local histogram</dd></dl>