Systems and methods for assessing the authenticity of dynamic handwritten signature

The present invention relates to an interactive computer system and method for acquiring and processing bio-kinetic information of a signature to assess dynamic handwritten signature authentication. The present invention has applicability in the field of behavioral biometrics in situations where the user is interested in confirming his or her will by handwritten signature.

In addition to the computer-based methods and techniques of administrative physiology biometrics, other methods and techniques belonging to the field of behavioral biometrics are used as supplementary links in the procedure for confirming a person's open identity. What you can do is generally accepted. Acquisition and recognition of handwritten signature components represent a type of behavioral biometrics.

In Patent Document 1 and Patent Document 2 described herein by reference, solutions and processing methods for acquiring and processing dynamic information (acceleration) related to the signature process have already been developed for verifying handwritten signatures. Information is captured, processed and compared by computer-based systems. The inertial accelerometer integrated into the electronic pen, or MEMS, and the specific topology in which the sensors are placed, not only captures spatial dynamics information, but also captures fine contact vibrations generated at the paper level. make it easier. The processing method for the acquired acceleration signal is --Algorithmic, representing the time variation of the threshold distance to the paper combined with the variation of the start and end points of the signature, the frequency and amplitude parameters of the contact microvibrations generated by the subject / pen / paper interaction. The method realized through data processing and --Invariant sequences related to the initial acceleration signal and invariant sequences related to the derived components of the initial acceleration signal, --The distance between two signatures by an algorithmic comparison of a set of invariant sequences using two different methods, --- The final result is determined via a determination method in which the sample of the targeted subject registered in the signature database and the sample of another subject contribute through the result of comparison with the input signature. To judge.

The systems and methods of Patent Documents 1 and 2 deal with special kinetic phenomena combined with contact micro-vibrations by processing the acceleration captured by an accelerometer or MEMS. However, this process has the disadvantage that it does not include both the capture and processing of graphic information, and this capture and processing of the system when combined with the dynamic information by sensorial fusion. It should improve accuracy and quality of man-machine interaction. The present invention aims to eliminate the above-mentioned drawbacks by incorporating an optical navigation sensor together with an inertial accelerometer or MEMS in consideration of the conditions of the sensor fusion concept. This optimization is graphically and kinetic, on the one hand, in a new algorithmic multimodal processing method for captured signals, and on the other hand, for the achievement of visual feedback for the user with the benefits described in detail later. Used for.

Patent Document 3 (Microsoft Corporation) presents a method for interpreting handwriting thickness using an electronic method based on ballistic motion. To capture this information, the patent uses a pen with one or more (embedded) accelerometers. For the correct interpretation of this information, "pulse width variation or pen tilt angle provided by the accelerometer" is used. This method implies the use of a pen that includes an accelerometer. However, this method does not include any elements for the direct description of the graphic trajectory. This method proposes the use of a graphic tablet type complex and expensive external device to correlate information about handwriting thickness with the pen trajectory. The specific issue of handwritten signatures is not discussed in this patent.

In contrast to the patents mentioned above, the present invention provides a pen-type signature capture module that integrates not only an accelerometer but also an optical navigation sensor with the purpose of capturing graphic trajectories through sensor fusion. Including. The topological concept of sensor integration within the pen is essentially defined to ensure spatial synchronization of the accelerometer axis with the optical navigation sensor axis. The functional integration of both categories of sensors in the pen ensures synchronization in the time of acquisition of the two types of data: acceleration and self-referential motion. Spatio-temporal synchronization of information acquisition is the dominant principle of sensor fusion, and the application of this sensor fusion in the present invention provides better accuracy in signature authentication. The function of thickness handwriting interpretation is defined and is based not only on fluctuations in the contact microvibration frequency, but also on its amplitude fluctuations as an effect of the dynamic pressure generated between the pen and the paper in handwriting.

Patent Document 4 (DynaSig Corporation) is a signature authentication and acquisition system including an accelerometer and a pressure sensor incorporated in a writing device, in which the authentication method is achieved by data encryption and comparison of code results, and thus encryption. A system that prevents the storage of samples that are not encrypted will be described. This patent does not deal with any information about the graphic nature of writing devices or authentication methods.

Patent Document 5 (Microsoft Corporation) is a pen type in which the reference pattern analyzed by the integrated image sensor is a source that generates absolute or relative coordinates depending on the implementation options and the context of use. The system of graphic data input and the dual method will be described. The system produces graphic or location data, but does not include components and methods for handwritten signature authentication. Since the accelerometer is not integrated into the capture module (pen in this case), it is less applicable in signature authentication.

US Patent Document 6 (Data Research Inc.) describes a system and method of acquiring and recognizing hand movements when writing on a surface, that is, a "digital writing system". Dynamics Spatial data capture is performed using a 3-axis gyroscope, a 3-axis accelerometer integrated with the pen, and a proximity sensor associated with the writing support. This patent does not describe an capture sensor for graphic information or an algorithmic method of signature authentication. 3D trajectory data is indirectly estimated by dynamic linear filtering based on the Kalman estimator (filter).

Procedures that require handwritten signature verification are outlined and applied using concepts, principles, and conventions that belong to the social sciences, forensic sciences, behavioral psychology, and human neurophysiology. Electronics and information technology contribute to improving the security of checking procedures by acquiring, processing, and recognizing the form of signals and signatures.

<p num="0011"><patcit num="1"><text>Russian Patent No. 141297</text></patcit><patcit num="2"><text>European Patent No. 1846868</text></patcit><patcit num="3"><text>U.S. Pat. No. 7,176906 B2</text></patcit><patcit num="4"><text>U.S. Pat. No. 7433499 B2</text></patcit><patcit num="5"><text>U.S. Pat. No. 7483018</text></patcit><patcit num="6"><text>U.S. Pat. No. 7,508,384 B2</text></patcit></p>

<p num="0012"><nplcit num="1"><text>http://www.biometrics.org</text></nplcit><nplcit num="2"><text>J. Piaget, "Theory of Cognitive Development" (1952)</text></nplcit></p>

<p num="0013"> User ease of use, adaptation and acceptance of biometric procedures are essential elements and are defined in Non-Patent Document 1 which is used as the basis for the classification of biometric techniques. Most people should prefer handwritten signature biometrics to other biometric methods. Signatures are considered, first of all, a means of personalization and individualization of self-defense, due to their social usefulness and personal perception. Justification lies in the individually differentiated nature of the handwritten motive phenomenon, the acquisition reflex linked to self-interest and free will. The present invention accepts and adapts by addressing the acquisition and processing of signals specific to the psychotricity of signatures by maintaining paper or any other material with a similar texture as a support. The degree of is high.</p><p num="0014"> During the signature, the user's forearm stands on the elbow on a horizontal desk, so that only the palm and fingers perform the spatial dynamic gestures associated with the signature creation. In this context, a self-referential optical navigation sensor (ONS) placed inside the pen, quasi-parallel to the writing plan, and near the tip of the pen is printed on paper / support as a navigation reference. It has an element that spreads semi-uniformly in the pattern. The lens (L) captures a dynamic image of the elements of the pattern by projecting it onto the photosensitive area of the optical navigation sensor. The field of view of the lens includes a quasi-constant area containing sufficient pattern elements to allow the ONS sensor to determine relative motion from a previous moment (reported to its own axes x, y). Based on the difference between the contiguous images and the contiguous images sampled internally by ONS, the sensors are MEMS A sensor and MEMS. Milliseconds controlled and synchronized by a microcontroller located within the pen using acceleration signal capture and analog-to-digital conversion (see Patent Documents 1 and 2) sensed by the B-sensor. Relative motion of the scale over a period of time d_x, D_yGenerate a pair of. As a result, the computerized graphic representation of the ONS sensor trajectory resembles the graphic created by the pen tip. The graphic difference between the representation of the ONS trajectory and the graphic created by the pen's replacement core represents the morphism of the same motive phenomenon caused by the dynamic tilt of the pen. This morphism distinguishes signatures biometrically individually. In addition, due to the synchronization between the graphics on the paper and the capture of the trajectory through the ONS sensor, there are two representations: the trace left on the paper by the pen's core and the electronic trajectory captured by the system. ) Can be stored electronically, including the line segment of the trajectory where the tip of the pen does not touch the writing support. In situations where the execution of the signature implies the momentary lifting of the pen from the paper and the amplitude of these liftings being small enough, a few millimeters or less, the optical navigation sensor prints on the writing paper. Continue to focus on the pattern elements that have been made and continue to capture the trajectory. The system continues to acquire the information created within these sequences (when the pen does not touch the paper). This contributes to the individual distinction of signatures. This information is part of the signature and is displayed by the processing and graphic display module, firstly to achieve the user's visual and kinetic apperception, and secondly graphically. -Processed by handwriting screening and processing as line segments that do not require display to achieve the view.</p><p num="0015"> Handwriting a signature is a psychomotor act. Psychomotority is defined as the result of the integration of motical and psychological functions under the influence of nervous system maturation and relates to reporting between subjects and their bodies. Therefore, psychomotority is not only an ability, but also a complex function of individual behavior, including various psychomotor functions that provide both data receipt and correct execution of reaction actions, and participation of psychomotor processes. It is also an adjustment.</p><p num="0016"> As Non-Patent Document 2 argues, motric and psychic are single, rather than two separate categories, one being the subject of pure thinking and the other being the subject of physical and physiological mechanisms. It is a bipolar representation of efficient and flexible adaptation to processes, or external conditions. The act of signing between the external and subjective conditions of the motric act is not only the mechanism of execution, but all stages of the action, all details, semi-acquired reflexes between the individual and the environment, respectively. It is also a loop circuit, which is a direct representation of the relationship between the individual's will, psiychomotrically manifested by, and the support of the projection of the will, in this case the writing paper on which the signature is performed. ..</p><p num="0017"> By making the signature using a writing device, which is a device with an actual running subject, the inventive principle is the principle of capturing motrick variables by a self-reference sensor built into the pen (writing device). Heads towards (acceleration and relative motion), thereby emulating the non-entry integration of the sensor "inside the subject". The self-referential concept replaces the necessary practice of a close connection between the observer (sensor) and the psychometric act, and thus the purpose is to observe and assess the psychomotor act or subject's signature personality. Data is taken from the subject's sensorimotor perspective in the context of. Written paper has relatively constant standardized qualitative properties, which is why the expression of personality through the motivative act of signature is affected by potential differences in paper quality with low extend.</p><p num="0018"> The set of trajectory attributes captured by the ONS sensor, along with the spatial gestures captured by the sequencing and accelerometer MEMS, is an individual signature-distinguishing element created by the subject's free will.</p><p num="0019"> From the user's point of view, the role of integrating visual feedback into the system is dual. --This integration favors the cognitive mechanism of gesture apperception and gives the user the possibility of acquiring complex gestures that are not written on paper but belong to the dynamics of signatures through practice and visualization. The user can deliberately make these acquired gestures for the additional individual distinction of the signature. --This integration facilitates the user's adaptation to the specificity of the system due to the combination between the psychomotor mechanism and the display.</p><p num="0020"> The consistency and complexity of signature gestures naturally acquired over time or through practice depends on the motivative ability and the originality of the user who composes the signature with the gesture elements as individually as possible. Only restricted.</p><p num="0021"> The present invention is a data acquisition and processing system associated with dynamic handwritten signatures performed on a writing paper support with a printed semi-uniform pattern or any other material with a similar texture. Given two groups of inertial accelerator MEMS to capture dynamics data and data on microvibration in contact with the support, integrated into a personal computer to acquire and process signature-related signals. Personal computers, including electronic pens, can be networked together with other personal computers, each to which the electronic pen is connected as a peripheral, and the electronic pens are the momentary necessary to reconstruct the trajectory of the pen. Data pair as exercise (d_x, D_yA self-referential optical navigation sensor for capturing a series of) data pairs (d)._x, D_yThe series of) is the kinetic data (a) captured by the set of inertial sensors MEMS._x, A_y, B_x, B_y), And self-reference optical navigation to form personal computer input data for processing through sensor fusion and creating conditions for extracting information from sensor and psychomotor representations of the user's perspective. -Sensor, infrared LED to illuminate the pattern, and optical navigation positioned to allow projection of the pattern area image to be obtained on the sensor's sensing area. Pattern focus and current image on the sensor's sensing area. Consists of a sensor for projection, graphic type d_x, D_yAnd a_xAcceleration, a_yAcceleration, b_xAcceleration, b_yThe spatiotemporal synchronization conditions required for sensor fusion acquisition of the two data categories of acceleration are the axial directions of the three origins of the MEMS sensor, the optical navigation sensor, and the coordinate axes of the pen tip. Topological quasi-alignment, three sensor components with the writing support plane, ie two MEMS sensors and optical navigation sensor sensing / axes plane parallel quasi-alignment topology, and over high frequency graphics and dynamics Sampling of data captured by sensors with constant periodicity within regions 1-8 ms to ensure synchronization in time of detail acquisition, acquisition is visualization, writing thickness reconstruction, and processing. And to acquire data from the psychomotor and sensor representations of the user's perspective, achieved by sampling, controlled by an acquisition microcontroller that sends information in real time to a personal computer hosting an algorithmic method of comparison. In addition, a set of MEMS sensors and optical navigation sensors integrated within a writing device handled by humans are self-referenced coordinate systems, with six related signals a in their own x and y axes of sensitivity, respectively._x, A_y, B_x, B_y, D_x, D_yA system that incorporates, and a method of assessing dynamic handwritten signature authentication, on the one hand handwriting reconstruction and writing thickness visualization, and on the other hand, signature-related to assess signature authenticity a._xSignal, a_ySignal, b_xSignal, b_ySignal, d_xSignal, d_yReferences are made to methods consisting of various transformations and application of comparisons of signals.</p><p num="0022"> Later, as an example, the present invention will be described with reference to FIGS. 1 to 11 showing the following.</p>

<figref num="1">It is a functional block diagram of a system.</figref><figref num="2">It is a figure which shows the physical module of a system.</figref><figref num="3">It is a figure which shows a signature capture mode diagram and a pen.</figref><figref num="4">It is a function diagram of the signature import mode and the pen.</figref><figref num="5">It is a figure which shows ONS, MEMS A, MEMS B topology, and a replacement core.</figref><figref num="6">It is a figure which shows the topology detail from a pen module.</figref><figref num="7">It is a figure which shows the reconstruction of the whole locus described by a pen.</figref><figref num="8">It is a figure which shows the reconstruction of the graphic locus at the time of contact with a paper.</figref><figref num="9">It is a figure which shows the general framework of signature conversion and comparison.</figref><figref num="10">It is a figure which shows the summary of the signature analysis algorithm.</figref><figref num="11">It is a figure which shows the determination of the difference of the slope between two continuous line segments of a plane curve represented by a point.</figref>

The system described in the present invention physically includes an electronic pen module 1 that integrates a sensor assembly for data acquisition, and a personal computer 2 that processes data acquired by the pen via an algorithmic method. Consists of.

In terms of functionality, the system is shown in a block diagram in Figure 1. The system is driven by a hand pen assembly related to handwritten signatures through a sensor fusion of two categories of signals: acceleration signals captured by MEMS A, MEMS B and motion signals captured by the ONS optical navigation sensor. Incorporate the created motric phenomenon. The specific topology of the distribution of sensors within the pen 1 and the sampling principle of the acquired signal guarantee the conditions of spatiotemporal synchronization between signals of the same phenomenon. This condition is mandatory to achieve the concept of sensor fusion for multimodal algorithmic processing.

The present invention presents a self-referenced capture of the pen trajectory performed by an optical navigation sensor integrated within the pen type module, firstly a graphical display of the pen trajectory, and secondly the pen. Deales with the method of handwriting display of pen trajectories with respect to these lines when is in dynamic contact with paper during signature execution. Figure 7 shows the entire reconstruction of the trajectory described by the pen, and Figure 8 shows the various thicknesses associated with the moment of dynamic contact of the pen's core with the writing paper / support of the same signature. Represents a handwriting reconstruction with a pen. The electronic representation on the monitor 2 of the personal computer generally corresponds to the handwriting of the signature made by the pen's replacement core on the writing support.

According to the present invention, the method of assessing the authenticity of a signature refers to graphic and kinetic processing of the signal associated with the signature and multimodal comparison and will be described in detail in practical examples.

The captured and displayed signal is the first phase of visualization on monitor 2 of a personal computer, the difference between a fragment of an "air" signature, eg, Figures 7 and 8 representing the same signature. Also includes. These pieces, along with the pieces provided by the movement on paper, contribute to the authenticity of the signature.

The psychomotor manifestations involved in signing include invariant series that are automatically evaluated via acquisition / processing / comparison methods and elements of personality that are determined by the system as vector data. These methods can be added to Patent Document 1 and Patent Document 2 whose description is contained herein by reference. Further, the editions described in the connection, processing, and judgment hardware structures (computer, server) or earlier patents (Patent Documents 1 and 2) are the pen capture type modules described in the present invention. Includes necessary and sufficient elements to provide a driving platform for algorithmic elements that operate in unison with 1).

According to the present invention, the new concept of signature capture embodied by the electronic pen module and the method of processing and comparison will be described in detail later as an integral concept.

In order to capture and reconstruct the graphic trajectory of the pen, an optical navigation sensor ONS is installed in the pen for biometric verification and signature visualization instead of the C sensor for threshold distance evaluation described in Patent Document 1 and Patent Document 2. It was integrated. In the implementation example, an ADNS optical navigation sensor (AGILENT manufacturer) was used. The threshold distance detection function provides sufficient captured images for the sensor to analyze the pattern when the ONS sensor is facing paper with the pattern p at a distance within the d spacing determined by the optical properties of the lens L. It is taken over by the ONS sensor because of its property of producing output data only when it is focused on. In an embodiment, this d spacing is 5-20 mm for a lens made of polycarbonate (plexiglass) with a refraction coefficient of n = 1,5. The ONS sensor d of the image projected by the lens L over the sensitive area within a determined time period (sampling period)._x, D_yIt is a self-referential sensor that has the property of generating relative motion as an output. d_x, D_yThe movement corresponds to the movement of the projection of the image of the pattern printed on paper. The image captured at the beginning of each sampling cycle has the value d_x, D_yRepresents a momentary reference to a pair of.

d_x, D_yThe motion is estimated relative to the ONS sensor's self (X, Y) coordinate system. The ONS sensor has one overall plan parallel topology of MEMS accelerometer axes integrated within the pen (see MEMS A and MEMS B, Patent Documents 1 and 2) according to the topology of Figure 5. Form a piece. In this example, an inertial accelerometer from the MEMS ADXL class (Analog Devices manufacturer) with two axes, an output range of ± 2 g, and an analogical output was used. MEMS has the same role as the previous patent cited as a reference, and also from a frequency band above 150 Hz corresponding to micro-vibrations captured in contact with paper through the processing methods described. The fluctuation of the signal amplitude represents the input for handwriting interpretation / visualization of the trajectory created by the ONS sensor integrated in the pen.

D captured by ONS_x, D_yThe motion is evaluated by the sensor via internal processing of multiple consecutive images captured during the sampling period, which lasts only 2-3 ms, and in the embodiment, the ONS sensor has a sampling period of 2 ms. Is. The processed image represents the projection of the pattern P printed on the writing paper / support through the lens L over the photosensitive area of the sensor. The topology shown in Figure 5 (detailed in Figure 6) has three conditions: three sensor systems, MEMS A, MEMS B, and ONS x, y coordinate / sensitivity axis alignment conditions. Creates an optimal link between the non-intersection of the pen's peak with the lens's field of view and the condition that minimizes navigation errors caused by dynamic tilting during the writing process. The ONS sensor has the highest sensitivity for images captured in the near infrared band. Therefore, the pattern illumination is provided by an IR LED with radiation in the near infrared band, arranged as shown in FIG. 5 and detailed in FIG.

4 accelerations generated by the MEMS sensor and 2 d generated by the ONS sensor_x, D_yThe microcontroller μCA, which manages both synchronous sampling with exercise, triggers a sampling cycle over a predetermined time period.

At each moment n, within computer 2, d_x(n), d_yAn Algerbrick sum of the elements of the (n) series is created, which is the paper / within the focus retention interval (d = 5-20 mm) of the pattern P image through the lens L on the photosensitive area of the ONS sensor. Approximation of the writing projection of the ONS sensor contained within the pen 1 through a plane curve on the writing support in the context of dynamic and smooth execution of the signature, with the lifting of the pen peak from the support. Represent.

d_x(n), d_y(n) The graphic signal set is a_x(n), a_y(n), b_x(n), b_y(n) To computer 2 via the USB protocol via a specialized microcontroller μC-USB contained within the pen for acquisition, visualization, processing, and signature comparison with a digitized acceleration set. Will be sent. The graphic data set has a constant periodicity and well-defined intervals ensure phenomenal synchronization with the accelerometers obtained via the accelerometers MEMS A and MEMS, the latter: It has a constant periodicity, eg 1 mS. On the writing surface / paper, pattern P having a quasi-uniform distribution was printed. The role of this pattern is to explain the relative motion of the ONS sensor performed between two consecutive readings._x, D_yIt is the source of the static reference points needed for the ONS sensor to calculate the pair. Finally, at the output of pen module 1, six signals sent by the USB protocol to computer 2, each of the following four acceleration signals, ie a_x(n) -Digitalized signal generated by MEMS A in the x direction at point A a_y(n) -Digitalized signal generated by MEMS A in the y direction at point A b_x(n) -Digitalized signal generated by MEMS B in the x direction at point B b_y(n) -The digitized signal generated by MEMS B in the y direction at point B, and the two motion signals: d_x(n)-Signal generated by ONS in the x direction, d_yThe signal generated by ONS in the (n) -y direction corresponds to the sampling cycle. The x and y directions correspond to the internal axes of the MEMS and ONS sensors.

The system achieves the following specific functionality: --This system captures the trajectory projection of a writing device (electronic pen) in the context of performing signatures on writing paper printed with a diffuse pattern P with a quasi-uniform distribution or any other support with a similar texture. .. This function is achieved via an optical self-referential navigation sensor ONS integrated into a pen on which the user / subject signs with it. This functionality has dual utility, which produces graphic information that is manipulated by the system in an algorithmic way of determining the authenticity of a signature, which is a signature registration procedure or system. It also produces the visual feedback displayed on the monitor of the system, which is essential for the user in the adaptation procedure using and the visual confirmation by the subject to be authenticated. --This system has two categories of signals: motion signals captured by the ONS sensor and accelerometer MEMS integrated into the pen. The beginning and end of the signature are detected by combined processing with the signal corresponding to the minute vibration captured by A. This method is based on the analysis of the frequency and amplitude parameters of microvibrations that appear with data on the pen's self-reference trajectory in the dynamic interaction of gesture-pen-paper fiber elements. This functionality is implemented as an algorithmic method included in the application of the system existing on the computer 2. --This system visualizes the trajectory of the pen. This procedure is immediate, in fact, real-time, and after the process, it also displays a handwriting effect (information) about the thickness of the trajectory, similar to the graphic path created by the pen core on paper. Will be done. This functionality has the role of providing the human subject with the visual feedback needed for the acceptance of acquisitions made by the system. The algorithmic module corresponding to the visualization functionality is included by the signature authentication application of the system existing in the computer, and the visualization is performed on the monitor of the computer. --- Multimodal processing and comparison of the acquired signal / shape is performed via the algorithmic method described below with the acceleration captured by the MEMS and the signal / shape captured by the ONS sensor as input elements. The characteristics of the algorithmic method with six acquired signals and signal types (graphics and acceleration), especially for functioning in n-dimensional space (2D, 4D, 6D), determine the multimodal nature of the system's method. To do. The pens and methods used in the present invention are functionally and integrally associated and implemented as a set of algorithmic methods included in the application of modules and signature authentication systems.

The method of handwriting reconstruction of the writing thickness displayed on the personal computer monitor 2 is carried out through the next step of running in the personal computer 2.

In the first stage, for handwriting reconstruction of handwriting thickness, the trajectory obtains the coordinates of the instantaneous motion captured by the optical navigation sensor ONS._xSignal and d_yCalculated by the Algebrick sum of a series of signal values, this is the trajectory that is graphically reproduced on computer monitor 2.

In the second stage, a related to signature_xAcceleration and a_yFilter the acceleration. Digitized a from MEMS A_xAcceleration and a_yAcceleration is filtered using a high frequency filter with a cutoff frequency of 150 Hz, c_xSignal and c_yProduces a signal. a_xSignal, a_ySignal, c_xSignal, c_yEach of the signals is actually a sample vector represented as a positive integer.

In the third stage, RMS v_x(n) Value and v_y(n) The value is calculated within the time period (ni, n) and the resulting value is c_xSignal and c_yIt is related to the time point n corresponding to the instantaneous value of the signal. I mean,

<maths num="1"><img id="000002" he="23" wi="37" file="JP5740407B2_D0001.tif" img-format="tif" img-content="drawing" /></maths>and

<maths num="2"><img id="000003" he="23" wi="41" file="JP5740407B2_D0001.tif" img-format="tif" img-content="drawing" /></maths>Because, here, c_x(n) represents the nth sample of the signal.

Two valid values v_x(n) and v_yResulting value via the arithmetic mean of (n)

<maths num="3"><img id="000004" he="15" wi="39" file="JP5740407B2_D0001.tif" img-format="tif" img-content="drawing" /></maths>Contains information used to reconstruct the handwritten thickness. Within the range of m (n) signal values, the interval [a, b] is examined, where a and b represent the minimum and maximum thickness associated with handwriting at a given point in time, respectively. a and b were determined experimentally.

In the 4th stage, the situation where the line segment of the trajectory created when the pen is not in direct contact with the writing support is removed from the initial graphic representation is determined, and the thickness information association of the line segment whose trajectory is not removed is also executed. And therefore, handwriting interpretation is achieved. Therefore, the value v_x(n), v_y(n), or

<maths num="4"><img id="000005" he="15" wi="37" file="JP5740407B2_D0001.tif" img-format="tif" img-content="drawing" /></maths>If any of the above is less than the experimentally determined threshold p, then the handwriting-related thickness at time n is considered to be zero. At a time when this condition is satisfied, the line segment corresponding to the trajectory captured by the pen 1 (Fig. 7) is removed from the initial graphic representation (Fig. 8) displayed on the computer monitor 2 and is therefore written-supported. You can obtain a handwriting expression similar to that drawn on the body by the replacement core of the pen.

The present invention also provides the following method of signature recognition called SRA3, SRA5, SRA7, SRA8 (SRA = system recognition algorithm) for assessing dynamic handwritten signature authentication. See Figure 10.

These methods process the signal produced by the pen. To describe the processing algorithm, the signal generated by the pen is marked as follows: --a_x: Signal generated by MEMS A in the x direction at point A --a_y: Signal generated by MEMS A in the y direction of point A --b_x: Signal generated by MEMS B in the x direction at point B --b_y: Signal generated by MEMS B in the y direction at point B --d_x: Signal generated by ONS in the x direction --d_y: Signal generated by ONS in the y direction

Each signal is actually a vector of samples represented as a positive integer. This vector is a numerical representation of the waveform. All vectors with the same signature have the same length (same number of samples).

These four methods are based on the following general principles:

Each method involves two modules (see Figure 9). a) Input data processing module. Through a series of actions, representative information is extracted from the input data that stores data about the sample signature and is used to represent the data composed of the recognized input signatures (original or false). The assembly of these behaviors is named Signature Conversion Method 3. b) A signature comparison module between two signatures, the sample signature and the input signature. The assembly of this behavior is named Signature Comparison Method 4. The assembly formed by the signature conversion method and the signature comparison method is named the signature recognition method.

The conversion of the input signal (created by the input data processing module) to a format that can be used in the comparison process encloses the next step. a) Converting a signature signal into an invariant. b) Weighting invariant sequences.

The sample signature is converted and stored in the signature database. Then, when the input signature appears (original or fake), the input signature is converted and compared with the signature from the database to calculate the distance between the input signature and the sample signature, thus the input signature subject is sample-signed. It is established whether it is the same as the subject.

The four mentioned methods are various curves (straight line segments) to an invariant sequence, that is, a sequence of elements that do not change with respect to frequency or signal amplitude (when talking about acceleration) or scale (when talking about graphic signals). Based on the general conversion method (approximate by).

The SRA3 and SRA5 methods deal with plane curves (2D), the SRA7 method deals with curves represented in 6-dimensional space, and the SRA8 deals with curves represented in 4-dimensional space (Fig. 10).

First, the SRA3 method for handling a plane curve will be described.

The first step lies in the conversion of the signal into an invariant.

The curve (5) defined by the (x, y) coordinates according to Figure 11 is d_xSignal and d_yRepresents a signal.

On this curve, the coordinates (x)_i, y_i), (X_{i + 1}, y_{i + 1}), (X_{i + 2}, y_{i + 2}) (Where i = 1,2, ..., n-2, where n represents the total number of points that define the shape), three consecutive points T to be considered_i, T_{i + 1}, T_{i + 2}There is. Value x_i, X_{i + 1}, X_{i + 2}Is d_xBelongs to a curve, y_i, Y_{i + 1}, Y_{i + 2}Is d_yIt belongs to a curve.

T_iT_{i + 1}The slope of the line segment defined by p_iIs determined relative to the OX axis according to the following algorithm. If x_{i + 1}&gt; x_i and y_{i + 1}= y_i then p_i= 0 If x_{i + 1}&gt; x_i and y_{i + 1}&gt; y_i then p_i= atan ((y_{i + 1}-y_i) / (X_{i + 1}-x_i)). If x_{i + 1}= x_i and y_{i + 1}&gt; y_i then p_i= π / 2 If x_{i + 1}&lt;x_i and y_{i + 1}&gt; y_i then p_i= π / 2 + atan ((x)_i-x_{i + 1}) / (Y_{i + 1}-y_i)) If x_{i + 1}&lt;x_i and y_{i + 1}= y_i then p_i= π If x_{i + 1}&lt;x_i and y_{i + 1}&lt;y_i then p_i= π + atan ((y_i-y_{i + 1}) / (X_i-x_{i + 1})) If x_{i + 1}= x_i and y_{i + 1}&lt;y_i then p_i= 3 * π / 2 If x_{i + 1}&gt; x_i and y_{i + 1}&lt;y_i then p_i= 2 * π-atan ((y)_i-y_{i + 1}) / (X_{i + 1}-x_i)) If x_{i + 1}= x_i and y_{i + 1}= y_i then p_i= 0 (Representing the inverse trigonometric function by atan ())

By analogy, T_{i + 1}T_{i + 2}Line segment p defined by_{i + 1}The slope is calculated relative to the OX axis.

Then T_{i + 1}Is the new coordinate system (X'T)_{i + 1}A line segment T that is considered to define Y') and refers to the new coordinate system_{i + 1}T_{i + 2}Direction, r_i= | p₂-p₁| Is determined by calculating.

Here you can define two codes: ca_i= Absolute orientation code cr_i= Relative orientation code

Absolute orientation code ca_iIs T_{i + 1}S an imaginary circle with a center in_aCalculated by dividing into sectors. These sectors are numbered by trigonometry. ca_i= int (p_{i + 1}/ (2 * π / S_a)) (The notation int () means the integer part of the value.)

S_aThe value is determined experimentally (eg S)_aCan be 8).

Relative bearing code cr_iIs T_{i + 1}S the circle with the center in_rObtained by dividing into sectors. Sectors are first numerotated from the BX axis and therefore p_{i + 1}&gt; = p_iIn the case of cr_i= int (n / (2 * π / S)_r)) p₂&lt;p₁In the case of c_r= S_r-1-int (p / (2 * π / S)_r)) Have.

Finally, the two codes, the expression c_i= cr_i* S_a+ ca_iCombine into a single invariant using. S_rThe value is determined experimentally (eg S)_r=145)。

An explained analysis of three consecutive points starting at each point on the curve determines the sequence of invariants.

The second step is to assign weights to invariants.

The invariant is three consecutive points on the curve (T)_i, T_{i + 1}, T_{i + 2}), So the "invariant length" L_i(i = 1,2, ..., n-2) is

<maths num="5"><img id="000006" he="13" wi="109" file="JP5740407B2_D0001.tif" img-format="tif" img-content="drawing" /></maths>Is considered to be.

L_tIs defined as the "reference overall length" and the formula

<maths num="6"><img id="000007" he="15" wi="47" file="JP5740407B2_D0001.tif" img-format="tif" img-content="drawing" /></maths>Can be calculated using.

Where the weights of each invariant w_i, Invariant length L_iAnd standard length L_rCan be established as a ratio between. w_i= L_i/ L_r

Each invariant is an expression C_i= c_i+ w_i* S_a* S_rUsing a single C_iHas a value that can be packed into the code (c_i, w_i).

The SRA5 method reassembles by the principle of using the SRA3 method by analyzing a given curve in two-dimensional space. However, there are two curves that are analyzed as separate components. The first curve is a_xAnd a_yDefined by, the second curve is b_xAnd b_yDefined by. By performing this analysis, the curve corresponding to the acceleration (measured at point A or B) has a time-independent representation, but is analyzed with respect to the order given by the arrival of the sample (point of the curve) in time. It becomes a plane curve that must be done.

The SRA7 method will now be described using curves in 6-dimensional space.

The first step lies in converting the signal into an invariant.

Curves defined by points in 6-dimensional space (numbered 0, 1, 2, 3, 4, 5) and belonging to this shape and coordinates T_i:( u_{i, 0}, u_{i, 1}, u_{i, 2}, u_{i, 3}, u_{i, 4}, u_{i, 5}) T_{i + 1}:( u_{i + 1,0}, u_{i + 1,1}, u_{i + 1,2}, u_{i + 1,3}, u_{i + 1,4}, u_{i + 1,5}) T_{i + 2}:( u_{i + 2,0}, u_{i + 2,1}, u_{i + 2,2}, u_{i + 2,3}, u_{i + 2,4}, u_{i + 2,5}) (However, i = 1,2, ..., n-2 are the total number of points that determine the curve) Three consecutive points T_i, T_{i + 1}, T_{i + 2}To consider. u_{i, 0}, U_{i + 1,0}, U_{i + 2,0}The value of is a_xBelongs to the curve, uu_{i, 1}, U_{i + 1,1}, U_{i + 2,1}The value of is a_yBelongs to a curve and has a value u_{i, 2}, U_{i + 1,2}, U_{i + 2,2}Is b_xBelongs to a curve, u_{i, 3}, U_{i + 1,3}, U_{i + 2,3}The value of is b_yBelongs to a curve and has a value u_{i, 4}, U_{i + 1,4}, U_{i + 2,4}Is d_xBelongs to a curve and has a value u_{i, 5}, U_{i + 1,5}, U_{i + 2,5}Is d_yIt belongs to a curve.

T_{i + 1}T_{i + 2}T with the line segment defined by_iT_{i + 1}Angle p composed of line segments defined by_iIs determined according to the following algorithm.

First, the two values s₁And s₂Is calculated using the following formula. s₁= (u_{i + 1,0}-u_{i, 0}) * (U_{i + 2,0}-u_{i + 1,0}) + (u_{i + 1,1}-u_{i, 1}) * (U_{i + 2,1}-u_{i + 1,1}) + (u_{i + 1,2}-u_{i, 2}) * (U_{i + 2,2}-u_{i + 1,2}) + (u_{i + 1,3}-u_{i, 3}) * (U_{i + 2,3}-u_{i + 1,3}) + (u_{i + 1,4}-u_{i, 4}) * (U_{i + 2,4}-u_{i + 1,4}) + (u_{i + 1,5}-u_{i, 5}) * (U_{i + 2,5}-u_{i + 1,5}) s₂²= ((u_{i + 1,0}-u_{i, 0})²+ (U_{i + 1,1}-u_{i, 1})²+ (U_{i + 1,2}-u_{i, 2})²+ (u_{i + 1,3}-u_{i, 3})²+ (U_{i + 1,4}-u_{i, 4})²+ (U_{i + 1,5}-u_{i, 5})²) * ((u_{i + 2,0}-u_{i + 1,0})²+ (U_{i + 2,1}-u_{i + 1,1})²+ (U_{i + 2,2}-u_{i + 1,2})²+ (u_{i + 2,3}-u_{i + 1,3})²+ (U_{i + 2,4}-u_{i + 1,4})²+ (U_{i + 2,5}-u_{i + 1,5})²) s₂# 0 and s₁If&gt; = 0,

<maths num="7"><img id="000008" he="12" wi="43" file="JP5740407B2_D0001.tif" img-format="tif" img-content="drawing" /></maths>And s₂# 0 and s₁If &lt;0,

<maths num="8"><img id="000009" he="14" wi="48" file="JP5740407B2_D0001.tif" img-format="tif" img-content="drawing" /></maths>And s₂If = 0, p_i= 0.

Where the i-th invariant code c_iTo judge. c_i= p_i/ (2 * π / S_r) However, S_rThe value is determined experimentally (eg S)_r=7)。

The second step is to assign weights to each invariant.

The invariant is three consecutive points T on the curve_i, T_{i + 1}, T_{i + 2}As defined by, "invariant length" L_i(i = 1,2, ..., n-2) is calculated using the following equation.

<maths num="9"><img id="000010" he="19" wi="158" file="JP5740407B2_D0001.tif" img-format="tif" img-content="drawing" /></maths>

L_tIs defined as the reference overall length according to the following equation.

<maths num="10"><img id="000011" he="16" wi="40" file="JP5740407B2_D0001.tif" img-format="tif" img-content="drawing" /></maths>

Where the weights of each invariant w_i, Invariant length L_iAnd the standard total length L_rIt is calculated as if it is paid between. w_i= L_i/ L_r

The value of each invariant is an expression C_i= c_i+ w_i* S Using a single C_iCan determine the code (c_i, w_i) Represented by a pair. S is c_iGreater value (for example, S = 32768).

The SRA8 method is similar to SRA7, but a_x, A_y, B_x, B_yUse only the 4th dimension given by.

Using SRA3, SRA5, SRA7, and SRA8, the input signal is transformed into an invariant array, each of which has a weight (or cost). A means of determining the distance between two strings of symbols defined by the two signatures being compared in order to compare the two signatures according to the chosen algorithm. The most appropriate method for calculating this distance is the "Levenshtein" algorithm. Finally, if the result (Levenshtein distance) is D, then the (normalized) distance d to take into account is:

<maths num="11"><img id="000012" he="20" wi="60" file="JP5740407B2_D0001.tif" img-format="tif" img-content="drawing" /></maths>And here, cost_i= w_iAnd cost_j= w_jRepresents the cost of the invariants of the two components.