Title of Invention	SYSTEM AND METHOD FOR AUTOMATICALLY VERIFYING THE AUTHENTICITY OF SIGNATURES
Abstract	System and method of authentication are described. The instant invention presents a new handwritten signature verification system where the characteristics of online signature are also used for the verification of offline signatures. The system has a unique new method to handle the Bank Mandate scenarios wherein there can be multiple signatories for a single account. The instant invention is an intelligent system with built-in learning capabilities, which enables it to successfully detect intra-person variations to mark the signature as genuine, while at the same time discarding the signature by detecting inter-personal variances. The invention automatically recognizes changes in a person's signature owing to various factors such as aging, time, illness, type of writing instrument, differences in signature size (due to space allotted for signature), etc., and updating these changes in a database to verify the authenticity of the signature.

Title of Invention

SYSTEM AND METHOD FOR AUTOMATICALLY VERIFYING THE AUTHENTICITY OF SIGNATURES

Abstract

System and method of authentication are described. The instant invention presents a new handwritten signature verification system where the characteristics of online signature are also used for the verification of offline signatures. The system has a unique new method to handle the Bank Mandate scenarios wherein there can be multiple signatories for a single account. The instant invention is an intelligent system with built-in learning capabilities, which enables it to successfully detect intra-person variations to mark the signature as genuine, while at the same time discarding the signature by detecting inter-personal variances. The invention automatically recognizes changes in a person's signature owing to various factors such as aging, time, illness, type of writing instrument, differences in signature size (due to space allotted for signature), etc., and updating these changes in a database to verify the authenticity of the signature.

Full Text	System and method for automatically verifying the authenticity of signatures Field of Invention The present invention relates to an efficient and cost-effective method and system for signature verification which is largely dependent on offline features but also uses online features for automatic signature verification. The present invention relates generally to the field of automatic image processing, focusing in particular on financial instruments such as bank checks and drafts. Background of the Invention A large volume of work has been performed to evolve efficient methods and systems of accurate identity verification. In general, the problem attempted to be solved in identity verification is: given a plurality of identifying reference samples and a test sample, does the test sample belong to the owner of the reference samples? The samples may be signatures, biometric data such as voice samples, face pictures or other such relevant information. Modern technology offers a number of new, sophisticated and secure authentication mechanisms such as biometrics. However, these are restricted to only specific and specialized segments of particular industries critical to human life and safety such as nuclear programs, space programs, intelligence agencies, and other capital-intensive program which can justify the costs and the efforts required for their implementation. For other environments which do not necessitate the use of such complex means, signature authentication offers a quick, simple and cost-effective means for correctly establishing identity of an individual. In particular, financial institutions such as banks and insurance companies still rely heavily on traditional method of signature verification for effecting transactions done through paper instruments such as checks, drafts, etc. The sheer volume of such instruments makes manual verification a time and cost-intensive process and mandates companies to have an automated system, which could automatically and quickly establish the veracity of the signatures, and thereby affect a valid transaction. Two main approaches have been adopted for automated signature verification - Offline Signature Verification and Online Signature Verification. Offline Signature is the most popular approach for signature verification owing to its lower costs. However, conventional offline approach of verification offers results of relatively low accuracy. Unconventional and individual writing styles make detection and segmenting of signature strokes difficult. The problem is compounded if signature images are of low resolution. Absence of any dynamic and temporal information makes offline signatures much harder to detect. In contrast to the offline approach, online signature verification provides relatively better results in establishing the authenticity of the signature. This is because online verification uses a plurality of parameters associated with stylus and an electronic writing pad for determining the authenticity of the signature. These parameters may include speed, direction, pressure of the stylus, number and order of the strokes, etc. These properties impart certain distinctive features to the signature of each individual, which are quite difficult to forge. However, the practical implementation of this approach requires the use of stylus and electronic pads along with other relevant instruments for signature capture which add up to high infrastructure costs. Moreover, paper instruments such as checks and documents still form the majority of volume. In light of the above, several approaches have been suggested to counter the disadvantages of various offline and online signature verification systems at various points. U.S. Patent No. 5,251,265 describes a method and system for automatic offline signature verification. The apparatus digitizes an image of the signature and provides a gray scale representation of each pixel. Next, the apparatus determines parameters of this digitized image and compares them to corresponding reference parameter values, which were determined from a valid signature. The results of the comparison indicate if the signature to be verified is valid. U.S. Patent No. 5,995,953 describes a method for verification of signatures and handwriting based on comparison of extracted features instead of the images themselves. Thus it focuses on the offline method of verification. It uses a neural net for image recognition. This leads to a significant reduction of storage space and calculation time needed. U.S. Patent No. 4,985,928 describes a signature forgery detection device. The system relies on the collection of measurements of the optical density of a plurality of elements in the specimen and selectively retaining said measurements and then comparing each density measurement to a threshold value. Only those density measurements are retained which are located within a bounded locus. After the retained density measurements have been established, they are automatically compared to the measurements of the specimen signature to permit the apparatus to accept or reject the signature on the presented document U.S. Patent No. 6,424,728 describes an automatic signature verification system that utilizes a main routine for comparing signatures using forensic hand writing methodology. A secondary program is used to modify the algorithms used by the main program for making adjustments thereto based on either additional data consisting of a plurality of genuine or authenticated signatures or changes in a person's signature due to ageing or some other physical change resulting in a change in signature features. The above-mentioned patents suggest various approaches of online or offline verification of signatures. However, these methods are more or less akin to traditional approaches of signature verification using simple signature matching or feature extraction techniques. This limits the ability of the existing techniques to accurately verify an individual's signature. Moreover, such techniques are quite time-intensive. Also, some of these methods involve a lot of manual intervention, thereby making the process inefficient for being applicable in an industrial environment. Further, these systems also do not have a built-in intelligence to detect the changes in the signature of a user due to natural factors like changes in handwriting patterns due to age, time, illness, etc. With the growing volume of checks and other similar, signature-bearing instruments, these methods, thus, may not be able to process the instruments as per the exacting requirements of the banks. Hence, there is a need for a method and system that provides an efficient and cost effective system for automatic signature verification. The system and method should be able to provide accurate results by taking into account person-specific confidence thresholds. The system should also have in-built intelligence for constant learning to handle scenarios where a valid signature is identified as invalid by the system on account of changes induced in the signature image due to factors such as age, illness, orientation, etc. Such a method should incorporate techniques a plurality of unique offline characteristics and preferably various online verification techniques as well to provide better results in all operating environments. Such system should also be able to handle scenarios such as a Bank Mandate which requires verification of more than one person's signature. In addition, said system should be language independent. Objects and Summary of the Invention The present invention has the objective to provide an efficient and cost effective system and method, which is largely dependent on offline features but also uses online features for automatic signature verification. It is an objective of the present invention to provide a system and method for automatic signature verification with in-built intelligence to detect both inter-person and intra-person changes and ascertain if the variations in the signature are induced owing to natural parameters such as age, health and orientation and use this information for continual learning. It is also an objective of the present invention to provide a means to authenticate a signature against more than one reference signatures in case multiple users are authorized to access and operate the relevant information such as in a Bank Mandate scenario prevalent in operating corporate bank accounts. It is yet another objective of the present invention to provide a language independent method and system of automatic signature verification. To achieve the foregoing and other objectives obvious to a person skilled in the art, the present invention provides a method of automatic signature verification comprising the steps of: - receiving user input of at least one signature; - extracting one or more features and associated information from the signature using offline and online extraction techniques; - determining if the information can be used for updating a reference database; - verifying the extracted features against predefined information stored in the reference database for authenticating the signature; and - updating the reference database. The present invention further provides for a system for automatic signature verification comprising: - acquisition means for receiving one or more input signatures; - image processing means coupled to acquisition means for extraction of one or more features from the signature and verifying said features against one or more predetermined features associated with at least one reference signature; - storage means coupled to image processing means and acquisition means for storing the input signatures, extracted features, reference signatures, associated information for each user account, and operating instructions for automatic signature verification. Brief Description of Drawings The detailed description is described with reference to the accompanying figures. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. The same numbers are used throughout the drawings to reference like features and components. Figure 1 illustrates an exemplary system for automatic signature verification as disclosed by the present invention. Figure 2 illustrates a flow diagram of an exemplary method of pre processing an acquired image. Figure 3 illustrates an exemplary method of feature extraction from a pre processed image. Figure 4 illustrates an exemplary method of extraction of information from a reference database required by a user of the disclosed system. Figure 5 illustrates an exemplary method for verification of signatures as disclosed by the present invention. Figure 6 illustrates an exemplary method for dynamically updating the signature database. Figure 7 illustrates an exemplary method for manually verifying and updating of the signature database. Detailed Description of the Invention Systems and methods for efficient means of automatic signature verification are described. The description of the system or method shown herein below is intended only for illustration and disclosure of an operative embodiment and not to show all of the various forms or modifications in which this invention might be embodied or operated, since the same may be modified in various particulars or relations without departing from the spirit or scope of the claimed invention. The system and method of the present invention aim to provide automatic signature verification, which is largely dependent on offline features but also uses online features for verifying signatures in particular borne by financial instruments such as bank checks. The financial instrument bearing the signature is scanned. Before extracting the relevant features, the scanned image undergoes one or more preprocessing steps for noise removal and orientation correction. Subsequendy, a plurality of features is extracted from the processed image using offline as well as online feature extraction techniques. The features extracted from the signature image are compared against pre-determined features from one or more reference signatures stored earlier in a reference database. While comparing a signature specimen against reference signatures, if certain variations in specific features are determined which are within predefined permissible limits, the variations are incorporated into the reference database for future processing. A further verification is done to check if the request conforms to a predetermined pattern for the specific user. On positive verification of both, the user signature is authenticated. In case the variations in the features do not fall within the predefined permissible limits, the signature is sent for manual verification. On positive verification, the sample is added to the signature database for future processing. Thus, the method and system of the present invention display inherent intelligence by constant learning and also facilitates development of a dynamic signature database using both offline and/or online feature extraction techniques with time. Moreover, the disclosed method of the present invention is also language independent. Further, the present invention can be easily adapted to operate in environments requiring one-to-many signature verification. An exemplary implementation of the system and method of the present invention is discussed in the following section with respect to the accompanying figures. Figure 1 illustrates and exemplary system 100 for automatic signature verification. In one implementation, system 100 can include acquisition means 102 coupled to image processing means 104 and reference database 106. In some embodiments, image processing means 104 and reference database 106 may be communicatively coupled to acquisition system 102 over a wired or wireless communication network. Acquisition means 102 can further include one or more digital input devices 108 and one or more imaging devices 110. One or more interfaces for coupling of said devices may also be part of the acquisition system 102. A signature specimen to be authenticated can be input into system 100 by using digital input devices 108 such as tablet pens and transducer pads. Further, imaging devices 110 such as a camera and/or scanner can be used to receive signature specimens from a user. The information acquired by the acquisition system 102 can be stored in reference database 106 along with operating instructions for facilitating signature authentication as disclosed by the method of the present invention described herein below. Image processing means 104 can further include means to process the acquired/scanned signature image information in order to remove noisy elements, conversion of said image into binary format, thinning, orientation detection, rotation and scaling. Image processing means 104 can facilitate extraction of relevant features from the processed image using one or more offline as well as online feature extraction techniques as described below. Further, the extracted features can be verified against predetermined features of reference signatures stored in the reference database 106 to authenticate the input signature as a valid means for authorizing a user transaction. To this end, reference database 106 stores predetermined reference signature specimens and values of associated features of the same for each user account. Further, specific information regarding the type of the account being individual or Bank Mandate, sufficiency of data for verification, specific set of features to be used for verification, whether automatic verification can be performed, and other transactional information for specific user accounts can be stored. Based on the comparison of the stored information for each user account and the relevant received user input, a comparison of relevant features can be performed to authenticate the received user signatures. In an embodiment, a signature card can be signed by the depositor upon opening an account at a financial institution. The card establishes the type of account ownership and sets forth the account terms and the obligations of the customer and the institution. Companies for subsequent identification of the customer use signature cards. The user provides his signatures using a digital input device 108 and/or the same is captured by an imaging device 110. A plurality of relevant features are extracted by image processing means 104 from the reference signatures received using digital input device 108 using online extraction techniques namely, "Pen out-of or in-contact with paper", "Pen-down time to total time ratio", "Current and previous position of pen", "Pen down, Pen up", "Length of signature", "Total time of signature", "Tangent at each point in Signature", "Velocity and acceleration at each point of signature" and so on. Further features can also be extracted by image processing means 104 using offline extraction techniques described below such as height, curve comparison, critical points using contours, projection, holes, smoothness index, shape matrix, grid density, radial transition, wrinkleless and so on from the scanned image. One or more features are determined to be relevant for signature verification for each user account. Values of the extracted features and information regarding which features are to be used for verification for each user account is stored in the reference database 106. When a signature specimen is received for authentication, relevant features are captured by acquisition system 102 using a digital input device 108 and/or imaging device 110 the values of the features are then compared with the stored values for the specific user. System 100 has built-in learning capabilities, which enables it to successfully detect intra-person variations to mark the signature as genuine, while at the same time discarding the signature by detecting inter-personal variances using operating instructions stored in reference database 106. The system 100 automatically recognizes changes in a person's signature owing to various factors such as aging, time, illness, type of writing instrument, differences in signature size (due to space allotted for signature), etc., and updates these changes in a database to verify the authenticity of the signature for future processing. Further, based on the account type, system 100 can also authenticate more than one signatures associated with a single account if the Bank Mandate feature is enabled for the same. Figure 2 illustrates an exemplary method of processing a scanned image prior to application of a plurality of feature extraction techniques as disclosed by the present invention. At 202, scanned image acquired by an acquisition means. At 204, the acquired image is processed by a preprocessing system. At 206, the noisy elements in the image are removed. At 208, the image is converted into binary format. At 210, using one or more techniques the binary image undergoes thinning. At 212, the angle of orientation of the image is determined. Subsequently, at step 214, the image is rotated by an appropriate angle as determined during orientation detection. At 216, the appropriately oriented image is scaled. The scaled image undergoes further binarization and thinning at steps 218 and 220. The resultant image is subjected to a plurality of feature extraction techniques as described below. Figure 3 illustrates an exemplary method of application of static and dynamic feature extraction techniques for feature extraction as used in the present invention. At 302, a user's signature specimen is received for authentication. At step 304, the user signature is inputted using a digital pad while at step 306 a signature card with reference signatures is used for verification. The information received from digital pad as well as signature card or similar means can be used for applying a combination of online and offline techniques of feature extraction to the signature specimen. At 308, a plurality of dynamic feature extraction techniques is employed. One or more parameters are considered for the determination and extraction of relevant features from the signature specimen to be authenticated. In an embodiment, one of the parameters can be the pressure induced by the signatories' pen when signing on transducer pad produces an analogue voltage, which is converted to digital form. These digital values are processed to determine numerical parameter values, each representing a feature of the signature. The user input of signature on a digital pad can facilitate extraction of a plurality of features from the specimen for verification against one or more reference signatures stored in a signature database. These can be "Pen out-of or in-contact with paper", "Pen-down time to total time ratio", "Current and previous position of pen", "Pen down, Pen up", "Length of signature", "Total time of signature", "Tangent at each point in Signature", "Velocity and acceleration at each point of signature" etc. At 310, various static feature extraction techniques are employed, particularly tangent of the signature is extracted and analyzed. This feature traces the path of the test signature. Once the path of the signature is known, it becomes easy to determine the authenticity of the signature. Further a plurality o{ novel techniques as described below are used to verify a given signature specimen against stored reference signatures. HEIGHT: Height of individual parts of the signature image, divided along the signature breadth into equal parts is used for verification. The average of heights of individual parts of all the sample images is compared with the height of the corresponding part of the signature image to be verified. CURVE COMPARISON: Curve Comparison works by generating chain codes. A plurality of points is chosen randomly on the trajectory of the signatures and these two-dimensional coordinates are then reduced to one-dimensional codes and the values are compared to that of the database values. CRITICAL POINTS USING CONTOURS: Pixel density, for various regions of the signature if vertically incident light rays are hypothetically thrown on the signature image, is determined. The pixel density of the various regions is compared to the average pixel density of the sample authentic images. This feature tests the similarity or confidence value against a threshold value. Critical points are chosen on the trajectory of the signature based on the deviation angle. A LCS (Longest Common Sub String) is formed from these critical points. Based on the comparison of this length with a certain threshold, the authenticity of the check image is checked. PROJECTION FEATURE: Projection feature also works by comparing the similarity or confidence value against a threshold value. This feature is extracted by rotating the signature from -X° to +X° with a variation of 8 each time. The width of the image is accordingly adjusted to a fixed size and a Normalized Cross Correlation is calculated for all the authentic images. HOLE FEATURE: Another unique feature is the Hole feature in which the number of closed loops in the signature are verified. Comparison of the check image with the authentic image is done based on the number of holes, positions and sizes of the holes. SMOOTHNESS INDEX: Smoothness is a vital parameter in ascertaining if the signature is genuine. This feature is mainly used to detect skilled forgery. Signature forgeries are broadly classified into three categories — Random Forgery, Simple Forgery and Skilled Forgery. Random forgery involves forged signature without any knowledge of the original signature. Simple Forgery involves forged signature with a little knowledge of original signature — the knowledge may involve only a cursory glance at the original signature, imitating the original signature based on memory, and other such limited knowledge — but done without practice. Skilled Forgery involves forged signature with good knowledge of original signature — the knowledge may involve constant availability of the original signature to minutely study the nuances of the signature, well-practiced imitation of the original signature, and so on — and close imitation of the original signature after practice. Therefore, Skilled Forgery is hardest to detect. The basic aim behind employing the feature, Smoothness Index, is to compare the steadiness of the hand while signing. While doing Skilled Forgery, the imitator tries to closely match the original signature; however, he/she essentially loses out on smoothness of curves present in the signature - even if he/she succeeds in correctly imitating smoothness of certain curves, it is nearly impossible to perfectly imitate the smoothness for all the curves present in the signatures. Also, signatures are highly behavior oriented as they can vary greatly depending on external factors like age, physical comfort, and illness etc. So, the signature of an aged man if forged by a young man will have a vast difference in their smoothness. The system of the present invention measures this smoothness index for the various signatures to determine their authenticity. However, at the same time, the disclosed system is able to determine the minor variations in various original signatures of an individual that may result due to factors previously mentioned. SHAPE FEATURE: The shape feature is based on combination of well-known features, such as baseline value, image area, aspect ratio, etc., which are extensively used by current signature verification methods. For calculating the baseline value, first the centroid of the whole image is calculated. Then, the image is divided vertically along its centroid, and the centroids of these two parts are also calculated. The angle subtended by the line joining the centroids of both the parts with the x-axis is called the base angle and the line is called the base line. Aspect ratio is defined as the height to width ratio of the image whose orientation has been corrected. This feature filters out a large number of random forgeries. The calculation of the image area normalizes against variations in scale of the signature. It is calculated by the number of black pixels in the image divided by the width of the image. DIRECTIONAL PROBABILITY DENSITY FUNCTION: The directional probability function determines the angle of a point with all the other points on the signature. Based on a sampling done at regular intervals of angles, the probability of going in a particular direction is determined. The process is repeated for all the points on the signature and a graph denoting the probability density is drawn. This graph provides the overall representation of the directional probability for various points on the signature. CROSS POINTS: The visible overlaps of a person's signatures are known as cross points. All the major cross points in a signature are extracted and matched with those of original signatures using graph matching and triangle matching technique to ascertain the authenticity. GRID ANALYSIS: This involves dividing the signature into horizontal and vertical boxes such that the numbers of pixels are approximately the same. After the division, the signature is analyzed for the extraction of the following values: a) pixel density, b) 2nd order moment along X-axis, c) 2nd order moment along Y-axis, d) distance of center of mass from the bottom left corner divided by the diagonal of the box, and e) average angle of each pixel with respect to the bottom left corner. Once the above features are extracted, using a modified version of regression, the test signature is compared using the sample signature. NOVEL TECHNIQUE: The signature image is divided into Vertical Cut First and Horizontal Cut First, based on center of mass of the box surrounding the entire signature image. Each cut provides a box covering a certain portion of the signature. The subsequent center of mass is calculated for every box. A vector based on Euclidean distance classification is formed with these center of mass values for six different sample images. This vector is compared with that of the test signature. If the value is above a certain prerecorded threshold, then the signature is authentic else it is a forgery. SHAPE MATRIX: The signature is smeared and loops are filled to enhance the efficiency of the verification process. The centroid of the signature image is calculated and the distance between the farthest pixel on the signature and this centroid is considered as the radius. A circle of this radius is drawn and divided into 128 equal parts and the line having maximum number of black pixels is found. A vector corresponding to this line is obtained. The radius of the circle is reduced by two pixels and another circle is drawn, and the same procedure is repeated till the centroid of the image is reached. For each circle, a vector corresponding to line containing maximum number of black pixels is obtained. All the 128 vectors are represented using a matrix known as shape matrix. The shape matrix for the sample image is compared with that of the original image to ascertain the authenticity of the signature. GRID BASED DENSITY FEATURE: The signature image is first divided into n number of buckets horizontally such that each bucket contains equal number of pixels. The same process is repeated when the signature is divided vertically. Both the images are then superimposed and the ratio of black pixels in the box to the black pixels in the entire image is calculated and a feature vector is obtained. This feature vector for the signature is compared with the vector of the original signature. RADIAL TRANSITION: Concentric circles are drawn at an increment of every five pixels starting from the centroid of the image as the center of the circles. The circles are drawn till the farthest pixel in the image is reached. The black and white transitions on the circumference on each circle are used to form a matrix. This matrix is compared with that of the test signature to determine the authenticity of the signature. WRINKLINESS FEATURE: The inner boundary of the signature is taken and logically divided (horizontally and vertically) at an interval of two pixels. The number of squares (2x2) in which the boundary of the signature lies is counted and the log of the number is calculated. The same procedure is repeated at the interval of four pixels. Division of the first number by the second number gives the measure of the wrinkliness of the signature. GRAPH FEATURE: Three sample signatures are taken. Value of all the pixels on the image is found and a matrix is computed for every sample signature compared with the other, considering all the possible combinations. The maximum value matrix is then taken from the three matrices. The cost matrix of this signature is calculated and compared with that of the original signature. If the cost of the signature is less than that of the original image then the signature is authentic, else it is fake. IMPRINT DETECTION: In case of skilled forgery, when signatures are traced/topographed, imprints appear at the backside of the check because of the pressure of the pen. When a check with a forged signature appears for verification, by observing the image of the check, the forgery can be detected with the help of these imprints. RATIO OF SUCCESSIVE PEAKS: The scanned image of the check is first aligned horizontally and then the peaks in the signature are determined. The ratios of each peak with its adjacent peak are calculated for the entire signature image. These ratios are compared with the corresponding ratios of the adjacent peaks of the sample signatures to determine the signature authenticity. One or more of the above features along with several other such features characteristic of a user's handwriting are extracted from the received signatures are compared against predetermined values of one or more reference signatures as stored in the signature database. Figure 4 illustrates a flow diagram of an exemplary method of extraction of relevant information to process the verification of the signature parameters against those of reference signatures stored in the signature database. At 402 is represented the database of signatures which includes a plurality of reference signatures and their associated parameters for various users. At 404, the disclosed system on the basis of information acquired from a financial instrument such as a check or draft, the user account is determined. Further it is determined if the user account is an individual account or of the bank mandate type. Further information regarding if sufficient training data is available for verification, the set of features which work for a particular account, whether automatic verification can be carried out on the account, whether the check conforms to regular expenditure pattern and other such verifications is extracted. Figure 5 illustrates a flow diagram of an exemplary method of signature verification as disclosed by the instant invention. At 502, the database of signatures is employed. At 504, a plurality of relevant features is extracted from the signature image using techniques disclosed in the description of figures 2-4. At 506, it is determined if sufficient training material is available to update the database. At 508, the value of the extracted feature is matched against the predefined threshold value of the said feature as present in one or more associated reference signatures stored in the signature database. In case, the training material is not sufficient, the information is added to the system for dynamic update as disclosed below in the description of Figure 6 and subsequently manual verification is employed at step 516. At 512, it is determined if the threshold values of the feature extracted from the test signature matches the threshold value of the same feature present in the reference signature. In case the direshold values do not match, manual verification is employed at step 516. If the threshold values are the same, the database is updated with the sample signature at 514. At 518, it is verified if the check conforms to regular expenditure pattern. If it does, the check is cleared without verification at step 520 else manual verification is employed and the database is updated with the decision for future use in step 522. If in the manual verification, the signature is found a valid match, this knowledge is recorded in the signature database for later use. In such a case, whenever another check is received in the future with the same sample signatures that were earlier rejected are verified and cleared on the basis of the combination of the sample signature and the check amount. Thus, if the amount is small, the signature is verified as authentic automatically by the system, whereas in case the amount is substantial, the signature is referred for manual verification. Figure 6 illustrates a flow diagram of the technique of dynamically updating the signature database. At 602, sample signatures are stored in the signature database. At 604(a), 604(b) and 604(c) one or more features are extracted from the sample signatures. At 606, each feature is compared with the relevant feature in the sample signatures. At 608, validity is determined on the basis of said comparison. In case, the feature is determined to be invalid, it is not used for the particular user at 610. If the feature is valid, it is stored along with its associated value for a specific user in the signature database at 612. At 614, it is determined if all features have been validated. At 616, the plurality of features determined to be relevant for a specific user and their associated values are used to verify the signatures for that user. Figure 7 illustrates an exemplary method of manual verification of the financial instrument, in case data for dynamic verification is insufficient, and subsequendy updating the database. With its inherent capability to build intelligence by constant learning, the system provides the capability to handle scenarios where a valid signature is identified as invalid by the system. There might be some changes that have been induced in the signature image due to factors such as age, illness, orientation, etc. These types of cases are ripe for automatic rejection, in spite of them being valid signatures. The system refers such cases for manual verification. At 702, the check is sent for manual verification subsequent to the determination that data for dynamic verification is insufficient. At 704, a valuator verifies the check. A 706, the check is rejected if the valuator finds some discrepancy in the same. At 708, the valuator finds the check in order and approves the same. A further decision on whether to use the signature for future transactions is made. At 710, the sample is added to the signature database for future reference. At 712, in case the signature sample is determined not to be used for future transactions, the check is passed but the signatures are not added to the database for future reference. The existing signature verification systems use a signature card for the identification purpose. A signature card is a form signed by the depositor upon opening an account at a financial institution. The card establishes the type of account ownership and sets forth the account terms and the obligations of the customer and the institution. Companies for subsequent identification of the customer use signature cards. However, the banks might not have a consolidated reference database with them to start with for the deployment of such systems. For such a scenario, the disclosed system has in built intelligence to enable the banks to develop a dynamic database gradually with time. The signature samples will be regularly added to the database as and when the bank receives checks for a particular account. In case sufficient signature samples are not present to extract the features required by the system to run in automated verification mode, the disclosed system by default will operate in manual verification mode for that particular account. The system will continue to operate in die manual verification mode until it has enough samples required to switch to automatic verification mode. Once the system has enough samples, without human intervention, system would set the account to operate in automatic verification mode. A further advantage of our system is that the signature verification system according to the invention is independent of a transducer pad or the like in cases where there is unavailability of a transducer pad. For such cases, the system is able to verify a signature only with the comparison of a plurality of new and unique offline features as disclosed above. Another major advantage of our invention is that it allows to automatically set an acceptance level for every individual's signature. This acceptance level is set on the basis of the kind of input that is being considered, that is, whether the input signature was acquired offline or online. It also handles special cases of the customers who due to their different handwriting patterns always sign in a different way or have a fraudulent check history. For such customers, the acceptance level for clearing of their checks can be kept sufficiently high. In addition to the above, a notable advantage of the disclosed method and system is to provide a highly efficient method of signature verification. Towards this end, the system incorporates capabilities to handle the actual scenarios faced by the banks with regards to automatic signature verification as against a plain-vanilla offering that is equipped to handle only standardized or generic signature-verification scenarios. The instant invention offers a one-to-many signature matching capability as against the one-to-one signature verification. In the current scenario, whenever a signed check reaches a bank, there is normally only a one-to-one signature verification of the user signature by the bank, that is, whenever a particular check reaches a bank, the signatures on that particular check are verified only against the reference signature of the same user that is stored in the database. This is one-to-one signature verification. The one-to-many signature verification capability makes the disclosed system fully capable of handling the bank mandate feature usually offered by banks to their corporate customers. This feature makes a single account operable by more than one user's signature. A Bank Mandate form provided by the concerned bank to its corporate customers is to be filled by the Company in the presence of its Board of Directors. This form has the signatures of all the signatories who will be responsible for carrying out any transaction for the account. The existing signature verification systems are not able to handle such a case. The disclosed system helps match a single signature with all the reference signatures of the users who are responsible for operating that particular account. Also, whenever a 'Notice of Variation', for updating/changing the signatories, is received by the bank from any of the signatories for the account, our system automatically updates the database of signatures. Further, the disclosed method and system of the instant invention also offer language independence as novel offline and online feature extraction techniques as described above are employed. This is an added advantage as many individuals, whether on account of their personal choices or unfamiliarity with the language, sign in their regional languages. The embodiments described above and illustrated in the figures are presented by way of example only and are not intended as a limitation upon the concepts and principles of the present invention. As such, it will be appreciated by one having ordinary skill in the art that various changes in the elements and their configuration and arrangement are possible without departing from the spirit and scope of the present invention as set forth in the appended claims. It will readily be appreciated by those skilled in the art that the present invention is not limited to the specific embodiments shown herein. The techniques described hereinabove can be adapted for verifying any handwriting sample of a specific user and the scope of the same is not restricted to just signature verification. Thus variations may be made with in the scope and spirit of the accompanying claims without sacrificing the principal advantages of the invention. We claim: 1. A method of automatic signature verification comprising the steps of: - receiving user input of at least one signature; - extracting one or more features and associated information from the signature using offline and online extraction techniques; determining if the information can be used for updating a reference database; - verifying the extracted features against predefined information stored in the reference database for authenticating the signature; and - updating the reference database. 2. The method as claimed in claim 1, wherein the step of receiving includes receiving the signature using a digital input device and an imaging device. 3. The method as claimed in claim 1, wherein the step of extracting includes the steps of: - removing noisy elements, - conversion of the signature image into binary format; - thinning and orientation detection; and - rotating and scaling of the signature image. 4. The method as claimed in claim 1, wherein the step of determining includes the step of manual verification if the information is insufficient for updating the reference database. 5. The method as claimed in claim 1, wherein the step of verifying includes the step of: - determining if variations in the features are within predefined threshold values; and - determining if the variations in the features are due to one or more predefined factors; and - determining if the associated information conforms to a predetermined pattern. 6. The method as claimed in claim 1, wherein the step of verifying includes the step of positively authenticating a signature against more than one reference signatures stored in the reference database for a user account of predefined account type. 7. The method as claimed in claim 1, wherein the step of verifying includes the step of determining if the variations in the features are due to one or more predefined factors without limitations including aging, time, illness, type and orientation of writing instrument and differences in signature size. 8. The method as claimed in claim 1, wherein the online feature extraction techniques without limitation include pen out-of contact, pen in-contact with paper, pen-down time to total time ratio, current and previous position of pen, pen down, pen up, length of signature, total time of signature, tangent at each point in Signature and Velocity and acceleration at each point of signature. 9. The method as claimed in claim 1, wherein the offline feature extraction techniques without limitation include height, curve, critical points using contours, projection, hole, smoothness index, shape, direction probability density function, cross points, grid analysis, vector based distance classification over vertical and horizontal cut sections, shape matrix, rid density, radial transition, wrinkliness, graph using pixel values of signatures, imprint detection, ratios of successive peaks. 10. A system for automatic signature verification using both offline and online feature extraction techniques comprising: - acquisition means for receiving one or more input signatures; - image processing means coupled to acquisition means for extraction of one or more features from the signature and verifying said features against one or more predetermined features associated with at least one reference signature; - storage means coupled to image processing means and acquisition means for storing the input signatures, extracted features, reference signatures, associated information for each user account, and operating instructions for automatic signature verification. 11. The system as claimed in claim 10, wherein the acquisition means includes one or more digital input devices and one or more imaging devices. 12. The system as claimed in claim 10, wherein the image processing means further includes means to: - remove noisy elements; - convert signature image into binary format; - orientation detection; and - rotating and scaling the image. 13. The system as claimed in claim 10, wherein the storage means is updated based on variation in one or more extracted features within a predefined threshold. 14. The system as claimed in claim 10, wherein the image processing means verifies if variation in the extracted features is due to one or more predefined factors without limitations including aging, time, illness, type and orientation of writing instrument and differences in signature size. 15. A computer program product for automatic signature verification, comprising one or more computer readable media configured to perform the method as claimed in any of the claims 1-9.

Full Text

System and method for automatically verifying the authenticity of
signatures
Field of Invention
The present invention relates to an efficient and cost-effective method and system for signature verification which is largely dependent on offline features but also uses online features for automatic signature verification. The present invention relates generally to the field of automatic image processing, focusing in particular on financial instruments such as bank checks and drafts.
Background of the Invention
A large volume of work has been performed to evolve efficient methods and systems of accurate identity verification. In general, the problem attempted to be solved in identity verification is: given a plurality of identifying reference samples and a test sample, does the test sample belong to the owner of the reference samples? The samples may be signatures, biometric data such as voice samples, face pictures or other such relevant information.
Modern technology offers a number of new, sophisticated and secure authentication mechanisms such as biometrics. However, these are restricted to only specific and specialized segments of particular industries critical to human life and safety such as nuclear programs, space programs, intelligence agencies, and other capital-intensive program which can justify the costs and the efforts required for their implementation. For other environments which do not necessitate the use of such complex means, signature authentication offers a quick, simple and cost-effective means for correctly establishing identity of an individual. In particular, financial institutions such as banks and insurance companies still rely heavily on traditional method of signature verification for effecting transactions done through paper instruments such as checks, drafts, etc. The sheer volume of such instruments makes manual verification a time and cost-intensive process and mandates companies to have an automated system, which could automatically and quickly establish the veracity of the signatures, and thereby affect a valid transaction.
Two main approaches have been adopted for automated signature verification - Offline Signature Verification and Online Signature Verification.
Offline Signature is the most popular approach for signature verification owing to its lower costs. However, conventional offline approach of verification

offers results of relatively low accuracy. Unconventional and individual writing styles make detection and segmenting of signature strokes difficult. The problem is compounded if signature images are of low resolution. Absence of any dynamic and temporal information makes offline signatures much harder to detect.
In contrast to the offline approach, online signature verification provides relatively better results in establishing the authenticity of the signature. This is because online verification uses a plurality of parameters associated with stylus and an electronic writing pad for determining the authenticity of the signature. These parameters may include speed, direction, pressure of the stylus, number and order of the strokes, etc. These properties impart certain distinctive features to the signature of each individual, which are quite difficult to forge. However, the practical implementation of this approach requires the use of stylus and electronic pads along with other relevant instruments for signature capture which add up to high infrastructure costs. Moreover, paper instruments such as checks and documents still form the majority of volume.
In light of the above, several approaches have been suggested to counter the disadvantages of various offline and online signature verification systems at various points.
U.S. Patent No. 5,251,265 describes a method and system for automatic offline signature verification. The apparatus digitizes an image of the signature and provides a gray scale representation of each pixel. Next, the apparatus determines parameters of this digitized image and compares them to corresponding reference parameter values, which were determined from a valid signature. The results of the comparison indicate if the signature to be verified is valid.
U.S. Patent No. 5,995,953 describes a method for verification of signatures and handwriting based on comparison of extracted features instead of the images themselves. Thus it focuses on the offline method of verification. It uses a neural net for image recognition. This leads to a significant reduction of storage space and calculation time needed.
U.S. Patent No. 4,985,928 describes a signature forgery detection device. The system relies on the collection of measurements of the optical density of a plurality of elements in the specimen and selectively retaining said measurements and then comparing each density measurement to a threshold value. Only those density measurements are retained which are located within a bounded locus. After the retained density measurements have been established,

they are automatically compared to the measurements of the specimen signature to permit the apparatus to accept or reject the signature on the presented document
U.S. Patent No. 6,424,728 describes an automatic signature verification system that utilizes a main routine for comparing signatures using forensic hand writing methodology. A secondary program is used to modify the algorithms used by the main program for making adjustments thereto based on either additional data consisting of a plurality of genuine or authenticated signatures or changes in a person's signature due to ageing or some other physical change resulting in a change in signature features.
The above-mentioned patents suggest various approaches of online or offline verification of signatures. However, these methods are more or less akin to traditional approaches of signature verification using simple signature matching or feature extraction techniques. This limits the ability of the existing techniques to accurately verify an individual's signature. Moreover, such techniques are quite time-intensive. Also, some of these methods involve a lot of manual intervention, thereby making the process inefficient for being applicable in an industrial environment. Further, these systems also do not have a built-in intelligence to detect the changes in the signature of a user due to natural factors like changes in handwriting patterns due to age, time, illness, etc. With the growing volume of checks and other similar, signature-bearing instruments, these methods, thus, may not be able to process the instruments as per the exacting requirements of the banks.
Hence, there is a need for a method and system that provides an efficient and cost effective system for automatic signature verification. The system and method should be able to provide accurate results by taking into account person-specific confidence thresholds. The system should also have in-built intelligence for constant learning to handle scenarios where a valid signature is identified as invalid by the system on account of changes induced in the signature image due to factors such as age, illness, orientation, etc. Such a method should incorporate techniques a plurality of unique offline characteristics and preferably various online verification techniques as well to provide better results in all operating environments. Such system should also be able to handle scenarios such as a Bank Mandate which requires verification of more than one person's signature. In addition, said system should be language independent.

Objects and Summary of the Invention
The present invention has the objective to provide an efficient and cost effective system and method, which is largely dependent on offline features but also uses online features for automatic signature verification.
It is an objective of the present invention to provide a system and method for automatic signature verification with in-built intelligence to detect both inter-person and intra-person changes and ascertain if the variations in the signature are induced owing to natural parameters such as age, health and orientation and use this information for continual learning.
It is also an objective of the present invention to provide a means to authenticate a signature against more than one reference signatures in case multiple users are authorized to access and operate the relevant information such as in a Bank Mandate scenario prevalent in operating corporate bank accounts.
It is yet another objective of the present invention to provide a language independent method and system of automatic signature verification.
To achieve the foregoing and other objectives obvious to a person skilled in the art, the present invention provides a method of automatic signature verification comprising the steps of:
- receiving user input of at least one signature;
- extracting one or more features and associated information from the signature using offline and online extraction techniques;
- determining if the information can be used for updating a reference database;
- verifying the extracted features against predefined information stored in the reference database for authenticating the signature; and
- updating the reference database.
The present invention further provides for a system for automatic signature verification comprising:
- acquisition means for receiving one or more input signatures;
- image processing means coupled to acquisition means for extraction of one or more features from the signature and verifying said features against one or more predetermined features associated with at least one reference signature;
- storage means coupled to image processing means and acquisition means for storing the input signatures, extracted features, reference signatures,

associated information for each user account, and operating instructions for automatic signature verification.
Brief Description of Drawings
The detailed description is described with reference to the accompanying figures. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. The same numbers are used throughout the drawings to reference like features and components.
Figure 1 illustrates an exemplary system for automatic signature verification as disclosed by the present invention.
Figure 2 illustrates a flow diagram of an exemplary method of pre processing an acquired image.
Figure 3 illustrates an exemplary method of feature extraction from a pre processed image.
Figure 4 illustrates an exemplary method of extraction of information from a reference database required by a user of the disclosed system.
Figure 5 illustrates an exemplary method for verification of signatures as disclosed by the present invention.
Figure 6 illustrates an exemplary method for dynamically updating the signature database.
Figure 7 illustrates an exemplary method for manually verifying and updating of the signature database.
Detailed Description of the Invention
Systems and methods for efficient means of automatic signature verification are described. The description of the system or method shown herein below is intended only for illustration and disclosure of an operative embodiment and not to show all of the various forms or modifications in which this invention might be embodied or operated, since the same may be modified in various particulars or relations without departing from the spirit or scope of the claimed invention.

The system and method of the present invention aim to provide automatic signature verification, which is largely dependent on offline features but also uses online features for verifying signatures in particular borne by financial instruments such as bank checks. The financial instrument bearing the signature is scanned. Before extracting the relevant features, the scanned image undergoes one or more preprocessing steps for noise removal and orientation correction. Subsequendy, a plurality of features is extracted from the processed image using offline as well as online feature extraction techniques. The features extracted from the signature image are compared against pre-determined features from one or more reference signatures stored earlier in a reference database. While comparing a signature specimen against reference signatures, if certain variations in specific features are determined which are within predefined permissible limits, the variations are incorporated into the reference database for future processing. A further verification is done to check if the request conforms to a predetermined pattern for the specific user. On positive verification of both, the user signature is authenticated. In case the variations in the features do not fall within the predefined permissible limits, the signature is sent for manual verification. On positive verification, the sample is added to the signature database for future processing. Thus, the method and system of the present invention display inherent intelligence by constant learning and also facilitates development of a dynamic signature database using both offline and/or online feature extraction techniques with time. Moreover, the disclosed method of the present invention is also language independent. Further, the present invention can be easily adapted to operate in environments requiring one-to-many signature verification.
An exemplary implementation of the system and method of the present invention is discussed in the following section with respect to the accompanying figures.
Figure 1 illustrates and exemplary system 100 for automatic signature verification. In one implementation, system 100 can include acquisition means 102 coupled to image processing means 104 and reference database 106. In some embodiments, image processing means 104 and reference database 106 may be communicatively coupled to acquisition system 102 over a wired or wireless communication network. Acquisition means 102 can further include one or more digital input devices 108 and one or more imaging devices 110. One or more interfaces for coupling of said devices may also be part of the acquisition system 102. A signature specimen to be authenticated can be input into system 100 by using digital input devices 108 such as tablet pens and transducer pads. Further, imaging devices 110 such as a camera and/or scanner can be used to receive signature specimens from a user. The information

acquired by the acquisition system 102 can be stored in reference database 106 along with operating instructions for facilitating signature authentication as disclosed by the method of the present invention described herein below.
Image processing means 104 can further include means to process the acquired/scanned signature image information in order to remove noisy elements, conversion of said image into binary format, thinning, orientation detection, rotation and scaling. Image processing means 104 can facilitate extraction of relevant features from the processed image using one or more offline as well as online feature extraction techniques as described below. Further, the extracted features can be verified against predetermined features of reference signatures stored in the reference database 106 to authenticate the input signature as a valid means for authorizing a user transaction.
To this end, reference database 106 stores predetermined reference signature specimens and values of associated features of the same for each user account. Further, specific information regarding the type of the account being individual or Bank Mandate, sufficiency of data for verification, specific set of features to be used for verification, whether automatic verification can be performed, and other transactional information for specific user accounts can be stored. Based on the comparison of the stored information for each user account and the relevant received user input, a comparison of relevant features can be performed to authenticate the received user signatures.
In an embodiment, a signature card can be signed by the depositor upon opening an account at a financial institution. The card establishes the type of account ownership and sets forth the account terms and the obligations of the customer and the institution. Companies for subsequent identification of the customer use signature cards. The user provides his signatures using a digital input device 108 and/or the same is captured by an imaging device 110. A plurality of relevant features are extracted by image processing means 104 from the reference signatures received using digital input device 108 using online extraction techniques namely, "Pen out-of or in-contact with paper", "Pen-down time to total time ratio", "Current and previous position of pen", "Pen down, Pen up", "Length of signature", "Total time of signature", "Tangent at each point in Signature", "Velocity and acceleration at each point of signature" and so on.
Further features can also be extracted by image processing means 104 using offline extraction techniques described below such as height, curve comparison, critical points using contours, projection, holes, smoothness index, shape matrix, grid density, radial transition, wrinkleless and so on from the

scanned image. One or more features are determined to be relevant for signature verification for each user account. Values of the extracted features and information regarding which features are to be used for verification for each user account is stored in the reference database 106. When a signature specimen is received for authentication, relevant features are captured by acquisition system 102 using a digital input device 108 and/or imaging device 110 the values of the features are then compared with the stored values for the specific user. System 100 has built-in learning capabilities, which enables it to successfully detect intra-person variations to mark the signature as genuine, while at the same time discarding the signature by detecting inter-personal variances using operating instructions stored in reference database 106. The system 100 automatically recognizes changes in a person's signature owing to various factors such as aging, time, illness, type of writing instrument, differences in signature size (due to space allotted for signature), etc., and updates these changes in a database to verify the authenticity of the signature for future processing. Further, based on the account type, system 100 can also authenticate more than one signatures associated with a single account if the Bank Mandate feature is enabled for the same.
Figure 2 illustrates an exemplary method of processing a scanned image prior to application of a plurality of feature extraction techniques as disclosed by the present invention. At 202, scanned image acquired by an acquisition means. At 204, the acquired image is processed by a preprocessing system. At 206, the noisy elements in the image are removed. At 208, the image is converted into binary format. At 210, using one or more techniques the binary image undergoes thinning. At 212, the angle of orientation of the image is determined. Subsequently, at step 214, the image is rotated by an appropriate angle as determined during orientation detection. At 216, the appropriately oriented image is scaled. The scaled image undergoes further binarization and thinning at steps 218 and 220. The resultant image is subjected to a plurality of feature extraction techniques as described below.
Figure 3 illustrates an exemplary method of application of static and dynamic feature extraction techniques for feature extraction as used in the present invention. At 302, a user's signature specimen is received for authentication. At step 304, the user signature is inputted using a digital pad while at step 306 a signature card with reference signatures is used for verification. The information received from digital pad as well as signature card or similar means can be used for applying a combination of online and offline techniques of feature extraction to the signature specimen.

At 308, a plurality of dynamic feature extraction techniques is employed. One or more parameters are considered for the determination and extraction of relevant features from the signature specimen to be authenticated. In an embodiment, one of the parameters can be the pressure induced by the signatories' pen when signing on transducer pad produces an analogue voltage, which is converted to digital form. These digital values are processed to determine numerical parameter values, each representing a feature of the signature. The user input of signature on a digital pad can facilitate extraction of a plurality of features from the specimen for verification against one or more reference signatures stored in a signature database. These can be "Pen out-of or in-contact with paper", "Pen-down time to total time ratio", "Current and previous position of pen", "Pen down, Pen up", "Length of signature", "Total time of signature", "Tangent at each point in Signature", "Velocity and acceleration at each point of signature" etc.
At 310, various static feature extraction techniques are employed, particularly tangent of the signature is extracted and analyzed. This feature traces the path of the test signature. Once the path of the signature is known, it becomes easy to determine the authenticity of the signature. Further a plurality o{ novel techniques as described below are used to verify a given signature specimen against stored reference signatures.
HEIGHT: Height of individual parts of the signature image, divided along the signature breadth into equal parts is used for verification. The average of heights of individual parts of all the sample images is compared with the height of the corresponding part of the signature image to be verified.
CURVE COMPARISON: Curve Comparison works by generating chain codes. A plurality of points is chosen randomly on the trajectory of the signatures and these two-dimensional coordinates are then reduced to one-dimensional codes and the values are compared to that of the database values.
CRITICAL POINTS USING CONTOURS: Pixel density, for various regions of the signature if vertically incident light rays are hypothetically thrown on the signature image, is determined. The pixel density of the various regions is compared to the average pixel density of the sample authentic images. This feature tests the similarity or confidence value against a threshold value. Critical points are chosen on the trajectory of the signature based on the deviation angle. A LCS (Longest Common Sub String) is formed from these critical points. Based on the comparison of this length with a certain threshold, the authenticity of the check image is checked.

PROJECTION FEATURE: Projection feature also works by comparing the similarity or confidence value against a threshold value. This feature is extracted by rotating the signature from -X° to +X° with a variation of 8 each time. The width of the image is accordingly adjusted to a fixed size and a Normalized Cross Correlation is calculated for all the authentic images.
HOLE FEATURE: Another unique feature is the Hole feature in which the number of closed loops in the signature are verified. Comparison of the check image with the authentic image is done based on the number of holes, positions and sizes of the holes.
SMOOTHNESS INDEX: Smoothness is a vital parameter in ascertaining if the signature is genuine. This feature is mainly used to detect skilled forgery. Signature forgeries are broadly classified into three categories — Random Forgery, Simple Forgery and Skilled Forgery. Random forgery involves forged signature without any knowledge of the original signature. Simple Forgery involves forged signature with a little knowledge of original signature — the knowledge may involve only a cursory glance at the original signature, imitating the original signature based on memory, and other such limited knowledge — but done without practice. Skilled Forgery involves forged signature with good knowledge of original signature — the knowledge may involve constant availability of the original signature to minutely study the nuances of the signature, well-practiced imitation of the original signature, and so on — and close imitation of the original signature after practice. Therefore, Skilled Forgery is hardest to detect. The basic aim behind employing the feature, Smoothness Index, is to compare the steadiness of the hand while signing. While doing Skilled Forgery, the imitator tries to closely match the original signature; however, he/she essentially loses out on smoothness of curves present in the signature - even if he/she succeeds in correctly imitating smoothness of certain curves, it is nearly impossible to perfectly imitate the smoothness for all the curves present in the signatures. Also, signatures are highly behavior oriented as they can vary greatly depending on external factors like age, physical comfort, and illness etc. So, the signature of an aged man if forged by a young man will have a vast difference in their smoothness. The system of the present invention measures this smoothness index for the various signatures to determine their authenticity. However, at the same time, the disclosed system is able to determine the minor variations in various original signatures of an individual that may result due to factors previously mentioned.
SHAPE FEATURE: The shape feature is based on combination of well-known features, such as baseline value, image area, aspect ratio, etc., which are extensively used by current signature verification methods. For calculating the

baseline value, first the centroid of the whole image is calculated. Then, the image is divided vertically along its centroid, and the centroids of these two parts are also calculated. The angle subtended by the line joining the centroids of both the parts with the x-axis is called the base angle and the line is called the base line. Aspect ratio is defined as the height to width ratio of the image whose orientation has been corrected. This feature filters out a large number of random forgeries. The calculation of the image area normalizes against variations in scale of the signature. It is calculated by the number of black pixels in the image divided by the width of the image.
DIRECTIONAL PROBABILITY DENSITY FUNCTION: The directional probability function determines the angle of a point with all the other points on the signature. Based on a sampling done at regular intervals of angles, the probability of going in a particular direction is determined. The process is repeated for all the points on the signature and a graph denoting the probability density is drawn. This graph provides the overall representation of the directional probability for various points on the signature.
CROSS POINTS: The visible overlaps of a person's signatures are known as cross points. All the major cross points in a signature are extracted and matched with those of original signatures using graph matching and triangle matching technique to ascertain the authenticity.
GRID ANALYSIS: This involves dividing the signature into horizontal and vertical boxes such that the numbers of pixels are approximately the same. After the division, the signature is analyzed for the extraction of the following values: a) pixel density, b) 2nd order moment along X-axis, c) 2nd order moment along Y-axis, d) distance of center of mass from the bottom left corner divided by the diagonal of the box, and e) average angle of each pixel with respect to the bottom left corner. Once the above features are extracted, using a modified version of regression, the test signature is compared using the sample signature.
NOVEL TECHNIQUE: The signature image is divided into Vertical Cut First and Horizontal Cut First, based on center of mass of the box surrounding the entire signature image. Each cut provides a box covering a certain portion of the signature. The subsequent center of mass is calculated for every box. A vector based on Euclidean distance classification is formed with these center of mass values for six different sample images. This vector is compared with that of the test signature. If the value is above a certain prerecorded threshold, then the signature is authentic else it is a forgery.

SHAPE MATRIX: The signature is smeared and loops are filled to enhance the efficiency of the verification process. The centroid of the signature image is calculated and the distance between the farthest pixel on the signature and this centroid is considered as the radius. A circle of this radius is drawn and divided into 128 equal parts and the line having maximum number of black pixels is found. A vector corresponding to this line is obtained. The radius of the circle is reduced by two pixels and another circle is drawn, and the same procedure is repeated till the centroid of the image is reached. For each circle, a vector corresponding to line containing maximum number of black pixels is obtained. All the 128 vectors are represented using a matrix known as shape matrix. The shape matrix for the sample image is compared with that of the original image to ascertain the authenticity of the signature.
GRID BASED DENSITY FEATURE: The signature image is first divided into n number of buckets horizontally such that each bucket contains equal number of pixels. The same process is repeated when the signature is divided vertically. Both the images are then superimposed and the ratio of black pixels in the box to the black pixels in the entire image is calculated and a feature vector is obtained. This feature vector for the signature is compared with the vector of the original signature.
RADIAL TRANSITION: Concentric circles are drawn at an increment of every five pixels starting from the centroid of the image as the center of the circles. The circles are drawn till the farthest pixel in the image is reached. The black and white transitions on the circumference on each circle are used to form a matrix. This matrix is compared with that of the test signature to determine the authenticity of the signature.
WRINKLINESS FEATURE: The inner boundary of the signature is taken and logically divided (horizontally and vertically) at an interval of two pixels. The number of squares (2x2) in which the boundary of the signature lies is counted and the log of the number is calculated. The same procedure is repeated at the interval of four pixels. Division of the first number by the second number gives the measure of the wrinkliness of the signature.
GRAPH FEATURE: Three sample signatures are taken. Value of all the pixels on the image is found and a matrix is computed for every sample signature compared with the other, considering all the possible combinations. The maximum value matrix is then taken from the three matrices. The cost matrix of this signature is calculated and compared with that of the original signature. If the cost of the signature is less than that of the original image then the signature is authentic, else it is fake.

IMPRINT DETECTION: In case of skilled forgery, when signatures are traced/topographed, imprints appear at the backside of the check because of the pressure of the pen. When a check with a forged signature appears for verification, by observing the image of the check, the forgery can be detected with the help of these imprints.
RATIO OF SUCCESSIVE PEAKS: The scanned image of the check is first aligned horizontally and then the peaks in the signature are determined. The ratios of each peak with its adjacent peak are calculated for the entire signature image. These ratios are compared with the corresponding ratios of the adjacent peaks of the sample signatures to determine the signature authenticity.
One or more of the above features along with several other such features characteristic of a user's handwriting are extracted from the received signatures are compared against predetermined values of one or more reference signatures as stored in the signature database.
Figure 4 illustrates a flow diagram of an exemplary method of extraction of relevant information to process the verification of the signature parameters against those of reference signatures stored in the signature database. At 402 is represented the database of signatures which includes a plurality of reference signatures and their associated parameters for various users. At 404, the disclosed system on the basis of information acquired from a financial instrument such as a check or draft, the user account is determined. Further it is determined if the user account is an individual account or of the bank mandate type. Further information regarding if sufficient training data is available for verification, the set of features which work for a particular account, whether automatic verification can be carried out on the account, whether the check conforms to regular expenditure pattern and other such verifications is extracted.
Figure 5 illustrates a flow diagram of an exemplary method of signature verification as disclosed by the instant invention. At 502, the database of signatures is employed. At 504, a plurality of relevant features is extracted from the signature image using techniques disclosed in the description of figures 2-4. At 506, it is determined if sufficient training material is available to update the database. At 508, the value of the extracted feature is matched against the predefined threshold value of the said feature as present in one or more associated reference signatures stored in the signature database. In case, the training material is not sufficient, the information is added to the system for

dynamic update as disclosed below in the description of Figure 6 and subsequently manual verification is employed at step 516.
At 512, it is determined if the threshold values of the feature extracted from the test signature matches the threshold value of the same feature present in the reference signature. In case the direshold values do not match, manual verification is employed at step 516. If the threshold values are the same, the database is updated with the sample signature at 514. At 518, it is verified if the check conforms to regular expenditure pattern. If it does, the check is cleared without verification at step 520 else manual verification is employed and the database is updated with the decision for future use in step 522. If in the manual verification, the signature is found a valid match, this knowledge is recorded in the signature database for later use. In such a case, whenever another check is received in the future with the same sample signatures that were earlier rejected are verified and cleared on the basis of the combination of the sample signature and the check amount. Thus, if the amount is small, the signature is verified as authentic automatically by the system, whereas in case the amount is substantial, the signature is referred for manual verification.
Figure 6 illustrates a flow diagram of the technique of dynamically updating the signature database. At 602, sample signatures are stored in the signature database. At 604(a), 604(b) and 604(c) one or more features are extracted from the sample signatures. At 606, each feature is compared with the relevant feature in the sample signatures. At 608, validity is determined on the basis of said comparison. In case, the feature is determined to be invalid, it is not used for the particular user at 610. If the feature is valid, it is stored along with its associated value for a specific user in the signature database at 612. At 614, it is determined if all features have been validated. At 616, the plurality of features determined to be relevant for a specific user and their associated values are used to verify the signatures for that user.
Figure 7 illustrates an exemplary method of manual verification of the financial instrument, in case data for dynamic verification is insufficient, and subsequendy updating the database. With its inherent capability to build intelligence by constant learning, the system provides the capability to handle scenarios where a valid signature is identified as invalid by the system. There might be some changes that have been induced in the signature image due to factors such as age, illness, orientation, etc. These types of cases are ripe for automatic rejection, in spite of them being valid signatures. The system refers such cases for manual verification. At 702, the check is sent for manual verification subsequent to the determination that data for dynamic verification is insufficient. At 704, a valuator verifies the check. A 706, the check is rejected

if the valuator finds some discrepancy in the same. At 708, the valuator finds the check in order and approves the same. A further decision on whether to use the signature for future transactions is made. At 710, the sample is added to the signature database for future reference. At 712, in case the signature sample is determined not to be used for future transactions, the check is passed but the signatures are not added to the database for future reference.
The existing signature verification systems use a signature card for the identification purpose. A signature card is a form signed by the depositor upon opening an account at a financial institution. The card establishes the type of account ownership and sets forth the account terms and the obligations of the customer and the institution. Companies for subsequent identification of the customer use signature cards. However, the banks might not have a consolidated reference database with them to start with for the deployment of such systems. For such a scenario, the disclosed system has in built intelligence to enable the banks to develop a dynamic database gradually with time. The signature samples will be regularly added to the database as and when the bank receives checks for a particular account. In case sufficient signature samples are not present to extract the features required by the system to run in automated verification mode, the disclosed system by default will operate in manual verification mode for that particular account. The system will continue to operate in die manual verification mode until it has enough samples required to switch to automatic verification mode. Once the system has enough samples, without human intervention, system would set the account to operate in automatic verification mode.
A further advantage of our system is that the signature verification system according to the invention is independent of a transducer pad or the like in cases where there is unavailability of a transducer pad. For such cases, the system is able to verify a signature only with the comparison of a plurality of new and unique offline features as disclosed above.
Another major advantage of our invention is that it allows to automatically set an acceptance level for every individual's signature. This acceptance level is set on the basis of the kind of input that is being considered, that is, whether the input signature was acquired offline or online. It also handles special cases of the customers who due to their different handwriting patterns always sign in a different way or have a fraudulent check history. For such customers, the acceptance level for clearing of their checks can be kept sufficiently high.
In addition to the above, a notable advantage of the disclosed method and system is to provide a highly efficient method of signature verification.

Towards this end, the system incorporates capabilities to handle the actual scenarios faced by the banks with regards to automatic signature verification as against a plain-vanilla offering that is equipped to handle only standardized or generic signature-verification scenarios. The instant invention offers a one-to-many signature matching capability as against the one-to-one signature verification. In the current scenario, whenever a signed check reaches a bank, there is normally only a one-to-one signature verification of the user signature by the bank, that is, whenever a particular check reaches a bank, the signatures on that particular check are verified only against the reference signature of the same user that is stored in the database. This is one-to-one signature verification.
The one-to-many signature verification capability makes the disclosed system fully capable of handling the bank mandate feature usually offered by banks to their corporate customers. This feature makes a single account operable by more than one user's signature. A Bank Mandate form provided by the concerned bank to its corporate customers is to be filled by the Company in the presence of its Board of Directors. This form has the signatures of all the signatories who will be responsible for carrying out any transaction for the account. The existing signature verification systems are not able to handle such a case. The disclosed system helps match a single signature with all the reference signatures of the users who are responsible for operating that particular account. Also, whenever a 'Notice of Variation', for updating/changing the signatories, is received by the bank from any of the signatories for the account, our system automatically updates the database of signatures.
Further, the disclosed method and system of the instant invention also offer language independence as novel offline and online feature extraction techniques as described above are employed. This is an added advantage as many individuals, whether on account of their personal choices or unfamiliarity with the language, sign in their regional languages.
The embodiments described above and illustrated in the figures are presented by way of example only and are not intended as a limitation upon the concepts and principles of the present invention. As such, it will be appreciated by one having ordinary skill in the art that various changes in the elements and their configuration and arrangement are possible without departing from the spirit and scope of the present invention as set forth in the appended claims.
It will readily be appreciated by those skilled in the art that the present invention is not limited to the specific embodiments shown herein. The

techniques described hereinabove can be adapted for verifying any handwriting sample of a specific user and the scope of the same is not restricted to just signature verification. Thus variations may be made with in the scope and spirit of the accompanying claims without sacrificing the principal advantages of the invention.
We claim:
1. A method of automatic signature verification comprising the steps of:
- receiving user input of at least one signature;
- extracting one or more features and associated information from the signature using offline and online extraction techniques; determining if the information can be used for updating a reference database;
- verifying the extracted features against predefined information stored in the reference database for authenticating the signature; and
- updating the reference database.

2. The method as claimed in claim 1, wherein the step of receiving includes receiving the signature using a digital input device and an imaging device.
3. The method as claimed in claim 1, wherein the step of extracting includes the steps of:

- removing noisy elements,
- conversion of the signature image into binary format;
- thinning and orientation detection; and
- rotating and scaling of the signature image.

4. The method as claimed in claim 1, wherein the step of determining includes the step of manual verification if the information is insufficient for updating the reference database.
5. The method as claimed in claim 1, wherein the step of verifying includes the step of:

- determining if variations in the features are within predefined threshold values; and
- determining if the variations in the features are due to one or more predefined factors; and
- determining if the associated information conforms to a predetermined pattern.

6. The method as claimed in claim 1, wherein the step of verifying includes the step of positively authenticating a signature against more than one reference signatures stored in the reference database for a user account of predefined account type.
7. The method as claimed in claim 1, wherein the step of verifying includes the step of determining if the variations in the features are due to one or more predefined factors without limitations including aging, time, illness, type and orientation of writing instrument and differences in signature size.
8. The method as claimed in claim 1, wherein the online feature extraction techniques without limitation include pen out-of contact, pen in-contact with paper, pen-down time to total time ratio, current and previous position of pen, pen down, pen up, length of signature, total time of signature, tangent at each point in Signature and Velocity and acceleration at each point of signature.
9. The method as claimed in claim 1, wherein the offline feature extraction techniques without limitation include height, curve, critical points using contours, projection, hole, smoothness index, shape, direction probability density function, cross points, grid analysis, vector based distance classification over vertical and horizontal cut sections, shape matrix, rid density, radial transition, wrinkliness, graph using pixel values of signatures, imprint detection, ratios of successive peaks.
10. A system for automatic signature verification using both offline and online feature extraction techniques comprising:

- acquisition means for receiving one or more input signatures;
- image processing means coupled to acquisition means for extraction of one or more features from the signature and verifying said features against one or more predetermined features associated with at least one reference signature;
- storage means coupled to image processing means and acquisition means for storing the input signatures, extracted features, reference signatures, associated information for each user account, and operating instructions for automatic signature verification.
11. The system as claimed in claim 10, wherein the acquisition means includes
one or more digital input devices and one or more imaging devices.

12. The system as claimed in claim 10, wherein the image processing means
further includes means to:
- remove noisy elements;
- convert signature image into binary format;
- orientation detection; and
- rotating and scaling the image.
13. The system as claimed in claim 10, wherein the storage means is updated
based on variation in one or more extracted features within a predefined
threshold.
14. The system as claimed in claim 10, wherein the image processing means
verifies if variation in the extracted features is due to one or more
predefined factors without limitations including aging, time, illness, type
and orientation of writing instrument and differences in signature size.
15. A computer program product for automatic signature verification,
comprising one or more computer readable media configured to perform
the method as claimed in any of the claims 1-9.

Documents:

http://ipindiaonline.gov.in/patentsearch/GrantedSearch/viewdoc.aspx?id=aqa86wC0STKtk6Tm7qIiog==&loc=egcICQiyoj82NGgGrC5ChA==

« Previous Patent

Next Patent »

Patent Number

279299

Indian Patent Application Number

1189/CHE/2008

PG Journal Number

03/2017

Publication Date

20-Jan-2017

Grant Date

17-Jan-2017

Date of Filing

15-May-2008

Name of Patentee

NEWGEN SOFTWARE TECHNOLOGIES LIMITED,

Applicant Address

BROOKLYN BUSINESS CENTRE,5TH FLOOR,EAST WING, 103-105 PERIYAR E V R ROAD, CHENNAI - 600084.

Inventors:

#	Inventor's Name	Inventor's Address
1	MR.VIRENDER JEET	C/O NEWGEN SOFTWARE TECHNOLIGIES LIMITED, OF BROOKLYN BUSINESS CENTRE, 5TH FLOOR, EAST WING, 103-105 PERIYAR E V R ROAD, CHENNAI - 600084,
2	MRS. PUJA LAL	C/O NEWGEN SOFTWARE TECHNOLIGIES LIMITED, OF BROOKLYN BUSINESS CENTRE, 5TH FLOOR, EAST WING, 103-105 PERIYAR E V R ROAD, CHENNAI - 600084,
3	MR. SIDDHARTH CHABRA	C/O NEWGEN SOFTWARE TECHNOLIGIES LIMITED, OF BROOKLYN BUSINESS CENTRE, 5TH FLOOR, EAST WING, 103-105 PERIYAR E V R ROAD, CHENNAI - 600084,
4	MR. RAJU GUPTA	C/O NEWGEN SOFTWARE TECHNOLIGIES LIMITED, OF BROOKLYN BUSINESS CENTRE, 5TH FLOOR, EAST WING, 103-105 PERIYAR E V R ROAD, CHENNAI - 600084,
5	MR. DINESH GANOTRA,	C/O NEWGEN SOFTWARE TECHNOLIGIES LIMITED, OF BROOKLYN BUSINESS CENTRE, 5TH FLOOR, EAST WING, 103-105 PERIYAR E V R ROAD, CHENNAI - 600084,
6	MR. PROMOD KUMAR,	C/O NEWGEN SOFTWARE TECHNOLIGIES LIMITED, OF BROOKLYN BUSINESS CENTRE, 5TH FLOOR, EAST WING, 103-105 PERIYAR E V R ROAD, CHENNAI - 600084,
7	MR. NEERAJ JAIN,	C/O NEWGEN SOFTWARE TECHNOLIGIES LIMITED, OF BROOKLYN BUSINESS CENTRE, 5TH FLOOR, EAST WING, 103-105 PERIYAR E V R ROAD, CHENNAI - 600084,
8	MR. SOPAN SANT,	C/O NEWGEN SOFTWARE TECHNOLIGIES LIMITED, OF BROOKLYN BUSINESS CENTRE, 5TH FLOOR, EAST WING, 103-105 PERIYAR E V R ROAD, CHENNAI - 600084,

PCT International Classification Number

H04L09/00

PCT International Application Number

N/A

PCT International Filing date

PCT Conventions:

#	PCT Application Number	Date of Convention	Priority Country
1			NA