WO2018117791A1 - Method for pre-processing the image of a signature using artificial vision - Google Patents

Abstract

The verification of the authenticity of a signature captured on an official banking document is an important activity, given that the possibility of fraud is high. The present application shows a method used prior to the analysis of a signature so as to subsequently confirm the legitimacy of same. The method corresponds to pre-processing the image.

Description

 METHOD FOR PRE-PROCESSAM I AM ON THE IMAGE OF A SIGNATURE

USING ARTIFICIAL VISION.

TECHNICAL FIELD OF THE INVENTION

The present invention has its preponderant field of application in the field of banking security, more specifically in the process of pre-processing the image of a signature in an official document in preparation for the verification of authenticity.

BACKGROUND OF THE INVENTION

The pre-processing stage in a signature verification system can consist of a wide variety of operations that depend, in some cases, on the way in which the information has been captured, that is the signature. These operations include the elimination of noise derived from the scanning process or the medium where the signature was captured, the amplification of the captured signal to make it easier to process, data filtering, image conditioning, signal truncation, normalization , coding of the signature or signed address, detection of the beginning and end of the signature, detection of strokes up or down, segmentation of the signature into components, among others. Some of the pre-processing activities depend on the type of verification system, for example, in dynamic systems (online verification), where the signature information is captured by electronic devices with sensors, it is possible to identify the direction and speed of the strokes, to detect strokes made downwards, upwards, pressure exerted by the signer, etc. which are captured regularly as signals that can be subject to algorithms of filtering, amplification, conditioning, etc. other than those that can be applied in static processes (offline verification). In off-line verification processes, pre-processing usually includes binarization, noise elimination, orientation, reduction or segmentation, normalization, among others. One of the crucial parts of a firm's preprocessing is segmentation. In the case of off-line verification, segmentation is usually done through the use of algorithms that detect the contour of a signature or the use of vertical or horizontal histograms. For its part, the elimination of noise is usually done by applying a media filter. Below are some patents related to the preparation of a signature for analysis:

 US8320674B2 shows a procedure to recognize a text from an image or a video, accurately locating the position of the text. If the localized text is of low resolution it can be extracted, improved and binarized. Finally, OCR technology can be applied to binarized text for recognition.

 Patent No. US20060262352 describes a mixed media reality system (MMR) and associated techniques. The MMR system provides mechanisms to form a mixed media document that includes media of at least two types (printed paper and digital content). The MMR system provides a match-processing of images in a document to compare with data in a database.

 In the invention US7869634 a signature authentication system of a user electronically introduced into the system by a mouse or other manual input device is provided that provides an output indicative of its location when manipulated by the user. The system serves to extract angle and distance data that relate different parts of the user's signature entered into the system and to store corresponding angle and distance data relative to a reference signature previously entered into the system during a training procedure. The extracted data is then compared by the system with the reference data stored by the system and, when appropriate, an indicative output of an appropriate match between the entered signature and the reference signature is provided depending on the result of the comparison. This system provides a dynamic online biometric verification system that can be customized for multiple Internet-based applications that require secure authentication. The system does not require specialized equipment at the point of use, which allows access from any computer capable of the Internet with a mouse and browser compatible with Java, for example.

Patent No. US20060050962 claims a system, process and software for recognizing manuscript characters, where character data is obtained that includes information indicative of at least one manuscript character. The data from Characters include a set of segmentation points for the manuscript character. A score for each particular character of a previously stored character set can then be provided, based on a comparison between the character data and the particular character previously stored.

 Patent No. US20060002593, details in its invention an information processing apparatus that detects the sampling rate of a coordinate input device (digitizer) and normalizes the writing data that is entered from the base coordinate input device at the sampling rate detected. Standardized writing data is also used for signature verification or for the recognition of handwritten characters.

 The invention CN 103593673 is based on the online handwritten signature authentication method for dynamic threshold, pertaining to the field of information security, the method of the present invention is used to coordinate the online information model of signature, pressure, Integrated speed and consider several factors to improve the accuracy of a firm. The method of the present invention uses a signature segment discrimination policy, and in accordance with the signature and writing habits, two strategy options are presented for the automatic segmentation of the segment method that can improve the signing process. The threshold signature method of the present invention is determined in the test procedure to help reflect the bias characteristics of each signature, taking into account the stability characteristics of each person's signature.

In the invention US Pat. No.4710822 describes a method for image processing in which a figure is divided into a plurality of blocks. Each block comprises a variety of image elements that are classified into a first group consisting of image elements of densities not less than a reference density and a second group consisting of image elements of densities lower than the reference density. A representative density of image elements of the first group and the second group in each block is obtained. An image discrimination is made in accordance with representative densities and then a determination of a Threshold value according to representative densities and the result of image discrimination. Image discrimination and determination of threshold values are carried out using histograms made from the representative density of image elements of the first group in each block and the representative density of image elements of the second group in each block. The image information is then converted into binary signals according to the threshold value.

 In US patent 5077805A a character recognition identifier is developed based on the function and the level of confidence that are determined for an unknown symbol. If the level of trust is within an intermediate range, the feature-based identification is confirmed by matching the unknown character with a reference template corresponding to the feature-based identification. If the confidence level is below the intermediate range, template matching character recognition is replaced instead of feature-based identification. If template match recognition identifies more than one symbol, corresponding templates from a second set of templates that have thicker character strokes are used to resolve ambiguity.

 In US4594732A a method and an apparatus for determining the passage of a character of an aligned group of images on the surface of a paper are described, where the group contains several characters that touch or merge with each other. Separable letter blocks are extracted from the aligned group of characters and the size and location of each is detected and stored. Based on this information, an allowable range for a block width of one letter is determined and the width of letter blocks that occurs most frequently in that allowable range is determined. Based on the width of the block of letters that occurs most frequently. An effective interval is determined to determine the pitch of the letter and a numerical method is used to estimate the tone of the letter.

In US5859929A an effective system and method is mentioned for reliably identifying guidelines, governed lines and the like, in text images, to distinguish those portions of lines that touch or intersect character strokes in the image. The system facilitates the elimination of guide segments between character strokes without deleting character strokes. The system works effectively on both machine printed text and handmade text. A text image has the most separate characters for more efficient subsequent OCR processing.

 In US6438265B1, a binary method used in an OCR system to determine the pixels of a text is presented, checking for each pixel the difference between its value and the values of a plurality of pixels located at a predetermined distance thereof is greater than a relative threshold corresponding to the difference in intensities between the text and the background of the image. The image is subsampled at a corresponding rate of at least two pixels to detect text cores and then binarize the image pixels only in multi-sided mosaics of the stroke containing text cores, using in each frame an absolute threshold estimated in said picture. The determination of pixels in a text includes, for each pixel analyzed, the verification that any of the differences between the value of the analyzed pixel and the value of the two pixels located at each intersection of a circle with each of the row line , column line and both lines at the angle of 45 degrees, is greater than the relative threshold at which said circle is centered on the position of the analyzed pixel and has a radius equal to the width of the stroke.

A method of interpreting visual information with alphanumeric characters is presented in WO2011080361 A1. The method begins with a digital image, converted to grayscale, segmented so that a black and white image formed by a plurality of particles is obtained; filtering said plurality of particles removes particles that do not contain information associated with a character of the original image; a dilated image is obtained to select segments, trying that each segment corresponds to a character of the original image. Finally, the information of these segments is interpreted by means of a character recognition algorithm.

The invention US20100303356 provides a method for an Optical Character Recognition (OCR) system that provides recognition of characters that are partially hidden by outputs due to, for example, a stamp print, handwritten signatures, etc. The method establishes a set of template images recognized from the image of the text that is being processed by the OCR system, in which the effect of the crossed out section is modeled on the template images before comparing these images with the image of a visually impaired crossed out character. The modeled template image that has the highest similarity with the crossed out character with visual impairments is the correct identification for the instance of the visually impaired character.

DETAILED DESCRIPTION OF THE INVENTION

The characteristic details of the present invention are clearly shown in the following description and in the accompanying figures, which are mentioned by way of example, and therefore should not be considered as a limitation for said invention.

Brief description of the figures:

Figure 1 shows the signature location window;

Figure 2 shows an example of deletion of check template information important features in a signature;

Figure 3 shows the activities of the pre-processing stage;

Figure 4 shows the resulting image at the end of the preprocessing process;

Figure 1 shows the result of applying the Signature Extraction software to an official document, in this case a check. The first element is Signature Window Manager is responsible for identifying a window in the check that corresponds to the area where the signature is expected, as well as making adjustments to that window according to user interaction. Once the window has been confirmed, the Signature Window Cleaner class is responsible for deleting the check template information found in the signature area, so that it can be processed later. When identifying the template, the user is shown an image of the check with a box that marks the window where the signature is expected, and is asked to confirm that the box encloses the signature in its entirety. Figure 2 shows the next step in the process: once the signature area is extracted, the same template area is extracted of the check, to identify the existing elements in the template and eliminate them from the signature area. It identifies the places where it is expected to have lines, or texts, and seeks to eliminate such information. However, it is important to indicate that the removal of all existing information in the template could also remove information from the firm (See figure 2b). To reduce the risk of loss of such information, a simple algorithm has been applied that verifies adjacent pixels in search of information from the firm, so that the result resembles that shown in (Figure 2 c). Figure 3 exemplifies the process of removing lines belonging to the template. In figure 3 a), the section corresponding to the template is shown, indicating the line to be deleted; Figure 3 b) shows the section of the signature from which you want to delete template information, and Figure 3 c) shows the result once the template information has been deleted. Figure 4 shows the 4 stages of the pre-processing process of the firm: binarization, thinning, orientation and edge removal. Once the edge removal process is applied, the resulting image is like the one shown in Figure 5.

Claims

 1. -A procedure developed in different stages for the preparation of the image of a firm, for which its authenticity will be verified later, characterized by:  • A binarization method for converting the signature image to a black and white scale;  • A method of thinning the signature for the normalization of the strokes that are part of a signature;  • A signature orientation method;  • A method of removing edges to eliminate unnecessary areas in the signature strokes;  2. The system according to claim No. 1, wherein an electronic device is used to perform the binarization process on a black and white scale;  3. The system according to claim No. 2, wherein the selection of a black and white format is justified by occupying less space, since each pixel encodes it with one of two values (ie 1 bit) 1 to indicate the white color, and 0 for the black color;  4. The system according to claim No. 3, wherein one of the central points in the black and white scale coding is the limit or "threshold"; 5. The system according to claim No. 4, wherein to perform the conversion a limit must be established that represents a reference value from which the pixels are considered white or black;  6. The system according to claim No. 5, wherein a limit of 0.5 is selected, where the values below said limit are considered black, and the rest white or vice versa as the case may be;  7. The system according to claim No. 1, wherein the method of thinning or "Thinning" allows to eliminate the problems derived from the comparison of signatures with different line thicknesses; 8. The system according to claim No. 7, wherein the process "thins" the lines to occupy a single pixel in width, while maintaining the main characteristics of the signature form; 9. The system according to claim No. 1, wherein perfectly oriented signatures can greatly assist in the operation of the validation algorithm, so seeking adequate guidance is an important task during pre-processing;  10. The system in accordance with claim No. 9, where a check would expect the signature to be oriented, but in practice situations that lead to the signer may occur to minor or major changes in the orientation of your signature, by example, by signing in different positions;  11. The system according to claim No. 10, wherein it is chosen to use the algorithm that seeks a baseline of the signature and rotates it to place it parallel to the x-axis;  12. The system according to claim No. 11, wherein the algorithm assumes three things: 1) that the signature must be written from right to left or from left to right, 2) that the background noise has been eliminated more possible, and 3) that the input image has already been binarized;  13. The system according to claim No. 12, wherein the orientation of an image as a function of the pixels it contains, can be obtained by the least squares technique, for example, applied in the approximation of a line, a polynomial or an ellipse;  14. The system according to claim No. 1, wherein the last step in the pre-processing stage is the removal of edges;  15. The system according to claim No. 14, wherein the process of removing edges causes a reduction in the size of the image, eliminating unnecessary areas within it;  16. - The system according to claim No. 1, where after applying the four methods the signature is ready for validation.