Signature recognition method and device, electronic equipment, storage medium and program product

Abstract

The invention provides a signature recognition method and device, electronic equipment, a storage medium and a program product, and belongs to the technical field of financial science and technology. The method comprises the following steps: acquiring a complete signature sample set, a first incomplete signature sample set obtained by removing strokes from a complete signature image sample for each incomplete signature image sample, and a second incomplete signature sample set obtained by removing an occlusion data set from the complete signature image sample for each incomplete signature image sample; according to the three signature sample sets, generating a target signature sample set in which each target signature image sample comprises signature image samples corresponding to the same complete signature in the three signature sample sets; training a signature integrity recognition model by using the target signature sample set; in the training process, the signature integrity recognition model carries out missing feature transfer learning on each target signature image sample. According to the invention, the accuracy of identifying incomplete signatures by the signature integrity identification model can be improved.

Description

Technical Field

[0001] The present application relates to the field of financial technology, and in particular to a signature recognition method, device, electronic device, storage medium and program product. Background Art

[0002] In the field of financial technology, with the standardization of business processes, such as leasing, most require users to sign business documents. The integrity of the user's signature directly determines the legal validity of the signature. Currently, most signatures are identified through signature integrity recognition models to verify the integrity of the user's signature.

[0003] However, the signature integrity recognition models currently used are mostly trained and developed based on historical signature data. Incomplete signatures are a risk point for default and are extremely rare and difficult to collect in historical signature data. As a result, the current signature integrity recognition models have poor accuracy in identifying incomplete signatures. Summary of the Invention

[0004] The main purpose of the embodiments of the present application is to propose a signature recognition method, device, electronic device, storage medium and program product, aiming to improve the accuracy of the signature integrity recognition model in identifying incomplete signatures.

[0005] To achieve the above-mentioned objectives, a first aspect of an embodiment of the present application proposes a signature recognition method, comprising: obtaining a complete signature sample set, a first incomplete signature sample set, and a second incomplete signature sample set; wherein, each incomplete signature image sample in the first incomplete signature sample set is obtained by removing strokes from the complete signature image sample; and each incomplete signature image sample in the second incomplete signature sample set is obtained by removing an occlusion data set from the complete signature image sample; generating a target signature sample set based on the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set; wherein, each target signature image sample in the target signature sample set includes signature image samples corresponding to the same complete signature in the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set; using the target signature sample set, training a signature integrity recognition model to obtain the trained signature integrity recognition model; wherein, in the process of training the signature integrity recognition model, the signature integrity recognition model performs missing feature transfer learning on each of the target signature image samples; and the signature integrity recognition model is used to identify signature integrity.

[0006] In some embodiments, before obtaining the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set, the signature recognition method further includes: obtaining pixel sets corresponding to each stroke in the first signature image; wherein the first signature image is a complete signature image sample; obtaining pixel values of each pixel in the first pixel set corresponding to the first stroke; changing the pixel values of each pixel in the first pixel set corresponding to the first stroke to background pixel values of the first signature image to obtain a second signature image; and determining the second signature image as an incomplete signature image sample in the first incomplete signature sample set.

[0007] In some embodiments, obtaining a pixel set corresponding to each stroke in the first signature image includes: obtaining each target pixel in the first signature image whose pixel value is within a target range; the target range is determined based on the color channel attributes and background pixel values of the first signature image; and using a region growing algorithm to group each target pixel based on the similarity between the target pixels to obtain the pixel set.

[0008] In some embodiments, before obtaining the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set, the signature recognition method further includes: using a mask to block the fourth signature image, removing the second stroke corresponding to the blocked area of the mask from the fourth signature image, and obtaining a fifth signature image; wherein the fourth signature image is a complete signature image sample; and determining the fifth signature image as an incomplete signature image sample in the second incomplete signature sample set.

[0009] In some embodiments, the masking of the fourth signature image using a mask to remove the second stroke corresponding to the masked area from the fourth signature image to obtain the fifth signature image includes: masking the fourth signature image using the mask to remove the target portion of the second stroke corresponding to the masked area from the fourth signature image to obtain the sixth signature image; and performing an erosion and dilation operation on the sixth signature image using an opening operation to remove the remaining portion of the second stroke to obtain the fifth signature image.

[0010] In some embodiments, the occluded area of the mask is a rectangular area.

[0011] To achieve the above-mentioned purpose, the second aspect of the embodiment of the present application proposes a signature recognition device, including: a first acquisition module, a second acquisition module and a training module; the first acquisition module is used to acquire a complete signature sample set, a first incomplete signature sample set and a second incomplete signature sample set; wherein, each incomplete signature image sample in the first incomplete signature sample set is obtained by removing strokes from a complete signature image sample; each incomplete signature image sample in the second incomplete signature sample set is obtained by removing an occlusion data set from the complete signature image sample; the second acquisition module is used to train a signature recognition device based on the complete signature sample set, the first incomplete signature sample set and the second incomplete signature sample set. A complete signature sample set is used to generate a target signature sample set; wherein, each target signature image sample in the target signature sample set includes signature image samples corresponding to the same complete signature in the complete signature sample set, the first incomplete signature sample set and the second incomplete signature sample set; the training module is used to use the target signature sample set to train the signature integrity recognition model to obtain the trained signature integrity recognition model; wherein, in the process of training the signature integrity recognition model, the signature integrity recognition model performs missing feature transfer learning on each of the target signature image samples; the signature integrity recognition model is used to identify signature integrity.

[0012] To achieve the above-mentioned purpose, the third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the method described in the first aspect when executing the computer program.

[0013] To achieve the above-mentioned purpose, the fourth aspect of the embodiments of the present application proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method described in the first aspect.

[0014] To achieve the above-mentioned purpose, the fifth aspect of the embodiments of the present application proposes a computer program product, which is stored in a storage medium and implements the method described in the first aspect when the computer program product is executed by at least one processor.

[0015] The present application proposes a signature recognition method, device, electronic device, storage medium and program product. On the basis that each incomplete signature image sample in the first incomplete signature sample set is obtained by removing strokes from the complete signature image sample, and each incomplete signature image sample in the second incomplete signature sample set is obtained by removing an occlusion data set from the complete signature image sample, a target signature sample set is generated based on the complete signature sample set, the first incomplete signature sample set and the second incomplete signature sample set, in which each target signature image sample includes signature image samples in the aforementioned three signature sample sets corresponding to the same complete signature, and a signature integrity recognition model is trained using the target signature sample set. During the training process, the signature integrity recognition model performs missing feature transfer learning on each target signature image sample, which can improve the accuracy of the signature integrity recognition model in recognizing incomplete signatures. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Schematic diagram of the signature recognition method provided in the embodiment of the present application;

[0017] Figure 2 This is a schematic diagram of a process for generating an incomplete signature image sample provided by an embodiment of the present application;

[0018] Figure 3 is a schematic diagram of a process for obtaining a pixel set provided in an embodiment of the present application;

[0019] Figure 4 This is another flowchart of generating an incomplete signature image sample provided by an embodiment of the present application;

[0020] Figure 5 Schematic diagram of the structure of the signature recognition device provided in the embodiment of the present application;

[0021] Figure 6 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0023] It should be noted that although the device schematics illustrate functional module divisions and the flowcharts illustrate logical sequences, in certain circumstances, the steps shown or described may be performed in a sequence that differs from the module divisions in the device or the sequence in the flowcharts. The terms "first," "second," and so on, in the specification, claims, and drawings, are used to distinguish similar items and are not necessarily used to describe a specific sequence or precedence.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0025] In the field of financial technology, the signature integrity recognition models currently used are mostly trained and developed based on historical signature data. Incomplete signatures are a risk point for default and are extremely rare and difficult to collect in historical signature data. As a result, the current signature integrity recognition models have poor accuracy in identifying incomplete signatures.

[0026] Based on this, the embodiments of the present application provide a signature recognition method, apparatus, electronic device, storage medium, and program product to improve the accuracy of the signature integrity recognition model in identifying incomplete signatures. This is specifically illustrated by the following embodiments, first describing the signature recognition method in the embodiments of the present application.

[0027] The signature recognition method provided in the embodiments of the present application can be applied to electronic devices or software of electronic devices, and the electronic devices can be terminals or servers. In some embodiments, the terminal can be a smartphone, tablet computer, laptop computer, desktop computer, etc.; the server can be configured as an independent physical server, or as a server cluster or distributed system composed of multiple physical servers, or as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the signature recognition method, etc., but is not limited to the above forms.

[0028] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0029] Figure 1This is an optional flowchart of the signature recognition method provided in the embodiment of the present application. Figure 1 The signature recognition method may include but is not limited to steps S101 to S103.

[0030] Step S101: Obtain a complete signature sample set, a first incomplete signature sample set, and a second incomplete signature sample set; wherein each incomplete signature image sample in the first incomplete signature sample set is obtained by removing strokes from a complete signature image sample; and each incomplete signature image sample in the second incomplete signature sample set is obtained by removing an occlusion dataset from a complete signature image sample;

[0031] Optionally, the complete signature sample set includes multiple complete signature image samples; the complete signature image samples can be obtained by photographing the complete signature signed by the user, or by extracting the complete signature image from historical signature data.

[0032] Furthermore, a complete signature image sample can be randomly selected from the complete signature sample set, and stroke removal processing and occlusion removal data set processing can be performed based on the same complete signature image sample to obtain an incomplete signature image sample in the first incomplete signature sample set and an incomplete signature image sample in the second incomplete signature sample set.

[0033] For each incomplete signature image in the first incomplete signature sample set, the complete signature image sample can be cropped based on the pixel coordinates of each pixel in the first pixel set corresponding to the first stroke in the complete signature image sample, so that the first stroke is removed from the complete signature image sample; the first stroke can also be removed from the complete signature image sample by changing the transparency of each pixel in the first pixel set corresponding to the first stroke in the complete signature image sample to 0; the first stroke can also be removed from the complete signature image sample by changing the pixel value of each pixel in the first pixel set corresponding to the first stroke in the complete signature image sample to the background pixel value of the complete signature image sample. The specific implementation can be found in the relevant description below and is not described here.

[0034] Optionally, the occluded dataset includes at least one occluded stroke in the incomplete signature image sample. For each incomplete signature image sample in the second incomplete signature sample set, the complete signature image sample may be occluded using at least one of mosaic occlusion, blur occlusion, graffiti occlusion, and mask occlusion. The specific implementation is described below and is not described here.

[0035] In addition, when classified according to color channel attributes, the above-mentioned complete signature image samples and the above-mentioned incomplete signature image samples can be any type of binary images, grayscale images, red, green, and blue three-channel (Red Green Blue, RGB) images, and red, green, blue transparency four-channel (Red Green Blue Alpha, RGBA) images.

[0036] Step S102: generating a target signature sample set based on the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set; wherein each target signature image sample in the target signature sample set includes signature image samples corresponding to the same complete signature in the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set;

[0037] In actual implementation, for each target signature image sample in the target signature sample set, a complete signature image sample can be selected from the complete signature sample set. Accordingly, an incomplete signature image sample 1 obtained by removing strokes from the complete signature image sample can be selected from the first incomplete signature sample set, and an incomplete signature image sample 2 obtained by removing the occlusion data set from the complete signature image sample can be selected from the second incomplete signature sample set. Since the incomplete signature image sample 1 and the incomplete signature image sample 2 are both obtained by processing the complete signature image sample, the complete signature image sample, the incomplete signature image sample 1 and the incomplete signature image sample 2 must correspond to the same complete signature. Furthermore, the complete signature image sample, the incomplete signature image sample 1 and the incomplete signature image sample 2 can be combined to obtain a target signature image sample of the above-mentioned target signature sample set.

[0038] Step S103: Using the target signature sample set, the signature integrity recognition model is trained to obtain a trained signature integrity recognition model; wherein, during the process of training the signature integrity recognition model, the signature integrity recognition model performs missing feature transfer learning on each target signature image sample; the signature integrity recognition model is used to identify signature integrity.

[0039] Transfer learning is a machine learning method whose core idea is to improve the performance of a related task (the target task) by reusing knowledge learned in one task (the source task). It aims to address problems with scarce data or limited computing resources, and is particularly effective in cross-domain and cross-task scenarios.

[0040] In actual implementation, each target signature image sample in the target signature sample set can be input into the signature integrity recognition model. The signature integrity recognition model can learn the first stroke missing feature of the incomplete signature image sample from the first incomplete signature sample set and the second stroke missing feature of the incomplete signature image sample from the second incomplete signature sample set for each target signature image sample; then, the signature integrity recognition model can use transfer learning to perform various combinations of the first stroke missing feature and the second stroke missing feature, and can transfer the combined missing feature to the complete signature sample from the complete signature sample set to obtain a new incomplete signature image sample; then, the signature integrity recognition model can continue to be trained based on the new incomplete signature image sample, and finally obtain a trained signature integrity recognition model. The specific implementation can be found in the relevant description below and is not described here.

[0041] In steps S101 to S103 illustrated in the embodiment of the present application, based on the fact that each incomplete signature image sample in the first incomplete signature sample set is obtained by removing strokes from the complete signature image sample, and each incomplete signature image sample in the second incomplete signature sample set is obtained by removing the occlusion data set from the complete signature image sample, a target signature sample set is generated based on the complete signature sample set, the first incomplete signature sample set and the second incomplete signature sample set, in which each target signature image sample includes signature image samples in the aforementioned three signature sample sets corresponding to the same complete signature, and the target signature sample set is used to train the signature integrity recognition model. During the training process, the signature integrity recognition model performs missing feature transfer learning on each target signature image sample, which can improve the accuracy of the signature integrity recognition model in recognizing incomplete signatures.

[0042] See also Figure 2 Before obtaining the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set in step S101, the signature recognition method provided in the embodiment of the present application may further include, but is not limited to:

[0043] Step S201: Obtaining a pixel set corresponding to each stroke in a first signature image; wherein the first signature image is a complete signature image sample;

[0044] Optionally, the first signature image can be obtained by photographing the complete signature signed by the user, or by extracting the complete signature image from historical signature data; when classified according to color channel attributes, the first signature image can be any type of binary image, grayscale image, red, green, and blue three-channel (Red Green Blue, RGB) image, and red, green, and blue four-channel transparency (Red Green Blue Alpha, RGBA) image.

[0045] In actual implementation, the first signature image can be input into the trained stroke segmentation model to obtain the information of each stroke in the first signature image. Further, based on the information of each stroke, the pixels corresponding to each stroke can be extracted to obtain the pixel sets corresponding to each stroke. The region growing algorithm can also be used to group the target pixels based on the similarity between the target pixels in the first signature image to obtain the pixel sets corresponding to the strokes in the first signature image. The specific implementation can be found in the relevant description below and will not be described here.

[0046] Step S202: obtaining pixel values of each pixel in a first pixel set corresponding to a first stroke;

[0047] Optionally, the first stroke may be any one stroke in the first signature image; or may be any multiple strokes in the first signature image.

[0048] In the case where the first stroke is any stroke in the first signature image, the embodiment of the present application can remove the stroke from the first signature image based on information of each pixel in the pixel set corresponding to the stroke to obtain a second signature image missing one stroke.

[0049] In the case where the first stroke is any multiple strokes in the first signature image, the embodiment of the present application can remove the multiple strokes from the first signature image based on information of each pixel in the pixel set corresponding to the multiple strokes, thereby obtaining a second signature image lacking the multiple strokes.

[0050] Step S203: changing the pixel value of each pixel in the first pixel set corresponding to the first stroke to the background pixel value of the first signature image to obtain a second signature image;

[0051] Optionally, the information of each pixel includes at least one of a pixel value, a coordinate value, and transparency of each pixel.

[0052] In actual implementation, the first signature image can be cropped based on the coordinate values of each pixel in the first pixel set corresponding to the first stroke to remove the first stroke from the first signature image to obtain a second signature image; the first stroke can also be removed from the first signature image by setting the transparency of each pixel in the first pixel set corresponding to the first stroke to 0 to obtain a second signature image; the first stroke can also be removed from the first signature image by changing the pixel value of each pixel in the first pixel set corresponding to the first stroke to the background pixel value of the first signature image. The specific implementation can be found in the relevant description below and is not described here.

[0053] Step S204: Determine the second signature image as an incomplete signature image sample in the first incomplete signature sample set.

[0054] In steps S201 to S204 shown in the embodiment of the present application, based on the information of each pixel in the first pixel set corresponding to the first stroke, the first stroke is removed from the first signature image to obtain the second signature image, which can improve the efficiency of obtaining the incomplete signature image samples in the first incomplete signature image samples and reduce the difficulty of obtaining the incomplete signature image samples in the first incomplete signature image samples, so that based on the incomplete signature image samples in the first incomplete signature sample set, the accuracy of the signature integrity recognition model in identifying incomplete signatures can be improved.

[0055] In one embodiment, obtaining pixel sets corresponding to respective strokes in the first signature image includes:

[0056] Obtaining target pixels in the first signature image whose pixel values are within a target range; the target range is determined based on color channel attributes and background pixel values of the first signature image;

[0057] The region growing algorithm is used to group target pixels based on the similarity between them to obtain a pixel set.

[0058] In actual implementation, when the user signs on a preset signature template, the background pixel value of the first signature image can be obtained by obtaining the preset value of the background pixel of the pre-stored preset signature template; when the user signs on a non-preset signature template, the background pixel value of the first signature image can be obtained by performing background pixel analysis on the first signature image.

[0059] Furthermore, the values of each pixel in the pixel set corresponding to other strokes in the first signature image can be kept unchanged, and only the pixel value of each pixel in the first pixel set corresponding to the first stroke can be changed to the background pixel value, so that the first stroke is hidden in the first signature image while the display effect of other strokes in the first signature image remains unchanged, and a second signature image is obtained in which the first stroke is missing compared to the first signature image.

[0060] The embodiment of the present application obtains the background pixel value of the first signature image and changes the pixel value of each pixel in the first pixel set to the background pixel value to obtain the second signature image. This can improve the efficiency of generating incomplete signature image samples in the first incomplete signature image sample and reduce the difficulty of generating incomplete signature image samples in the first incomplete signature image sample. Therefore, the signature integrity recognition model can be trained based on the incomplete signature image samples in the first incomplete signature sample set, thereby improving the accuracy of the signature integrity recognition model in identifying incomplete signatures.

[0061] In some embodiments, the above step S201: obtaining pixel sets corresponding to respective strokes in the first signature image includes:

[0062] Obtaining target pixels in the first signature image whose pixel values are within a target range; the target range is determined based on color channel attributes and background pixel values of the first signature image;

[0063] The region growing algorithm is used to group target pixels based on the similarity between them to obtain a pixel set.

[0064] Optionally, when the color channel attributes of the first signature image indicate that the first signature image is a binary image and the background pixel values of the first signature image are 0, the target range can be determined as pixel values equal to 255, and pixels in the first signature image with pixel values equal to 255 can be determined as target pixels. When the color channel attributes of the first signature image indicate that the first signature image is an RGB image and the background pixel values of each of the R, G, and B channels of the first signature image are all 255, the target range can be determined as pixel values of each of the R, G, and B channels within the range of [0, 255), and pixels in the first signature image with pixel values of each of the R, G, and B channels within the range of [0, 255) can be determined as target pixels.

[0065] In actual implementation, the i-th pixel can be first selected from each target pixel according to the pixel arrangement order of the first signature image. Then, for the selected i-th pixel, a region growing algorithm can be used to group the i-th pixel and the j-th pixel into the same group based on the similarity between the i-th pixel and other pixels in the target pixels. The j-th pixel is a pixel among the other pixels whose similarity to the i-th pixel is greater than a preset value. Then, i can be updated to i+1, and the step of selecting the i-th pixel from each target pixel is returned to execute until all target pixels are grouped to obtain a pixel set. Optionally, the preset value can be determined through experimental calibration to define the degree of similarity between the i-th pixel and the j-th pixel.

[0066] In order to further improve the efficiency of grouping each target pixel, it is possible to determine whether to execute the above-mentioned region growing algorithm for the i-th pixel based on the access status of the i-th pixel, and to divide the i-th pixel and the j-th pixel into the same group based on the similarity between the i-th pixel and other pixels in each target pixel, so as to improve the efficiency of grouping each target pixel. The specific implementation can be found in the relevant description below and is not described here.

[0067] The embodiment of the present application obtains each target pixel in the first signature image whose pixel values are within the target range, and uses a region growing algorithm to group each target pixel based on the similarity between the target pixels. This can accurately determine the pixel sets corresponding to each stroke in the first signature image, thereby improving the effect of generating incomplete signature image samples in the first incomplete signature image sample based on the pixel sets corresponding to each stroke. The signature integrity recognition model is trained based on the incomplete signature image samples in the first incomplete signature sample set, which can improve the accuracy of the signature integrity recognition model in identifying incomplete signatures.

[0068] In some embodiments, the region growing algorithm is used to group target pixels based on similarities between target pixels to obtain a pixel set, including:

[0069] According to the pixel arrangement order of the first signature image, select the i-th pixel from each target pixel;

[0070] When the access state of the i-th pixel is the pending access state, the i-th pixel and the j-th pixel are divided into the same group based on the similarity between the i-th pixel and other pixels in each target pixel using a region growing algorithm; wherein the j-th pixel is a pixel among other pixels whose similarity with the i-th pixel is greater than a preset value; the access state of the i-th pixel and the j-th pixel is updated to the visited state; i is updated to i+1, and the step of selecting the i-th pixel from each target pixel is returned to be executed until all target pixels are grouped to obtain a pixel set;

[0071] When the access state of the i-th pixel is the visited state, i is updated to i+1, and the step of selecting the i-th pixel from each target pixel is returned to be executed.

[0072] Optionally, the pixels of the first signature image may be arranged in a row order or a column order. In actual implementation, the i-th pixel may be selected from the target pixels of the first signature image by traversing the target pixels in each row one by one according to the row order of the first signature image; or the i-th pixel may be selected from the target pixels of the first signature image by traversing the target pixels in each column one by one according to the column order of the first signature image.

[0073] Optionally, the initial access state of each target pixel in the first signature image is a state to be visited. For the i-th pixel, the access state of the i-th pixel and the j-th pixel can be updated to a visited state when executing the step of using a region growing algorithm to divide the i-th pixel and the j-th pixel into the same group based on the similarity between the i-th pixel and other pixels in each target pixel. In actual implementation, a visited matrix of the same size as the first signature image can be created, and based on the execution of the above steps, the visit matrix values corresponding to the i-th pixel and the j-th pixel can be updated in real time. In this way, the access state of each target pixel can be determined by obtaining the visit matrix value corresponding to each target pixel. Taking the i-th pixel as an example, the access matrix value corresponding to the i-th pixel can be true or false, true represents that the access state of the i-th pixel is a visited state, and false represents that the access state of the i-th pixel is a state to be visited.

[0074] Optionally, for the first target pixel selected according to the pixel arrangement order of the first signature image, before grouping the first target pixel, a pixel set (regions) corresponding to the first target pixel can be created, and starting from the first target pixel, other target pixels whose similarity with the first target pixel is higher than a preset value are filled into the pixel set corresponding to the first target pixel through a region growing algorithm (for example, a boundary tracking and segmentation (BTS) method) until the pixel set corresponding to the first target pixel is filled; then, for the next target pixel, a pixel set corresponding to the next target pixel is created, and starting from the next target pixel, other target pixels whose similarity with the next target pixel is higher than a preset value are filled into the pixel set corresponding to the next target pixel through a region growing algorithm until the pixel set corresponding to the next target pixel is filled; the above steps are repeated until the target pixels in the first signature image are grouped, and the pixel sets corresponding to the strokes in the first signature image are obtained.

[0075] See also Figure 3 The signature recognition method provided in the embodiment of the present application may also include, but is not limited to, the following steps:

[0076] Step S301: Acquire target pixels in the first signature image whose pixel values are within a target range;

[0077] Step S302: selecting the i-th pixel from each target pixel according to the pixel arrangement order of the first signature image;

[0078] Step S303: determining whether the access state of the i-th pixel is a pending access state;

[0079] If the access state of the i-th pixel is the pending access state, step S304 is executed: using a region growing algorithm, based on the similarity between the i-th pixel and other pixels in each target pixel, the i-th pixel and the j-th pixel are divided into the same group; wherein the j-th pixel is a pixel among the other pixels whose similarity with the i-th pixel is greater than a preset value;

[0080] Step S305: updating the access status of the i-th pixel and the j-th pixel to the visited status;

[0081] Step S306: determining whether all target pixels in the first signature image have been grouped;

[0082] If all target pixels have not been grouped, execute step S307: update i to i+1, and return to step S302;

[0083] If all target pixels have been grouped, the process ends;

[0084] When the access state of the i-th pixel is the accessed state, step S306 is executed.

[0085] The specific implementation process of the above steps S301 to S307 is described in the above embodiment and will not be repeated here.

[0086] In the embodiment of the present application, when selecting the i-th pixel from each target pixel, by judging the access status of the i-th pixel, it is determined whether to execute the step of using the region growing algorithm to divide the i-th pixel and the j-th pixel into the same group based on the similarity between the i-th pixel and other pixels in each target pixel. This can improve the efficiency and accuracy of determining the pixel sets corresponding to each stroke in the first signature image, and further improve the efficiency and effect of generating the incomplete signature image samples in the first incomplete signature image samples based on the pixel sets corresponding to each stroke. Therefore, based on the incomplete signature image samples in the first incomplete signature image samples, the signature integrity recognition model is trained, which can improve the accuracy of the signature integrity recognition model in identifying incomplete signatures.

[0087] In one embodiment, after the above step S201: obtaining pixel sets corresponding to respective strokes in the first signature image, the signature recognition method further comprises:

[0088] Obtaining information of each pixel in a second pixel set corresponding to a second stroke;

[0089] removing the second stroke from the first signature image based on information of each pixel in the second pixel set to obtain a third signature image;

[0090] The third signature image is determined as an incomplete signature image sample in the first incomplete signature sample set.

[0091] Optionally, the second stroke is any one stroke in the first signature image; or any multiple strokes in the first signature image; the second stroke may be completely different from or partially different from the first stroke.

[0092] In addition, based on the information of each pixel in the second pixel set, the second stroke is removed from the first signature image to obtain the third signature image. The specific implementation process can refer to the specific implementation process of the above step S203 and will not be repeated here.

[0093] In the embodiment of the present application, based on the above-mentioned method of removing the first stroke from the first signature image based on the information of each pixel in the first pixel set corresponding to the first stroke to obtain the second signature image, the second stroke can be further removed from the first signature image based on the information of each pixel in the second pixel set to obtain a third signature image. This is equivalent to performing multiple stroke removal operations on the first signature image to obtain different stroke-missing signature images; since the strokes in the first signature image are randomly removed each time, the repetition rate of the removed strokes is low. Such an operation can greatly increase the number of incomplete signature image samples in the first incomplete signature sample set.

[0094] See also Figure 4 Before obtaining the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set in step S101, the signature recognition method provided in the embodiment of the present application may further include, but is not limited to:

[0095] Step S401: Using a mask to block the fourth signature image, the second stroke corresponding to the blocked area of the mask is removed from the fourth signature image to obtain a fifth signature image; wherein the fourth signature image is a complete signature image sample;

[0096] In actual implementation, a randomly generated mask can be used to block the fourth signature image to remove the second stroke in the fourth signature image corresponding to the blocked area of the mask; or based on the font shape corresponding to the fourth signature image, a mask with a specific shape in the blocked area can be used to block the fourth signature image to remove the second stroke corresponding to the blocked area of the mask. For the specific implementation, please refer to the relevant description below and will not be described here.

[0097] Furthermore, the signature image obtained by removing the second stroke corresponding to the occluded area of the mask from the fourth signature image can be directly determined as the fifth signature image; the signature image obtained by removing the second stroke corresponding to the occluded area of the mask from the fourth signature image can also be further processed to avoid the presence of residual parts of the second stroke that are not removed, to obtain the fifth signature image. The specific implementation can be found in the relevant description below and is not described here.

[0098] Step S402: Determine the fifth signature image as an incomplete signature image sample in the second incomplete signature sample set.

[0099] In some embodiments, the fifth signature image can be directly determined as an incomplete signature image sample in the second incomplete signature sample set; or the fifth signature image can be compared with the fourth signature image to determine the stroke removal ratio of the fifth signature image relative to the fourth signature image, and when the stroke removal ratio is higher than the preset ratio, the fifth signature image can be determined as an incomplete signature image sample in the second incomplete signature sample set, so as to avoid the situation where incomplete signature image samples (also called negative samples) in the second incomplete signature sample set are mixed into complete signature image samples (also called positive samples) due to invalid stroke removal, thereby enabling batch generation of incomplete signature image samples with nearly real incomplete signature effects.

[0100] The embodiment of the present application utilizes a mask to block the fourth signature image, removes the second stroke corresponding to the blocked area of the mask from the fourth signature image, obtains the fifth signature image, and determines the fifth signature image as an incomplete signature image sample in the second incomplete signature sample set. This can improve the efficiency and effect of generating the incomplete signature image samples in the second incomplete signature image sample, so that the signature integrity recognition model can be trained based on the incomplete signature image samples in the second incomplete signature image sample, thereby improving the accuracy of the signature integrity recognition model in identifying incomplete signatures.

[0101] In one embodiment, the step S401 of shielding the fourth signature image with the mask and removing the second stroke corresponding to the shielded area of the mask from the fourth signature image to obtain the fifth signature image includes:

[0102] Using the mask to block the fourth signature image, the target portion of the second stroke corresponding to the blocked area is removed from the fourth signature image to obtain a sixth signature image;

[0103] The sixth signature image is subjected to an erosion and dilation process by using an opening operation to remove the remaining portion of the second stroke, thereby obtaining a fifth signature image.

[0104] In actual implementation, to ensure that the strokes are removed cleanly, the fourth signature image can be first masked using a mask, and the second stroke in the fourth signature image can be removed to obtain a sixth signature image. Then, the sixth signature image can be eroded and expanded using an opening operation to effectively remove small objects in the sixth signature image, smooth the boundaries in the sixth signature image, and convert the sharp edges in the sixth signature image into softer pen-shaped structures, so that the remaining part of the second stroke is removed from the fourth signature image, thereby obtaining a fifth signature image with the second stroke completely removed.

[0105] The embodiment of the present application uses a mask to block the fourth signature image, removes the target portion of the second stroke corresponding to the blocked area from the fourth signature image, and obtains a sixth signature image. The sixth signature image is then subjected to corrosion and dilation processing using an opening operation to remove the remaining portion of the second stroke, thereby obtaining a fifth signature image with the second stroke removed. This can improve the efficiency and effect of generating incomplete signature image samples in the second incomplete signature image sample, so that a signature integrity recognition model can be trained based on the incomplete signature image samples in the second incomplete signature image sample, thereby improving the accuracy of the signature integrity recognition model in recognizing incomplete signatures.

[0106] In one embodiment, the shielding area of the mask is a rectangular area.

[0107] In actual implementation, considering that Chinese has the characteristic of square character distribution, which means that square characters lack strokes and are prone to "hollow" shapes, which are closer to rectangles, the embodiment of the present application can set the shape of the occlusion area of the above-mentioned mask to a rectangle, and can remove the second stroke corresponding to the rectangular occlusion area of the mask by using a mask with a rectangular occlusion area to mask the fourth signature image, thereby obtaining a fifth signature image in which at least one stroke corresponding to the rectangular occlusion area is missing.

[0108] The embodiment of the present application obtains the second incomplete signature sample set by determining the fifth signature image that is missing the strokes corresponding to the rectangular occlusion area as an incomplete signature image sample in the second incomplete signature sample set, and trains the signature integrity recognition model through the incomplete signature image samples in the second incomplete signature sample set and the incomplete signature image samples in the first incomplete signature sample set. In this way, during the training process, the interference of the missing stroke features corresponding to other shape occlusion areas on the missing feature transfer learning of the signature integrity recognition model can be eliminated, so that the signature integrity recognition model can focus on the transfer learning of the missing stroke features of the incomplete signature image samples in the first incomplete signature sample set and the missing stroke features of the incomplete signature image samples in the second incomplete signature sample set corresponding to the rectangular occlusion area, thereby improving the efficiency and effect of the missing feature transfer learning of the signature integrity recognition model, and ensuring that the missing features obtained after transfer learning can cover most actual scenarios where strokes are missing.

[0109] In one embodiment, the signature recognition method provided in the embodiment of the present application may also include, but is not limited to, the following steps:

[0110] Obtaining a training data set, wherein the training data set includes a plurality of training samples, each training sample including the target signature image sample and its corresponding signature integrity label result;

[0111] The preset signature integrity recognition model is trained using the training data set until the training stop condition is met, thereby obtaining a trained signature integrity recognition model.

[0112] Optionally, the aforementioned method of using the training data set to train a preset signature integrity recognition model until a training stop condition is met, thereby obtaining a trained signature integrity recognition model, including:

[0113] For each training sample, perform the following steps:

[0114] The training sample is input into the preset signature integrity recognition model, and the missing feature transfer learning is performed on the target signature image sample through the preset signature integrity recognition model to obtain a new target signature image sample and its signature integrity label result;

[0115] Identify the new target signature image sample using a preset signature integrity recognition model to obtain a signature integrity recognition result of the new target signature image sample;

[0116] Determine the loss function value of the preset signature integrity recognition model based on the signature integrity recognition result of the new target signature image sample and the signature integrity label result corresponding to the second signature image;

[0117] When the loss function value does not meet the training stop condition, the model parameters of the preset signature integrity recognition model are adjusted to obtain an updated signature integrity recognition model, and the updated signature integrity recognition model is trained using the next training sample until the training stop condition is met, thereby obtaining a trained signature integrity recognition model.

[0118] In an embodiment of the present application, based on the fact that each target signature sample in the target signature sample set includes signature image samples corresponding to the same complete signature in a complete signature sample set, a first incomplete signature sample set, and a second incomplete signature sample set, for each target signature image sample, the target signature sample and its corresponding signature integrity label result are input into a preset signature integrity recognition model, and missing feature transfer learning is performed on the target signature image sample through the preset signature integrity recognition model to obtain a new target signature image sample and its signature integrity label result, and the preset signature integrity recognition model is continued to be trained with the new target signature image sample and its signature integrity label result, which can improve the training effect of the signature integrity recognition model and thereby improve the accuracy of the signature integrity recognition model in recognizing incomplete signatures.

[0119] In one embodiment, after obtaining the trained signature integrity recognition model in step S103, the signature recognition method provided in the embodiment of the present application further includes:

[0120] Obtaining a target signature image whose signature integrity is to be identified;

[0121] The target signature image is input into the trained signature integrity recognition model to obtain the signature integrity recognition result of the target signature image.

[0122] Optionally, the target signature image is obtained by photographing the target signature whose signature integrity is to be identified; similar to the above-mentioned complete signature image samples and incomplete signature image samples, when classified according to color channel attributes, the target signature image can also be any type of binary image, grayscale image, red, green, and blue three-channel (Red Green Blue, RGB) image, and red, green, and blue four-channel transparency (Red Green Blue Alpha, RGBA) image.

[0123] In actual implementation, the signature integrity recognition result of the target signature image can be that the target signature corresponding to the target signature image is complete or incomplete; it can also be that the target signature corresponding to the target signature image is complete, or the target signature corresponding to the target signature image is incomplete and there is an explanation of missing strokes corresponding to the target signature.

[0124] In addition, compared with training a signature integrity recognition model solely based on the incomplete signature image samples in the first incomplete signature sample set, the signature integrity recognition model trained by the embodiments of the present application based on the target signature image samples in the target signature sample set can enable the signature integrity recognition model to accurately recognize incomplete signatures in the case where there are missing strokes in connected characters in the signature, or multiple strokes are missing in the same area of the signature, or the target stroke that should originally pass through other strokes is missing.

[0125] Compared with training a signature integrity recognition model solely based on the incomplete signature image samples in the second incomplete signature sample set, the signature integrity recognition model trained by the embodiments of the present application based on the target signature image samples in the target signature sample set can avoid the occlusion pattern being captured by the model, thereby improving the training effect of the signature integrity recognition model, and can enable the signature integrity recognition model to accurately recognize incomplete signatures in the case where the target stroke that should originally pass through other strokes is missing in the signature.

[0126] Taking the character "德" in the signature as an example, the signature integrity recognition model trained by the embodiments of the present application can not only recognize that the signature is an incomplete signature when a certain connected stroke (such as "一") is missing in the character "德", but also recognize that the signature is an incomplete signature when multiple strokes corresponding to the occlusion data set (such as "心") are missing in the character "德", and can also recognize that the signature is an incomplete signature when a certain connected stroke (such as "一") and multiple strokes corresponding to the occlusion data set (such as "心") are missing simultaneously in the character "德".

[0127] Through actual use and comparison, the accuracy rate of the signature integrity recognition model trained solely based on the incomplete signature image samples in the above first incomplete signature sample set for recognizing incomplete signatures is 77%; the accuracy rate of the signature integrity recognition model trained solely based on the incomplete signature image samples in the above second incomplete signature sample set for recognizing incomplete signatures is 64.8%; in the case of not restricting the shape of the mask occlusion area, the accuracy rate of the signature integrity recognition model trained by the embodiments of the present application for recognizing incomplete signatures is 92.55%; in the case of restricting the shape of the mask occlusion area to a rectangle, the accuracy rate of the signature integrity recognition model trained by the embodiments of the present application for recognizing incomplete signatures is 96.7%.

[0128] In a specific embodiment, the signature recognition method provided by the embodiments of the present application may further include, but is not limited to, the following steps:

[0129] Step 1: Determine the original complete signature image as the complete signature image sample in the complete signature sample set (which can be abbreviated as the complete class sample);

[0130] Step 2: Based on information about each pixel in a first pixel set corresponding to the first stroke, remove the first stroke from the complete signature image to obtain an incomplete signature image sample in a first incomplete signature sample set (which may be referred to as mask_region); wherein the first stroke is any one or more strokes in the complete signature image;

[0131] Step 3: masking the complete signature image based on the mask having the rectangular masking area to remove the second stroke corresponding to the rectangular masking area from the complete signature image, thereby obtaining an incomplete signature image sample in a second incomplete signature sample set (which may be referred to as mask_rectangle);

[0132] Step 4: Take the complete class sample as the positive sample, and mix the incomplete signature image sample in mask_reg i on and the incomplete signature image sample in mask_rectangle as the negative sample;

[0133] Step 5. Select a classification model as the signature integrity recognition model. The classification model can be any of resnet18 and resnet50 models. Input positive samples and negative samples into the classification model, train the classification model, and obtain a trained classification model. The trained classification model can identify signature integrity by classifying signatures into complete signatures and incomplete signatures.

[0134] The specific implementation process of this embodiment can refer to the description of the above embodiment and will not be repeated here.

[0135] See also Figure 5 , the embodiment of the present application further provides a signature recognition device 500, which can implement the above-mentioned signature recognition method, and the device includes: a first acquisition module 501, a second acquisition module 502 and a training module 503;

[0136] The first acquisition module 501 is used to acquire a complete signature sample set, a first incomplete signature sample set, and a second incomplete signature sample set; wherein each incomplete signature image sample in the first incomplete signature sample set is obtained by removing strokes from a complete signature image sample; and each incomplete signature image sample in the second incomplete signature sample set is obtained by removing an occlusion dataset from a complete signature image sample;

[0137] The second acquisition module 502 is configured to generate a target signature sample set based on the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set; wherein each target signature image sample in the target signature sample set includes signature image samples corresponding to the same complete signature in the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set;

[0138] The training module 503 is used to train the signature integrity recognition model using the target signature sample set to obtain a trained signature integrity recognition model; wherein, during the process of training the signature integrity recognition model, the signature integrity recognition model performs missing feature transfer learning on each target signature image sample; the signature integrity recognition model is used to identify signature integrity.

[0139] The specific implementation of the signature recognition device is basically the same as the specific embodiment of the above-mentioned signature recognition method, and will not be repeated here.

[0140] The present application also provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-mentioned signature recognition method when executing the computer program. The electronic device can be any smart terminal including a tablet computer, an in-vehicle computer, or the like.

[0141] See also Figure 6 , Figure 6 The hardware structure of an electronic device according to another embodiment is shown. The electronic device includes:

[0142] The processor 601 may be implemented as a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is configured to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0143] The memory 602 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 602 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 602 and is called by the processor 601 to execute the signature recognition method of the embodiments of this application.

[0144] Input / output interface 603, used to implement information input and output;

[0145] Communication interface 604, used to implement communication interaction between this device and other devices, which can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.);

[0146] Bus 605 , which transmits information between various components of the device (e.g., processor 601 , memory 602 , input / output interface 603 , and communication interface 604 );

[0147] The processor 601 , the memory 602 , the input / output interface 603 and the communication interface 604 are connected to each other in communication within the device via a bus 605 .

[0148] An embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned signature recognition method is implemented.

[0149] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0150] An embodiment of the present application also provides a computer program product, which is stored in a storage medium and implements the above-mentioned signature recognition method when the computer program product is executed by at least one processor.

[0151] The signature recognition method, device, electronic device, storage medium and program product provided in the embodiments of the present application train a signature integrity recognition model using a target signature sample set, and during the training process, the signature integrity recognition model performs missing feature transfer learning on each target signature image sample. This can organically combine the features of different handwritten signature artificial data, so that the missing features learned by the signature integrity recognition model focus on a first stroke missing feature obtained by removing the stroke based on pixel information corresponding to each stroke, a second stroke missing feature obtained by removing the stroke based on occlusion of a mask in a rectangular occlusion area, and a migration derivative feature of the first stroke missing feature and the second stroke missing feature. This can significantly reduce the dimensionality of the features learned by the signature integrity recognition model, thereby reducing the training cost and deployment cost of the signature integrity recognition model, and can ensure that the stroke missing features learned by the signature integrity recognition model are closer to the actual stroke missing scenario, thereby improving the accuracy of using the signature integrity recognition model to recognize incomplete signatures in the field of financial technology, especially in the signing scenario of leasing business documents.

[0152] The embodiments described in the embodiments of this application are intended to more clearly illustrate the technical solutions of the embodiments of this application and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0153] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0154] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0155] Those skilled in the art will appreciate that all or some of the steps in the methods, systems, and functional modules / units in the devices disclosed above may be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0156] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0157] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0158] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the above-mentioned units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0159] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0160] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0161] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes multiple instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store programs.

[0162] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and essence of the present invention should be within the scope of the present invention.

Claims

1. A signature recognition method, characterized in that: include: Obtaining a complete signature sample set, a first incomplete signature sample set, and a second incomplete signature sample set; wherein each incomplete signature image sample in the first incomplete signature sample set is obtained by removing strokes from a complete signature image sample; and each incomplete signature image sample in the second incomplete signature sample set is obtained by removing an occlusion dataset from the complete signature image sample; generating a target signature sample set based on the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set; wherein each target signature image sample in the target signature sample set includes signature image samples corresponding to the same complete signature from the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set; The target signature sample set is used to train the signature integrity recognition model to obtain the trained signature integrity recognition model; wherein, during the training of the signature integrity recognition model, the signature integrity recognition model performs missing feature transfer learning on each target signature image sample; the signature integrity recognition model is used to identify signature integrity.

2. The signature recognition method according to claim 1, wherein: Before obtaining the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set, the signature recognition method further includes: Obtaining a pixel set corresponding to each stroke in a first signature image; wherein the first signature image is a sample of the complete signature image; Obtaining pixel values of each pixel in a first pixel set corresponding to the first stroke; changing the pixel value of each pixel in the first pixel set corresponding to the first stroke to the background pixel value of the first signature image to obtain a second signature image; The second signature image is determined as an incomplete signature image sample in the first incomplete signature sample set.

3. The signature recognition method according to claim 2, wherein: The obtaining of pixel sets corresponding to the strokes in the first signature image includes: Acquire target pixels in the first signature image whose pixel values are within a target range; the target range is determined based on color channel attributes and background pixel values of the first signature image; The target pixels are grouped based on the similarity between the target pixels using a region growing algorithm to obtain the pixel set.

4. The signature recognition method according to claim 1, wherein: Before obtaining the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set, the signature recognition method further includes: Using the mask to block the fourth signature image, the second stroke corresponding to the blocked area of the mask is removed from the fourth signature image to obtain a fifth signature image; wherein the fourth signature image is a sample of the complete signature image; The fifth signature image is determined as an incomplete signature image sample in the second incomplete signature sample set.

5. The signature recognition method according to claim 4, wherein: The method of shielding the fourth signature image with the mask and removing the second stroke corresponding to the shielded area of the mask from the fourth signature image to obtain the fifth signature image includes: Using the mask to block the fourth signature image, a target portion of the second stroke corresponding to the blocked area is removed from the fourth signature image to obtain a sixth signature image; The sixth signature image is subjected to an erosion and dilation process by using an opening operation to remove the remaining portion of the second stroke, thereby obtaining the fifth signature image.

6. The signature recognition method according to claim 4, wherein: The shading area of the mask is a rectangular area.

7. A signature recognition device, characterized in that: include: A first acquisition module, a second acquisition module and a training module; The first acquisition module is used to acquire a complete signature sample set, a first incomplete signature sample set, and a second incomplete signature sample set; wherein each incomplete signature image sample in the first incomplete signature sample set is obtained by removing strokes from a complete signature image sample; and each incomplete signature image sample in the second incomplete signature sample set is obtained by removing an occlusion dataset from the complete signature image sample; The second acquisition module is configured to generate a target signature sample set based on the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set; wherein each target signature image sample in the target signature sample set includes signature image samples corresponding to the same complete signature from the complete signature sample set, the first incomplete signature sample set, and the second incomplete signature sample set; The training module is used to train the signature integrity recognition model using the target signature sample set to obtain the trained signature integrity recognition model; wherein, during the process of training the signature integrity recognition model, the signature integrity recognition model performs missing feature transfer learning on each of the target signature image samples; the signature integrity recognition model is used to identify signature integrity.

8. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the signature recognition method according to any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the signature recognition method according to any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The computer program product is stored in a storage medium, and when the computer program product is executed by at least one processor, the signature recognition method according to any one of claims 1 to 6 is implemented.

Images