CN114049686A - Signature recognition model training method and device and electronic equipment - Google Patents

Abstract

The present disclosure provides a training method, device and electronic device for a signature recognition model, and relates to the technical field of artificial intelligence, in particular to the technical fields of deep learning and computer vision. The specific scheme is: obtaining samples to be used and corresponding name annotation information from a predetermined sample library; wherein, the sample library includes the first type of samples and the corresponding name annotation information; the first type of samples is identified based on the signature. When the model performs signature recognition, the signature picture with the wrong recognition result is fed back; the text line picture corresponding to the to-be-used sample is obtained, wherein the text-line picture is the signature area in the to-be-used sample; based on the text line picture and name corresponding to the to-be-used sample Label information, and update and train the signature recognition model. Through this solution, the iterative signature recognition model can be automatically trained, which greatly reduces the labor cost of the iterative signature recognition model.

Description

Signature recognition model training method, device and electronic device

technical field

The present disclosure relates to the technical field of artificial intelligence, in particular to the technical fields of deep learning and computer vision, and in particular to a training method, apparatus and electronic device for a signature recognition model.

Background technique

For digital enterprises, there is a high demand for intelligent recognition of handwritten signatures. The so-called intelligent recognition of handwritten signatures is to automatically recognize the signature content of handwritten signatures.

Using the signature recognition model to recognize signature pictures containing handwritten signatures is a common recognition method. In order to ensure the accuracy of the signature recognition model, the signature recognition model needs to be iterated, that is, the model is updated and trained. In the related art, the training model is manually updated, for example, a new signature image is manually selected as a sample, and the signature area and name information in the sample are manually marked.

SUMMARY OF THE INVENTION

The present disclosure provides a method, apparatus and electronic device for training a signature recognition model.

According to an aspect of the present disclosure, a method for training a signature recognition model is provided, the method comprising:

Obtain the samples to be used and the corresponding name annotation information from a predetermined sample library; wherein, the sample library includes the first type of samples and the corresponding name annotation information; the first type of samples is based on the signature recognition model When performing signature recognition, the signature picture with the wrong recognition result is fed back;

acquiring the text line picture corresponding to the sample to be used, wherein the text line picture is the signature area in the sample to be used;

Based on the text line picture and name annotation information corresponding to the sample to be used, the signature recognition model is updated and trained.

According to another aspect of the present disclosure, an information processing system is provided, including: a signature recognition subsystem and a model training subsystem;

The signature recognition subsystem is used to obtain the name of the to-be-recognized image when receiving the user's feedback that the recognition result is incorrect for the to-be-recognized image, take the to-be-recognized image as a first-type sample, and label the name as a corresponding name The information is stored in a predetermined sample library; wherein, the recognition result is the result obtained by recognizing the to-be-recognized picture based on a pre-trained signature recognition model;

The model training subsystem is used for obtaining samples to be used and corresponding name annotation information from a predetermined sample library; obtaining text line pictures corresponding to the samples to be used; based on the text line pictures corresponding to the samples to be used and name annotation information to update and train the signature recognition model.

According to another aspect of the present disclosure, there is provided an apparatus for training a signature recognition model, the apparatus comprising:

a first acquisition module, configured to acquire samples to be used and corresponding name annotation information from a predetermined sample library; wherein the sample library at least includes a first type of samples and corresponding name annotation information; the first type of samples When performing signature recognition based on the signature recognition model, a signature picture with an incorrect recognition result is fed back;

a second acquisition module, configured to acquire a text line picture corresponding to the to-be-used sample, wherein the text-line picture is a signature area in the to-be-used sample;

A training module, configured to update and train the signature recognition model based on the text line picture and name annotation information corresponding to the to-be-used sample.

According to another aspect of the present disclosure, there is provided an electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described signature recognition model training method.

According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to perform the steps of the above-mentioned signature recognition model training method.

According to another aspect of the present disclosure, there is provided a computer program product comprising a computer program that, when executed by a processor, implements the steps of the above-described signature recognition model training method.

It should be understood that what is described in this section is not intended to identify key or critical features of embodiments of the disclosure, nor is it intended to limit the scope of the disclosure. Other features of the present disclosure will become readily understood from the following description.

Description of drawings

The accompanying drawings are used for better understanding of the present solution, and do not constitute a limitation to the present disclosure. in:

1 is a flowchart of a method for training a signature recognition model provided by an embodiment of the present disclosure;

FIG. 2( a ) is a sample to be used in the embodiment of the present disclosure, and FIG. 2( b ) is the minimum circumscribed rectangle including the signature text in the embodiment of the present disclosure;

3 is a schematic structural diagram of an information processing system provided by an embodiment of the present disclosure;

4 is a schematic structural diagram of a signature recognition model training device provided by an embodiment of the present disclosure;

FIG. 5 is a block diagram of an electronic device used to implement the signature recognition model training method according to an embodiment of the present disclosure.

Detailed ways

Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding and should be considered as exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted from the following description for clarity and conciseness.

At present, contracts and confirmation documents in various fields such as finance, government affairs, and office require the applicant's handwritten signature. After that, it is necessary to verify whether the applicant's handwritten signature is consistent with the person's real name.

In the related art, a manual review method is used for signature recognition, that is, it is recognized whether the handwritten signature of the manager is consistent with the person's real name. However, for a signature scenario that generates a large number of signatures, a large amount of manual participation is required for review, which is inefficient. In addition, the manual review method requires frequent communication, waiting for review, etc., which will affect the user experience of the signature user.

Based on this, the signature recognition service came into being. The signature recognition service can recognize the text content of handwritten signatures and participate in personal identity verification, which can save more than 90% of business processing time, such as contract signing time. Moreover, reducing manual guidance can not only save the workload of the staff, but also reduce interference and improve the user's signature experience.

For the signature recognition service, the signature recognition model is an important means to realize the service. For the signature recognition model, after collecting the signature image containing the handwritten signature, the signature image can be annotated with names, and then the signature recognition model can be trained supervised to recognize the text in the signature image through the trained signature recognition model. Content, i.e. handwritten signature. It is understandable that each person's handwriting style is different, the distribution of signature image features is scattered, and the recognition accuracy is limited by the amount of labeled data. In this way, a large number of signature images are required for model training, and the signature images also need to be marked one by one, wherein the large number of signature images cover various types of handwritten data, and the various types of handwritten data are various text contents and handwriting. style.

However, signature recognition models tend to have accuracy guarantees only for samples that can be covered or similar in the training set. In order to ensure the accuracy of the signature recognition model, the signature recognition model needs to be iterated, that is, the model is updated and trained. In the related art, the training model is updated manually, for example: manually selecting a large number of new signature pictures as samples, manually labeling the signature area and name information in the samples, etc., as a new training set for the model, and constantly iterating the model. .

It can be seen that updating and training the signature recognition model manually requires a lot of labor costs.

In order to solve the above problems and realize the automatic update training of the signature recognition model, the embodiments of the present disclosure provide a signature recognition model training method, system, apparatus, device and storage medium.

The following first introduces a method for training a signature recognition model provided by an embodiment of the present disclosure.

Among them, a method for training a signature recognition model provided by an embodiment of the present disclosure is applied to an electronic device. In practical applications, the electronic device may be a server or a terminal device, which is reasonable. In addition, the scenarios to which the signature recognition model training method is applied may be counter signature scenarios, entry signature scenarios, OCR (Optical Character Recognition, Optical Character Recognition) scenarios, etc., but are not limited thereto.

A method for training a signature recognition model provided by an embodiment of the present disclosure may include the following steps:

Obtain the samples to be used and the corresponding name annotation information from a predetermined sample library; wherein, the sample library at least includes the first type of samples and the corresponding name annotation information; the first type of samples is identified based on the signature. When the model performs signature recognition, the signature picture with the wrong recognition result is fed back;

acquiring the text line picture corresponding to the sample to be used, wherein the text line picture is the signature area in the sample to be used;

Based on the text line picture and name annotation information corresponding to the sample to be used, the signature recognition model is updated and trained.

In this embodiment, the predetermined sample library includes the first type of samples, that is, the signature pictures that are wrongly identified based on the signature recognition model, and the corresponding name annotation information; the samples to be used and the corresponding name annotation information are obtained from the sample library; Further, obtain the text line picture corresponding to the sample to be used, wherein the text line picture is the signature area in the to-be-used sample; finally, based on the text line picture and name annotation information corresponding to the to-be-used sample, the signature The recognition model is updated for training. It can be seen that this scheme does not need to manually select a new training set and manually mark the signature area, etc. Therefore, through this scheme, the automatic update and training of the signature recognition model can be realized, which greatly reduces the labor cost of manually iterating the signature recognition model.

The following describes a method for training a signature recognition model provided by the present disclosure with reference to the accompanying drawings.

As shown in FIG. 1 , a method for training a signature recognition model provided by the present disclosure may include the following steps:

S101: Obtain samples to be used and corresponding name annotation information from a predetermined sample library; wherein, the sample library at least includes a first type of samples and corresponding name annotation information; the first type of samples is based on the When the signature recognition model performs signature recognition, the signature picture with the wrong recognition result is fed back;

The signature recognition model is an artificial intelligence model pre-trained based on a certain order of magnitude (eg, one million) signature image samples and name annotation information. The signature recognition model can be obtained by training any deep learning model, and the structure of the deep learning model can be set according to the actual situation.

After the signature recognition model training is completed, in order to improve the signature recognition accuracy, the signature recognition model can be iterated, that is, the signature recognition model is updated and trained. In order to update and train the signature recognition model, the model update conditions for the signature recognition model can be preset, so that when the update conditions for the signature recognition model are satisfied, the update training for the model can be started. Exemplarily, the model update condition may be, every predetermined time interval; the number of collected samples reaches a predetermined number, etc., which are all possible.

Moreover, the sample library is a database, that is, a collection of large amounts of data stored in the computer for a long time, organized and uniformly managed, used to store the signature pictures as samples and their corresponding name annotation information, wherein the signature pictures are Include a picture of the written signature. In this embodiment, the predetermined sample library includes a plurality of samples of the first type and corresponding name annotation information. In this way, the samples to be utilized obtained from the predetermined sample library may include samples of the first type. It can be understood that the number of samples to be used may be multiple, and the processing process for each sample to be used is the same.

It can be understood that the preliminarily trained signature recognition model is applied to the signature recognition subsystem that provides the signature recognition service, so that the signature recognition subsystem can use the trained signature recognition model to recognize signature information. In order to build a predetermined sample library, the signature recognition subsystem also has a confirmation feedback function, which is used to record user feedback on the signature recognition model, and identify correct or incorrect signature pictures and name annotation information; The signature image of , as the first-class sample, and the first-class sample and its corresponding name annotation information will be automatically stored in the sample library. In addition, in an implementation manner, the signature recognition subsystem can also record the label corresponding to the correctly identified or incorrectly identified signature image, so that the user feedback and identified incorrectly identified signature image and corresponding name annotation information can be found according to the label.

It should be noted that the name annotation information is the real name of the signer corresponding to the signature picture, and there are many ways to determine the name annotation information of the first type of samples, for example: when the user feedbacks the wrong recognition result for the signature picture. , instruct the user to manually enter the real name, or obtain the pre-recorded user's name in the database.

Optionally, the sample library further includes a second type of sample and corresponding name annotation information; wherein the second type of sample is a signature picture about a predetermined signature scene generated within a predetermined period of time.

Wherein, the predetermined signature scene may include but is not limited to: entry signature, counter signature, and the like. It can be understood that in these scenarios, there will be large-scale signature pictures and are constantly updated. Therefore, in the predetermined signature scene, the signature pictures and corresponding annotation information generated within a predetermined period of time can be collected and added. into a predetermined sample library.

In this way, the predetermined sample library may include, in addition to the first type of samples for which the user feedback recognition errors, the second type of samples, that is, new samples about the predetermined signature scene that are continuously collected.

It can be seen that the first type of samples and the second type of samples are constantly updated samples. The first type of samples and the second type of samples and their corresponding name annotation information are continuously stored in the sample database, which can continuously update and enrich the sample database. The signature images generated in the predetermined signature scenarios are fully utilized, thereby avoiding data waste and increasing the range of handwritten data types covered by the sample library.

S102, acquiring a text line picture corresponding to the to-be-used sample, wherein the text-line picture is a signature area in the to-be-used sample;

Since the sample to be used may contain non-signature areas, such as blank areas, and the training of the signature recognition model only needs to use the text content in the signature area, after obtaining the sample to be used, it is necessary to extract the signature area of the sample to be used. Text line image.

In an implementation manner, acquiring the text line picture corresponding to the to-be-used sample may include steps A1-A2:

A1, the sample to be used is tested for the signature area, and the test result is obtained;

The detection result may be the coordinate information of the signature area, as shown in FIG. 2(a) for the sample to be used, and FIG. 2(b) for the minimum circumscribed rectangle containing the signature text. The coordinate information can be generated according to the four end points of the minimum circumscribed rectangle of the signature text. At this time, the coordinate information can uniquely characterize the signature area. There may be many ways to perform signature area detection on the to-be-used sample.

Exemplarily, the performing signature area detection on the to-be-used sample to obtain a detection result may include:

Perform binarization processing on the to-be-used sample, and determine that the binarized sample contains the smallest circumscribed rectangle of each character area, and obtain the detection result of the signature area;

or,

Based on the pre-trained text detection model for detecting the signature area, the signature area in the to-be-used sample is detected to obtain the detection result.

In order to save the cost of manual labeling, the detection result can be obtained by binarizing the sample to be used or using a pre-trained text detection model to detect the signature area in the sample to be used.

It can be understood that the specific process of the binarization process is to first perform grayscale processing on the color picture of the sample to be used to obtain a grayscale image, and then divide each pixel of the obtained grayscale image into two levels. Exemplarily, an appropriate threshold can be set for a grayscale image to determine whether a pixel is a target or a background, and the pixels exceeding the threshold are set to 1, and the pixels below the threshold are set to 0, so as to obtain a binarized image. image. After that, it is determined that the sample after the binarization process contains the smallest circumscribed rectangle of each character area, and the detection result of the signature area is obtained.

In addition, the text detection model is a model for locating text regions that is pre-trained based on each sample and text annotation information, and the present disclosure does not limit the specific training process of the text detection model. Among them, the text annotation information corresponding to each sample can be obtained by manual annotation, or obtained by binarizing the sample, which is reasonable. Exemplarily, in an implementation manner, a certain number (such as 5,000) of signature pictures can be selected, and a label label is set to mark the coordinates of the four corners of the rectangles circumscribed by the text in these signature pictures, so as to be based on these signatures. The label of the image and its 4 corner coordinates is used to train the text detection model.

A2, based on the detection result, extract a signature area from the to-be-used sample as a text line picture corresponding to the to-be-used sample.

If the extracted signature area is the smallest circumscribed rectangle containing each text area, considering that the rectangle is often inclined, you can perform affine transformation on the signature area after extracting the signature area to correct the inclined rectangle, which is convenient for Identification of the signature recognition model, so as to better train the signature recognition model.

It can be understood that the above step of generating a text line picture can be performed after obtaining a sample to be used when the model update condition for the signature recognition model is satisfied. At this time, the step of obtaining the text line picture corresponding to the sample to be used includes: Steps A1-A2. In addition, in order to improve the efficiency of updating and training the signature recognition model, when the sample to be used is stored in the sample library, the corresponding text line picture can be automatically generated, and the text line picture can be stored in the sample library. When the model update condition of the model is set, you can directly obtain the generated text line picture from the sample library.

S103, based on the text line picture and name annotation information corresponding to the sample to be used, update and train the signature recognition model.

In an implementation manner, updating and training the signature recognition model may include: inputting the text line picture corresponding to the sample to be used into the signature recognition model, and obtaining an output result of the signature recognition model for the sample to be used; based on the output result and The name annotation information of the sample to be used is used to determine the loss value; the model parameters of the signature recognition model under training are continuously adjusted, and the training is continued, thereby continuously reducing the loss value of the model and achieving the purpose of updating and iterating.

In addition, when training the initially constructed basic signature recognition model, the collected signature image samples can also be used for signature area detection, and the signature area can be extracted as the basic text line image, and then the basic text line image and its name can be extracted. The annotation information is used as a training set for the initially constructed signature recognition model.

In the specific implementation process, for the training of the basic signature recognition model, a certain number of signature pictures with signer's name information can be extracted from the database, and the signature area in these signature pictures can be extracted as the basic text line picture. By selecting the signature picture with the signer's name information, it is possible to avoid manually labeling the signature picture with the name information. Of course, if the number of signature pictures with the signer's name information in the database is insufficient, a certain amount of signature pictures without the signer's name information can also be selected, and the name information of the certain amount of signature pictures can be marked manually. For example, set the label label for the signature pictures without the signer's name information, and mark the text in these signature pictures, that is, the actual name of the signer.

In this embodiment, the predetermined sample library includes the first type of samples, that is, the signature pictures that are wrongly identified based on the signature recognition model, and the corresponding name annotation information; the samples to be used and the corresponding name annotation information are obtained from the sample library; Further, obtain the text line picture corresponding to the sample to be used, wherein the text line picture is the signature area in the to-be-used sample; finally, based on the text line picture and name annotation information corresponding to the to-be-used sample, the signature The recognition model is updated for training. It can be seen that this scheme does not need to manually select a new training set and manually mark the signature area, etc. Therefore, through this scheme, the automatic update and training of the signature recognition model can be realized, which greatly reduces the labor cost of manually iterating the signature recognition model.

Optionally, in another embodiment, the number of the samples to be used is multiple; before the signature recognition model is updated and trained based on the text line pictures and name annotation information corresponding to the samples to be used , also including steps B1-B4:

B1, count the frequency of occurrence of each character in the name annotation information corresponding to each sample to be used;

Since various characters with different frequencies appear in the signature, such as frequently used characters appear frequently, uncommon characters appear less frequently, and the frequency of appearance of each character presents a long-tailed distribution, that is, a large number of characters appear less frequently, and only each character appears in a low frequency. The signature recognition model can achieve a high recognition accuracy only when the samples of type characters are sufficient. Therefore, in the name annotation information corresponding to each sample to be used, the occurrence frequency of each character can be counted, and the low-frequency character can be processed accordingly, thereby improving the recognition accuracy of the signature recognition model for the low-frequency character.

B2, detecting the target text whose frequency of occurrence is lower than the predetermined frequency threshold;

Use the text whose frequency is lower than the threshold in the statistical result as the target text.

B3, determine the to-be-used sample containing the target text in the corresponding name annotation information, as the to-be-processed sample;

The to-be-used samples containing the target text, that is, the samples containing the text whose frequency of occurrence is lower than the predetermined frequency threshold, can be used as the samples to be processed.

B4, performing data enhancement processing on the text line picture corresponding to the sample to be processed to obtain the target picture;

Exemplarily, the data enhancement process may perform image resizing (resize), image rotation (rotate), adding pixels on each side (padding), and the like, for the text line image corresponding to the sample to be processed. In addition, it is also possible to manually collect samples for low-frequency characters, or manually write low-frequency characters of different handwriting types as target samples, and extract text line pictures of the manually collected and written samples, which is also possible.

On the basis of including B1-B4, correspondingly, the updating and training of the signature recognition model based on the text line picture and name annotation information corresponding to the to-be-used sample may include:

Based on the text line picture and name annotation information corresponding to the sample to be used, as well as the target sample and the corresponding name annotation information, the signature recognition model is updated and trained; wherein, the name annotation information corresponding to the target sample is: : Name annotation information corresponding to the sample to be processed.

At this time, the samples used in the update and training of the signature recognition model include the text line pictures and name annotation information corresponding to the samples to be used, the target samples obtained after performing data enhancement processing on the samples to be processed corresponding to the low-frequency characters, and Corresponding name annotation information, wherein the name annotation information is the name annotation information corresponding to the specified sample.

In addition, the frequency of occurrence of characters in the first type of samples that are incorrectly recognized by the user feedback can be counted, and the above steps are performed for the low-frequency characters in the first type of samples, and the signature can be trained more accurately for the low-frequency characters in the wrongly recognized characters by the model. Identify the model.

Similarly, when training the initially constructed basic signature recognition model, the low-frequency text in the training set can also be counted, and the basic target sample obtained after data enhancement processing is performed on the to-be-processed sample corresponding to the low-frequency text, and then the basic target sample and its name annotation information as the training set of the initially constructed signature recognition model to train the basic signature recognition model, thereby improving the long-tailed distribution of the frequency of occurrence of different characters in the training set.

In this embodiment, the to-be-used samples of the target text whose frequency of occurrence is lower than the predetermined frequency threshold are taken as the to-be-processed samples, and data enhancement processing is performed on the to-be-processed samples to obtain the target samples, and based on the text line pictures and name annotations corresponding to the to-be-used samples information, as well as the target sample and the corresponding name annotation information, to update and train the signature recognition model. It can be seen that the identification accuracy of the signature identification model can be improved through the solution of this embodiment.

It can be seen that through the signature recognition model training method provided by the present disclosure, it is possible to continuously supplement and update the first type of samples that are wrongly identified by the signature recognition model, and import new second type samples, and based on the first type of samples and the second type of samples. The corresponding text line pictures and name annotation information are automatically updated for the training signature recognition model, which reduces the cost of manual collection of training sets, manual annotation, and manual iteration, and at the same time ensures that the real data can be fully utilized in the training process. Therefore, as the amount of data continues to cover more signature handwriting and text samples, the generalization and recognition accuracy of the signature recognition model will continue to improve, achieving the effect of online use.

In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of the user's personal information involved are all in compliance with the relevant laws and regulations, and do not violate public order and good customs.

According to an embodiment of the present disclosure, the present disclosure also provides an information processing system, as shown in FIG. 3 , the system includes: a signature recognition subsystem 310 and a model training subsystem 320;

The signature recognition subsystem 310 is configured to obtain the name of the to-be-recognized image when receiving an incorrect recognition result from the user for the to-be-recognized image, take the to-be-recognized image as a first-type sample, and use the name as the corresponding name The labeling information is stored in a predetermined sample library; wherein, the recognition result is a result obtained by recognizing the to-be-recognized picture based on a pre-trained signature recognition model;

The model training subsystem 320 is configured to obtain samples to be used and corresponding name annotation information from a predetermined sample library; obtain text line pictures corresponding to the samples to be used; based on the text lines corresponding to the samples to be used The image and name annotation information is used to update and train the signature recognition model.

For the specific implementation of each function of the model training subsystem, reference may be made to the corresponding content in the foregoing method embodiments, which will not be repeated here.

In this embodiment, when the signature recognition subsystem receives feedback from the user that the recognition result of the picture to be recognized is incorrect, it obtains the name of the picture to be recognized, takes the picture to be recognized as the first type sample, and uses the name as the corresponding name The annotation information is stored in a predetermined sample library; the model training subsystem obtains the sample to be used and the corresponding name annotation information from the predetermined sample library; and based on the text line picture and name annotation information corresponding to the sample to be used, the The signature recognition model is updated and trained. It can be seen that in this solution, the error samples fed back by the user are automatically stored in the sample database, and the signature recognition model is automatically updated and trained based on the text line pictures and name annotation information corresponding to the samples to be used. Therefore, through this solution, the iterative signature recognition model can be automatically trained, which greatly reduces the labor cost of the iterative signature recognition model.

According to an embodiment of the present disclosure, the present disclosure also provides a signature recognition model training device, as shown in FIG. 4 , the device includes:

The first acquisition module 410 is configured to acquire samples to be used and corresponding name annotation information from a predetermined sample library; wherein, the sample library at least includes samples of the first type and corresponding name annotation information; the first type of The sample is a signature picture with an incorrect recognition result after feedback when performing signature recognition based on the signature recognition model;

The second obtaining module 420 is configured to obtain a text line picture corresponding to the to-be-used sample, wherein the text-line picture is a signature area in the to-be-used sample;

The training module 430 is configured to update and train the signature recognition model based on the text line picture and name annotation information corresponding to the to-be-used sample.

Optionally, the sample library further includes a second type of sample and corresponding name annotation information; wherein the second type of sample is a signature picture about a predetermined signature scene collected within a predetermined period of time.

Optionally, the second obtaining module 420 includes:

a detection submodule, used for performing signature area detection on the to-be-used sample to obtain a detection result;

The extraction sub-module is configured to extract a signature area from the to-be-used sample based on the detection result, as a text line picture corresponding to the to-be-used sample.

Optionally, the detection submodule performs signature area detection on the to-be-used sample to obtain a detection result, including:

Perform binarization processing on the to-be-used sample, and determine that the binarized sample contains the smallest circumscribed rectangle of each character area, and obtain the detection result of the signature area;

or,

Based on the pre-trained text detection model for detecting the signature area, the signature area in the to-be-used sample is detected to obtain the detection result.

Optionally, the number of the samples to be utilized is multiple; the device further includes:

The statistics module is used to count the frequency of occurrence of each character in the name annotation information corresponding to each sample to be used;

a detection module, used to detect target characters whose frequency of occurrence is lower than a predetermined frequency threshold;

A determination module, configured to determine the to-be-used sample containing the target text in the corresponding name annotation information as a designated sample;

an enhancement module, configured to perform data enhancement processing on the text line picture corresponding to the specified sample to obtain a target picture;

The training module is specifically used for:

Based on the text line image and name annotation information corresponding to the sample to be used, and the target image and the corresponding name annotation information, the signature recognition model is updated and trained;

Wherein, the name annotation information corresponding to the target image is: name annotation information corresponding to the specified sample.

According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

An electronic device provided by the present disclosure may include:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one process The device can perform the steps of the above-mentioned signature recognition model training method.

The present disclosure provides a computer-readable storage medium, where a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned signature recognition model training method are implemented.

In yet another embodiment provided by the present disclosure, there is also provided a computer program product containing instructions, which, when running on a computer, cause the computer to execute the steps of the above-mentioned signature recognition model training method in the above-mentioned embodiment.

5 shows a schematic block diagram of an example electronic device 500 that may be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are by way of example only, and are not intended to limit implementations of the disclosure described and/or claimed herein.

As shown in FIG. 5 , the device 500 includes a computing unit 501 that can be executed according to a computer program stored in a read only memory (ROM) 502 or loaded from a storage unit 508 into a random access memory (RAM) 503 Various appropriate actions and handling. In the RAM 503, various programs and data necessary for the operation of the device 500 can also be stored. The computing unit 501 , the ROM 502 , and the RAM 503 are connected to each other through a bus 504 . An input/output (I/O) interface 505 is also connected to bus 504 .

Various components in the device 500 are connected to the I/O interface 505, including: an input unit 506, such as a keyboard, mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a magnetic disk, an optical disk, etc. ; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit 509 allows the device 500 to exchange information/data with other devices through a computer network such as the Internet and/or various telecommunication networks.

Computing unit 501 may be various general-purpose and/or special-purpose processing components with processing and computing capabilities. Some examples of computing units 501 include, but are not limited to, central processing units (CPUs), graphics processing units (GPUs), various specialized artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processing processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above. For example, in some embodiments, the signature recognition model training method may be implemented as a computer software program tangibly embodied on a machine-readable medium, such as storage unit 508 . In some embodiments, part or all of the computer program may be loaded and/or installed on device 500 via ROM 502 and/or communication unit 509 . When the computer program is loaded into RAM 503 and executed by computing unit 501, one or more steps of the signature recognition model training method described above may be performed. Alternatively, in other embodiments, the computing unit 501 may be configured to perform the signature recognition model training method by any other suitable means (eg, by means of firmware).

Various implementations of the systems and techniques described herein above may be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips system (SOC), load programmable logic device (CPLD), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpretable on a programmable system including at least one programmable processor that The processor, which may be a special purpose or general-purpose programmable processor, may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device an output device.

Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer or other programmable data processing apparatus, such that the program code, when executed by the processor or controller, performs the functions/functions specified in the flowcharts and/or block diagrams. Action is implemented. The program code may execute entirely on the machine, partly on the machine, partly on the machine and partly on a remote machine as a stand-alone software package or entirely on the remote machine or server.

In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in connection with the instruction execution system, apparatus or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media may include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media would include one or more wire-based electrical connections, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), fiber optics, compact disk read only memory (CD-ROM), optical storage, magnetic storage, or any suitable combination of the foregoing.

To provide interaction with a user, the systems and techniques described herein may be implemented on a computer having a display device (eg, a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user ); and a keyboard and pointing device (eg, a mouse or trackball) through which a user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (eg, visual feedback, auditory feedback, or tactile feedback); and can be in any form (including acoustic input, voice input, or tactile input) to receive input from the user.

The systems and techniques described herein may be implemented on a computing system that includes back-end components (eg, as a data server), or a computing system that includes middleware components (eg, an application server), or a computing system that includes front-end components (eg, a user's computer having a graphical user interface or web browser through which a user may interact with implementations of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system. The components of the system may be interconnected by any form or medium of digital data communication (eg, a communication network). Examples of communication networks include: Local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.

A computer system can include clients and servers. Clients and servers are generally remote from each other and usually interact through a communication network. The relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, a distributed system server, or a server combined with blockchain.

It should be understood that steps may be reordered, added or deleted using the various forms of flow shown above. For example, the steps described in the present disclosure can be executed in parallel, sequentially, or in different orders. As long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, there is no limitation herein.

The above-mentioned specific embodiments do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present disclosure should be included within the protection scope of the present disclosure.

Claims (14)

1. A method of signature recognition model training, the method comprising:

obtaining a sample to be utilized and corresponding name marking information from a preset sample library; the sample library comprises a first type of sample and corresponding name labeling information; the first type of sample is a signature picture with an incorrect feedback identification result when signature identification is carried out based on the signature identification model;

acquiring a text line picture corresponding to the sample to be utilized, wherein the text line picture is a signature area in the sample to be utilized;

and updating and training the signature recognition model based on the text line picture and the name marking information corresponding to the sample to be utilized.

2. The method of claim 1, wherein the sample library further comprises a second type of sample and corresponding name tagging information; wherein the second type of sample is a signature picture generated in a predetermined time period and related to a predetermined signature scene.

3. The method according to claim 1 or 2, wherein obtaining the text line picture corresponding to the sample to be utilized comprises:

carrying out signature area detection on the sample to be utilized to obtain a detection result;

and extracting a signature region from the sample to be utilized as a text line picture corresponding to the sample to be utilized based on the detection result.

4. The method of claim 3, wherein the performing signature region detection on the sample to be utilized to obtain a detection result comprises:

carrying out binarization processing on the sample to be utilized, and determining the minimum circumscribed rectangle of each character area in the sample after binarization processing to obtain the detection result of the signature area;

or,

and detecting the signature area in the sample to be utilized based on a pre-trained character detection model for detecting the signature area to obtain a detection result.

5. The method according to claim 1 or 2, wherein the number of samples to be utilized is plural; before updating and training the signature recognition model based on the text line picture and the name marking information corresponding to the sample to be utilized, the method further comprises the following steps:

counting the occurrence frequency of each character in the name marking information corresponding to each sample to be utilized;

detecting target characters with the occurrence frequency lower than a preset frequency threshold;

determining a to-be-utilized sample containing the target characters in the corresponding name labeling information as a to-be-processed sample;

performing data enhancement processing on the text line picture corresponding to the sample to be processed to obtain a target sample;

the updating and training of the signature recognition model based on the text line picture and the name marking information corresponding to the sample to be utilized comprises the following steps:

updating and training the signature recognition model based on the text line picture and the name marking information corresponding to the sample to be utilized, the target sample and the corresponding name marking information;

the name marking information corresponding to the target sample is as follows: and name marking information corresponding to the sample to be processed.

6. An information processing system comprising: the system comprises a signature identification subsystem and a model training subsystem;

the signature identification subsystem is used for acquiring the name of the picture to be identified when receiving that the identification result fed back by a user aiming at the picture to be identified is wrong, taking the picture to be identified as a first type sample, and storing the name as corresponding name marking information into a preset sample library; the identification result is a result obtained by identifying the picture to be identified based on a signature identification model trained in advance;

the model training subsystem is used for acquiring a sample to be utilized and corresponding name marking information from a preset sample library; acquiring a text line picture corresponding to the sample to be utilized; and updating and training the signature recognition model based on the text line picture and the name marking information corresponding to the sample to be utilized.

7. A signature recognition model training apparatus, the apparatus comprising:

the first acquisition module is used for acquiring a sample to be utilized and corresponding name marking information from a preset sample library; the sample library at least comprises a first type of sample and corresponding name labeling information; the first type of sample is a signature picture with an incorrect feedback identification result when signature identification is carried out based on the signature identification model;

the second obtaining module is used for obtaining a text line picture corresponding to the sample to be utilized, wherein the text line picture is a signature area in the sample to be utilized;

and the training module is used for updating and training the signature recognition model based on the text line picture and the name marking information corresponding to the sample to be utilized.

8. The apparatus of claim 7, wherein the sample library further comprises a second type of sample and corresponding name tagging information; wherein the second type of sample is a signature picture generated in a predetermined time period and related to a predetermined signature scene.

9. The apparatus of claim 7 or 8, wherein the second obtaining means comprises:

the detection submodule is used for carrying out signature region detection on the sample to be utilized to obtain a detection result;

and the extraction submodule is used for extracting a signature area from the sample to be utilized as a text line picture corresponding to the sample to be utilized based on the detection result.

10. The apparatus of claim 9, wherein the detection sub-module performs signature region detection on the sample to be utilized to obtain a detection result, and the detection sub-module includes:

carrying out binarization processing on the sample to be utilized, and determining the minimum circumscribed rectangle of each character area in the sample after binarization processing to obtain the detection result of the signature area;

or,

and detecting the signature area in the sample to be utilized based on a pre-trained character detection model for detecting the signature area to obtain a detection result.

11. The apparatus according to claim 7 or 8, wherein the number of samples to be utilized is plural; the device further comprises:

the statistical module is used for counting the occurrence frequency of each character in the name marking information corresponding to each sample to be utilized;

the detection module is used for detecting the target characters with the occurrence frequency lower than a preset frequency threshold;

the determining module is used for determining a sample to be utilized containing the target characters in the corresponding name marking information as an appointed sample;

the enhancement module is used for carrying out data enhancement processing on the text line picture corresponding to the specified sample to obtain a target picture;

the training module is specifically configured to:

updating and training the signature recognition model based on the text line picture and the name marking information corresponding to the sample to be utilized, and the target picture and the corresponding name marking information;

the name marking information corresponding to the target picture is as follows: and name marking information corresponding to the specified sample.

12. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.

13. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any one of claims 1-5.

14. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1-5.

Images