Determination method and apparatus for artificial palm print generation model, and determination method and apparatus for palm print recognition model

Abstract

The present application provides a determination method and apparatus for an artificial palm print generation model, and a determination method and apparatus for a palm print recognition model. The determination method for an artificial palm print generation model comprises: acquiring a first training set, wherein the first training set comprises real palm print images and identity control curves of a plurality of users; performing data augmentation on a first real palm print image of a first user in the first training set to obtain a first real palm print image subjected to the data augmentation; obtaining a first artificial palm print image corresponding to the first user and generated by a deep learning model by using the first real palm print image subjected to the data augmentation and a first identity control curve of the first user; when the first artificial palm print image subjected to the data augmentation is the same as the first real palm print image subjected to the data augmentation or the similarity between the first artificial palm print image subjected to the data augmentation and the first real palm print image subjected to the data augmentation is not greater than a preset similarity threshold, adjusting parameters of the deep learning model; and when the first artificial palm print image subjected to the data augmentation is different from the first real palm print image subjected to the data augmentation or the similarity between the first artificial palm print image subjected to the data augmentation and the first real palm print image subjected to the data augmentation is less than the similarity threshold, determining the deep learning model as an artificial palm print generation model.

Description

Method for determining synthetic palmprint generation model, method and device for determining palmprint recognition model

[0001] This application claims priority to the Chinese patent application filed with the China Patent Office on November 23, 2023, with application number 202311582374.4 and invention name “Method, device and storage medium for determining synthetic palmprint generation model”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present application relates to the field of artificial intelligence technology, and in particular to a method and device for determining a synthetic palmprint generation model and a method for determining a palmprint recognition model.

[0003] Background of the Invention

[0004] In the field of biometrics, palmprint recognition technology is gradually emerging as a new identification method, following fingerprint and facial recognition. Palmprint recognition offers significant advantages in protecting user privacy. This technology has been widely used in mobile payments, identity verification, and other areas, playing a significant role in protecting users' personal privacy and financial security. However, the high degree of privacy inherent in palmprints makes large-scale data collection difficult. Consequently, the lack of public palmprint datasets has become a significant constraint on technological development. The quantity and quality of existing public palmprint datasets are limited. Deep learning models are prone to overfitting and training instability when trained on small samples of real-world data.

[0005] Summary of the Invention

[0006] In view of this, embodiments of the present application provide a method, device, and storage medium for determining an artificial palm print generation model.

[0007] The present application provides a method for determining a synthetic palmprint generation model, performed by a computer device. The method comprises: obtaining a first training set, the first training set comprising real palmprint images and identity control curves of multiple users, wherein the real palmprint image of each user corresponds to the user's unique user identity, and the identity control curve of each user is a parameterized palmprint curve pre-generated for the user's unique user identity to simulate the palmprint appearance; performing data augmentation on a first real palmprint image of a first user in the first training set to obtain a data-enhanced first real palmprint image; obtaining a first synthetic palmprint image corresponding to the first user generated by a deep learning model using the data-enhanced first real palmprint image and the first identity control curve of the first user; adjusting parameters of the deep learning model when the data-enhanced first synthetic palmprint image is identical to the data-enhanced first real palmprint image or the degree of similarity between the two is not greater than a preset similarity threshold; and determining the deep learning model as the synthetic palmprint generation model when the data-enhanced first synthetic palmprint image is different from the data-enhanced first real palmprint image and the degree of similarity between the two is less than the similarity threshold.

[0008] An embodiment of the present application also provides a method for determining a palmprint recognition model, the method comprising: obtaining a third training set, the third training set comprising real palmprint images and synthetic palmprint images, each of the real palmprint images and the synthetic palmprint images corresponding to a known user identity, wherein the synthetic palmprint images are generated using the synthetic palmprint generation model of each embodiment; inputting the real palmprint images and the synthetic palmprint images into a palmprint recognition model to obtain a predicted user identity identified by the palmprint recognition model; when the degree of similarity between the known user identity and the predicted user identity is not greater than a preset similarity threshold, adjusting parameters of the palmprint recognition model; and when the degree of similarity between the known user identity and the predicted user identity is greater than the similarity threshold, providing the palmprint recognition model for identifying the user identity corresponding to the palmprint image.

[0009] An embodiment of the present application also provides a device for determining a synthetic palmprint generation model, the device comprising: an acquisition module configured to acquire a first training set, the first training set comprising real palmprint images and identity control curves of multiple users, the real palmprint image of each user in the multiple users corresponding to the user's unique user identity, and the identity control curve of each user being a parameterized palmprint curve pre-generated for the user's unique user identity for simulating palmprint appearance; an iteration module configured to perform data enhancement on a first real palmprint image of a first user in the first training set to obtain a data-enhanced first real palmprint image; obtain a first synthetic palmprint image corresponding to the first user generated by a deep learning model using the data-enhanced first real palmprint image and the first identity control curve of the first user; when the data-enhanced first synthetic palmprint image is identical to the data-enhanced first real palmprint image or the degree of similarity between them is not greater than a preset similarity threshold, adjusting parameters of the deep learning model; and when the data-enhanced first synthetic palmprint image is different from the data-enhanced first real palmprint image and the degree of similarity between them is less than the similarity threshold, determining the deep learning model as the synthetic palmprint generation model.

[0010] An embodiment of the present application also provides a device for determining a palmprint recognition model, the device comprising: an acquisition module configured to acquire a third training set, the third training set comprising real palmprint images and synthetic palmprint images, each of the real palmprint images and the synthetic palmprint images corresponding to a known user identity, wherein the synthetic palmprint images are generated by the synthetic palmprint generation model of each embodiment; an update module configured to input the real palmprint images and the synthetic palmprint images into a palmprint recognition model to obtain a predicted user identity identified by the palmprint recognition model; when the degree of similarity between the known user identity and the predicted user identity is not greater than a preset similarity threshold, adjusting parameters of the palmprint recognition model; and when the degree of similarity between the known user identity and the predicted user identity is greater than the similarity threshold, providing the palmprint recognition model for identifying the user identity corresponding to the palmprint image.

[0011] An embodiment of the present application further provides a computing device comprising: a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor is prompted to execute the method for determining a synthetic palmprint generation model and / or the method for determining a palmprint recognition model according to each embodiment of the present application.

[0012] The embodiments of the present application further provide a computer-readable storage medium storing computer-readable instructions, which, when executed, implement the method for determining a synthetic palmprint generation model and / or the method for determining a palmprint recognition model of each embodiment of the present application.

[0013] The embodiments of the present application provide a computer program product, including a computer program. When executed by a processor, the computer program implements the method for determining a synthetic palmprint generation model and / or the method for determining a palmprint recognition model of each embodiment of the present application.

[0014] The present application provides a method, device, and storage medium for determining a synthetic palmprint generation model. This method effectively improves the training stability and palmprint synthesis quality in small sample scenarios by performing data enhancement processing on real palmprints and synthetic palmprints, thereby improving the performance of the synthetic palmprint generation model. By generating palmprint images with high accuracy and controllable user identity, a large amount of labeled data is provided for deep learning model training, thereby meeting the demand for large-scale data sets based on deep learning palmprint recognition technology, significantly improving the performance of the model, and reducing the dependence of palmprint recognition tasks on real samples. As a result, this method solves the technical problem that the synthetic palmprint generation model is prone to overfitting and unstable training process when the number and quality of public palmprint data sets are limited.

[0015] These and other aspects of the application will be apparent from and elucidated with reference to the embodiments described hereinafter.

[0016] BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Further details, features and advantages of the present application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0018] FIG1 is a schematic diagram showing an exemplary implementation environment of an embodiment of the present application;

[0019] FIG2 shows a flowchart of palmprint recognition in an exemplary mobile payment scenario according to an embodiment of the present application;

[0020] FIG3A is a schematic diagram of the palmprint recognition process according to an embodiment of the present application;

[0021] FIG3B shows a flow chart of a method for determining a synthetic palmprint generation model according to an embodiment of the present application;

[0022] FIG3C shows a flow chart of a method for determining a synthetic palmprint generation model according to an embodiment of the present application;

[0023] FIG3D shows a method for determining a palmprint recognition model according to an embodiment of the present application;

[0024] FIG3E shows a schematic diagram of a device for determining a synthetic palmprint generation model according to an embodiment of the present application;

[0025] FIG3F shows a schematic diagram of a palmprint recognition model determination device according to an embodiment of the present application;

[0026] FIG4 is a schematic diagram of a training system for a synthetic palmprint generation model according to an embodiment of the present application;

[0027] 5A and 5B are schematic diagrams showing the effect of data enhancement in a scenario where the discriminator is overfitted according to an embodiment of the present application;

[0028] 6A and 6B are schematic diagrams of palm print templates according to an embodiment of the present application;

[0029] FIG7 is a schematic diagram of quality comparison of synthetic palmprint images according to an embodiment of the present application;

[0030] FIG8 is a flow chart of a method for determining a synthetic palmprint generation model according to an embodiment of the present application;

[0031] FIG9 is a flow chart of a method for determining a palmprint recognition model according to an embodiment of the present application;

[0032] FIG10 is a schematic block diagram of a device for determining a synthetic palmprint generation model according to an embodiment of the present application;

[0033] FIG11 shows a schematic block diagram of a device for determining a palmprint recognition model;

[0034] FIG12 is an example block diagram of a computing device according to an embodiment of the present application.

[0035] Modes for Carrying Out the Invention

[0036] The following will provide a clear and complete description of the technical solutions in this application in conjunction with the accompanying drawings. The embodiments described are only some of the embodiments of this application, not all of them. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0037] A region of interest (ROI) refers to a specific area in an image or video that is important for a task or focus. When performing tasks such as image processing and target detection, it is usually only necessary to process and analyze the region of interest to improve computational efficiency and accuracy. The region of interest can be manually selected and annotated in an image or video or automatically detected. The definition and selection of the region of interest vary for different tasks and applications. In image processing, the region of interest may be a local area of ​​the image to be enhanced, repaired, or transformed. In target detection, the region of interest is usually the target object to be detected. By limiting the processing scope, energy and computing resources can be focused on the most relevant areas, improving the efficiency and performance of the algorithm.

[0038] A generative adversarial network (GAN) is a generative model that learns through a game of two neural networks. A GAN consists of a generator and a discriminator. The generator takes random samples from the latent space as input, and its output is required to closely mimic real samples from the training set. The discriminator takes real samples or the generator's output as input, and its goal is to distinguish the generator's output from real samples as closely as possible. The generator and discriminator compete with each other and continuously learn, ultimately making it impossible for the discriminator to determine whether the generator's output is realistic.

[0039] Palmprint recognition technology, used in mobile payment scenarios, directly affects user wealth, requiring extremely high accuracy. Traditional palmprint recognition methods use feature extraction to measure the similarity between different user identities. Deep learning-based palmprint recognition technology has become a popular solution in recent years. By designing an efficient network structure and implementing loss function constraints, it extracts features from palmprint data and then uses a similarity matching algorithm to verify identity. Within deep learning-based palmprint recognition technology, methods based on adversarial neural networks can generate highly accurate synthetic palmprint images with controllable user identities, thus meeting the large-scale dataset requirements of deep learning-based palmprint recognition technology.

[0040] Deep learning algorithm models rely on large-scale, annotated data for training to extract effective features and achieve results close to those of real palm prints. Palm print data synthesis techniques, for example, include generative adversarial networks (GANs) and diffusion models. These two methods can achieve good synthesis results on large-scale and diverse training data. However, due to concerns about palm print data privacy, the availability and quality of existing public palm print datasets are limited.

[0041] Existing palmprint synthesis models can effectively generate a large amount of palmprint information with the same valid user identity (also known as user identity-constrained palmprints) when there are sufficient training samples, thereby expanding the training dataset for palmprint recognition. However, due to the limitations in the quantity and quality of public palmprint datasets, existing deep learning models for palmprint recognition are prone to problems such as discriminator overfitting, loss of user identity information, and unstable training process under conditions of small sample real data. Due to the particularity of palmprint data, existing small sample generation models cannot be optimized for palmprint synthesis scenarios and cannot meet existing needs. In scenarios with small samples (i.e., limited real data), data augmentation and transfer learning models can alleviate the need for real data. The emergence of synthetic data provides a large amount of labeled data for the training of deep learning models for palmprint recognition, thereby significantly improving the performance of palmprint recognition models.

[0042] In the embodiments of the present application, the degree of overfitting of the model is effectively measured through the image quality comparison module and statistical data during the training process, and then the training stability and image synthesis quality of the model in small sample scenarios are effectively improved through dynamic data enhancement and regularization methods.

[0043] 1 shows a schematic diagram of an exemplary computer system 100 according to an embodiment of the present application. The implementation environment may include: a collection terminal 110 , a server 120 , and a registration terminal 130 .

[0044] Collection terminal 110 has application 111 installed and running. When application 111 is running, collection terminal 110 activates its camera to capture the user's palm print image in real time and sends the image to server 120 for payment verification. Upon receiving the payment result from server 120, it displays it to inform the user whether the payment was successful. Application 111 can be a shopping application, a payment application, or a financial transaction application.

[0045] The registration terminal 130 has an application 131 installed and running. In some embodiments, the applications installed on the acquisition terminal 110 and the registration terminal 130 are the same. In some embodiments, the applications installed on the two terminals are the same type of applications on different operating system platforms (Android or IOS). In some embodiments, the applications installed on the two terminals are different versions of the same type of application (for example, an application that supports payment and an application that supports collection). When the registration terminal 130 receives an account registration operation, it turns on the camera to capture the palm print image of the user 132, and sends the captured image and the registered account to the server 120, so that the server 120 updates the user database 123, which stores the user account and the corresponding palm print features.

[0046] Figure 1 shows only one payment terminal and one account registration terminal, but in different embodiments, multiple other terminals may be connected to server 120. In some embodiments, one or more terminals are provided for developers, on which an application development and editing platform is installed. Developers can edit and update applications on these terminals and transmit the updated application installation packages to server 120 via wireless or wired network 140. Collection terminal 110 and registration terminal 130 can download application installation packages from server 120 to update the applications.

[0047] The collection terminal 110 , the registration terminal 130 and other terminals are connected to the server 120 via a wireless network or a wired network.

[0048] Server 120 includes at least one of a single server, a server cluster consisting of multiple servers, a cloud computing platform, and a virtualization center. In some embodiments, server 120 includes memory 121, a processor 122, a user database 123, a verification module 124, and a user-facing input / output interface (I / O interface) 125. Processor 122 is used to load instructions stored in server 120 and process data in user database 123 and verification module 124. User database 123 is used to store user account data used by registration terminal 130 and other terminals, including user accounts and corresponding palm print features. Verification module 124 is used to determine the target account for payment verification after obtaining the palm print features and to make payments using the target account. User-facing I / O interface 125 is used to establish communication and exchange data with collection terminal 110 and / or registration terminal 130 via a wireless or wired network.

[0049] FIG2 illustrates a palmprint recognition flowchart 200 for an exemplary mobile payment scenario according to an embodiment of the present application. As will be appreciated by those skilled in the art, the mobile payment scenario herein is merely an example, and the application scope of the palmprint recognition method is not limited to this scenario. In step 210, a user's hand image is captured via a terminal payment device; in step 220, a region of interest is extracted from the hand image. In step 230, the region of interest is input into a palmprint recognition model to obtain an identity recognition result. Thus, the palmprint recognition model is used to identify the user identity corresponding to the palmprint image. In some embodiments, as shown in the dashed box in FIG2 , in step 240, a feature vector of the palm corresponding to the user identity is extracted. In step 250, the similarity between the palm feature vector and the underlying database features is calculated. This similarity may be cosine similarity. The identity information corresponding to the underlying database image 260 with the highest similarity is used as the recognition result.

[0050] Figure 3A shows a schematic diagram of a palmprint recognition process 300 according to an embodiment of the present application. This process uses a synthetic palmprint generation model 320 to generate a synthetic palmprint image. The screened synthetic palmprint image is input into a palmprint recognition model 340 as a pseudo-positive sample. In some embodiments, the synthetic palmprint generation model 320 includes a structure 400 of a deep learning model. For a detailed description of the deep learning model 400, see Figure 4. In block 310, an identity-controlled palmprint curve (hereinafter referred to as the identity-controlled curve) is synthesized, and the generated Bezier curve is used as the input of the synthetic palmprint generation model 320 to impose identity constraints on the synthetic palmprint generation model 320, so that it can generate identity-controllable synthetic palmprints. In block 330, latent variables are input into the synthetic palmprint generation model 320 to make the synthetic palmprint images generated by the trained synthetic palmprint generation model more robust. In some embodiments, the latent variables 330 can be random Gaussian noise. When training the palmprint recognition model 340, the real palmprint image and a portion of the synthesized palmprint images selected according to predetermined conditions are input into the palmprint recognition model 340 for palmprint recognition. The method for determining the synthesized palmprint generation model 320 and the palmprint recognition model 340 is described in detail with reference to Figures 3B, 3C, 3D, 8, and 9.

[0051] In the embodiments of the present application, the synthetic palmprint generation model 320 is optimized for scenarios with small sample training data, and a data augmentation component is added to the synthetic palmprint generation model 320. Furthermore, in some embodiments, the overfitting problem in the generative adversarial network is effectively addressed based on statistical data from the training process and an image quality comparison module.

[0052] Figure 3B shows a flow chart of a method for determining a synthetic palmprint generation model according to an embodiment of the present application. As shown in Figure 3B , the method 30 may include the following steps.

[0053] Step S31: Obtain a first training set, comprising real palm print images and identity control curves of multiple users. The real palm print image of each user corresponds to the user's unique user identity, and the identity control curve of each user is a parameterized palm print curve pre-generated for the user's unique user identity to simulate the palm print appearance.

[0054] Step S32: performing data enhancement on the first real palmprint image of the first user in the first training set to obtain the first real palmprint image after data enhancement.

[0055] Step S33: Obtain a first synthetic palmprint image corresponding to the first user generated by the deep learning model using the first real palmprint image after data enhancement and the first identity control curve of the first user.

[0056] Step S34: When the first synthetic palmprint image after data enhancement is identical to the first real palmprint image after data enhancement or the degree of similarity between them is not greater than a preset similarity threshold, adjust the parameters of the deep learning model.

[0057] Step S35: When the first synthetic palmprint image after data enhancement is different from the first real palmprint image after data enhancement and the degree of similarity between them is less than a similarity threshold, the deep learning model is determined as the synthetic palmprint generation model.

[0058] In each embodiment, steps S32-S34 can be performed repeatedly, i.e., iteratively training the deep learning model. Each iteration can use different training data from the first training set, i.e., {real palmprint image, identity control curve} data pairs. After each synthetic palmprint image is generated, the parameters of the deep learning model are adjusted based on the similarity between the real palmprint image and the synthetic palmprint image after data augmentation. This repeated iteration and adjustment process, i.e., the training process, enables the deep learning model to generate the desired synthetic palmprint image, i.e., meet the conditions of step S35, and training is complete.

[0059] In some embodiments, the data enhancement includes at least one of a geometric transformation, a color transformation, and a filtering transformation.

[0060] In some embodiments, the data enhancement includes data enhancement processing based on an enhancement probability, wherein the enhancement probability is determined based on the following parameters: the degree of overfitting between the data-enhanced second real palmprint image previously input into the deep learning model and the data-enhanced second synthetic palmprint image previously output by the deep learning model.

[0061] In some embodiments, the enhancement probability is determined according to the following steps: adjusting the enhancement probability by a preset amplitude value based on the degree of overfitting between the second synthetic palmprint image after data enhancement and the second real palmprint image after data enhancement, and adjusting the enhanced probability so that the adjusted enhanced probability falls within a preset first interval. When the enhancement probability is adjusted for the first time, the value of the enhanced probability is a preset initial value.

[0062] In some embodiments, step S33 may include:

[0063] Random Gaussian noise is input into the deep learning model to obtain the first synthetic palmprint image generated by the deep learning model using the random Gaussian noise to change the image features of the first real palmprint image.

[0064] In some embodiments, the identity control curve is a parameterized palmprint curve generated using a palmprint template to simulate the appearance of a palmprint. Each palmprint template includes a first number of first palmprint sub-curves and a second number of second palmprint sub-curves parameterized using a Bezier curve. In this case, step S33 may include obtaining the first synthetic palmprint image generated by the deep learning model using the first number of first palmprint sub-curves and the second number of second palmprint sub-curves in the identity control curve, where the palmprint lines in the first synthetic palmprint image correspond one-to-one to the palmprint sub-curves in the first palmprint sub-curve and the second palmprint sub-curve.

[0065] Figure 3C shows a flow chart of a method for determining a synthetic palmprint generation model according to an embodiment of the present application. As shown in Figure 3C, the method 30 shown in Figure 3B may further include the following steps.

[0066] Step S36, determining whether the number of times the parameters of the deep learning model have been adjusted has reached a predetermined number.

[0067] If the number of adjustments does not reach the predetermined number, step S32 is executed to obtain data of another user from the first training set to perform the next training on the deep learning model.

[0068] In step S37, after the parameters of the deep learning model have been adjusted a predetermined number of times, a synthetic palmprint image is selected from the multiple synthetic palmprint images generated by the deep learning model based on a reversal probability and added to the first training set for training the deep learning model. The reversal probability is determined based on the following parameters: the degree of overfitting between the data-augmented third real palmprint image previously input to the deep learning model and the data-augmented third synthetic palmprint image previously output by the deep learning model, and the reversal probability is within a second interval.

[0069] In some embodiments, adding a synthetic palmprint image selected from a plurality of synthetic palmprint images generated by a deep learning model with an inverted probability to the first training set comprises:

[0070] performing a quality comparison between an image in a first set of composite palmprint images and an image in a second set of composite palmprint images output by the deep learning model, where the second set of composite palmprint images is generated before the first set of composite palmprint images, and the deep learning model generates a predetermined number of composite palmprint images after the second set of composite palmprint images is generated and before the first set of composite palmprint images is generated;

[0071] In response to the image quality of the first set of composite palmprint images being higher than that of the second set of composite palmprint images, images selected from the first set of composite palmprint images with inverted probability are added to a first training set for training the deep learning model.

[0072] In some embodiments, comparing the quality of the images in the first group of composite palmprint images and the images in the second group of composite palmprint images output by the deep learning model includes:

[0073] Acquire a second training set, where the second training set includes a plurality of image pairs and corresponding quality comparison labels, each image pair includes a first image and a second image, and the quality comparison label indicates a quality comparison result of the first image and the second image;

[0074] extracting a first eigenvector of the first image and a second eigenvector of the second image;

[0075] Obtaining a concatenated eigenvector by concatenating the first eigenvector and the second eigenvector;

[0076] Inputting the spliced ​​feature vectors into the first neural network to obtain a prediction quality comparison result of the first image and the second image;

[0077] When the difference between the predicted quality comparison result and the quality comparison result indicated by the quality comparison label does not meet a preset minimization condition, or the distribution of the first eigenvector and the second eigenvector does not conform to a normal distribution, adjusting the parameters of the first neural network model;

[0078] When the difference between the predicted quality comparison result and the quality comparison result indicated by the quality comparison label satisfies the preset minimization condition and the distributions of the first eigenvector and the second eigenvector are each normally distributed, determining the first neural network model as a trained first neural network model;

[0079] The feature vectors of the image in the first group of composite palmprint images and the image in the second group of composite palmprint images output by the deep learning model are spliced ​​and input into the trained first neural network model to obtain a quality comparison result.

[0080] Figure 3D shows a method for determining a palmprint recognition model according to an embodiment of the present application. As shown in Figure 3D, the method 90 may include the following steps.

[0081] S91: Obtain a third training set, the third training set comprising real palmprint images and synthetic palmprint images. Each of the real palmprint images and synthetic palmprint images corresponds to a known user identity. The synthetic palmprint images are generated using a synthetic palmprint generation model determined by the method of any embodiment of the present application.

[0082] S92: Input the real palmprint image and the synthesized palmprint image into a palmprint recognition model to obtain a predicted user identity recognized by the palmprint recognition model.

[0083] S93: When the similarity between the known user identity and the predicted user identity is not greater than a preset similarity threshold, adjust the parameters of the palmprint recognition model.

[0084] S94: When the similarity between the known user identity and the predicted user identity is greater than the similarity threshold, providing the palmprint recognition model for identifying the user identity corresponding to the palmprint image.

[0085] Accordingly, the embodiment of the present application further provides a device for determining a synthetic palmprint generation model. FIG3E shows a schematic diagram of the device 40 for determining a synthetic palmprint generation model according to the embodiment of the present application. As shown in FIG3E , the device 40 for determining a synthetic palmprint generation model includes the following modules.

[0086] The acquisition module 41 may acquire a first training set. The first training set includes real palm print images and identity control curves of multiple users, wherein the real palm print image of each user corresponds to the user's unique user identity, and the identity control curve of each user is a parameterized palm print curve pre-generated for the user's unique user identity to simulate the palm print appearance.

[0087] The iteration module 42 can input the first real palmprint image of the first user in the first training set and the first identity control curve of the first user into the deep learning model to obtain a first synthetic palmprint image corresponding to the first user generated by the deep learning model using the first identity control curve and the first real palmprint image after data enhancement; when the first synthetic palmprint image after data enhancement is identical to the first real palmprint image after data enhancement or the degree of similarity between them is not greater than a preset similarity threshold, adjust the parameters of the deep learning model; when the first synthetic palmprint image after data enhancement is different from the first real palmprint image after data enhancement and the degree of similarity between them is less than the similarity threshold, determine the deep learning model as the synthetic palmprint generation model.

[0088] The present invention also provides a palmprint recognition model determination device. Figure 3F shows a schematic diagram of the palmprint recognition model determination device 50 of the present invention. As shown in Figure 3F, the palmprint recognition model determination device 50 includes the following modules.

[0089] The acquisition module 51 may acquire a third training set. The third training set includes real palmprint images and synthetic palmprint images, each of which corresponds to a known user identity. The synthetic palmprint images may be generated by the synthetic palmprint generation model determined in any embodiment of the present application.

[0090] The updating module 52 may input the real palmprint image and the synthesized palmprint image into a palmprint recognition model to obtain a predicted user identity identified by the palmprint recognition model; when the degree of similarity between the known user identity and the predicted user identity is not greater than a preset similarity threshold, adjust the parameters of the palmprint recognition model; when the degree of similarity between the known user identity and the predicted user identity is greater than the similarity threshold, provide the palmprint recognition model for identifying the user identity corresponding to the palmprint image.

[0091] The aforementioned device 40 for determining a synthetic palmprint generation model and the device 50 for determining a palmprint recognition model can be implemented by a computer device. The computer device includes a processor and a memory. The modules in the device can be implemented by software modules stored in the memory, for example, as computer-executable instructions. When the processor executes the instructions stored in the memory, the functions of the corresponding modules are realized.

[0092] The above technical solution will be described in detail in the following embodiments.

[0093] The training system for the synthetic palmprint generation model in each embodiment can also be implemented using a deep learning model, such as a generative adversarial network or a diffusion model. Figure 4 shows a schematic diagram of a training system 400 for the synthetic palmprint generation model in an embodiment of the present application. Training system 400 includes a generator and a discriminator. In scenarios where the training set consists of small samples, the discriminator may overfit. This can lead to unstable training and difficulty in convergence. To address the discriminator overfitting issue, a dynamic, differentiable data augmentation module is integrated into training system 400.

[0094] In each embodiment, the training system 400 can obtain a real palmprint image 410 and perform data enhancement processing on the real palmprint image 410 using a data enhancement module 420 to obtain an enhanced real palmprint curve. The enhanced real palmprint image is input into an encoder 430 to obtain vectorized real data features 440. The real data features 440, random noise features 450, and a Bezier curve 460 used as an identity control curve are input into a generator 470 to obtain a synthetic palmprint curve 480 for a unique user identity. Here, the identity control curve can be a Bezier curve, which is a parameterized palmprint curve pre-generated for a unique user identity to simulate the palmprint appearance. The identity control curve is pre-generated for a unique user identity based on a palmprint template. Each of the palmprint templates includes a first palmprint sub-curve parameterized using a Bezier curve and a second number of second palmprint sub-curves. For details about the palmprint templates, see FIG7 . Similarly, the synthetic palmprint image is also input into the data enhancement module 490 and subjected to the aforementioned data enhancement processing to obtain an enhanced synthetic palmprint image.

[0095] In some embodiments, a deep learning model serving as a synthetic palmprint generation model (hereinafter collectively referred to as a synthetic palmprint generation model) may include the encoder 430 and generator 470 in the aforementioned training system 400. In some embodiments, the synthetic palmprint generation model may include the data enhancement module 420, encoder 430, and generator 470 in the aforementioned training system 400. In some embodiments, the synthetic palmprint generation model may include the data enhancement module 420, encoder 430, generator 470, and data enhancement module 490 in the aforementioned training system 400.

[0096] The parameters of the synthetic palmprint generation model are iteratively updated so that the enhanced synthetic palmprint image and the enhanced real palmprint image are different and have a degree of similarity greater than a similarity threshold. The synthetic palmprint generation model with updated parameters is determined as the trained synthetic palmprint generation model. The similarity threshold can be a value set by those skilled in the art based on empirical values, and the value is greater than 0.5 and less than 1. In another example, the similarity threshold can be a value greater than 0.8 and less than 1. As will be appreciated by those skilled in the art, the similarity threshold can be any other suitable value. The similarity threshold can be set so that the enhanced synthetic palmprint image and the enhanced real palmprint image are different and have the greatest degree of similarity.

[0097] In some embodiments, the data enhancement process may include performing at least one of geometric transformation, color transformation, and filtering transformation on the real / synthetic palmprint image to obtain an enhanced real palmprint image.

[0098] In some embodiments, the data enhancement performs data enhancement processing on the user's real palmprint image according to a predetermined enhancement probability, where the enhancement probability depends on the degree of overfitting between the enhanced synthetic palmprint image and the enhanced real palmprint image, to obtain the enhanced real palmprint image.

[0099] The loss functions of the generator 470 and the discriminator 4100 are as follows:

[0100] G is the generator, D is the discriminator, is the loss function of the generator G, is the transformation, and z is Gaussian noise.

[0101] G is the generator, D is the discriminator, is the loss function of the discriminator D, For transformation, A real is a real image randomly drawn with inverted probability.

[0102] here This stands for Random Data Augmentation Graphics Transformation. It applies geometric and color transformations to randomly sampled images. These transformations include isotropic scattering, rotation, translation, color brightness change, and contrast adjustment.

[0103] In some embodiments, the data enhancement further includes filtering the sample space in which the image resides. After Fourier transforming the image, the signal strengths of different frequencies are sampled, thereby achieving enhancement for a specific frequency band.

[0104] Figures 5A and 5B show the effect of data augmentation in the scenario of discriminator overfitting. Figure 5A uses a fixed data augmentation probability, and Figure 5B uses a dynamic data augmentation probability. The degree of discriminator overfitting varies greatly at different stages, and dynamic data augmentation can solve this problem. In Figures 5A and 5B, the vertical axis represents the expected value of the discriminator for real images and synthetic images, and the horizontal axis represents the number of training rounds (epochs). The expected value of the real image is The expected value of the synthetic image is exist Approaching 1, When it approaches 0, it means that the discriminator has experienced severe overfitting. The main reason for overfitting is that the discriminator has memorized most of the real image samples, making it unable to provide effective feedback to the generator. In this case, increasing the probability of data augmentation will effectively alleviate the discriminator's overfitting problem. The degree of overfitting r can be expressed as follows:

[0105] Where r is the degree of overfitting, D is the discriminator, G is the generator, z is Gaussian noise, A real For real pictures.

[0106] In some embodiments, the enhancement probability is determined based on the following steps: an initial value of the enhancement probability is set as the initial enhancement probability of the deep learning model. For example, the initial data enhancement probability is set to p1 equal to 0.2. After each iteration, the enhancement probability is adjusted based on the degree of overfitting between the enhanced synthetic palmprint image and the enhanced real palmprint image, and the adjusted enhancement probability is used as the enhancement probability for the next iteration. When the overfitting degree r exceeds a certain threshold, the probability of data enhancement is dynamically adjusted. Specifically, the dynamic adjustment is performed according to the following formula.

[0107] Where p1 is the data augmentation probability, and the expected value of G(z) and the expected value of D Range constraints are applied. p1 lower bound is the lower bound of the data augmentation probability, p1 upper bound is the upper bound of the data augmentation probability, and α is the change in the expected value of a single time Δ.

[0108] Due to the characteristics of palmprint images, an excessively high data enhancement probability will cause the generated image to produce color collapse, while an initially too small enhancement probability will not be able to effectively solve the overfitting problem. Therefore, an interval constraint is added to the data enhancement here. In one embodiment, the enhancement probability is determined according to the following steps: based on the degree of overfitting between the enhanced synthetic palmprint image after the last iterative update and the enhanced real palmprint image (that is, the synthetic palmprint image generated by the synthetic palmprint generation model last time and the real palmprint image as input after data enhancement), the enhancement probability is determined; wherein, when the last iteration is the first iteration, the enhancement probability is set to the initial value, and the enhancement probability is within the first interval.

[0109] Figures 6A and 6B show schematic diagrams of a palmprint template. As shown in Figure 6A, the palmprint template can be composed of a first number of first palmprint sub-curves and a second number of second palmprint sub-curves. As shown in Figure 6A, the palmprint is basically composed of a first number (for example, 3 to 5) of main lines and a second number of wrinkles. In order to fit the geometric appearance of the palmprint, the Bezier function is used to parameterize the palmprint. For simplicity, a second-order Bezier curve with three parameter points on a two-dimensional plane is used in Figure 6B. Here, the three control points are: the starting point marked by a triangle, the control point marked by a five-pointed star, and the end point marked by a circle. Here, the three main lines of the left hand are taken as an example. In one example, the situation on the right hand can be regarded as a mirror image of the left hand.

[0110] Figure 7 shows a schematic diagram 700 of a quality comparison of synthetic palmprint images in an embodiment of the present application. In a deep learning model, the generator can effectively offset the discriminator's overfitting problem. When the discriminator's overfitting exceeds a predetermined threshold, a portion of the synthetic palmprint image can be dynamically (e.g., with probability p2) incorporated into the real image during training.

[0111] In each embodiment, after the parameters of the synthetic palmprint generation model are iteratively updated for a predetermined number of iterations, an image in the synthetic palmprint image is selected with a reversal probability p2; the synthetic palmprint images selected with the reversal probability p2 are merged into a training set for training the synthetic palmprint generation model.

[0112] In some embodiments, for an image in a synthetic palmprint image output at a first iteration (e.g., multiple iterations in a certain iteration round) selected with a reversal probability p2, a quality comparison is performed between the image in the synthetic palmprint image output at the first iteration and a different synthetic palmprint image output at a second iteration, where the second iteration is subsequent to the first iteration and a predetermined number of iterations separate the second iteration. In response to the selected image in the synthetic palmprint image output at the first iteration having higher quality than the different synthetic palmprint image output at the second iteration, the selected image in the synthetic palmprint image is added to a training set for training a synthetic palmprint generation model.

[0113] In some embodiments, for an image in a synthetic palmprint image output at a first iteration selected with an inversion probability p2, performing a quality comparison between the image in the synthetic palmprint image output at the first iteration and a different synthetic palmprint image output at a second iteration includes: obtaining a second training set, the second training set including a plurality of image pairs and their corresponding quality comparison labels, each image pair including a first image and a second image, the quality comparison label indicating a quality comparison result of the first image and the second image; encoding the first image and the second image respectively to obtain a first feature vector and a second feature vector; concatenating the first feature vector and the second feature vector to obtain a concatenated feature vector; inputting the concatenated feature vector into a first neural network to obtain a predicted quality comparison result of the first image and the second image; iteratively updating the parameters of the first neural network to minimize the difference between the predicted quality comparison result and the quality comparison result indicated by the quality comparison label, updating the parameters of the first neural network model, and determining the first neural network model after the updated parameters as the trained first neural network model; inputting the image in the synthetic palmprint image output at the first iteration and the different synthetic palmprint images output at the second iteration into the trained first neural network model to obtain a quality comparison result.

[0114] Here, the loss function can be expressed as follows:

[0115] is the loss function of the discriminator D, A real is a real image randomly extracted with an inversion probability, and the inversion probability p2 is the probability of extracting from the synthetic palmprint image and incorporating it into the training set.

[0116] The update probability of p2 can be defined as follows:

[0117] where p2 is the probability of reversal, and the expected value of G(z) is and the expected value of D Range constraints are applied. p2 lower bound is the lower bound of the reversal probability, p2 upper bound is the upper bound of the reversal probability, and α is the change in the expected value of a single time, Δ.

[0118] The probability interval serves as a hyperparameter for the model. In one embodiment, after the parameters of the synthetic palmprint generation model are iteratively updated for a predetermined number of iterations, images from the synthetic palmprint images are selected using an inversion probability. The synthetic palmprint images selected using the inversion probability are then merged into the training set for training the synthetic palmprint generation model, where the inversion probability is determined based on the degree of overfitting between the enhanced synthetic palmprint image and the enhanced real palmprint image after the iterative update prior to the current iteration, and the inversion probability lies within a second interval. Initially, the generated image quality is poor, and prematurely introducing positive samples can lead to difficulties in model convergence and a decrease in the quality of the generated images. Initially, the discriminator's overfitting is not significant. Therefore, a constraint is proposed: only after a certain number of rounds should the optimization method be introduced into model training. This strategy helps improve the performance and stability of generative adversarial networks.

[0119] To further ensure image quality in the pseudo-positive dataset, this method designs an efficient image quality comparison module. This module compares newly generated images with other images from the previous training round, ensuring that only high-quality newly generated images are included in the pseudo-positive dataset. This approach helps improve the performance of the generator, thereby offsetting the overfitting problem of the discriminator to a certain extent. The training set is derived from synthetic palmprint images during the training of the generative model. Figure 7 assumes that within a certain range, the image quality of the adversarial generative network increases with the number of training rounds. By pairing palmprint images synthesized later in the training round with images synthesized earlier in the training round, a large number of image pairs of varying quality can be constructed, thereby forming the image quality comparison module. It is worth noting that a certain distance between the previous and next training rounds is required for effective results; otherwise, the quality of the two will be relatively close.

[0120] Image A and image B, which follows the training round number, are each passed through an encoder to extract real image features. In one example, the dimension of the extracted feature vector is 16*16. The encoder can use a 4-layer convolutional structure to effectively capture the features in the real palm print image. Next, the input multi-dimensional tensor is flattened into a one-dimensional tensor. Constraints are added to the feature vector. The distribution of the eigenvector N(μ Q ,σ Q ) is normally distributed, where μ Q is the mean value, σ Qis the standard deviation. Loss function as follows:

[0121] By splicing the feature vectors of image A and image B and inputting them into a multi-layer neural network, the result generated by the activation function sigmoid is the final image quality comparison result.

[0122] FIG8 is a flow chart of a method 800 for determining a synthetic palmprint generation model according to an embodiment of the present application. The method 800 is executed by a computer device. The method includes:

[0123] In step 810, a training set is obtained, which includes real palmprint images and identity control curves of multiple users. As understood by those skilled in the art, the real palmprint image here is not a picture of the user's hand, but an image of a hand region of interest extracted from the hand picture. Each user's real palmprint image corresponds to a unique user identity, and each user's identity control curve is a parameterized palmprint curve pre-generated for the unique user identity to simulate the palmprint appearance. In one embodiment, the control curve for each user identity is pre-generated for the unique user identity and is used to simulate the parameterized palmprint curve of the palmprint appearance. In one embodiment, each palmprint template includes a first number of first palmprint sub-curves and a second number of second palmprint sub-curves parameterized using a Bezier curve.

[0124] Palm prints are essentially composed of a first number (e.g., 3-5) of primary lines and a second number of wrinkles. To fit the geometric appearance of palm prints, a Bezier function is used to parameterize the palm prints. A second-order Bezier curve with three parameter points on a two-dimensional plane is used. Here, the three control points are: a starting point marked by a triangle, a control point marked by a five-pointed star, and an end point marked by a circle. Here, the three primary lines on the left hand are used as an example. In one example, the situation on the right hand can be considered a mirror image of the left hand.

[0125] In step 820, data enhancement processing is performed on the real palmprint image to obtain an enhanced real palmprint image.

[0126] In one embodiment, at least one of geometric transformation, color transformation, and filtering transformation is performed on the user's real palm print image to obtain an enhanced real palm print image. The geometric and color transformations include methods such as isotropic scattering, rotation, translation, color brightness change, and contrast adjustment.

[0127] In one embodiment, the data enhancement further includes filtering the sample space where the image is located. After Fourier transforming the image, the signal strengths of different frequencies are sampled, thereby achieving enhancement for a specific frequency band.

[0128] In another embodiment, data augmentation processing is performed on the user's real palmprint image according to a predetermined augmentation probability, where the augmentation probability depends on the degree of overfitting between the enhanced synthetic palmprint image and the enhanced real palmprint image, to obtain an augmented real palmprint image. Additionally, the augmentation probability is determined based on the following steps: setting an initial value of the augmentation probability as the initial augmentation probability of the deep learning model; adjusting the augmentation probability after each iteration based on the degree of overfitting between the enhanced synthetic palmprint image and the augmented real palmprint image, and using the adjusted augmentation probability as the augmentation probability for the next iteration.

[0129] For example, the initial data augmentation probability is set to p1 equal to 0.2. After each iteration, the augmentation probability is adjusted based on the degree of overfitting between the enhanced synthetic palmprint image and the enhanced real palmprint image, and the adjusted augmentation probability is used as the augmentation probability for the next iteration. When the overfitting degree r exceeds a certain threshold, the data augmentation probability is dynamically adjusted.

[0130] Specifically, dynamic adjustment is performed according to the following formula.

[0131] Where p1 is the data augmentation probability, and the expected value of G(z) and the expected value of D Range constraints are applied. p1 lower bound is the lower bound of the data augmentation probability, p1 upper bound is the upper bound of the data augmentation probability, and α is the change in the expected value of a single time Δ.

[0132] Due to the characteristics of palm print images, excessively high data augmentation probabilities can cause color collapse in the generated images, while initially too low an augmentation probability cannot effectively address overfitting. Therefore, we add interval constraints to data augmentation.

[0133] In step 830, the enhanced real palmprint image and the identity control curve are input into the deep learning model to obtain a synthetic palmprint image for a unique user identity.

[0134] In step 840, data enhancement processing is performed on the synthetic palmprint image to obtain an enhanced synthetic palmprint image. The data enhancement processing here is the same as the data enhancement in step 820. Specifically, data enhancement can include at least one of geometric transformation, color transformation, and filtering transformation to obtain an enhanced real palmprint image. The geometric and color transformations include methods such as isotropic scattering, rotation, translation, color brightness change, and contrast adjustment. In one embodiment, the data enhancement also includes filtering processing on the sample space where the image is located. After the image is Fourier transformed, the signal intensities of different frequencies are sampled, thereby achieving enhancement for specific frequency bands.

[0135] In step 850, the parameters of the deep learning model are iteratively updated so that the enhanced synthetic palmprint image is different from the enhanced real palmprint image and the degree of similarity is greater than a similarity threshold, and the deep learning model with updated parameters is determined as the synthetic palmprint generation model. The similarity threshold can be a value set by those skilled in the art based on empirical values, and the value is greater than 0.5 and less than 1. In another example, the similarity threshold can be a value greater than 0.8 and less than 1. As will be understood by those skilled in the art, the similarity threshold can be any other suitable value. The similarity threshold can be set so that the enhanced synthetic palmprint image is different from the enhanced real palmprint image and the degree of similarity is maximized.

[0136] In the case where the above-mentioned deep learning model is a generative adversarial network, the following loss function can be constructed to ensure that the enhanced synthetic palmprint image is different from the enhanced real palmprint image and the degree of similarity is greater than the similarity threshold:

[0137] The loss functions of the generator and discriminator are as follows:

[0138] G is the generator, D is the discriminator, is the loss function of the generator G, is transformation, and z is Gaussian noise.

[0139] G is the generator, D is the discriminator, is the loss function of the discriminator D, is the transformation, A real is a real image randomly drawn with inverted probability.

[0140] In one embodiment, the method further includes: after iteratively updating the parameters of the synthetic palmprint generation model for a predetermined number of iterations, selecting an image from the synthetic palmprint image with an inversion probability; and merging the synthetic palmprint images selected with the inversion probability into a training set for training the synthetic palmprint generation model.

[0141] Specifically, the synthetic palmprint images selected with the inversion probability are merged into a training set for training the synthetic palmprint generation model, including: for an image in the synthetic palmprint image output at a first iteration number selected with the inversion probability, comparing the quality of the image in the synthetic palmprint image output at the first iteration number with a different synthetic palmprint image output at a second iteration number, the second iteration number being after the first iteration number and separated by a predetermined number of iterations; and in response to the quality of the image in the synthetic palmprint image output at the first iteration number being higher than the different synthetic palmprint images output at the second iteration number, adding the selected image in the synthetic palmprint image to the training set for training the synthetic palmprint generation model.

[0142] In one embodiment, the enhanced real palmprint image, the identity control curve, and random Gaussian noise are input into a deep learning model to obtain a synthetic palmprint image for a unique user identity. The addition of random Gaussian noise enhances the robustness of the deep learning model.

[0143] This method effectively improves training stability and palmprint synthesis quality in small sample scenarios by performing data augmentation on real and synthetic palmprints, thereby enhancing the performance of the synthetic palmprint generation model. By generating highly accurate and user-identity-controllable palmprint images, it provides a large amount of labeled data for deep learning model training, thereby meeting the demand for large-scale datasets for deep learning-based palmprint recognition technology, significantly improving model performance, and reducing the palmprint recognition task's reliance on real samples. This method thus addresses the technical issues of overfitting and unstable training processes in synthetic palmprint generation models when the quantity and quality of public palmprint datasets are limited. Furthermore, by combining an image quality comparison module with statistical data from the training process, the degree of model overfitting is effectively measured. Through dynamic data augmentation and regularization methods, the model's training stability and image synthesis quality in small sample scenarios are significantly improved.

[0144] FIG9 illustrates a method 900 for determining a palmprint recognition model according to an embodiment of the present application. In step 910, a training set is obtained. The training set includes real palmprint images and synthetic palmprint images. Each of the real palmprint images and synthetic palmprint images corresponds to a known user identity. The synthetic palmprint images are generated using a synthetic palmprint generation model such as that shown in FIG8 . The determination of the synthetic palmprint generation model is described in detail in connection with FIG8 and will not be further elaborated here.

[0145] In step 920, both the real palmprint image and the synthesized palmprint image are input into a palmprint recognition model to obtain an identified predicted user identity.

[0146] In step 930, the parameters of the palmprint recognition model are iteratively updated so that the similarity between the known user identity and the predicted user identity is greater than a similarity threshold, the parameters of the palmprint recognition model are updated, and the palmprint recognition model with updated parameters is determined as the palmprint recognition model.

[0147] FIG10 schematically shows a block diagram of a device 1000 for determining a synthetic palmprint generation model. The apparatus 1000 for determining a synthetic palmprint generation model includes: an acquisition module 1010 configured to acquire a first training set, the training set including real palmprint images and identity control curves of multiple users, wherein the real palmprint image of each user corresponds to a unique user identity, and the identity control curve of each user is a parameterized palmprint curve pre-generated for the unique user identity to simulate the palmprint appearance; a first enhancement module 1020 configured to perform data enhancement processing on the real palmprint image to obtain an enhanced real palmprint image; an input module 1030 configured to input the enhanced real palmprint image and the identity control curve into a deep learning model to obtain a synthetic palmprint image for the unique user identity; a second enhancement module 1040 configured to perform data enhancement processing on the synthetic palmprint image to obtain an enhanced synthetic palmprint image; and an iteration module 1050 configured to iteratively update parameters of the deep learning model so that the enhanced synthetic palmprint image is different from the enhanced real palmprint image and has a degree of similarity greater than a similarity threshold, and determine the deep learning model with the updated parameters as the synthetic palmprint generation model.

[0148] It should be understood that the apparatus 1000 for determining a synthetic palmprint generation model can be implemented in software, hardware, or a combination of software and hardware. Multiple different modules in the apparatus can be implemented in the same software or hardware structure, or one module can be implemented by multiple different software or hardware structures.

[0149] Furthermore, the synthetic palmprint generation model determination device 1000 can be used to implement the synthetic palmprint generation model determination method described above, the relevant details of which have been described in detail above and will not be repeated here for the sake of brevity. In addition, these devices can have the same features and advantages as those described for the corresponding methods.

[0150] FIG11 schematically illustrates a block diagram of a palmprint recognition model determination apparatus 1100. The palmprint recognition model determination apparatus 1100 includes: an acquisition module 1110 configured to acquire a third training set, the training set comprising real palmprint images and synthetic palmprint images, each of the real palmprint images and the synthetic palmprint images corresponding to a known user identity, wherein the synthetic palmprint images are generated using the synthetic palmprint generation model according to claim 1; an input module 1120 configured to input both the real palmprint image and the synthetic palmprint image into the palmprint recognition model to obtain an identified predicted user identity; and an update module 1130 configured to iteratively update parameters of the palmprint recognition model so that the degree of similarity between the known user identity and the predicted user identity exceeds a similarity threshold, update the parameters of the palmprint recognition model, and determine the palmprint recognition model with the updated parameters as the palmprint recognition model, so that the palmprint recognition model is used to identify the user identity corresponding to the palmprint image.

[0151] It should be understood that the palmprint recognition model determination device 1100 can be implemented in software, hardware, or a combination of software and hardware. Multiple different modules in the device can be implemented in the same software or hardware structure, or one module can be implemented by multiple different software or hardware structures.

[0152] In addition, the palmprint recognition model determination device 1100 can be used to implement the palmprint recognition model determination method described above, the relevant details of which have been described in detail above and will not be repeated here for the sake of brevity. In addition, these devices can have the same features and advantages as those described in the corresponding method.

[0153] FIG12 illustrates an example system 1200, which includes an example computing device 1210 that represents one or more systems and / or devices that can implement the various methods described herein. Computing device 1210 can be, for example, a server of a service provider, a device associated with a server, a system on a chip, and / or any other suitable computing device or computing system. The apparatus 1000 for determining a synthetic palmprint generation model and the apparatus 1100 for determining a palmprint recognition model described above with reference to FIG10-11 can take the form of computing device 1210. Alternatively, the apparatus 1000 for determining a synthetic palmprint generation model and the apparatus 1100 for determining a palmprint recognition model described in FIG10-11 can be implemented as a computer program in the form of an application 1216.

[0154] The example computing device 1210 as shown includes a processing system 1211, one or more computer-readable media 1212, and one or more I / O interfaces 1213 that are communicatively coupled to each other. Although not shown, the computing device 1210 may also include a system bus or other data and command transmission system that couples various components to each other. The system bus may include any one or a combination of different bus structures, such as a memory bus or memory controller, a peripheral bus, a universal serial bus, and / or a processor or local bus utilizing any one of a variety of bus architectures. Various other examples are also contemplated, such as control and data lines.

[0155] Processing system 1211 represents functionality that uses hardware to perform one or more operations. Thus, processing system 1211 is illustrated as including hardware elements 1214 that may be configured as processors, functional blocks, and the like. This may include hardware implemented as application-specific integrated circuits or other logic devices formed using one or more semiconductors. Hardware elements 1214 are not limited by the materials from which they are formed or the processing mechanisms employed therein. For example, a processor may be comprised of (a plurality of) semiconductors and / or transistors (e.g., electronic integrated circuits (ICs)). In such a context, processor-executable instructions may be electronically executable instructions.

[0156] Computer-readable media 1212 is illustrated as including memory / storage 1215. Memory / storage 1215 represents memory / storage capacity associated with one or more computer-readable media. Memory / storage 1215 may include volatile media (such as random access memory (RAM)) and / or non-volatile media (such as read-only memory (ROM), flash memory, optical disks, magnetic disks, etc.). Memory / storage 1215 may include fixed media (e.g., RAM, ROM, fixed hard drives, etc.) and removable media (e.g., flash memory, removable hard drives, optical disks, etc.). Computer-readable media 1212 may be configured in various other ways, as further described below.

[0157] One or more I / O interfaces 1213 represent functionality that allows a user to input commands and information to the computing device 1210 using various input devices, and optionally also allows information to be presented to the user and / or other components or devices using various output devices. Examples of input devices include a keyboard, a cursor control device (e.g., a mouse), a microphone (e.g., for voice input), a scanner, touch functionality (e.g., a capacitive or other sensor configured to detect physical touch), a camera (e.g., that can detect motion that does not involve touch as gestures using visible or invisible wavelengths (such as infrared frequencies)), and the like. Examples of output devices include a display device, a speaker, a printer, a network card, a tactile response device, and the like. Thus, the computing device 1210 can be configured in various ways, as further described below, to support user interaction.

[0158] The computing device 1210 also includes an application 1216. The application 1216 may be, for example, a software instance of the apparatus 1000 for determining a synthetic palmprint generation model and the apparatus 1100 for determining a palmprint recognition model described in FIG10-11, and implements the technology described herein in combination with other elements in the computing device 1210.

[0159] The present application provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computing device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computing device to perform the methods for presenting visual data provided in the various optional implementations described above.

[0160] Various techniques may be described herein in the general context of software, hardware, or program modules. Generally, these modules include routines, programs, objects, elements, components, data structures, and the like that perform specific tasks or implement specific abstract data types. As used herein, the terms "module," "function," and "component" generally refer to software, firmware, hardware, or a combination thereof. The techniques described herein are platform-independent, meaning that these techniques can be implemented on a variety of computing platforms with a variety of processors.

[0161] An implementation of the described modules and techniques may be stored on or transmitted across some form of computer-readable media. Computer-readable media may include various media accessible by computing device 1210. By way of example, and not limitation, computer-readable media may include "computer-readable storage media" and "computer-readable signal media."

[0162] As opposed to a mere signal transmission, carrier wave, or signal itself, "computer-readable storage medium" refers to a medium and / or device, and / or tangible storage device, capable of persistently storing information. Thus, computer-readable storage media refers to non-signal-bearing media. Computer-readable storage media include hardware such as volatile and non-volatile, removable and non-removable media and / or storage devices implemented in a method or technology suitable for storing information (such as computer-readable instructions, data structures, program modules, logic elements / circuits, or other data). Examples of computer-readable storage media may include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical storage devices, hard disks, cassettes, magnetic tape, magnetic disk storage devices or other magnetic storage devices, or other storage devices, tangible media, or articles of manufacture suitable for storing desired information and accessible by a computer.

[0163] "Computer-readable signal media" refers to signal-bearing media that is configured to send instructions to the hardware of the computing device 1210, such as via a network. Signal media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave, data signal, or other transport mechanism. Signal media also includes any information delivery media. The term "modulated data signal" refers to a signal that has one or more of its characteristics set or changed so as to encode information in the signal. By way of example, and not limitation, communication media includes wired media such as a wired network or direct connection, and wireless media such as acoustic, RF, infrared, and other wireless media.

[0164] As before, hardware elements 1214 and computer-readable media 1212 represent instructions, modules, programmable device logic and / or fixed device logic implemented in hardware form, which in some embodiments can be used to implement at least some aspects of the technology described herein. Hardware elements can include integrated circuits or systems on a chip, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), complex programmable logic devices (CPLDs) and other implementations in silicon or components of other hardware devices. In this context, hardware elements can be used as processing equipment for executing program tasks defined by the instructions, modules and / or logic embodied by the hardware elements, as well as hardware devices for storing instructions for execution, such as the computer-readable storage media described previously.

[0165] The aforementioned combination may also be used to implement various techniques and modules herein. Thus, software, hardware or program modules and other program modules may be implemented as one or more instructions and / or logic embodied on some form of computer-readable storage medium and / or by one or more hardware elements 1214. Computing device 1210 may be configured to implement specific instructions and / or functions corresponding to software and / or hardware modules. Thus, for example, by using a computer-readable storage medium and / or hardware elements 1214 of a processing system, a module may be implemented as a module executable by computing device 1210 as software, at least in part, in hardware. Instructions and / or functions may be executable / operable by one or more articles of manufacture (e.g., one or more computing devices 1210 and / or processing systems 1211) to implement the techniques, modules and examples herein.

[0166] In various embodiments, computing device 1210 can be implemented in a variety of different configurations. For example, computing device 1210 can be implemented as a computer-type device including a personal computer, a desktop computer, a multi-screen computer, a laptop computer, a netbook, etc. Computing device 1210 can also be implemented as a mobile device-type device including mobile devices such as mobile phones, portable music players, portable gaming devices, tablet computers, multi-screen computers, etc. Computing device 1210 can also be implemented as a television-type device, which includes devices having or connected to generally larger screens in casual viewing environments. These devices include televisions, set-top boxes, game consoles, etc.

[0167] The techniques described herein can be supported by these various configurations of computing device 1210 and are not limited to the specific examples of the techniques described herein. Functionality can also be implemented in whole or in part on the "cloud" 1220 using a distributed system, such as through platform 1222 as described below.

[0168] Cloud 1220 includes and / or represents a platform 1222 for resources 1224. Platform 1222 is the underlying functionality of the hardware (e.g., servers) and software resources of cloud 1220. Resources 1224 may include applications and / or data that can be used when executing computer processing on servers remote from computing device 1210. Resources 1224 may also include services provided over the Internet and / or over a subscriber network such as a cellular or Wi-Fi network.

[0169] The platform 1222 can abstract resources and functionality to connect the computing device 1210 with other computing devices. The platform 1222 can also be used to abstract hierarchies of resources to provide a corresponding level of hierarchy in the demand encountered for resources 1224 implemented via the platform 1222. Thus, in an interconnected device embodiment, the implementation of the functionality described herein can be distributed throughout the system 1200. For example, functionality can be implemented partially on the computing device 1210 and through the platform 1222 that abstracts the functionality of the cloud 1220.

[0170] Should be understood that, for the sake of clarity, the embodiments of the present application are described with reference to different functional units. However, it will be apparent that, without departing from the present application, the functionality of each functional unit can be implemented in a single unit, implemented in multiple units or implemented as a part for other functional units. For example, the functionality that is described as being performed by a single unit can be performed by multiple different units. Therefore, reference to a specific functional unit is only considered as a reference to the appropriate unit for providing the described functionality, rather than indicating strict logical or physical structure or organization. Therefore, the application can be implemented in a single unit, or can be physically and functionally distributed between different units and circuits.

[0171] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program that has a predetermined function and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0172] Although the present application has been described in conjunction with some embodiments, it is not intended to be limited to the specific forms set forth herein. On the contrary, the scope of the present application is limited only by the appended claims. Additionally, although individual features may be included in different claims, these may possibly be advantageously combined, and inclusion in different claims does not imply that a combination of features is not feasible and / or advantageous. The order of the features in the claims does not imply any specific order in which the features must work. Furthermore, in the claims, the word "comprising" does not exclude other elements, and the term "a" or "an" does not exclude a plurality. The reference numerals in the claims are provided merely as clear examples and should not be construed as limiting the scope of the claims in any way.

[0173] It is understood that in the specific implementation of this application, entity-related data such as entity default information is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use, and processing of relevant data must comply with relevant laws, regulations, and standards of relevant countries and regions.

[0174] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0175] In conclusion, the scope of the claims should not be limited to the exemplary embodiments described above, but should be given the broadest interpretation of the specification as a whole.

Claims

1. A method for determining a synthetic palm print generation model, executed by a computer device, the method comprising: Acquire a first training set, the first training set comprising real palm print images and identity control curves of a plurality of users, the real palm print image of each user among the plurality of users corresponding to the unique user identity of the user, and the identity control curve of each user being a parameterized palm print curve generated in advance for the unique user identity of the user for simulating the palm print appearance; Performing data enhancement on a first real palmprint image of a first user in the first training set to obtain a first real palmprint image after data enhancement; Obtain a first synthetic palmprint image corresponding to the first user generated by a deep learning model using the first real palmprint image after data enhancement and the first identity control curve of the first user; When the first synthetic palmprint image after data enhancement is identical to the first real palmprint image after data enhancement or the degree of similarity between them is not greater than a preset similarity threshold, adjusting the parameters of the deep learning model; When the first synthetic palmprint image after data enhancement is different from the first real palmprint image after data enhancement and the degree of similarity between them is less than the similarity threshold, the deep learning model is determined as the synthetic palmprint generation model.

2. The method according to claim 1, wherein: The data enhancement includes: At least one of a geometric transform, a color transform, and a filtering transform.

3. The method according to claim 1, wherein: The data enhancement includes: Data enhancement processing based on enhancement probability, wherein the enhancement probability is determined according to the following parameters: the degree of overfitting between the second real palmprint image that has been data-enhanced and input into the deep learning model last time and the second synthetic palmprint image that has been data-enhanced and output by the deep learning model last time.

4. The method according to claim 3, wherein: The enhancement probability is determined according to the following steps: Based on the overfitting degree between the second synthetic palmprint image after data enhancement and the second real palmprint image after data enhancement, adjusting the enhancement probability by a preset amplitude value, and making the adjusted enhancement probability fall within a preset first interval; Wherein, when the enhancement probability is adjusted for the first time, the value of the enhancement probability is a preset initial value.

5. The method according to claim 1, further comprising: After the number of adjustments to the parameters of the deep learning model reaches a predetermined number of times, a synthetic palmprint image is selected from the multiple synthetic palmprint images generated by the deep learning model with a reversal probability and added to the first training set for training the deep learning model; wherein the reversal probability is determined according to the following parameters: the degree of overfitting between the third real palmprint image that has been data-enhanced and input into the deep learning model last time and the third synthetic palmprint image that has been data-enhanced and output by the deep learning model last time, and the reversal probability is within a second interval.

6. The method according to claim 5, wherein: Adding a synthetic palmprint image selected from a plurality of synthetic palmprint images generated by the deep learning model with an inversion probability into the first training set comprises: Comparing the quality of an image in a first group of composite palmprint images and an image in a second group of composite palmprint images output by the deep learning model, wherein the second group of composite palmprint images is generated before the first group of composite palmprint images, and the deep learning model generates a predetermined number of composite palmprint images after the second group of composite palmprint images are generated and before the first group of composite palmprint images are generated; In response to the image quality of the first set of composite palmprint images being higher than the image quality of the second set of composite palmprint images, The images selected from the first group of synthetic palmprint images with inversion probability are added to the first training set for training the deep learning model.

7. According to the method of claim 6, comparing the quality of the images in the first group of composite palmprint images and the images in the second group of composite palmprint images output by the deep learning model comprises: Acquire a second training set, the second training set comprising a plurality of image pairs and corresponding quality comparison labels, each image pair comprising a first image and a second image, and the quality comparison label indicating a quality comparison result of the first image and the second image; Extracting a first eigenvector of the first image and a second eigenvector of the second image; Obtaining a concatenated feature vector by concatenating the first feature vector and the second feature vector; Inputting the spliced ​​feature vectors into the first neural network to obtain a prediction quality comparison result between the first image and the second image; When the difference between the predicted quality comparison result and the quality comparison result indicated by the quality comparison label does not satisfy a preset minimization condition, or the distribution of the first eigenvector and the second eigenvector does not conform to a normal distribution, adjusting the parameters of the first neural network model; When the difference between the predicted quality comparison result and the quality comparison result indicated by the quality comparison label satisfies the preset minimization condition and the distributions of the first eigenvector and the second eigenvector are each normally distributed, determining the first neural network model as a trained first neural network model; The feature vectors of the image in the first group of composite palmprint images and the image in the second group of composite palmprint images output by the deep learning model are spliced ​​and input into the trained first neural network model to obtain a quality comparison result.

8. The method according to claim 1, wherein: Obtaining a first synthetic palmprint image corresponding to the first user generated by a deep learning model using the first real palmprint image after data enhancement and the first identity control curve of the first user includes: Random Gaussian noise is input into the deep learning model to obtain the first synthetic palmprint image generated by the deep learning model using the random Gaussian noise to change the image features of the first real palmprint image.

9. The method according to any one of claims 1 to 8, wherein: The identity control curve is a parameterized palmprint curve for simulating the palmprint appearance generated by using a palmprint template, each palmprint template includes a first number of first palmprint sub-curves and a second number of second palmprint sub-curves parameterized by using a Bezier curve; Obtaining a first synthetic palmprint image corresponding to the first user generated by a deep learning model using the first real palmprint image after data enhancement and the first identity control curve of the first user includes: The first synthetic palmprint image generated by the deep learning model using a first number of first palmprint sub-curves and a second number of second palmprint sub-curves in the identity control curve is obtained, wherein the palmprint lines in the first synthetic palmprint image correspond one-to-one to the palmprint sub-curves in the first palmprint sub-curve and the second palmprint sub-curve.

10. A method for determining a palmprint recognition model, executed by a computer device, the method comprising: Acquire a third training set, the third training set comprising real palmprint images and synthetic palmprint images, each of the real palmprint images and the synthetic palmprint images corresponds to a known user identity, wherein the synthetic palmprint images are generated by a synthetic palmprint generation model determined by the method according to any one of claims 1 to 9; Inputting the real palmprint image and the synthetic palmprint image into a palmprint recognition model to obtain a predicted user identity recognized by the palmprint recognition model; When the similarity between the known user identity and the predicted user identity is not greater than a preset similarity threshold, adjusting the parameters of the palmprint recognition model; When the similarity between the known user identity and the predicted user identity is greater than the similarity threshold, the palmprint recognition model is provided for identifying the user identity corresponding to the palmprint image.

11. A device for determining a synthetic palmprint generation model, comprising: an acquisition module configured to acquire a first training set, wherein the first training set includes real palm print images and identity control curves of a plurality of users, wherein the real palm print image of each user among the plurality of users corresponds to a unique user identity of the user, and the identity control curve of each user is a parameterized palm print curve generated in advance for the unique user identity of the user for simulating the palm print appearance; An iteration module is configured to perform data enhancement on the first real palmprint image of the first user in the first training set to obtain the first real palmprint image after data enhancement; Obtain a first synthetic palmprint image corresponding to the first user generated by a deep learning model using the first real palmprint image after data enhancement and the first identity control curve of the first user; When the first synthetic palmprint image after data enhancement is the same as the first real palmprint image after data enhancement or the degree of similarity between them is not greater than a preset similarity threshold, the parameters of the deep learning model are adjusted; when the first synthetic palmprint image after data enhancement is different from the first real palmprint image after data enhancement and the degree of similarity between them is less than the similarity threshold, the deep learning model is determined as the synthetic palmprint generation model.

12. A palmprint recognition model determination device, the device comprising: an acquisition module, configured to acquire a third training set, wherein the third training set includes a real palmprint image and a synthetic palmprint image, each of the real palmprint image and the synthetic palmprint image corresponds to a known user identity, wherein the synthetic palmprint image is generated by a synthetic palmprint generation model determined by the method according to any one of claims 1 to 9; The updating module is configured to input the real palmprint image and the synthetic palmprint image into a palmprint recognition model to obtain a predicted user identity recognized by the palmprint recognition model; when the similarity between the known user identity and the predicted user identity is not greater than a preset similarity threshold, adjust the parameters of the palmprint recognition model; when the similarity between the known user identity and the predicted user identity is greater than the similarity threshold, provide the palmprint recognition model for identifying the user identity corresponding to the palmprint image.

13. A computing device comprising: a memory configured to store computer-executable instructions; A processor configured to perform a method according to any one of claims 1-10 when the computer executable instructions are executed by the processor.

14. A computer-readable storage medium storing computer-executable instructions, which when executed implement the method according to any one of claims 1 to 10.

15. A computer program product comprising a computer program which, when executed, implements the steps of the method according to any one of claims 1 to 10.

Images