Palmprint picture generating method, palmprint picture generating apparatus, computer program, and electronic device

Abstract

Each embodiment of the present application discloses a palm print picture generating method and apparatus, a storage medium, a program product, and an electronic device. The method includes: obtaining a simulated palm print picture, where the simulated palm print picture includes a simulated palm print curve combined with a target type curve; inputting the simulated palm print picture and a predetermined first noise vector into a trained target palm print picture generator, and generating the target palm print picture by sequentially performing multiple downsampling operations and multiple upsampling operations on the simulated palm print picture by the target palm print picture generator.

Description

[Technical Field]

[0001] This application claims priority from a Chinese patent application filed with the China Patent Office on June 20, 2023, bearing application number 2023107446120 and entitled "Method and apparatus for generating palm print pictures, storage medium and electronic device."

[0002] The present application relates to the technical fields of computers and communications, and in particular to a palm print picture generating method, apparatus, storage medium, program product, and electronic device. [Background technology]

[0003] Palm print recognition technology is the next generation of biometric authentication technology following fingerprint and face recognition technologies. Compared with fingerprint and face recognition technologies, palm prints are advantageous in protecting user privacy. Palm print recognition is closely related to users' personal privacy and property security in fields including mobile payment and identity verification, so recognition accuracy is extremely important.

[0004] Currently, sample palm print pictures can be used to train a palm print picture matching model that is waiting to be trained. However, in the actual use process, due to differences in the lighting conditions of the picture collection device or collection environment, the collected palm print pictures usually have different modalities, such as infrared modality and visible light modality, and the palm print picture matching model may have errors in recognizing palm print pictures of different modalities.

[0005] In order to minimize the recognition error, the training stage needs to be performed using a large number of sample palmprint pictures of different modalities having the same palmprint lines. Therefore, the training process of the palmprint picture matching model is highly dependent on the size and diversity of the sample palmprint pictures. However, due to the privacy concerns of palmprints, palmprint pictures are difficult to obtain, and multi-modality palmprint pictures that meet the above conditions are even rarer. Therefore, due to the low efficiency of palmprint picture generation, the size and diversity of the sample palmprint pictures are insufficient during the training process of the palmprint picture matching model. As a result, the trained palmprint picture matching model has low recognition ability for multi-modality palmprint pictures.

[0006] Currently, no effective solution has been proposed for the above problem. Summary of the Invention [Problem to be solved by the invention]

[0007] SUMMARY OF THE INVENTION The embodiments of the present application provide a palmprint picture generating method and apparatus, a storage medium, a program product and an electronic device that at least solve the technical problem of low palmprint picture generating efficiency. [Means for solving the problem]

[0008] According to one aspect of an embodiment of the present application, there is provided a method for generating a palm print picture, the method comprising: obtaining a simulated palm print picture, the simulated palm print picture including a simulated palm print curve combined with a target type curve; inputting the simulated palm print picture and a predetermined first noise vector into a trained target palm print picture generator, and generating a target palm print picture by sequentially performing multiple downsampling operations and multiple upsampling operations on the simulated palm print picture by the target palm print picture generator; the downsampling operation is for performing a downsampling process on an input picture representation vector to obtain a downsampled picture representation vector, and performing a noise addition process on the downsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector, wherein the input picture representation vector in the first downsampling operation is an initial picture representation vector of the simulated palmprint picture; The upsampling operation is for performing an upsampling process on an input picture representation vector to obtain an upsampled picture representation vector, and for performing a noise addition process on the upsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector.

[0009] According to another aspect of an embodiment of the present application, there is further provided an apparatus for generating a palm print picture, the apparatus comprising: An acquisition unit for acquiring a simulated palm print picture, wherein the simulated palm print picture includes a simulated palm print curve combined with a target type curve; an input unit for inputting the simulated palmprint picture and a predetermined first noise vector into a trained target palmprint picture generator, and generating a target palmprint picture by the target palmprint picture generator by sequentially performing multiple downsampling operations and multiple upsampling operations on the simulated palmprint picture; the downsampling operation is for performing a downsampling process on an input picture representation vector to obtain a downsampled picture representation vector, and performing a noise addition process on the downsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector, wherein the input picture representation vector in the first downsampling operation is an initial picture representation vector of the simulated palmprint picture; The upsampling operation is for performing an upsampling process on an input picture representation vector to obtain an upsampled picture representation vector, and for performing a noise addition process on the upsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector.

[0010] According to another aspect of the present invention, there is further provided a computer-readable storage medium having a computer program stored therein, the computer program being configured to, when executed, perform the above-described method for generating a palm print picture.

[0011] According to another aspect of the present invention, there is provided a computer program product or a computer program including computer instructions stored in a computer-readable storage medium, the computer instructions being read by a processor of a computing device from the computer-readable storage medium, and the processor executing the computer instructions causing the computing device to perform the method for generating a palm print picture as described above.

[0012] According to another aspect of the present invention, there is further provided an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the palm print picture generation method according to the computer program. [Brief explanation of the drawings]

[0013] The drawings described herein are intended to provide further understanding of the present application and constitute a part of the present application. The schematic examples and the description thereof are intended to aid in the interpretation of the present application and are not to be construed as undue limitations of the present application. [Figure 1] 1 is a flowchart of optional palm print recognition according to an embodiment of the present application. [Figure 2] 1 is a schematic diagram of an application environment of the method for generating optional palm print pictures according to an embodiment of the present application; FIG. [Figure 3] 1 is a flowchart of a method for generating an optional palm print picture according to an embodiment of the present application; [Figure 4] FIG. 2 is a schematic diagram of an optional picture representation vector according to an embodiment of the present application; [Figure 5] FIG. 10 is a schematic diagram of the generation of the i-th picture representation vector after optional noise addition according to an embodiment of the present application; [Figure 6] FIG. 10 is a schematic diagram illustrating a condition generation sub-module in the optional i-th downsampling module performing noise addition processing on a picture representation vector according to an embodiment of the present application; [Figure 7] FIG. 10 is a schematic diagram of the configuration of an optional condition generation sub-module according to an embodiment of the present application. [Figure 8] FIG. 10 is a schematic diagram of the generation of a picture representation vector after optional j-th noise addition according to an embodiment of the present application; [Figure 9] FIG. 10 is a schematic diagram illustrating a condition generation sub-module in the optional j-th down-sampling module performing noise addition processing on a picture representation vector according to an embodiment of the present application; [Figure 10] FIG. 2 is a schematic diagram of an optional first convolution process and a second convolution process according to an embodiment of the present application. [Figure 11] FIG. 1 is a schematic diagram of the training process of an optional training-ready palm print picture generator according to an embodiment of the present application; [Figure 12] FIG. 10 is a schematic diagram of determining a target loss value based on an optional first loss value in an embodiment of the present application. [Figure 13] FIG. 10 is a schematic diagram of determining a target loss value based on an optional first loss value and a second loss value in an embodiment of the present application. [Figure 14] FIG. 10 is a schematic diagram of determining a target loss value based on an optional first loss value, a second loss value, and a third loss value in an embodiment of the present application. [Figure 15] FIG. 10 is a schematic diagram of determining a target loss value based on an optional first loss value, a second loss value, a third loss value, and a fourth loss value in an embodiment of the present application. [Figure 16] FIG. 10 is a schematic diagram of determining a target loss value based on an optional first loss value, a second loss value, a third loss value, a fourth loss value, and a fifth loss value in an embodiment of the present application. [Figure 17] FIG. 1 is a schematic diagram of an overall block diagram of an algorithm for the generation of optional palm print pictures according to an embodiment of the present application. [Figure 18] FIG. 10 is a schematic diagram of an optional target palm print picture generator configuration according to an embodiment of the present application; [Figure 19] FIG. 2 is a schematic diagram of an optional encoder configuration according to an embodiment of the present application; [Figure 20] FIG. 10 is a schematic diagram of an optional generated target palm print picture according to an embodiment of the present application; [Figure 21] FIG. 1 is a block diagram of a configuration of a palm print picture generating device according to an embodiment of the present application. [Figure 22] 1 is a schematic diagram of an optional electronic device configuration according to an embodiment of the present application. [Figure 23] FIG. 2 is a block diagram of a computer system configuration of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION OF THE INVENTION

[0014] In order to help those skilled in the art understand the present invention better, the present invention will be described in detail below with reference to the drawings of the present invention. It is clear that the described embodiments are only some of the embodiments of the present invention, and not all of the embodiments. All other embodiments that those skilled in the art can obtain from the embodiments of the present invention without creative efforts should fall within the scope of protection of the present invention.

[0015] First, the following interpretations apply to some of the nouns and terms that appear in the description of the embodiments of the present application.

[0016] ROI: Region of Interest.

[0017] LReLU: An activation function, also known as Leaky ReLU. LReLU is similar to the traditional ReLU (Rectified Linear Unit), but differs in that when the input x is less than 0, it does not output 0, but instead has a small slope, which can improve the training effect of the model in some cases.

[0018] Flatten: A function used in neural networks to flatten a multidimensional array into a one-dimensional array. In deep learning models, the Flatten layer is typically used to flatten input data into a single vector for further processing.

[0019] BN: Batch Normalization is a widely used neural network layer that can speed up the training process and improve the accuracy of the model.

[0020] The palm print picture generation can be applied to palm print recognition technology. Below, we briefly introduce a palm print recognition scenario using palm print recognition technology. As shown in Figure 1, a palm print recognition scenario usually includes the following steps:

[0021] In step 1, a user hand picture 102 (specifically, it may be a user palm picture) is collected by a terminal device (e.g., an edge Internet of Things (IOT) device 101 including a terminal payment device and an identity collection device) and a region of interest (ROI) is extracted.

[0022] In step 2, the user hand picture 102 is input into a feature extraction model 103 (which can be understood as the palm print picture matching model described above) to obtain hand features (eg, palm feature vectors).

[0023] In step 3, the cosine similarity between the user hand features and the base features 105 is calculated (104), similarity ranking is performed (106), and the identity information corresponding to the base feature with the highest similarity (e.g., a base photo) is used as the recognition result, which is returned to the user (107). Here, the base features are features corresponding to a palm picture stored by the user to represent the user's identity, and it can be understood that one user corresponds to a unique set of base features.

[0024] As can be seen from the above process, the extraction ability of a feature extraction model for a user's hand picture directly determines the accuracy of the recognition result. In an actual palmprint recognition scenario, different types of terminal devices have different modalities for collecting user's hand pictures, so a feature extraction model is required to accurately extract features from user's hand pictures of various modalities. Therefore, during the training phase of the feature extraction model, training must be performed using a large number of sample palmprint pictures of different modalities that have the same palmprint lines. Due to palmprint privacy concerns, it is difficult to obtain sample palmprint pictures that meet the above conditions. In related art, the sample size obtained by artificially collecting human palmprints is small, and multi-modality palmprint pictures are more rare, resulting in low recognition ability of the trained palmprint picture matching model for multi-modality palmprint pictures.

[0025] It should be noted that the relevant information (including, but not limited to, user device information, user personal information, etc.) and data (including, but not limited to, display data, analysis data, etc.) related to the present application are all information and data authorized by the user or fully authorized by each party. For example, an interface is provided between the present system and the relevant user or institution, and before obtaining the relevant information, an acquisition request must be sent to the user or institution via the interface. Furthermore, the relevant information is obtained only after receiving approval information fed back from the user or institution.

[0026] According to one aspect of the embodiment of the present application, there is provided a palm print picture generating method. Optionally, as an optional embodiment, the above palm print picture generating method can be applied to devices such as, but not limited to, a terminal device or a server. The palm print picture generating method applied to a terminal device can be taken as an example for interpretation and explanation, but is not limited thereto.

[0027] As shown in FIG. 2, a method for generating a palm print picture will be described using an example in which P takes a value of 3 and Q takes a value of 2.

[0028] First, a simulated palm-print picture 201 is obtained, where the simulated palm-print picture 201 includes a simulated palm-print curve combined with a target-type curve.

[0029] The simulated palmprint picture 201 and the first noise vector 202 are input to a trained target palmprint picture generator 203, where the target palmprint picture generator 203 is for sequentially passing the simulated palmprint picture through P(3) downsampling modules (downsampling module 1, downsampling module 2, and downsampling module 3) and Q(2) upsampling modules (upsampling module 1 and upsampling module 2).

[0030] Here, the downsampling operation in the target palmprint picture generator 203 can be performed by a downsampling module, and the upsampling operation in the target palmprint picture generator can be performed by an upsampling module. The downsampling process can be performed by a downsampling sub-module, and the upsampling process can be performed by an upsampling sub-module. The condition generation sub-module can use the first noise vector to perform a noise addition process on the picture representation vector after the sampling process (downsampling process or upsampling process), thereby obtaining a picture representation vector after noise addition.

[0031] It should be noted that each of the P(3) downsampling modules includes a downsampling sub-module and a first condition generation sub-module. Taking downsampling module 1 as an example, downsampling sub-module 1 is for performing a downsampling process on an input picture representation vector 204 to obtain a downsampled picture representation vector 205, and first condition generation sub-module 1 is for using the first noise vector 202 to perform a noise addition process 206 on the downsampled picture representation vector 205 to obtain a noise-added picture representation vector 207.

[0032] Each of the Q(2) upsampling modules (upsampling module 1 and upsampling module 2) includes an upsampling sub-module and a second condition generation sub-module. The upsampling sub-module is for performing an upsampling process on an input picture representation vector to obtain an upsampled picture representation vector, and the second condition generation sub-module is for performing the noise addition process on the upsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector.

[0033] The flow of data (picture representation vectors) in the target palmprint picture generator is downsampling module 1, downsampling module 2, downsampling module 3, upsampling module 1, and upsampling module 2 in that order, ultimately producing the target palmprint picture 208. The picture representation vector after noise addition output from the previous sampling module (downsampling module or upsampling module) is the picture representation vector input to the next sampling module, and the one input to downsampling module 1 is the picture representation vector without noise addition.

[0034] Optionally, in this embodiment, the terminal device may be a terminal device on which a target client is located, and may include, but is not limited to, at least one of a mobile phone (e.g., an Android mobile phone, an iOS mobile phone, etc.), a laptop, a tablet computer, a palmtop computer, a mobile Internet device (MID), a PAD, a desktop computer, a smart TV, etc. The target client may be a video client, an instant communication client, a browser client, an education client, etc. The network may include, but is not limited to, a wired network and a wireless network. Here, the wired network includes a local area network, a metropolitan area network, and a wide area network, and the wireless network includes Bluetooth, WIFI, and other networks that realize wireless communication. The server may be a single server, a server cluster consisting of multiple servers, or a cloud server. The above is just an example. However, this embodiment is not limited to this.

[0035] Optionally, as an optional embodiment, as shown in FIG. 3 , the above palm print picture generating method comprises: Step S12 of acquiring a simulated palm print picture, in which the simulated palm print picture includes a simulated palm print curve combined with a target type curve; a step S14 of inputting the simulated palm print picture and a predetermined first noise vector into a trained target palm print picture generator, and generating a target palm print picture by the target palm print picture generator by sequentially performing multiple downsampling operations and multiple upsampling operations on the simulated palm print picture, the downsampling operation is for performing a downsampling process on an input picture representation vector to obtain a downsampled picture representation vector, and performing a noise addition process on the downsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector, wherein the input picture representation vector in the first downsampling operation is an initial picture representation vector of the simulated palmprint picture; The upsampling operation includes step S14, in which an upsampling process is performed on the input picture representation vector to obtain an upsampled picture representation vector, and a noise addition process is performed on the upsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector.

[0036] Optionally, in this embodiment, the process of generating the target palmprint picture only needs to input the simulated palmprint picture and the first noise vector into a trained target palmprint picture generator.

[0037] It is worth noting that the related art relies on artificially collecting human palmprints, and for the same human palmprint, pictures of different modalities (which can be understood as the above-mentioned target palmprint pictures) must be collected multiple times as a palmprint picture set to train a palmprint picture matching model. In contrast, the present application does not require artificial collection, and only uses simulated palmprint pictures and a first noise vector to generate a palmprint picture set of pictures of different modalities with matching palmprint line features (which can be understood as the above-mentioned target palmprint pictures). This significantly improves the efficiency of generating target palmprint pictures, and solves the problem in the prior art that, due to the difficulty of obtaining palmprints due to privacy concerns, the sample size obtained by artificially collecting human palmprints is small, and multi-modality palmprint pictures are rare, resulting in low recognition ability of the trained palmprint picture matching model for multi-modality palmprint pictures.

[0038] Optionally, in this embodiment, the target palm print picture matching model can be obtained by training a palm print picture matching model to be trained using the target palm print picture as a palm print picture set, and the target palm print picture matching model is required to be able to accurately extract features from user hand pictures (palm print pictures) of various modalities. Therefore, in the training stage of the feature extraction model, it is necessary to use a large number of target palm print pictures of different modalities with the same palm print lines as the palm print picture set for training.

[0039] According to an embodiment provided herein, an acquired simulated palmprint picture and a first noise vector are input to a trained target palmprint picture generator. Here, the simulated palmprint picture includes a simulated palmprint curve combined with a target type curve. The target palmprint picture generator generates a target palmprint picture by sequentially performing multiple downsampling operations and multiple upsampling operations on the simulated palmprint picture. Here, the downsampling operation is to perform a downsampling process on the input picture representation vector to obtain a downsampled picture representation vector, and to perform a noise addition process on the downsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector. The input picture representation vector in the first downsampling operation is the initial picture representation vector of the simulated palmprint picture. Here, the upsampling operation is to perform an upsampling process on the input picture representation vector to obtain an upsampled picture representation vector, and to perform a noise addition process on the upsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector. After the above processing, the target palm print picture obtained by processing the simulated palm print picture with the target palm print picture generator retains the characteristics of the simulated palm print curve of the simulated palm print picture. Furthermore, since the above downsampling operation and upsampling operation both use the first noise vector to perform noise addition processing on the sampled picture representation vector, the generated target palm print picture can have different modalities, and it is possible to generate a large number of target palm print pictures of different modalities having the same palm print lines based on the simulated palm print picture. This achieves the technical effect of improving the efficiency of palm print picture generation and further solves the technical problem of low efficiency of palm print picture generation.

[0040] In one optional configuration, the step of inputting the simulated palm print picture and a first noise vector into a trained target palm print picture generator, and generating a target palm print picture by the target palm print picture generator sequentially performing multiple downsampling operations and multiple upsampling operations on the simulated palm print picture, includes: a step S21 of determining an initial picture representation vector of the simulated palm-print picture based on the simulated palm-print picture; a step S22 of obtaining a P-th noise-added picture representation vector by sequentially using the P down-sampling modules to perform the down-sampling operation on the initial picture representation vector, wherein each of the P down-sampling modules includes a down-sampling sub-module and a first condition generating sub-module; the downsampling sub-module is for performing the downsampling process on an input picture representation vector to obtain a downsampled picture representation vector; Step S22, in which the first condition generation sub-module uses the first noise vector to perform a noise addition process on the down-sampled picture representation vector to obtain a noise-added picture representation vector; a step S23 of obtaining a Q-th noise-added picture representation vector by sequentially using the Q upsampling modules to perform the upsampling operation on the P-th noise-added picture representation vector, wherein each of the Q upsampling modules includes an upsampling sub-module and a second condition generating sub-module; the upsampling sub-module is for performing the upsampling process on an input picture representation vector to obtain an upsampled picture representation vector; Step S23, in which the second condition generating sub-module uses the first noise vector to perform the noise adding process on the up-sampled picture representation vector to obtain a noise-added picture representation vector; The method further includes a step S24 of generating the target palm print picture based on the Qth noise-added picture representation vector.

[0041] Optionally, in this embodiment, as shown in FIG. 4 , an initial picture representation vector 401, a Pth noise-added picture representation vector 402, and a Qth noise-added picture representation vector 403 are shown, taking the case where P takes the value of 3 and Q takes the value of 2 as an example. The target palmprint picture generator sequentially passes the simulated palmprint picture through P(3) downsampling modules (downsampling module 1, downsampling module 2, and downsampling module 3) and Q(2) upsampling modules (upsampling module 1 and upsampling module 2). Downsampling module 1 is the first sampling module to receive a picture representation vector, and the input picture representation vector is the initial picture representation vector 401 corresponding to the simulated palmprint picture. The picture representation vector input to each subsequent sampling module (upsampling module or upsampling module) is the noise-added picture representation vector output from the previous sampling module. For example, downsampling module 3, as the last downsampling module, outputs the P(3)th post-noise picture representation vector 402, which is the picture representation vector input to upsampling module 1. Upsampling module 2, as the last upsampling module, outputs the Q(2)th post-noise picture representation vector 403. Then, the target palmprint picture is generated based on the Q(2)th post-noise picture representation vector 403.

[0042] In one optional configuration, the step of obtaining a P-th noise-added picture representation vector by sequentially using the P down-sampling modules to perform the down-sampling operation on the initial picture representation vector includes: The i-th (i is a positive integer greater than or equal to 1 and less than or equal to P) noise-added picture representation vector is performing the downsampling process on the picture representation vector input to the i-th downsampling module by the downsampling sub-module of the i-th downsampling module to obtain an i-th downsampled picture representation vector; and a step S31 of obtaining an i-th noise-added picture representation vector by performing the noise addition processing on the i-th downsampled picture representation vector using the first noise vector by the first condition generation sub-module in the i-th downsampling module.

[0043] Optionally, in this embodiment, as shown in Fig. 5, the step of generating the i(2)th noise-added picture representation vector 502 is described by taking the case where i takes the value of 2 as an example. When i takes the value of 2, the downsampling sub-module 2 in the i(2)th downsampling module (downsampling module 2) performs a downsampling process on the input picture representation vector 501 of the i(2)th downsampling module to obtain the i(2)th downsampled picture representation vector 502, and the first condition generation sub-module 2 uses the first noise vector to perform the noise addition process on the i(2)th downsampled picture representation vector 502 to obtain the i(2)th noise-added picture representation vector 503.

[0044] In one optional configuration, the first condition generation sub-module includes a first set of fully connected layers and a second set of fully connected layers, and the step of obtaining the i-th noise-added picture representation vector by the first condition generation sub-module of the i-th downsampling module using the first noise vector to perform the noise addition process on the i-th downsampling picture representation vector includes: Step S41: outputting a first control vector by passing the first noise vector through the first set of fully connected layers, and outputting a second control vector by passing the first noise vector through the second set of fully connected layers; and step S42 of obtaining the i-th noise-added picture representation vector by performing the noise addition process on the i-th downsampled picture representation vector based on the first control vector and the second control vector.

[0045] Optionally, in this embodiment, as shown in FIG. 6, the case where i takes the value of 2 is taken as an example to describe the process in which the condition generation sub-module performs noise addition processing on the picture representation vector. The step of performing the noise addition process on the i(2)th downsampled picture representation vector 502 by the first condition generation sub-module 2 in the i(2)th downsampling module (downsampling module 2) using the first noise vector 202 to obtain the i(2)th noise-added picture representation vector 503 includes the steps of: passing the first noise vector 202 through a first set of fully connected layers 601 in the first condition generation sub-module 2 to obtain a first control vector 603; passing the first noise vector 202 through a second set of fully connected layers 602 in the first condition generation sub-module 2 to obtain a second control vector 604; and performing the noise addition process on the i(2)th downsampled picture representation vector 502 based on the first control vector 603 and the second control vector 604 to obtain the i(2)th noise-added picture representation vector 503.

[0046] Optionally, in this embodiment, as shown in FIG. 7, CAdaIN, i.e., the condition generation sub-module, includes four fully connected layers (FC), which are FC1, FC2, FC3, and FC4, respectively. N(z) is the first noise vector. N(z) passes through FC1 and FC2 sequentially, and after being output from FC2, it branches into two paths, which respectively enter FC3 and FC4. Here, FC1, FC2, and FC3 included in one branch constitute the first set of fully connected layers, and FC1, FC2, and FC4 included in another branch constitute the second set of fully connected layers. The i-th downsampled picture representation vector 701 is input to CAdaIN, and then noise is added sequentially using the first control vector output from the first set of fully connected layers (FC1, FC2, and FC3) and the second control vector output from the second set of fully connected layers (FC1, FC2, and FC4). In other words, a predetermined noise vector 703 is superimposed to obtain the i-th noise-added picture representation vector 702.

[0047] Optionally, the above N(z) is the above first noise vector. After N(z) is input into CAdaIN, the first noise vector N(z) may be sampled to obtain an 8-dimensional Gaussian noise sampled signal 704, and then the 8-dimensional Gaussian noise sampled signal 704 may be encoded into a 128-dimensional hidden control vector by four successive fully connected layers (FC1, FC2, FC3, and FC4), and the mean and variance of the input feature map may be adjusted by the hidden control vector.

[0048] Optionally, a sampling scheme for obtaining an eight-dimensional Gaussian noise sampled signal 704 by sampling the above-mentioned first noise vector N(z) is described herein. The dimension of the first noise vector N(z) is eight or more, for example, ten. Then, the order of the values ​​of each dimension of the original ten dimensions may be kept as it is, and eight-dimensional values ​​may be arbitrarily extracted as the eight-dimensional Gaussian noise sampled signal.

[0049] In one optional configuration, the step of obtaining the i-th noise-added picture representation vector by performing the noise addition process on the i-th down-sampled picture representation vector based on the first control vector and the second control vector includes: Step S51 of multiplying the first control vector by the i-th downsampled picture representation vector to obtain a first picture representation vector, and adding the first picture representation vector and the second control vector to obtain a second picture representation vector, wherein the first control vector is a control vector determined based on a sampling vector, the second control vector is a control vector determined based on the sampling vector, and the sampling vector is a vector sampled from the first noise vector; and step S52 of determining the second picture representation vector as the i-th noise-added picture representation vector or obtaining the i-th noise-added picture representation vector by adding the second picture representation vector and a predetermined noise vector, wherein the predetermined noise vector, the sampling vector, and the second picture representation vector have the same dimension.

[0050] Optionally, in this embodiment, the above process may be realized by, but is not limited to, the following calculation method:

[0051] By the following formula, Get TIFF2026505000000002.tif5170. TIFF2026505000000003.tif5170

[0052] where: TIFF2026505000000004.tif5170 is a sampling vector sampled from the first noise vector, TIFF2026505000000005.tif5170 is the first control vector, TIFF2026505000000006.tif5170 is the second control vector, TIFF2026505000000007.tif5170 is the i-th downsampled picture representation vector.

[0053] TIFF2026505000000008.tif5170 as the i-th noise-added picture representation vector, or The i-th noise-added picture representation vector is obtained by superimposing TIFF2026505000000009.tif5170 and a predetermined noise vector. TIFF2026505000000010.tif5170 has the same dimensions.

[0054] Optionally, in this embodiment: TIFF2026505000000011.tif5170 is a sampling vector sampled from the first noise vector, and this sampling vector is the above-mentioned eight-dimensional Gaussian noise sampling signal 704.

[0055] Optionally, in this embodiment, the manner of determining the i-th noise-added picture representation vector includes the following two:

[0056] 1) Without adding noise, TIFF2026505000000012.tif5170 is directly determined as the i-th noise-added picture representation vector.

[0057] 2) The i-th noise-added picture representation vector is obtained by overlapping TIFF2026505000000013.tif5170 with the predetermined noise vector 703 in Figure 7. Unlike 1), adding one predetermined noise vector of the same scale to the output position can further increase the diversity of the generated samples (which can be understood as noise-added picture representation vectors).

[0058] In one optional configuration, the step of performing the upsampling operation on the P noise-added picture representation vectors using the Q upsampling modules in sequence to obtain a Qth noise-added picture representation vector includes: The jth noise-added picture representation vector (j is a positive integer greater than or equal to 1 and less than or equal to Q) is performing the upsampling process on the picture representation vector input to the j-th upsampling module by the upsampling sub-module of the j-th upsampling module to obtain a j-th upsampled picture representation vector; and a step S61 of obtaining a j-th noise-added picture representation vector by performing the noise addition processing on the j-th upsampled picture representation vector using the first noise vector by the second condition generation sub-module of the j-th upsampling module.

[0059] Optionally, in this embodiment, as shown in Fig. 8, the step of generating the j(1)th post-noise picture representation vector will be described by taking the case where j takes the value of 1 as an example. When j takes the value of 1, an upsampling sub-module 1 in a j(1)th upsampling module (upsampling module 1) performs the upsampling process on the picture representation vector 801 input to the j(1)th upsampling module to obtain a j(1)th post-noise picture representation vector 802, and a second condition generation sub-module 1 in the j(1)th upsampling module uses the first noise vector 202 to perform the noise addition process on the j(1)th post-noise picture representation vector 802 to obtain a j(1)th post-noise picture representation vector 803.

[0060] In one optional configuration, the second condition generation sub-module includes a third set of fully connected layers and a fourth set of fully connected layers, and the step of obtaining a j-th upsampled picture representation vector by performing the upsampling process on the picture representation vector input to the j-th upsampling module by the upsampling sub-module of the j-th upsampling module includes: Step S71: outputting a third control vector by passing the first noise vector through the third set of fully connected layers, and outputting a fourth control vector by passing the first noise vector through the fourth set of fully connected layers; and step S72 of obtaining the j-th post-noise-added picture representation vector by performing the noise addition process on the j-th post-upsampled picture representation vector based on the third control vector and the fourth control vector.

[0061] Optionally, in this embodiment, as shown in FIG. 9, the process of the second condition generation sub-module performing noise addition processing on the picture representation vector is described by taking the case where j takes the value of 1 as an example. The step of performing the upsampling process on the picture representation vector 801 input to the j(1)th upsampling module by the upsampling sub-module 1 in the j(1)th upsampling module (upsampling module 1) to obtain the j(1)th upsampled picture representation vector 802 includes the steps of passing the first noise vector 202 through a third set of fully connected layers 901 in the second condition generation sub-module 1 to obtain a third control vector 903, and passing the first noise vector 202 through a fourth set of fully connected layers 902 in the second condition generation sub-module 1 to obtain a fourth control vector 904; and the step of performing the noise addition process on the j(1)th upsampled picture representation vector 802 based on the third control vector 903 and the fourth control vector 904 to obtain the j(1)th noise-added picture representation vector 803.

[0062] Optionally, in the embodiment of the present application, the configuration of the first condition generating sub-module and the configuration of the second condition generating sub-module may be the same (see FIG. 7), and the description will be omitted here.

[0063] In one optional configuration, the step of obtaining the j-th noise-added picture representation vector by performing the noise-adding process on the j-th upsampled picture representation vector based on the third control vector and the fourth control vector includes: Step S81 of multiplying the third control vector by the j-th upsampled picture representation vector to obtain a third picture representation vector, and adding the third picture representation vector and the fourth control vector to obtain a fourth picture representation vector, wherein the third control vector is a control vector determined based on a sampling vector, the fourth control vector is a control vector determined based on the sampling vector, and the sampling vector is a vector sampled from the first noise vector; The method further includes step S82 of determining the fourth picture representation vector as the j-th noise-added picture representation vector, or obtaining the j-th noise-added picture representation vector by adding the fourth picture representation vector and a predetermined noise vector, wherein the predetermined noise vector, the sampling vector, and the fourth picture representation vector have the same dimension.

[0064] Optionally, in this embodiment, the above process may be realized by, but is not limited to, the following calculation method:

[0065] By the following formula, Get TIFF2026505000000014.tif5170. TIFF2026505000000015.tif6170

[0066] where: TIFF2026505000000016.tif5170 is a sampling vector sampled from the first noise vector, TIFF2026505000000017.tif5170 is the third control vector, TIFF2026505000000018.tif5170 is the fourth control vector, TIFF2026505000000019.tif5170 is the picture representation vector after the j-th downsampling.

[0067] TIFF2026505000000020.tif5170 is determined as the j-th noise-added picture representation vector, or The j-th noise-added picture representation vector is obtained by superimposing TIFF2026505000000021.tif5170 and a predetermined noise vector. TIFF2026505000000022.tif5170 has the same dimensions.

[0068] Optionally, in this embodiment, the process of obtaining the j-th noise-added picture representation vector by performing the noise addition process on the j-th upsampled picture representation vector based on the third control vector and the fourth control vector is similar to the process of obtaining the i-th noise-added picture representation vector by performing the noise addition process on the i-th downsampled picture representation vector based on the first control vector and the second control vector described above, and therefore, description thereof will be omitted here.

[0069] In one optional configuration, the step of determining an initial picture representation vector of the simulated palm print picture based on the simulated palm print picture includes: The method further includes a step S91 of performing a first convolution process on the simulated palm print picture to obtain the initial picture representation vector.

[0070] generating the target palm print picture based on the Qth noise-added picture representation vector, The method further includes a step S92 of performing a second convolution process on the Qth noise-added picture representation vector to obtain the target palmprint picture.

[0071] Optionally, in this embodiment, as shown in Fig. 10, the down-sampling module 1 is the first sampling module in the target palmprint picture generator to receive a picture representation vector, and receives an initial picture representation vector 1001, while the remaining sampling modules all receive the noise-added picture representation vector 1002 output from the previous sampling module. The initial picture representation vector 1001 is generated based on the simulated palmprint picture. Specifically, the initial picture representation vector may be obtained by performing a first convolution process 1003 on the simulated palmprint picture.

[0072] Also, taking the case where Q takes a value of 2 as an example, upsampling module 2 is the last sampling module in the target palmprint picture generator that outputs a picture representation vector, and when the output Q(2)th noise-added picture representation vector 1004 passes through a second convolution process 1005, a target palmprint picture 1006 is generated.

[0073] In one optional configuration: S101: training a palmprint picture generator to be trained using a set of simulated palmprint sample pictures and a set of first real palmprint pictures for multiple rounds until a target loss value corresponding to the palmprint picture generator to be trained satisfies a predetermined convergence condition; determining the palm print picture generator at the end of training as the target palm print picture generator; In the training process, the training palm print picture generator sequentially performs multiple downsampling operations and multiple upsampling operations on the simulated palm print sample picture to generate a palm print picture; The downsampling operation in training is to perform a downsampling process on an input picture representation vector to obtain a downsampled picture representation vector, and to perform a noise addition process on the downsampled picture representation vector using an intermediate noise vector to obtain a noise-added picture representation vector; The input picture representation vector in the first down-sampling operation of training is the initial picture representation vector of the simulated palm print sample picture; The intermediate noise vector is a picture representation vector obtained by encoding the first true palmprint picture or a predetermined noise vector; The upsampling operation in training is to perform an upsampling process on an input picture representation vector to obtain an upsampled picture representation vector, and to perform a noise addition process on the upsampled picture representation vector using the intermediate noise vector to obtain a noise-added picture representation vector; the target loss value is a loss value determined based on at least a first loss value; the first loss value is a loss value between the generated palm print picture and the first true palm print picture; When determining the first loss value, the intermediate noise vector is a picture representation vector that encodes the first true palmprint picture.

[0074] Optionally, in this embodiment, as shown in Fig. 11 , a set of simulated palm print sample pictures (e.g., simulated palm print sample picture 1, simulated palm print sample picture 2, and simulated palm print sample picture 3) and a set of first true palm print pictures (first true palm print picture 1, first true palm print picture 2, and first true palm print picture 3) are used to train a palm print picture generator to be trained for multiple rounds until a target loss value corresponding to the palm print picture generator to be trained satisfies a predetermined convergence condition, and then the training is terminated, and the palm print picture generator to be trained at the end of training is determined as the target palm print picture generator. Here, the palm print picture generator to be trained includes P downsampling modules and Q upsampling modules to be trained.

[0075] In one optional configuration, the step of performing multiple rounds of training on the to-be-trained palm print picture generator using a set of simulated palm print sample pictures and a set of first real palm print pictures includes: The palm print picture generators waiting to be trained are trained in a k-th round (k is a positive integer of 2 or more), Obtaining a picture representation vector of the kth round by encoding the first true palmprint picture used in the kth round; inputting the picture representation vector of the kth round and the first simulated palm print sample picture used in the kth round into a palm print picture generator after training of a k-1th round to obtain a first palm print picture generated in the kth round; determining a first loss value for the kth round based on the first true palm print picture used in the kth round and the first palm print picture generated in the kth round; determining a target loss value for the kth round based on the first loss value for the kth round; If the target loss value of the kth round does not satisfy the convergence condition, adjusting parameters of the palmprint picture generator after the (k-1)th round of training to obtain a palmprint picture generator after the kth round of training; If the target loss value of the kth round satisfies the convergence condition, the training is terminated.

[0076] Optionally, in this embodiment, as shown in Fig. 12, a palm print picture generator (G) for training generates a palm print picture (B') based on an input simulated palm print sample picture (A) and an input first true palm print picture (B). The first loss value (L1) is a loss value between the palm print picture (B') and the first true palm print picture (B). The above-mentioned process of generating a palm print picture (B') by the palm print picture generator (G) for training based on the input simulated palm print sample picture (A) and the input first true palm print picture (B) may include: passing the first true palm print picture (B) through an encoder to obtain a picture representation vector Q(z|B) of the kth round after encoding; and inputting the picture representation vector and the simulated palm print sample picture (A) into the palm print picture generator (G) for training to obtain the palm print picture (B').

[0077] a step of obtaining a picture representation vector (Q(z|B)) of the kth round (k is a positive integer of 2 or more) of training for a palm print picture generator waiting for training by encoding (may be encoded by an encoder E) a first true palm print picture (B) used in the kth round; a step of obtaining a first palm print picture (B') generated in the kth round by inputting the picture representation vector (Q(z|B)) of the kth round and a first simulated palm print sample picture (A) used in the kth round into a palm print picture generator (G) after training for the k-1th round; The method includes the steps of: determining a first loss value (L1) for the kth round based on the actual palmprint picture (B) and the first palmprint picture (B') generated in the kth round; determining a target loss value for the kth round based on the first loss value (L1) for the kth round; if the target loss value for the kth round does not satisfy the predetermined convergence condition, adjusting parameters of the palmprint picture generator after the (k-1)th round of training to obtain a palmprint picture generator after the kth round of training; and terminating training if the target loss value for the kth round satisfies the predetermined convergence condition.

[0078] In one optional configuration, the step of determining a first loss value for the kth round based on the first true palm print picture used in the kth round and the first palm print picture generated in the kth round includes: Step S121: obtaining pixel differences between pixel values ​​of R pixel points at the same positions in the first true palmprint picture used in the kth round and the first palmprint picture generated in the kth round, thereby obtaining R pixel differences (R is a positive integer equal to or greater than 2); and step S122 of determining the sum of the R pixel differences as a first loss value of the kth round.

[0079] Optionally, in this embodiment, the above process may be realized by, but is not limited to, the following calculation method:

[0080] By the following formula, Get TIFF2026505000000023.tif5170. TIFF2026505000000024.tif11170

[0081] where: TIFF2026505000000025.tif5170 is the first pixel value of the r-th pixel point in the first true palmprint picture used in the kth round, TIFF2026505000000026.tif6170 is a second pixel value of the r-th pixel point of the first palm print picture generated in the k round, and both the first true palm print picture used in the k round and the first palm print picture generated in the k round include R pixel points; TIFF2026505000000027.tif6170 is the similarity between the first pixel value and the second pixel value.

[0082] TIFF2026505000000028.tif5170 is determined as the first loss value of the kth round.

[0083] Optionally, in this embodiment, a method for determining a first loss value of the kth round based on the first true palmprint picture used in the kth round and the first palmprint picture generated in the kth round is described herein. The first true palmprint picture and the first palmprint picture are both in picture format. In the formula for calculating TIFF2026505000000029.tif5170, R represents the number of pixels in the first true palm print picture and the second palm print picture, respectively; TIFF2026505000000030.tif5170 is the first pixel value of the r-th pixel point in the first true palmprint picture used in the kth round, TIFF2026505000000031.tif6170 is the second pixel value of the r-th pixel point in the first palm print picture generated in the k-th round. The first loss value is calculated by comparing the first and second pixel values ​​of any two corresponding pixel points among the R pixel points. Get TIFF2026505000000032.tif5170.

[0084] In one optional configuration, the step of performing the kth round of training on the palm print picture generators waiting to be trained includes: The method further includes step S131 of obtaining a first classification result of the kth round by using a trained target classifier to classify the first true palmprint picture used in the kth round and the first palmprint picture generated in the kth round, and determining a second loss value of the kth round based on the first classification result of the kth round.

[0085] In one optional configuration, the step of determining a target loss value for a kth round based on the first loss value for the kth round comprises: The method includes a step S132 of determining a target loss value for the kth round based on the first loss value for the kth round and the second loss value for the kth round.

[0086] Optionally, in this embodiment, as shown in Fig. 13, a palm print picture generator (G) for training generates a palm print picture (B') based on an input simulated palm print sample picture (A) and an input first true palm print picture (B). A first loss value (L1) is a loss value between the palm print picture (B') and the first true palm print picture (B). A trained target classifier (D) classifies the input first true palm print picture (B) and the generated palm print picture (B') to obtain a target identification result. The second loss value (LD1) is a loss value determined based on the target identification result. The process in which the palm print picture generator (G) waiting to train generates the palm print picture (B') based on the input simulated palm print sample picture (A) and the input first true palm print picture (B) may include passing the first true palm print picture (B) through an encoder to obtain a picture representation vector Q(z|B) of the kth round after encoding, and inputting the picture representation vector and the simulated palm print sample picture (A) into the palm print picture generator (G) waiting to train to obtain the palm print picture (B').

[0087] a step of obtaining a picture representation vector (Q(z|B)) of the kth round (k is a positive integer of 2 or more) of training for a palm print picture generator waiting for training by encoding (may be encoded by an encoder E) a first true palm print picture (B) used in the kth round; a step of obtaining a first palm print picture (B') generated in the kth round by inputting the picture representation vector (Q(z|B)) of the kth round and a first simulated palm print sample picture (A) used in the kth round into a palm print picture generator (G) after training for the k-1th round; a step of determining a first loss value (L1) of the kth round based on the first true palm print picture (B) used in the kth round and the first palm print picture (B') generated in the kth round; The method includes the steps of: obtaining a first classification result of the kth round by distinguishing between a first true palmprint picture (B) used in the kth round and a first palmprint picture (B') generated in the kth round; determining a second loss value (LD1) for the kth round based on the first classification result of the kth round; determining a target loss value for the kth round based on the first loss value (L1) for the kth round and the second loss value (LD1) for the kth round; if the target loss value for the kth round does not satisfy the predetermined convergence condition, adjusting parameters of the palmprint picture generator after training of the (k-1)th round to obtain a palmprint picture generator after training of the kth round; and terminating training if the target loss value for the kth round satisfies the predetermined convergence condition.

[0088] In one optional configuration, the step of determining a second loss value for a kth round based on the first identification result for the kth round comprises: a step S141 of obtaining a first probability by obtaining a probability that the first true palmprint picture used in the kth round, which is indicated by the first classification result, is true, and obtaining a second probability by obtaining a probability that the first palmprint picture generated in the kth round, which is indicated by the first classification result, is true; Step S142 of obtaining a first probability product by multiplying the first probability and the second probability by the probability of the same position; a step S143 of obtaining a third probability by obtaining a probability that the first palmprint picture generated in the kth round is false, which is indicated by the first classification result, and obtaining a fourth probability by obtaining a probability that the first true palmprint picture used in the kth round is false, which is indicated by the first classification result; Step S144 of multiplying the third probability by the probability of the same position in the fourth probability to obtain a second probability product; and a step S145 of determining a second loss value for the kth round based on the first probability product and the second probability product.

[0089] Optionally, in this embodiment, the above process may be realized by one of the following calculation methods, including but not limited to:

[0090] Calculation method 1: By the following formula, Get TIFF2026505000000033.tif5170. TIFF2026505000000034.tif11170

[0091] where: TIFF2026505000000035.tif5170 is the probability that the first true palm print picture used in the kth round is true, as indicated by the first classification result; TIFF2026505000000036.tif5170 is the probability that the first true palmprint picture used in the kth round is false, as indicated by the first classification result; TIFF2026505000000037.tif4170 is the probability that the first palm print picture generated in the kth round is true, as indicated by the first classification result; TIFF2026505000000038.tif5170 is the probability that the first classification result indicates that the first palm print picture generated in the kth round is false.

[0092] TIFF2026505000000039.tif5170 is determined as the second loss value of the kth round.

[0093] Calculation method 2: By the following formula, Get TIFF2026505000000040.tif5170. TIFF2026505000000041.tif17170

[0094] where: TIFF2026505000000042.tif5170 is used in the k-th round, as indicated by the first classification result. TIFF2026505000000043.tif4 is the probability that the t-th first true palm print picture is true among the 170 first true palm print pictures, TIFF2026505000000044.tif5170 is used in the k-th round, as indicated by the first classification result. TIFF2026505000000045.tif4 is the probability that the t-th first true palm print picture is false among the 170 first true palm print pictures, TIFF2026505000000046.tif4170 is generated in the kth round as indicated by the first classification result. TIFF2026505000000047.tif is the probability that the t-th first palm print picture out of 4170 first palm print pictures is true, TIFF2026505000000048.tif5170 is generated in the kth round as indicated by the first classification result. TIFF2026505000000049.tifThis is the probability that the tth first palm print picture out of 4170 first palm print pictures is false.

[0095] TIFF2026505000000050.tif5170 is determined as the second loss value of the kth round.

[0096] Optionally, in this embodiment, a target classifier is first described. The target classifier can classify an input palm print picture and determine whether the picture is a true picture or a false picture. The above-mentioned true picture can be understood as a picture that the target classifier recognizes as a true picture collected in the real world, for example, a palm print picture that has been collected truly. Conversely, a false picture can be understood as a picture that the target classifier recognizes as not a true picture collected in the real world, and is a picture synthesized by technical means.

[0097] In the embodiment of the present application, the key point is the training of the palm print picture generator. A palm print picture generator to be trained is trained, and the palm print picture generated by the palm print picture generator to be trained is passed through a target classifier. If the target classifier recognizes the palm print picture as a true picture, it indicates that the palm print picture generated by this palm print picture generator to be trained will be "realistic", and the palm print picture generator to be trained can be determined as the target palm print picture generator.

[0098] Optionally, in this embodiment, the above-mentioned true probability can be understood as the probability that the palm print picture is identified as a true picture by the target classifier, and the above-mentioned false probability can be understood as the probability that the palm print picture is identified as a false picture by the target classifier.

[0099] Optionally, in this embodiment, the first identification result of the kth round is: TIFF2026505000000051.tif 4170 first real palm print pictures, The first palm print images may include, but are not limited to, 4170 first palm print images identified by the target classifier. The above classification process can be performed in parallel using multiple processes simultaneously.

[0100] In one optional configuration, the step of performing the kth round of training on the palm print picture generators waiting to be trained includes: The method further includes a step S151 of inputting the noise vector of the kth round and the first simulated palm print sample picture used in the kth round into a palm print picture generator after training of the (k-1)th round to obtain a second palm print picture generated in the kth round, and determining a third loss value of the kth round based on the first palm print picture generated in the kth round and the second palm print picture generated in the kth round.

[0101] The step of determining a target loss value for the kth round based on the first loss value for the kth round and the second loss value for the kth round includes: The method further includes a step S152 of determining a target loss value for the kth round based on the first loss value for the kth round, the second loss value for the kth round, and the third loss value for the kth round.

[0102] Optionally, in this embodiment, the method of generating the first loss value and the second loss value may refer to the description in FIG. 13, but the description will be omitted here. Here, only the method of determining the third loss value will be described. As shown in FIG. 14, the noise vector (N(z)) of the kth round and the first simulated palm print sample picture (A) used in the kth round are input into the palm print picture generator (G) after training of the (k-1)th round to obtain the second palm print picture (B") generated in the kth round, and the third loss value (LID) of the kth round is determined based on the first palm print picture (B') generated in the kth round and the second palm print picture (B") generated in the kth round.

[0103] Here, a method for determining the third loss value (LID) of the kth round based on the first palmprint picture (B') generated in the kth round and the second palmprint picture (B") generated in the kth round may be, but is not limited to, inputting the first palmprint picture (B') and the second palmprint picture (B") into a palmprint recognition network, extracting feature vectors of the first palmprint picture (B') and the second palmprint picture (B"), and then determining the third loss value (LID) of the kth round based on the feature vectors of both the first and second palmprint pictures.

[0104] In one optional configuration, the step of determining a third loss value for the kth round based on the first palm print picture generated in the kth round and the second palm print picture generated in the kth round includes: A step S161 of calculating a vector dot product between a first feature vector of the first palmprint picture generated in the kth round and a second feature vector of the second palmprint picture generated in the kth round to obtain a vector dot product value; a step S162 of multiplying the norm of the first feature vector by the norm of the second feature vector to obtain a product value; The method further includes step S163 of obtaining a target ratio by dividing the vector dot product value by the multiplication product value, and obtaining a third loss value of the kth round by subtracting the target ratio from 1.

[0105] Optionally, in this embodiment, the above process may be realized by, but is not limited to, the following calculation method:

[0106] By the following formula, Get TIFF2026505000000053.tif5170. TIFF2026505000000054.tif17170

[0107] where: TIFF2026505000000055.tif7170 is the first feature vector of the first palmprint picture generated in the kth round, TIFF2026505000000056.tif7170 is the second feature vector of the second palmprint picture generated in the kth round, TIFF2026505000000057.tif7170 is the dot product of the first feature vector and the second feature vector, TIFF2026505000000058.tif7170 is the norm of the first feature vector, TIFF2026505000000059.tif7170 is the norm of the second feature vector.

[0108] TIFF2026505000000060.tif5170 is determined as the third loss value of the kth round.

[0109] Optionally, in this embodiment, the first palmprint picture (B') generated in the kth round and the second palmprint picture (B") generated in the kth round are input into a palmprint recognition network, and a first feature vector of the first palmprint picture (B') and the second palmprint picture (B") is obtained. TIFF2026505000000061.tif7170 and the second feature vector TIFF2026505000000062.tif7170 is extracted, and then the third loss value (LID) of the kth round is calculated according to the above formula. The calculation process is illustrated below:

[0110] TIFF2026505000000063.tif30170

[0111] In one optional configuration, the step of performing the kth round of training on the palm print picture generators waiting to be trained includes: The method further includes a step S171 of determining a fourth loss value of the kth round based on the noise vector of the kth round and the picture representation vector of the kth round.

[0112] determining a target loss value for the kth round based on the first loss value for the kth round, the second loss value for the kth round, and the third loss value for the kth round, The method further includes step S172 of determining a target loss value for the kth round based on the first loss value for the kth round, the second loss value for the kth round, the third loss value for the kth round, and the fourth loss value for the kth round.

[0113] Optionally, in this embodiment, the methods for generating the first loss value, the second loss value, and the third loss value may refer to the descriptions in FIGS. 13 and 14 above, but will not be described here. Here, only the method for determining the fourth loss value will be described. As shown in FIG. 15, the fourth loss value (LKL) of the kth round is determined based on the noise vector (N(z)) of the kth round and the picture representation vector (Q(z|B)) of the kth round. Here, the fourth loss value (LKL) can constrain the distribution of the picture representation vector (Q(z|B)) to be consistent with that of the noise vector (N(z)) of the kth round. Note that the noise vector (N(z)) of the kth round may be, but is not limited to, original Gaussian noise, i.e., white noise.

[0114] In one optional configuration, the step of determining the k-th round fourth loss value based on the k-th round noise vector and the k-th round picture representation vector comprises: Step S181 of acquiring a value of each dimension in the k-round picture representation vector and acquiring a value of each dimension in the k-round noise vector, wherein the total number of dimensions of the k-round picture representation vector and the k-round noise vector are the same; The method further includes a step S182 of determining a fourth loss value of the kth round based on the value of each dimension in the picture representation vector of the kth round and the value of each dimension in the noise vector of the kth round.

[0115] Optionally, in this embodiment, the above process may be realized by, but is not limited to, the following calculation method:

[0116] By the following formula, Get TIFF2026505000000064.tif5170. TIFF2026505000000065.tif16170

[0117] where: TIFF2026505000000066.tif5170 is the m-th dimension value in the k-th round picture representation vector, TIFF2026505000000067.tif5170 is the m-th dimension value in the k-th round noise vector, and the total number of dimensions of the k-th round picture representation vector and the k-th round noise vector is M.

[0118] TIFF2026505000000068.tif5170 is determined as the fourth loss value of the kth round.

[0119] Optionally, in this embodiment, The calculation process for TIFF2026505000000069.tif5170 is shown below as an example.

[0120] Take the case where M takes a value of 8 and m takes a value of 3 as an example. For example, the total number of dimensions of the k-th round picture representation vector and the k-th round noise vector is both 8, the k-th round picture representation vector is (c1, c2, c3, c4, c5, c6, c7, c8), and the k-th round noise vector is (d1, d2, d3, d4, d5, d6, d7, d8). In this case, TIFF2026505000000070.tif5170 has a value of c3, The value of TIFF2026505000000071.tif5170 is d3. Then, substitute them into the above formula and calculate. In this way, by adding 8-dimensional data, the fourth loss value Get TIFF2026505000000072.tif5170.

[0121] In one optional configuration, the step of performing the kth round of training on the palm print picture generators waiting to be trained includes: The method further includes step S191 of obtaining a second classification result of the kth round by using the target classifier to classify the second true palmprint picture used in the kth round and the second palmprint picture generated in the kth round, and determining a fifth loss value of the kth round based on the second classification result of the kth round.

[0122] determining a target loss value for the kth round based on a first loss value for the kth round, a second loss value for the kth round, a third loss value for the kth round, and a fourth loss value for the kth round, The method further includes a step S192 of determining a target loss value for the k round based on a first loss value for the k round, a second loss value for the k round, a third loss value for the k round, a fourth loss value for the k round, and a fifth loss value for the k round.

[0123] Optionally, in this embodiment, for the methods of generating the first loss value, the second loss value, the third loss value, and the fourth loss value, refer to the descriptions in FIGS. 13, 14, and 15 above, but the description will be omitted here. Here, only the method of determining the fifth loss value will be described. As shown in FIG. 16, the target classifier (D) classifies the second true palmprint picture (C) used in the kth round and the second palmprint picture (B") generated in the kth round to obtain a second classification result of the kth round, and determines a fifth loss value (LD2) for the kth round based on the second classification result of the kth round.

[0124] Optionally, in this embodiment, the second true palmprint picture (C) used in the kth round and the first true palmprint picture used in the kth round belong to true palmprint pictures of two different people, respectively, that is, their palmprint features are different.

[0125] In one optional configuration, the step of determining a fifth loss value of the kth round based on the second identification result of the kth round comprises: a step S201 of obtaining a fifth probability by obtaining a probability that the second true palmprint picture used in the kth round is true, which is indicated by the second classification result, and obtaining a sixth probability by obtaining a probability that the second palmprint picture generated in the kth round is true, which is indicated by the second classification result; Step S202 of multiplying the fifth probability by the probability of the same position in the sixth probability to obtain a third probability product; a step S203 of obtaining a seventh probability by obtaining a probability that the second true palmprint picture used in the kth round is false, which is indicated by the second classification result, and obtaining an eighth probability by obtaining a probability that the second palmprint picture generated in the kth round is false, which is indicated by the second classification result; Step S204 of multiplying the seventh probability by the probability of the same position in the eighth probability to obtain a fourth probability product; The method further includes a step S205 of determining a fifth loss value of the kth round based on the third probability product and the fourth probability product.

[0126] Optionally, in this embodiment, the above process may be realized by one of the following calculation methods, including but not limited to:

[0127] Calculation method 1: By the following formula, Get TIFF2026505000000073.tif5170. TIFF2026505000000074.tif10170

[0128] where: TIFF2026505000000075.tif5170 is the probability that the second true palm print picture used in the kth round is true, as indicated by the second classification result; TIFF2026505000000076.tif5170 is the probability that the second true palm print picture used in the kth round is false, as indicated by the second classification result; TIFF2026505000000077.tif4170 is the probability that the second palm print picture generated in the kth round is true, as indicated by the second classification result; TIFF2026505000000078.tif5170 is the probability that the second classification result indicates that the second palm print picture generated in the kth round is false.

[0129] TIFF2026505000000079.tif5170 is determined as the fifth loss value of the kth round.

[0130] Calculation method 2: By the following formula, Get TIFF2026505000000080.tif5170. TIFF2026505000000081.tif17170

[0131] where: TIFF2026505000000082.tif5170 is used in the k-th round, as indicated by the second classification result. TIFF2026505000000083.tif41702 of the 2nd true palm print pictures, TIFF2026505000000084.tif3170 is the probability that the second true palm print picture is true, TIFF2026505000000085.tif5170 is used in the k-th round, as indicated by the second classification result. TIFF2026505000000086.tif41702 of the 2nd true palm print pictures, TIFF2026505000000087.tif3170 is the probability that the second true palmprint picture is false, TIFF2026505000000088.tif4170 is generated in the kth round as indicated by the second classification result. TIFF2026505000000089.tif41702 of the second palm print pictures TIFF2026505000000090.tif3 is the probability that the 170th second palm print picture is true, TIFF2026505000000091.tif5170 is generated in the kth round as indicated by the second classification result. TIFF2026505000000092.tif41702 of the 2nd palm print pictures, TIFF2026505000000093.tif3 is the probability that the 170th second palm print picture is false.

[0132] TIFF2026505000000094.tif5170 is determined as the fifth loss value of the kth round.

[0133] Optionally, in this embodiment, a fifth loss value TIFF2026505000000095.tif5170 and the second loss value The calculation process for TIFF2026505000000096.tif5170 is similar, and the second loss value described above Refer to the calculation method of TIFF2026505000000097.tif5170 for the fifth loss value TIFF2026505000000098.tif5170 can be calculated, but the explanation will be omitted here.

[0134] In one optional configuration: The method further includes a step S212 of obtaining a target palm print picture matching model by training a palm print picture matching model waiting to be trained using the palm print picture set; the palm print picture set includes the target palm print picture and other palm print pictures, and the other palm print pictures are generated in the same manner as the target palm print picture; The target palm print picture matching model is for matching a set of input palm print pictures, where the set of palm print pictures includes palm print pictures awaiting confirmation and collected palm print pictures.

[0135] Optionally, in this embodiment, palm print pictures other than the target palm print picture in the palm print picture set are palm print pictures generated by the same generation method as the target palm print picture. The generation method may be, but is not limited to, adjusting parameters of the same conditional generation sub-modules in the P downsampling modules and the Q upsampling modules to output palm print pictures of different modalities. As the palm print picture set, the palm print line features of any two palm print pictures in one palm print picture set are kept consistent and are all taken from the simulated palm print curve in the simulated palm print picture. Since the parameters of the conditional generation sub-modules in each generation process are different, the output modalities can be different.

[0136] In one optional configuration: Step S222 of obtaining the simulated palmprint curve by combining Bessel curves, wherein the target type curve includes the Bessel curve; The method further includes a step S223 of generating the simulated palm-print picture including the simulated palm-print curve.

[0137] Optionally, in this embodiment, the simulated palm print picture includes a simulated palm print curve that is a combination of target-type curves, where the target-type curve may be, but is not limited to, a Bessel curve, and the combination of target-type curves may be, but is not limited to, a free combination of multiple Bessel curves or a combination of the Bessel curves arranged according to a rule similar to the palm print characteristics of a palm.

[0138] This application proposes a palmprint picture generation method, which can be understood as a multi-modality palmprint generation method. By adding a conditional generation submodule to each module of the generative adversarial network, the model's learning ability for diverse samples can be enhanced, and palmprint picture samples of different modalities can be simultaneously generated by controlling the conditional probability. This achieves multi-modality palmprint generation capabilities for visible light and infrared modalities. The generated palmprint picture samples are used to train a palmprint picture matching model, which enhances the adaptation effect of the palmprint picture matching model for images of different modalities. A twin identity protection module is also added at the same time to maintain the consistency of palmprint lines of the same identity.

[0139] As shown in Figure 17, it mainly consists of two modules.

[0140] One is a diversity generation module 1701 for increasing the intra-class diversity of the generated samples; The other is an identity consistency guarantee module (ID Consistency Constraint) 1702 for controlling the consistency of the intra-class identities of the samples.

[0141] For the imitation image B' (which can be understood as the first palm print picture above) generated by the palm print picture generator, L1 is calculated using the real palm print image B (which can be understood as the first real palm print picture above) to constrain the authenticity of the skin-muscle texture of the generated B'. To constrain Q(z|B), the KL divergence between it and N(z) is calculated as the loss function to constrain the consistency of the distribution between it and N(z). The obtained B' is passed through the target classifier D together with B. The target classifier consists of a three-layer convolutional neural network. The classifier extracts features of B' and B, respectively, and then optimizes the authenticity of the generated B' using an adversarial loss.

[0142] In order to increase the diversity of the generated samples, a condition generation sub-module is designed to improve the diversity of the generated samples. As shown in Figure 18, the overall structure of the target palmprint picture generator may consist of seven layers of downsampling modules and six layers of upsampling modules. Each downsampling module and each upsampling module includes one condition generation sub-module CAdaIN.

[0143] 1. Diversity generation module Training phase: In the diversity generation module, first, the encoder E maps the true palmprint image B to a Gaussian noise domain vector Q(z|B) (which can be understood as the above picture representation vector), and then the generator G (which can be understood as the above palmprint picture generator) remaps the encoded noise domain vector Q(z|B) to the palmprint image domain using the unpaired Bessel palmprint lines A (which can be understood as the above first simulated palmprint picture) as a condition. The Bessel palmprint lines A are quasi-palmprint curves constructed by combining different Bessel curves.

[0144] Inference stage: The customized Bessel palmprint lines A and Q(z|B) are input to a generator G to generate palmprint samples of different modalities (which can be understood as the target palmprint pictures above).

[0145] Here, as shown in Fig. 19, the structure of encoder E consists of four residual blocks (RB) and one fully connected layer (FC). In the figure, BN represents a batch normalization layer, LReLU represents a leakyReLU layer, conv represents a convolutional layer, avgpool represents an average pooling layer, FC represents a fully connected layer, and Flatten is a function for flattening a multi-dimensional array into a one-dimensional array.

[0146] 2. Identity Consistency Assurance Module An identity consistency assurance module is proposed to ensure the identity consistency of the generated imitation palm print within the same class. This module uses a twin structure. For the same input Bessel palm print line A, a new random control hidden vector N(z) is input to generator G to generate a new imitation palm print B'. B' and B' are input to a palm print feature extraction model pre-trained on real palm print images, respectively, to extract 512-dimensional feature vectors. The cosine similarity between the two is calculated based on the extracted feature vectors, and the LID is used to constrain the consistency of the samples within the class.

[0147] According to the palm print picture generation method proposed in this application, a single Bessel curve (which can be understood as the above-mentioned simulated palm print picture) is used to generate palm print effect diagrams of different modalities, and by adjusting the input random noise (N(z)), palms of various modalities (which can be understood as the above-mentioned target palm print pictures) including various visible light and infrared monochrome images can be generated. As shown in FIG. 20, according to the palm print picture generation method, target palm print pictures of different modalities can be generated based on a single simulated palm print picture. For example, 12 target palm print pictures can be generated using simulated palm print picture 1, simulated palm print picture 2, simulated palm print picture 3, and simulated palm print picture 4, respectively. Taking simulated palm print picture 1 as an example, after generating target palm print picture 1 using simulated palm print picture 1 and a first noise vector, another first noise vector can be input or the first noise vector can be adjusted to output target palm print picture 2, which has a different modality from target palm print picture 1 but whose palm print lines match (all match simulated palm print picture 1). By further inference, target palm print pictures of many other modalities can be generated. The effectiveness of the proposed CAdaIN module is verified. Furthermore, the palm print lines of the palms generated using a single Bessel curve all match, and no extraneous palms appear. The effectiveness of the proposed identity consistency assurance module is demonstrated.

[0148] Although the above-described method embodiments are expressed as a combination of a series of operations for ease of explanation, those skilled in the art should understand that the present application is not limited by the order of the operations described, since some steps may be performed in other orders or simultaneously. Next, those skilled in the art should also understand that the embodiments described in the specification belong to preferred embodiments, and that such operations and modules are not necessarily required for the present application.

[0149] According to another aspect of the embodiment of the present application, there is provided a palmprint picture generating device for implementing the above palmprint picture generating method. Figure 21 is a block diagram of the configuration of the palmprint picture generating device according to the embodiment of the present application. As shown in Figure 21, this device includes: an acquiring unit 2002 for acquiring a simulated palm print picture, the simulated palm print picture including a simulated palm print curve combined with a target type curve; an input unit 2004 for inputting the simulated palm print picture and a predetermined first noise vector into a trained target palm print picture generator, and generating a target palm print picture by the target palm print picture generator by sequentially performing multiple downsampling operations and multiple upsampling operations on the simulated palm print picture, the downsampling operation is for performing a downsampling process on an input picture representation vector to obtain a downsampled picture representation vector, and performing a noise addition process on the downsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector, wherein the input picture representation vector in the first downsampling operation is an initial picture representation vector of the simulated palmprint picture; The upsampling operation includes an input unit 2004 for performing an upsampling process on an input picture representation vector to obtain an upsampled picture representation vector, and performing a noise addition process on the upsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector.

[0150] In one optional configuration, the input unit comprises: a determination module for determining the initial picture representation vector based on the simulated palm print picture; a first sampling module for sequentially using the P downsampling modules to perform the downsampling operation on the initial picture representation vector to obtain a P-th noise-added picture representation vector, wherein each of the P downsampling modules includes a downsampling sub-module and a first condition generating sub-module; the downsampling sub-module is for performing the downsampling process on an input picture representation vector to obtain a downsampled picture representation vector; the first condition generation sub-module is for performing a noise addition process on the down-sampled picture representation vector using the first noise vector, to obtain a noise-added picture representation vector; and a second sampling module for sequentially using the Q upsampling modules to perform the upsampling operation on the P-th noise-added picture representation vector to obtain a Q-th noise-added picture representation vector, wherein each of the Q upsampling modules includes an upsampling sub-module and a second condition generating sub-module; the upsampling sub-module is for performing the upsampling process on an input picture representation vector to obtain an upsampled picture representation vector; the second condition generation sub-module is for performing the noise addition process on the up-sampled picture representation vector using the first noise vector, thereby obtaining a noise-added picture representation vector; and a determining module for generating the target palm print picture based on the Qth noise-added picture representation vector.

[0151] In one optional configuration, the first sampling module further comprises: The i-th (i is a positive integer greater than or equal to 1 and less than or equal to P) noise-added picture representation vector is performing the downsampling process on the picture representation vector input to the i-th downsampling module by the downsampling sub-module of the i-th downsampling module to obtain an i-th downsampled picture representation vector; and a step of using the first noise vector by the first condition generation sub-module in the i-th downsampling module to perform the noise addition process on the i-th downsampling picture representation vector to obtain the i-th noise-added picture representation vector.

[0152] In one optional configuration, the first condition generation sub-module includes a first set of fully connected layers and a second set of fully connected layers, and the first sampling module includes: Passing the first noise vector through the first set of fully connected layers to output a first control vector; Passing the first noise vector through the second set of fully connected layers to output a second control vector; The noise addition process is performed on the i-th downsampled picture representation vector based on the first control vector and the second control vector to obtain the i-th noise-added picture representation vector.

[0153] In one optional configuration, the second sampling module: The jth noise-added picture representation vector (j is a positive integer greater than or equal to 1 and less than or equal to Q) is performing the upsampling process on the picture representation vector input to the j-th upsampling module by the upsampling sub-module of the j-th upsampling module to obtain a j-th upsampled picture representation vector; and a step of using the first noise vector by the second condition generation sub-module in the jth upsampling module to perform the noise addition processing on the jth upsampling picture representation vector to obtain the jth noise-added picture representation vector.

[0154] In one optional configuration, the second condition generation sub-module includes a third set of fully connected layers and a fourth set of fully connected layers, and the second sampling module includes: Passing the first noise vector through the third fully connected layer to output a third control vector; Passing the first noise vector through the fourth fully connected layer to output a fourth control vector; The j-th noise-added picture representation vector is obtained by performing the noise addition process on the j-th upsampled picture representation vector based on the third control vector and the fourth control vector.

[0155] In one optional configuration, the determination module obtains the initial picture representation vector by performing a first convolution process on the simulated palm print picture; The determination module performs a second convolution process on the Qth noise-added picture representation vector to obtain the target palmprint picture.

[0156] In one optional configuration, the apparatus comprises: a first training unit for performing multiple rounds of training on a palmprint picture generator to be trained using a set of simulated palmprint sample pictures and a set of first real palmprint pictures until a target loss value corresponding to the palmprint picture generator to be trained satisfies a predetermined convergence condition, and determining the palmprint picture generator at the end of training as the target palmprint picture generator; In the training process, the training palm print picture generator sequentially performs multiple downsampling operations and multiple upsampling operations on the simulated palm print sample picture to generate a palm print picture; The downsampling operation in training is to perform a downsampling process on an input picture representation vector to obtain a downsampled picture representation vector, and to perform a noise addition process on the downsampled picture representation vector using an intermediate noise vector to obtain a noise-added picture representation vector; The input picture representation vector in the first down-sampling operation of training is the initial picture representation vector of the simulated palm print sample picture; The intermediate noise vector is a picture representation vector obtained by encoding the first true palmprint picture or a predetermined noise vector; The upsampling operation in training is to perform an upsampling process on an input picture representation vector to obtain an upsampled picture representation vector, and to perform a noise addition process on the upsampled picture representation vector using the intermediate noise vector to obtain a noise-added picture representation vector; the target loss value is a loss value determined based on at least a first loss value; the first loss value is a loss value between the generated palm print picture and the first true palm print picture; When determining the first loss value, the intermediate noise vector is a picture representation vector that encodes the first true palmprint picture.

[0157] In one optional configuration, the first training unit: The palm print picture generators waiting to be trained are trained in a k-th round (k is a positive integer of 2 or more), Obtaining a picture representation vector of the kth round by encoding the first true palmprint picture used in the kth round; inputting the picture representation vector of the kth round and the first simulated palm print sample picture used in the kth round into a palm print picture generator after training of a k-1th round to obtain a first palm print picture generated in the kth round; determining a first loss value for the kth round based on the first true palm print picture used in the kth round and the first palm print picture generated in the kth round; determining a target loss value for the kth round based on the first loss value for the kth round; If the target loss value of the kth round does not satisfy the convergence condition, adjusting parameters of the palmprint picture generator after the (k-1)th round of training to obtain a palmprint picture generator after the kth round of training; and terminating training if the target loss value of the kth round satisfies the convergence condition.

[0158] As an optional configuration, the training module further uses a trained target classifier to distinguish between a first true palmprint picture used in the kth round and a first palmprint picture generated in the kth round, to obtain a first classification result of the kth round; and determine a second loss value of the kth round based on the first classification result of the kth round; The training module further determines a target loss value for the kth round based on the first loss value for the kth round and the second loss value for the kth round.

[0159] As an optional configuration, the training module further inputs the noise vector of the kth round and the first simulated palm print sample picture used in the kth round into a palm print picture generator after training of the k-1th round to obtain a second palm print picture generated in the kth round; and determines a third loss value of the kth round based on the first palm print picture generated in the kth round and the second palm print picture generated in the kth round; The training module further determines a target loss value for the kth round based on the first loss value for the kth round, the second loss value for the kth round, and the third loss value for the kth round.

[0160] In one optional configuration, the training module further determines a fourth loss value for the k-th round based on the noise vectors for the k-th round and the picture representation vectors for the k-th round; The training module further determines a target loss value for the kth round based on the first loss value for the kth round, the second loss value for the kth round, the third loss value for the kth round, and the fourth loss value for the kth round.

[0161] As an optional configuration, the training module further uses the target classifier to distinguish between a second true palmprint picture used in the kth round and a second palmprint picture generated in the kth round to obtain a second classification result of the kth round; and determines a fifth loss value of the kth round based on the second classification result of the kth round; The training module further determines a target loss value for the kth round based on the first loss value for the kth round, the second loss value for the kth round, the third loss value for the kth round, the fourth loss value for the kth round, and the fifth loss value for the kth round.

[0162] In one optional configuration, the apparatus comprises: a second training unit for training a palm print picture matching model to be trained using the palm print picture set to obtain a target palm print picture matching model; the palm print picture set includes the target palm print picture and other palm print pictures, and the other palm print pictures are generated in the same manner as the target palm print picture; The target palm print picture matching model is for matching a set of input palm print pictures, where the set of palm print pictures includes palm print pictures awaiting confirmation and collected palm print pictures.

[0163] In one optional configuration, the apparatus comprises: a combination unit for obtaining the simulated palmprint curve by combining Bessel curves, wherein the target type curves include the Bessel curves; a generating unit for generating the simulated palm print picture including the simulated palm print curve.

[0164] According to another aspect of an embodiment of the present application, there is further provided an electronic device for implementing the above-described palm print picture generating method. The electronic device may be a terminal device or a server. In this embodiment, a case where the electronic device is a server will be described as an example. As shown in Fig. 22, the electronic device includes a memory 2102 and a processor 2104. A computer program is stored in the memory 2102, and the processor 2104 is configured to execute the steps of any one of the above-described method embodiments in accordance with the computer program.

[0165] Optionally, in this embodiment, the electronic device may be located in at least one network device of a plurality of network devices of a computer network.

[0166] Optionally, in this embodiment, the processor is configured by a computer program to: Step S1 of acquiring a simulated palm print picture, in which the simulated palm print picture includes a simulated palm print curve combined with a target type curve; a step S2 of inputting the simulated palm print picture and a predetermined first noise vector into a trained target palm print picture generator, and generating a target palm print picture by the target palm print picture generator by sequentially performing multiple downsampling operations and multiple upsampling operations on the simulated palm print picture, the downsampling operation is for performing a downsampling process on an input picture representation vector to obtain a downsampled picture representation vector, and performing a noise addition process on the downsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector, wherein the input picture representation vector in the first downsampling operation is an initial picture representation vector of the simulated palmprint picture; The upsampling operation may be configured to perform step S2, in which an upsampling process is performed on an input picture representation vector to obtain an upsampled picture representation vector, and a noise addition process is performed on the upsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector.

[0167] Optionally, as can be understood by those skilled in the art, the configuration shown in FIG. 22 is merely schematic, and the electronic device may be a terminal device such as a smartphone (e.g., an Android smartphone, an iOS smartphone, etc.), a tablet computer, a palmtop computer, or a mobile internet device (MID) or PAD. FIG. 22 does not limit the configuration of the electronic device. For example, the electronic device may include more or fewer components (e.g., a network interface, etc.) than those shown in FIG. 22 or may be arranged differently from those shown in FIG. 22.

[0168] The memory 2102 can be used to store software programs and modules, such as program instructions / modules corresponding to the palm print picture generation method and apparatus of the present application. The processor 2104 executes the software programs and modules stored in the memory 2102 to perform various functional applications and data processing, i.e., to realize the palm print picture generation method. The memory 2102 may include high-speed random access memory or non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 2102 may further include memory located remotely from the processor 2104, which can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, a corporate intranet, a local area network, a mobile communication network, and combinations thereof. The memory 2102 may be used to store information such as, but not limited to, sample characteristics of an item and target virtual resource accounts. 22, the memory 2102 may include, but is not limited to, the acquisition unit 2002 and the input unit 2004 in the palmprint picture generating device, and may further include, but is not limited to, other module units in the palmprint picture generating device, and the description thereof will be omitted in this example.

[0169] Optionally, the transmission device 2106 transmits and receives data via a network. Specific examples of the network may include a wired network and a wireless network. In one example, the transmission device 2106 includes a network interface controller (NIC) that can be connected to other network devices and routers via a network cable, thereby communicating with the Internet or a local area network. In one example, the transmission device 2106 is a radio frequency (RF) module that communicates with the Internet wirelessly.

[0170] The electronic device further includes a display 2108 for displaying the order information to be processed, and a connection bus 2110 for connecting the modular components of the electronic device.

[0171] In another embodiment, the terminal device or server may be a node of a distributed system. Here, the distributed system may be a blockchain system. The blockchain system may be a distributed system in which the plurality of nodes are connected in the form of network communication. Here, the nodes may form a peer-to-peer (P2P) network, and any type of computing device, for example, an electronic device such as a server or a terminal, may become a node of the blockchain system by joining the peer-to-peer network.

[0172] According to one aspect of the present application, a computer program product is provided that includes computer programs / instructions, the computer programs / instructions including program code for performing the methods illustrated in the flowcharts. In such an embodiment, the computer program may be downloaded and installed from a network via the communication unit 2209 and / or installed from a removable medium 2211. When executed by the central processing unit 2201, the computer program causes various functions provided in the embodiments of the present application to be performed.

[0173] The numbers of the above-mentioned embodiments of the present application are for illustrative purposes only and do not indicate the superiority or inferiority of the embodiments.

[0174] FIG. 23 is a block diagram illustrating the configuration of a computer system of an electronic device for implementing an embodiment of the present application.

[0175] It should be noted that the computer system 2200 of the electronic device shown in FIG. 23 is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present application.

[0176] 23, a computer system 2200 includes a central processing unit (CPU) 2201, which can execute various appropriate operations and processes based on programs stored in a read-only memory (ROM) 2202 or programs loaded from a storage unit 2208 into a random access memory (RAM) 2203. The random access memory 2203 also stores various programs and data necessary for system operation. The central processing unit 2201, the read-only memory 2202, and the random access memory 2203 are connected to one another via a bus 2204. An input / output interface (I / O interface) 2205 is also connected to the bus 2204.

[0177] The input / output interface 2205 is connected to an input unit 2206 including a keyboard, a mouse, etc., an output unit 2207 including, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, a storage unit 2208 including a hard disk, etc., and a communication unit 2209 including, for example, a network interface card such as a local area network card or a modem. The communication unit 2209 performs communication processing via a network such as the Internet. A driver 2210 is also connected to the input / output interface 2205 as needed. For example, a removable medium 2211 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory is attached to the driver 2210 as needed so that a computer program read from the removable medium 2211 is installed in the storage unit 2208 as needed.

[0178] In particular, according to embodiments of the present application, the processes illustrated in the flowcharts of each method may be implemented as a computer software program. For example, embodiments of the present application include a computer program product including a computer program stored on a computer-readable medium, the computer program including program code for executing the methods illustrated in the flowcharts. In such embodiments, the computer program may be downloaded and installed from a network via the communication unit 2209 and / or installed from removable media 2211. When executed by the central processing unit 2201, the computer program causes the system of the present application to perform various functions defined therein.

[0179] According to one aspect of the present application, a computer-readable storage medium is provided, wherein a processor of a computing device reads the computer instructions from the computer-readable storage medium and, when executed by the processor, causes the computing device to perform the methods provided in the various optional implementations described above.

[0180] Optionally, in this embodiment, as can be understood by those skilled in the art, all or part of the steps of each method in the above embodiment may be executed by instructing relevant hardware of the terminal device through a program. The program may be stored in a computer-readable storage medium. The storage medium may include a flash disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, etc.

[0181] The integrated units in the above embodiments may be realized in the form of a software functional unit and stored in the above computer-readable storage medium when sold or used as an independent product. Based on this understanding, the configurations of the present application may essentially, in other words, the portion contributing to the prior art, or all or part of the configurations, be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing one or more computer devices (which may be personal computers, servers, network devices, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application.

[0182] In the above embodiments of the present application, the description of each embodiment has its own focus, and for the parts not described in detail in one embodiment, reference can be made to the relevant descriptions of other embodiments.

[0183] In some embodiments provided herein, it should be understood that the displayed client may be implemented in other ways. The above-described device embodiments are merely schematic. For example, the division of the units is merely a logical division of functions, and other division methods may be used in actual implementation. For example, multiple units or components may be combined or incorporated into other systems, or some features may be ignored or not implemented. Furthermore, the shown or discussed mutual couplings, direct couplings, or communication connections may be via some interfaces, and the indirect couplings or communication connections of units or modules may be electrical or other types.

[0184] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple units on a network. Depending on actual needs, some or all of the units may be selected to achieve the objectives of the invention according to this embodiment.

[0185] Furthermore, the functional units in each embodiment of the present application may be integrated into one processing unit, each unit may exist physically as an individual unit, or two or more units may be integrated into one unit. The above-mentioned integrated units may be realized in the form of hardware or in the form of a software functional unit.

[0186] The above is only a preferred embodiment of the present application. It should be noted that those skilled in the art can make some improvements and modifications without departing from the principle of the present application. These improvements and modifications should also be considered as part of the protection scope of the present application.

Claims

1. 1. A method for generating a palm print picture, executed by an electronic device, comprising: obtaining a simulated palm print picture, the simulated palm print picture including a simulated palm print curve combined with a target type curve; inputting the simulated palm print picture and a predetermined first noise vector into a trained target palm print picture generator, and generating a target palm print picture by sequentially performing multiple downsampling operations and multiple upsampling operations on the simulated palm print picture by the target palm print picture generator; the downsampling operation is for performing a downsampling process on an input picture representation vector to obtain a downsampled picture representation vector, and performing a noise addition process on the downsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector, wherein the input picture representation vector in the first downsampling operation is an initial picture representation vector of the simulated palmprint picture; the upsampling operation is for performing an upsampling process on an input picture representation vector to obtain an upsampled picture representation vector, and performing a noise addition process on the upsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector; A method for generating palm print pictures.

2. The target palm print picture generator includes P downsampling modules and Q upsampling modules (P and Q are positive integers); the step of inputting the simulated palm print picture and a predetermined first noise vector into a trained target palm print picture generator, and using the target palm print picture generator to sequentially perform multiple downsampling operations and multiple upsampling operations on the simulated palm print picture to generate a target palm print picture, includes: determining the initial picture representation vector based on the simulated palm print picture; sequentially using the P downsampling modules to perform the downsampling operation on the initial picture representation vector to obtain a P-th noise-added picture representation vector, wherein each of the P downsampling modules includes a downsampling sub-module and a first condition generating sub-module; the downsampling sub-module is for performing the downsampling process on an input picture representation vector to obtain a downsampled picture representation vector; the first condition generation sub-module is for performing a noise addition process on the down-sampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector; sequentially using the Q upsampling modules to perform the upsampling operation on the P-th noise-added picture representation vector to obtain a Q-th noise-added picture representation vector, wherein each of the Q upsampling modules includes an upsampling sub-module and a second condition generating sub-module; the upsampling sub-module is for performing the upsampling process on an input picture representation vector to obtain an upsampled picture representation vector; the second condition generation sub-module is for performing the noise addition process on the up-sampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector; generating the target palm print picture based on the Qth noise-added picture representation vector; The method for generating palm print pictures according to claim 1 .

3. The step of obtaining a P-th noise-added picture representation vector by performing the downsampling operation on the initial picture representation vector using the P downsampling modules in sequence includes: The i-th noise-added picture representation vector (i is a positive integer greater than or equal to 1 and less than or equal to P) is defined as performing the downsampling process on the picture representation vector input to the i-th downsampling module by the downsampling sub-module of the i-th downsampling module to obtain an i-th downsampled picture representation vector; and performing the noise addition process on the i-th downsampled picture representation vector using the first noise vector by the first condition generation sub-module of the i-th downsampling module to obtain an i-th noise-added picture representation vector. The method for generating palm print pictures according to claim 2 .

4. The first condition generation submodule includes a first set of fully connected layers and a second set of fully connected layers; the step of obtaining an i-th noise-added picture representation vector by performing the noise addition process on the i-th downsampled picture representation vector using the first noise vector by the first condition generation sub-module of the i-th downsampling module, outputting a first control vector by passing the first noise vector through the first set of fully connected layers; outputting a second control vector by passing the first noise vector through the second set of fully connected layers; performing the noise addition process on the i-th downsampled picture representation vector based on the first control vector and the second control vector to obtain the i-th noise-added picture representation vector; The method for generating palm print pictures according to claim 3 .

5. The step of performing the upsampling operation on the Pth noise-added picture representation vector using the Q upsampling modules sequentially to obtain the Qth noise-added picture representation vector comprises: The jth noise-added picture representation vector (j is a positive integer greater than or equal to 1 and less than or equal to Q) is defined as performing the upsampling process on the picture representation vector input to the j-th upsampling module by the upsampling sub-module of the j-th upsampling module to obtain a j-th upsampled picture representation vector; and performing the noise addition process on the j-th upsampled picture representation vector using the first noise vector by the second condition generation sub-module of the j-th upsampling module to obtain a j-th noise-added picture representation vector. The method for generating palm print pictures according to claim 2 .

6. The second condition generation sub-module includes a third set of fully connected layers and a fourth set of fully connected layers, and the step of obtaining a j-th upsampled picture representation vector by performing the upsampling process on the picture representation vector input to the j-th upsampling module by the upsampling sub-module in the j-th upsampling module includes: outputting a third control vector by passing the first noise vector through the third set of fully connected layers; outputting a fourth control vector by passing the first noise vector through the fourth fully connected layer; performing the noise addition process on the j-th upsampled picture representation vector based on the third control vector and the fourth control vector to obtain the j-th noise-added picture representation vector; The method for generating palm print pictures according to claim 5.

7. The step of determining the initial picture representation vector based on the simulated palm print picture includes: performing a first convolution process on the simulated palm print picture to obtain the initial picture representation vector; The step of generating the target palm print picture based on the Qth noise-added picture representation vector comprises: performing a second convolution process on the Qth noise-added picture representation vector to obtain the target palm print picture; The method for generating palm print pictures according to claim 2 .

8. training a palmprint picture generator to be trained using a set of simulated palmprint sample pictures and a set of first real palmprint pictures for multiple rounds until a target loss value corresponding to the palmprint picture generator to be trained satisfies a predetermined convergence condition; determining the palm print picture generator at the end of training as the target palm print picture generator; In the training process, the training palm print picture generator sequentially performs multiple downsampling operations and multiple upsampling operations on the simulated palm print sample picture to generate a palm print picture; The downsampling operation in training is to perform a downsampling process on an input picture representation vector to obtain a downsampled picture representation vector, and to perform a noise addition process on the downsampled picture representation vector using an intermediate noise vector to obtain a noise-added picture representation vector; The input picture representation vector in the first down-sampling operation of training is the initial picture representation vector of the simulated palm print sample picture; The intermediate noise vector is a picture representation vector obtained by encoding the first true palmprint picture or a predetermined noise vector; The upsampling operation in training is to perform an upsampling process on an input picture representation vector to obtain an upsampled picture representation vector, and to perform a noise addition process on the upsampled picture representation vector using the intermediate noise vector to obtain a noise-added picture representation vector; the target loss value is a loss value determined based on at least a first loss value; the first loss value is a loss value between the generated palm print picture and the first true palm print picture; When determining the first loss value, the intermediate noise vector is a picture representation vector encoding the first true palmprint picture. A method for generating palm print pictures according to any one of claims 1 to 7.

9. The step of performing multiple rounds of training on a palm print picture generator to be trained using a set of simulated palm print sample pictures and a set of first real palm print pictures includes: The palm print picture generators waiting to be trained are trained in a k-th round (k is a positive integer of 2 or more), obtaining a picture representation vector for the k-th round by encoding the first true palmprint picture used for the k-th round; inputting the k-th round picture representation vector and the first simulated palm print sample picture used in the k-th round into a k-1-th round training palm print picture generator to obtain a first palm print picture generated in the k-th round; determining a first loss value for the kth round based on a first true palm print picture used in the kth round and a first palm print picture generated in the kth round; determining a target loss value for the kth round based on the first loss value for the kth round; If the k-th round target loss value does not satisfy the convergence condition, adjusting parameters of the k-1-th round training palmprint picture generator to obtain a k-th round training palmprint picture generator; if the target loss value of the kth round satisfies the convergence condition, terminating the training; The method for generating palm print pictures according to claim 8.

10. obtaining a first classification result of the kth round by classifying the first true palm print picture used in the kth round and the first palm print picture generated in the kth round using a trained target classifier; determining a second loss value for the kth round based on the first identification result for the kth round; The step of determining a target loss value for a kth round based on the first loss value for the kth round comprises: determining a target loss value for the kth round based on a first loss value for the kth round and a second loss value for the kth round; The method for generating palm print pictures according to claim 9.

11. obtaining a second palm print picture generated in the k-th round by inputting the noise vector of the k-th round and the first simulated palm print sample picture used in the k-th round into the palm print picture generator after training of the k-1-th round; determining a third loss value for the kth round based on the first palm print picture generated in the kth round and the second palm print picture generated in the kth round; The step of determining a target loss value for the kth round based on the first loss value for the kth round and the second loss value for the kth round includes: determining a target loss value for the kth round based on a first loss value for the kth round, a second loss value for the kth round, and a third loss value for the kth round; The method for generating palm print pictures according to claim 10.

12. determining a fourth loss value for the kth round based on the kth round noise vector and the kth round picture representation vector; determining a target loss value for the kth round based on a first loss value for the kth round, a second loss value for the kth round, and a third loss value for the kth round, determining a target loss value for the k round based on a first loss value for the k round, a second loss value for the k round, a third loss value for the k round, and a fourth loss value for the k round; The method for generating palm print pictures according to claim 11.

13. obtaining a second classification result of the kth round by using the target classifier to classify a second true palm print picture used in the kth round and a second palm print picture generated in the kth round; determining a fifth loss value of the kth round based on the second identification result of the kth round; determining a target loss value for the kth round based on a first loss value for the kth round, a second loss value for the kth round, a third loss value for the kth round, and a fourth loss value for the kth round, determining a target loss value for the k round based on a first loss value for the k round, a second loss value for the k round, a third loss value for the k round, a fourth loss value for the k round, and a fifth loss value for the k round; The method for generating palm print pictures according to claim 12.

14. The method further includes: training a waiting-to-train palm print picture matching model using the palm print picture set to obtain a target palm print picture matching model; the palm print picture set includes the target palm print picture and other palm print pictures, and the other palm print pictures are generated in the same manner as the target palm print picture; The target palm print picture matching model is for matching a set of input palm print pictures, and the set of palm print pictures includes palm print pictures awaiting confirmation and collected palm print pictures. A method for generating palm print pictures according to any one of claims 1 to 7.

15. obtaining the simulated palmprint curve by combining Bessel curves, wherein the target type curves include the Bessel curves; generating the simulated palm print picture including the simulated palm print curve. A method for generating palm print pictures according to any one of claims 1 to 7.

16. A palm print picture generation device, comprising: An acquisition unit for acquiring a simulated palm print picture, wherein the simulated palm print picture includes a simulated palm print curve combined with a target type curve; an input unit for inputting the simulated palm print picture and a predetermined first noise vector into a trained target palm print picture generator, and for generating a target palm print picture by the target palm print picture generator by sequentially performing multiple downsampling operations and multiple upsampling operations on the simulated palm print picture; the downsampling operation is for performing a downsampling process on an input picture representation vector to obtain a downsampled picture representation vector, and performing a noise addition process on the downsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector, wherein the input picture representation vector in the first downsampling operation is an initial picture representation vector of the simulated palmprint picture; the upsampling operation is for performing an upsampling process on an input picture representation vector to obtain an upsampled picture representation vector, and performing a noise addition process on the upsampled picture representation vector using the first noise vector to obtain a noise-added picture representation vector; Palm print picture generator.

17. 16. A computer readable storage medium containing a program stored thereon, the program, when executed, causing the method for generating palm print pictures according to any one of claims 1 to 15 to be carried out.

18. A computer program product comprising computer programs / instructions which, when executed by a processor, cause the computer program / instructions to implement the steps of the method for generating palm print pictures according to any one of claims 1 to 15.

19. 16. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the method for generating palm print pictures according to any one of claims 1 to 15 by means of the computer program.

Images