Palm liveness detection method, system, device and storage medium

Abstract

This invention discloses a method, system, device, and storage medium for palm liveness detection, belonging to the field of biometric recognition technology, and solving the problem of low accuracy in palm liveness detection in existing technologies. It involves acquiring an original palm image and converting it into a palm depth image and a palm frequency domain image; inputting the original palm image, depth image, and frequency domain image into three feature extraction networks to extract features from the three images; fusing these three image features through a fusion network to obtain deep texture and detail features of multiple modalities of the palm image; and finally outputting whether the original palm image is a real image. This invention can extract deep features from multiple modalities of palm images based on a single modal palm image, thereby amplifying the feature differences between real and fake palm images, improving the accuracy of judging the authenticity of palm images, and simultaneously performing palm print liveness detection and palm vein liveness detection with high efficiency and enhanced reliability.

Description

Technical Field

[0001] The present invention belongs to the technical field of biological sign recognition, and relates to a palm liveness detection method, system, equipment and storage medium. Background Art

[0002] In recent years, biometric recognition technology has developed rapidly and has been widely used in identity authentication systems such as financial payment and access control. However, at the same time, there are more and more attempts to attack biometric recognition with prostheses. With the promotion and application of palm recognition technology, how to effectively detect liveness of the palm has become a key issue in palm recognition technology.

[0003] Currently, many researchers have conducted research on palm liveness detection. However, for palm liveness detection methods under near-infrared light, most of them directly judge whether it is a real image based on the imaging principle of palm veins. However, if the palm vein image is directly printed, it is impossible to perform liveness detection based on the imaging principle of veins because the vein information already exists in the image. In addition, palm liveness detection under visible light is similar to liveness detection in face recognition. The original image is directly input into the neural network to obtain the result of whether it is real or not. However, this method only uses the RGB information of the palm image, and it is easy to get wrong results when the lighting is particularly complex. Summary of the invention

[0004] The purpose of the present invention is to solve the problem of low accuracy in palm liveness detection in the prior art, and to provide a palm liveness detection method, system, device and storage medium.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] The present invention provides a palm liveness detection method, comprising the following steps:

[0007] Acquire a palm image, and acquire a depth image and a frequency domain image according to the palm image;

[0008] Acquire palm image features according to the palm image, the depth image and the frequency domain image;

[0009] According to the palm image features, the living palm image and the non-living palm image are obtained to realize the living palm detection.

[0010] Preferably, the acquired palm image is a palm print image under visible light or a palm vein image under near infrared light.

[0011] Preferably, acquiring a depth image according to a palm image comprises:

[0012] The deep neural network is used to obtain the depth image, as follows:

[0013] When training a deep neural network, the mean square error loss L is used. MSEL Update the weights of the deep neural network, the mean square error loss L MSEL as follows:

[0014]

[0015] Among them, D P is the palm image, D G is the palm depth image.

[0016] Preferably, acquiring a frequency domain image according to a palm image comprises:

[0017] The frequency domain image is obtained by using two-dimensional discrete cosine transform, as follows:

[0018] The two-dimensional discrete cosine transform matrix C(m,n) is calculated as follows:

[0019]

[0020] Where C(i,j) is the pixel value of the grayscale image of the palm, a(m) and a(n) are coefficients that transform the two-dimensional discrete cosine transform into an orthogonal matrix, and the definitions of a(m) and a(n) are shown in formula (3) and formula (4):

[0021]

[0022] The two-dimensional discrete cosine transform matrix C(m,n) is then quantized and encoded to obtain a frequency domain image.

[0023] Preferably, the specific steps of inputting the palm image, the depth image and the frequency domain image into the feature extraction network to obtain the palm image features are as follows:

[0024] The output feature f of the feature extraction network out It is expressed as shown in formula (5):

[0025]

[0026] Among them, w c is the convolution kernel of the current feature extraction network, f in is the input of the feature extraction network, w l =[1,1,1; 1,-8,1; 1,1,1] is the Laplace operator, θ is the hyperparameter that controls the weights of the traditional Laplace operator and the convolution kernel, Represents a convolution operation.

[0027] Preferably, the feature extraction network includes a first network branch f1, a second network branch f2 and a third network branch f3, then the palm image f out1, depth image f out2 and the frequency domain image f out3 The characteristics are expressed as shown in formula (6):

[0028]

[0029] Among them, f in1 、f in2 、f in3 are the inputs of the first network branch f1, the second network branch f2 and the third network branch f3, respectively. c1 、w c2 、w c3 They are respectively represented as the convolution kernels of the first network branch f1, the second network branch f2 and the third network branch f3, and θ1, θ2 and θ3 are respectively represented as the hyperparameters for controlling the traditional Laplacian operator and the convolution kernel weights;

[0030] The first network branch f1, the second network branch f2 and the third network branch f3 combine the traditional Laplacian operator with the convolution kernel to extract the feature differences between the real palm image and the false palm image.

[0031] Preferably, the structures of the first network branch f1 , the second network branch f2 and the third network branch f3 are the same, but the parameters among the first network branch f1 , the second network branch f2 and the third network branch f3 are not shared.

[0032] The present invention provides a palm liveness detection system, comprising:

[0033] An image acquisition module, the image acquisition module is used to acquire a palm image, and acquire a depth image and a frequency domain image according to the palm image;

[0034] A feature extraction module, wherein the feature extraction module is used to obtain palm image features according to the palm image, the depth image and the frequency domain image;

[0035] The liveness detection module is used to obtain live palm images and non-live palm images according to palm image features to achieve liveness detection of the palm.

[0036] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a palm liveness detection method when executing the computer program.

[0037] A computer-readable storage medium stores a computer program, which implements the steps of a palm liveness detection method when executed by a processor.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention proposes a palm liveness detection method, which first generates a depth image and a frequency domain image using an original palm image, then extracts features based on the palm image, the depth image and the frequency domain image, and finally fuses the features of multiple palm images to fully explore the deep information of the image and predict the authenticity of the original palm image. Specifically, the present invention generates a depth image and a frequency domain image based on the original palm image, respectively, and then extracts the features of the palm image, the depth image and the frequency domain image respectively based on the palm image, the depth image and the frequency domain image, and then combines the extracted features to capture the difference between the deep features of the real palm image and the false palm image, which is not easily affected by external light and noise, and does not rely too much on the imaging principle of the image, so that the efficiency of liveness detection is higher, and it can further promote the application of palm recognition in actual scenes.

[0040] Furthermore, the acquired palm image is a palm print image under visible light or a palm vein image under near-infrared light, which can be used to process palm print liveness detection or palm vein liveness detection, and has a wider application range.

[0041] Furthermore, the present invention directly predicts a depth image based on a palm image, which has lower cost, higher efficiency, and is more in line with application scenario conditions of palm liveness detection.

[0042] Furthermore, the two-dimensional discrete cosine transform has better frequency domain energy concentration and can cut off unimportant frequency domain areas and coefficients, thereby retaining frequency domain features with more information. Combined with the characteristics of palm recognition, the main lines, textures, and wrinkles contained in the palm image have strong directionality, where coarse textures reflect low-frequency contour information and fine textures reflect high-frequency detail information. Therefore, using two-dimensional discrete cosine transform to transform the palm image can better characterize the palm features.

[0043] Furthermore, since palm images, depth images and frequency domain images belong to data in three different fields, the present invention sets the structures of f1, f2, and f3 to be the same, and uses non-shared parameters to extract features of data in these three fields, while previous methods mostly use shared parameters. Since data in different fields will introduce the influence of negative transfer, it will reduce the feature extraction ability of the model.

[0044] The present invention proposes a palm liveness detection system, which divides the estimation system into an image acquisition module, a feature extraction module and a liveness detection module, and adopts a modular concept to make each module independent of each other, thereby facilitating unified management of each module. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0046] Figure 1 This is a flow chart of the palm liveness detection method of the present invention.

[0047] Figure 2 Schematic diagram of the palm liveness detection method of the present invention.

[0048] Figure 3 Schematic diagram of the palm liveness detection method process of the present invention.

[0049] Figure 4 This is a diagram of the palm liveness detection system of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0051] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0053] In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. indicate an orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0054] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0055] In the description of the embodiments of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0056] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0057] like Figure 1 As shown, a palm liveness detection method proposed by the present invention comprises the following steps:

[0058] S1, obtaining a palm image, and obtaining a depth image and a frequency domain image according to the palm image;

[0059] The acquired palm image is a palm print image under visible light or a palm vein image under near infrared light.

[0060] Obtaining a depth image based on a palm image includes:

[0061] The deep neural network is used to obtain the depth image, as follows:

[0062] When training a deep neural network, the mean square error loss L is used. MSEL Update the weights of the deep neural network, the mean square error loss L MSEL as follows:

[0063]

[0064] Among them, D P Represented as a palm image, D G is the palm depth image.

[0065] Obtain frequency domain images based on palm images, including:

[0066] The frequency domain image is obtained by using two-dimensional discrete cosine transform, as follows:

[0067] The two-dimensional discrete cosine transform matrix C(m,n) is calculated as follows:

[0068]

[0069] Where C(i,j) is the pixel value of the grayscale image of the palm, a(m) and a(n) are coefficients that transform the two-dimensional discrete cosine transform into an orthogonal matrix, and the definitions of a(m) and a(n) are shown in formula (3) and formula (4):

[0070]

[0071] The two-dimensional discrete cosine transform matrix C(m,n) is then quantized and encoded to obtain a frequency domain image.

[0072] S2, obtaining palm image features according to the palm image, the depth image and the frequency domain image;

[0073] The specific steps of inputting the palm image, depth image and frequency domain image into the feature extraction network to obtain the palm image features are as follows:

[0074] The output feature f of the feature extraction network out It is expressed as shown in formula (5):

[0075]

[0076] Among them, w c is the convolution kernel of the current feature extraction network, f in is the input of the feature extraction network, w l =[1,1,1; 1,-8,1; 1,1,1] is the Laplace operator, θ is the hyperparameter that controls the weights of the traditional Laplace operator and the convolution kernel, Represents a convolution operation.

[0077] The feature extraction network includes a first network branch f1, a second network branch f2 and a third network branch f3, then the palm image f out1 , depth image f out2 and the frequency domain image f out3 The characteristics are expressed as shown in formula (6):

[0078]

[0079] Among them, f in1 、f in2 、f in3 are the inputs of the first network branch f1, the second network branch f2 and the third network branch f3, respectively. c1 、w c2 、w c3They are respectively represented as the convolution kernels of the first network branch f1, the second network branch f2 and the third network branch f3, and θ1, θ2 and θ3 are respectively represented as the hyperparameters for controlling the traditional Laplacian operator and the convolution kernel weights;

[0080] The first network branch f1, the second network branch f2 and the third network branch f3 combine the traditional Laplacian operator with the convolution kernel to extract the feature differences between the real palm image and the false palm image.

[0081] The first network branch f1 , the second network branch f2 , and the third network branch f3 have the same structure, but parameters among the first network branch f1 , the second network branch f2 , and the third network branch f3 are not shared.

[0082] S3. Acquire a live palm image and a non-live palm image according to the palm image features to implement liveness detection of the palm.

[0083] The present invention proposes a palm liveness detection method, such as Figure 2 and Figure 3 As shown, the specific steps are as follows:

[0084] Step 1: Convert the palm image to a depth image

[0085] To obtain the palm image, previous methods mostly rely on costly consumer-grade depth cameras such as structured light and binocular vision or use multiple cameras to obtain the depth image, while the present invention only uses a single ordinary camera to capture the palm image.

[0086] The palm depth image is obtained by using a deep neural network. The input is the palm image and the output is the palm depth image. This has lower cost, higher efficiency, and is more in line with the application scenario conditions of palm liveness detection. The specific steps for obtaining the palm depth image by using a deep neural network are as follows:

[0087] When training a deep neural network, the mean square error loss L is used. MSEL Update the weights of the deep neural network. Specifically, assuming that the palm image is represented by D P , the palm depth image is represented as D G , then the mean square error loss L MSEL The definition is shown in formula (1):

[0088]

[0089] Step 2: Convert the palm image to frequency domain image

[0090] Based on the two-dimensional discrete cosine transform, the palm frequency domain image is obtained. For the N×N dimensional palm image, the pixel value of the palm image is represented by c(i, j). The N×N dimensional palm grayscale image is obtained by grayscale operation. The pixel value of the palm grayscale image is represented by C(i, j). Then, the formula for calculating its two-dimensional discrete cosine transform matrix C(m, n) is shown in formula (2):

[0091]

[0092] Wherein, a(m) and a(n) are coefficients that make the two-dimensional discrete cosine transform into an orthogonal matrix. The definitions of a(m) and a(n) are shown in formula (3) and formula (4):

[0093]

[0094] Then the two-dimensional discrete cosine transform matrix C(m,n) can be quantized and encoded to obtain the frequency domain image. The two-dimensional discrete cosine transform has better frequency domain energy concentration, and can cut off unimportant frequency domain areas and coefficients, thereby retaining frequency domain features with more information. Combined with the characteristics of palm recognition, the main lines, textures, and wrinkles contained in the palm image have strong directionality, where coarse textures reflect low-frequency contour information and fine textures reflect high-frequency detail information. Therefore, the two-dimensional discrete cosine transform is used to transform the palm image to better characterize the palm features.

[0095] When the N×N dimensional palm image is not a square matrix, pixels need to be padded in order to obtain complete feature information.

[0096] Step 3: Feature extraction and fusion of palm image, depth image and frequency domain image, and output of liveness detection results

[0097] The original palm image, depth image and frequency domain image are input into the feature extraction network. The feature extraction network includes the first network branch f1, the second network branch f2 and the third network branch f3. The input is the original palm image, depth image and frequency domain image, and the output is three features. The first network branch f1, the second network branch f2 and the third network branch f3 have the same structure, but the parameters are not shared. The first network branch f1, the second network branch f2 and the third network branch f3 combine the traditional Laplacian operator with the convolution kernel to extract the feature differences in details and textures between the real palm image and the fake palm image. Specifically, assuming that the convolution kernel of the current feature extraction network is w c , the input of the feature extraction network is represented as f in , the Laplace operator is represented by w l =[1,1,1;1,-8,1;1,1,1], then we get the output feature f of the network out It is expressed as shown in formula (5):

[0098]

[0099] Among them, θ is a hyperparameter that controls the weights of the traditional Laplacian operator and the convolution kernel. Represents a convolution operation. Since the palm image, depth image and frequency domain image belong to three different fields of data, the present invention sets the structures of f1, f2 and f3 to be the same, and uses non-shared parameters to extract the features of the three fields of data, while the previous methods mostly use shared parameters. Since the data in different fields will introduce the influence of negative transfer, it will reduce the feature extraction ability of the model.

[0100] For the first network branch f1, the second network branch f2, and the third network branch f3, it is assumed that the inputs of the first network branch f1, the second network branch f2, and the third network branch f3 are respectively represented by f in1 、f in2 、f in3 The outputs of the first network branch f1, the second network branch f2 and the third network branch f3 are respectively represented by f out1 、f out2 、f out3 , the convolution kernels of the first network branch f1, the second network branch f2 and the third network branch f3 are represented as w c1 、w c2 、w c3 , the hyperparameters controlling the traditional Laplacian operator and the convolution kernel weights are denoted as θ1, θ2, and θ3, respectively. Then the palm image f out1 , depth image f out2 and the frequency domain image f out3 The characteristics are expressed as shown in formula (6):

[0101]

[0102] Further, after obtaining three features f out1 、f out2 、f out3 After that, they are input into the fusion network f to output the authenticity of the original palm image. Specifically, f out1 、f out2 、f out3 The channels are spliced ​​into a feature, and then the prediction result of the authenticity of the original palm image is directly obtained through the processing of the fusion network f, which is expressed as f(f out1 ,f out2 ,f out3 ).

[0103] The present invention proposes a palm liveness detection system, such as Figure 4 As shown, including:

[0104] An image acquisition module, the image acquisition module is used to acquire a palm image, and acquire a depth image and a frequency domain image according to the palm image;

[0105] A feature extraction module, wherein the feature extraction module is used to obtain palm image features according to the palm image, the depth image and the frequency domain image;

[0106] The liveness detection module is used to obtain live palm images and non-live palm images according to palm image features to achieve liveness detection of the palm.

[0107] A terminal device provided in one embodiment of the present invention comprises: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor executes the computer program, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0108] The computer program may be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to accomplish the present invention.

[0109] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0110] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0111] The memory may be used to store the computer programs and / or modules, and the processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory.

[0112] If the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0113] The present invention proposes a palm liveness detection method, system, device and storage medium. Based on the original palm image, a deep neural network is used to obtain a depth image, and a two-dimensional discrete cosine transform is used to obtain a frequency domain image. It can fuse information from multiple aspects, and with the help of a feature extraction network and a fusion network, the difference between the deep features of a real palm image and a false palm image is captured, and it is not easily affected by external light and noise. The depth image can extract the spatial structure information of the palm image, while the false palm image can only provide the plane features of the image. Therefore, the use of the depth image can easily identify false images in the form of paper printed photos, screens, etc., and the present invention does not rely on consumer-grade depth cameras such as structured light and binocular vision, which is lower in cost, higher in efficiency, and more in line with the application scenario conditions of palm liveness detection. In addition, converting the palm image into a frequency domain image can further utilize the frequency domain information of the image. Since the high-frequency information distribution of the false palm image is relatively simple, while the high-frequency information of the real palm image is divergent from the center of the image to the outside, the authenticity of the image can be accurately judged by the difference in the frequency domain information of the image. The present invention adopts an end-to-end method to simultaneously fuse the information of the original palm image, the depth image and the frequency domain image, and can capture the slight difference between the real image and the false image, thereby improving the accuracy of liveness detection. In addition, the present invention can also handle palm print recognition liveness detection and palm vein liveness detection in palm liveness detection, and does not rely too much on the imaging principle of the image, so that the liveness detection is more efficient and safer, and can further promote the application of palm recognition in actual scenarios.

[0114] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A palm liveness detection method, characterized in that: The following steps are involved: Acquire a palm image, and acquire a depth image and a frequency domain image according to the palm image; Acquire palm image features according to the palm image, the depth image and the frequency domain image; Input the palm image features into the fusion network, output the palm liveness detection result, and realize the palm liveness detection; The step of acquiring a frequency domain image according to a palm image comprises: The frequency domain image is obtained by using two-dimensional discrete cosine transform, as follows: 2D Discrete Cosine Transform Matrix The calculation of is as follows: in, is the pixel value of the grayscale image of the palm, and are the coefficients that make the two-dimensional discrete cosine transform into an orthogonal matrix. and The definition of is shown in formula (3) and formula (4): Then the two-dimensional discrete cosine transform matrix Through quantization and encoding processing, a frequency domain image is obtained; The specific steps of inputting the palm image, depth image and frequency domain image into the feature extraction network to obtain the palm image features are as follows: Output features of the feature extraction network It is expressed as shown in formula (5): in, is the convolution kernel of the current feature extraction network, is the input of the feature extraction network, is the Laplace operator, To control the hyperparameters of the traditional Laplacian operator and the weights of the convolution kernel, Represents a convolution operation.

2. The palm liveness detection method according to claim 1, characterized in that: The acquired palm image is a palm print image under visible light or a palm vein image under near infrared light.

3. The palm liveness detection method according to claim 1, characterized in that: The step of acquiring a depth image according to the palm image includes: The deep neural network is used to obtain the depth image, as follows: When training deep neural networks, the mean square error loss is used Update the weights of a deep neural network, mean squared error loss as follows: in, is the palm image, is the palm depth image.

4. The palm liveness detection method according to claim 1, characterized in that: The feature extraction network includes the first network branch , Second network branch and the third network branch , then the palm image , depth image and frequency domain images The characteristics are expressed as shown in formula (6): in, , , First Network Branch , Second network branch and the third network branch Input, , , Respectively represent the first network branch , Second network branch and the third network branch The convolution kernel, , , They are respectively represented as hyperparameters controlling the traditional Laplacian operator and the convolution kernel weights; First Network Branch , Second network branch and the third network branch The traditional Laplacian operator is combined with the convolution kernel to extract the feature differences between real palm images and fake palm images.

5. The palm liveness detection method according to claim 4, characterized in that: First Network Branch , Second network branch and the third network branch The structure is the same, but the first network branch , Second network branch and the third network branch The parameters are not shared.

6. A palm liveness detection system, characterized in that: The palm liveness detection method according to any one of claims 1 to 5 comprises: An image acquisition module, the image acquisition module is used to acquire a palm image, and acquire a depth image and a frequency domain image according to the palm image; A feature extraction module, wherein the feature extraction module is used to obtain palm image features according to the palm image, the depth image and the frequency domain image; The liveness detection module is used to input the palm image features into the fusion network, output the liveness detection result of the palm, and realize the liveness detection of the palm.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the palm liveness detection method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the palm liveness detection method according to any one of claims 1 to 5 are implemented.

Images