Method, apparatus and device for globally recognizing biometric feature image and storage medium

Abstract

The application discloses a method and device for globally identifying a biological feature image, equipment and a storage medium. The method comprises the following steps: acquiring a biological feature image in a four-dimensional tensor data format; inputting the biological feature image into a pre-constructed global branch network to output a global branch feature map; processing the biological feature image through a feature map module of a backbone network to obtain a local branch feature map; when the width of the local branch feature map is the same as the width of the global branch feature map and the height of the local branch feature map is the same as the height of the global branch feature map, fusing the global branch feature map and the local branch feature map to obtain a fused feature map. It can be seen that, while the backbone network analyzes the local features of the biological feature image with high precision, the global branch network extracts the global features of the biological feature image, so that the biological recognition features fuse the global features and the local features, thereby enhancing the biological recognition feature recognition capability.

Description

Technical Field

[0001] This application relates to the field of intelligent recognition technology, and more specifically, to methods, apparatus, devices, and storage media for globally recognizing biometric images. Background Technology

[0002] With the continuous development of information technology, information mining has attracted the interest of many researchers, who aim to extract useful information from massive amounts of data. Different objects possess unique characteristics, such as fingerprints, palm prints, and facial features. Researchers can extract these features from facial images or their complex details to identify a specific object. For image feature extraction methods, researchers can design algorithms or train specialized models to perform feature extraction. Typically, before inputting an image into an algorithm / model, it needs to be processed to conform to the algorithm / model's data input format, such as the format of four-dimensional tensor data.

[0003] Current biometric recognition models can perform high-precision recognition of local biometric features, but their ability to discriminate global features is weak.

[0004] How to enhance the global feature recognition of biometric images is an issue that needs attention. Summary of the Invention

[0005] In view of the above problems, this application is proposed to provide a method, apparatus, device and storage medium for global recognition of biometric images, so as to enhance the global feature recognition of biometric images and combine the high accuracy of local features to make the biometric recognition model more representative and descriptive.

[0006] To achieve the above objectives, the following specific solutions are proposed:

[0007] Methods for globally identifying biometric images include:

[0008] Acquire biometric images in a four-dimensional tensor format, wherein the biometric images include the number of image batches, the number of input channels, the image width, and the image height;

[0009] The biometric image is input into a pre-constructed global branch network, and a global branch feature map is output. The global branch feature map includes the number of global branch image batches, the number of global branch feature output channels, the width of the global branch feature map, and the height of the global branch feature map. The number of global branch feature images is equal to the number of image batches, the number of global branch feature output channels is greater than the number of input channels, the width of the global branch feature map is less than the width of the image, and the height of the global branch feature map is less than the height of the image.

[0010] The biometric image is processed by the feature map module of the backbone network to obtain a local branch feature map. The local branch feature map includes the number of local branch feature image batches, the number of local branch feature output channels, the width of the local branch feature map, and the height of the local branch feature map. The number of local branch feature image batches is the same as the number of image batches, and the number of local branch feature output channels is greater than the number of input channels.

[0011] When the width of the local branch feature map is the same as the width of the global branch feature map, and the height of the local branch feature map is the same as the height of the global branch feature map, the global branch feature map and the local branch feature map are fused to obtain a fused feature map. The fused feature map includes the number of fused feature image batches, the number of fused feature output channels, the width of the fused feature map, and the height of the fused feature map. The number of fused feature image batches is the same as the number of image batches. The number of fused feature output channels is the sum of the number of global branch feature output channels and the number of local branch feature output channels. The width of the fused feature map is equal to the width of the local branch feature map, and the height of the fused feature map is equal to the height of the local branch feature map.

[0012] Optionally, the biometric image is input into a pre-constructed global branch network, and a global branch feature map is output, including:

[0013] The biometric image is input into the embedding layer of a pre-constructed global branch network, and an intermediate feature map is output. The intermediate feature map includes the number of intermediate image batches, the number of intermediate feature output channels, the width of the intermediate feature map, and the height of the intermediate feature map. The number of intermediate image batches is the same as the number of image batches, the number of intermediate feature output channels is greater than the number of input channels, the width of the intermediate feature map is less than the width of the image, and the height of the intermediate feature map is less than the height of the image.

[0014] The intermediate feature map is converted into a global branch feature map by the global branch network generation module. The height of the global branch feature map is smaller than the height of the intermediate feature map, and the width of the global branch feature map is smaller than the width of the intermediate feature map.

[0015] Optionally, the embedding layer includes a first convolutional layer, a second convolutional layer, a BatchNorm network layer, and an activation function layer. The number of input channels of the first convolutional layer is a preset number of channels, the number of output channels of the first convolutional layer is half of the preset number of channels, the number of input channels of the second convolutional layer is half of the preset number of channels, and the number of output channels of the second convolutional layer is the preset number of channels.

[0016] The biometric image is input into the embedding layer of a pre-constructed global branch network, and intermediate feature maps are output, including:

[0017] The biometric image is input into the first convolutional layer to obtain the first intermediate result;

[0018] The first intermediate result is input into the BatchNorm network layer to obtain the second intermediate result;

[0019] The second intermediate result is input into the activation function layer to obtain the third intermediate result;

[0020] The third intermediate result is input into the second convolutional layer to obtain the fourth intermediate result;

[0021] The fourth intermediate result is input into the BatchNorm network layer to obtain an intermediate feature map.

[0022] Optionally, the intermediate feature map is converted into a global branch feature map by the generation module of the global branch network, including:

[0023] The intermediate feature map is deformed by the generation module of the global branch network to obtain a three-dimensional tensor;

[0024] The three-dimensional tensor is compressed to obtain a compressed three-dimensional tensor;

[0025] The compressed 3D tensor is converted into a global branch feature map.

[0026] Optionally, when the width of the local branch feature map is the same as the width of the global branch feature map, and the height of the local branch feature map is the same as the height of the global branch feature map, after fusing the global branch feature map and the local branch feature map to obtain the fused feature map, the method further includes:

[0027] The fused feature map is input into the feature recognition module of the backbone network, and the biometric features of the biometric image are output. The number of feature dimensions of the biometric features is the number of channels after convolving the number of output channels of the global branch features with the number of output channels of the local branch features.

[0028] A device for globally recognizing biometric images, comprising:

[0029] Tensor data acquisition unit is used to acquire biofeature images in four-dimensional tensor format, wherein the biofeature images include image batch size, number of input channels, image width and image height;

[0030] The branch network output unit is used to input the biometric image into a pre-constructed global branch network and output a global branch feature map. The global branch feature map includes the number of global branch image batches, the number of global branch feature output channels, the width of the global branch feature map, and the height of the global branch feature map. The number of global branch feature images is equal to the number of image batches, the number of global branch feature output channels is greater than the number of input channels, the width of the global branch feature map is less than the width of the image, and the height of the global branch feature map is less than the height of the image.

[0031] The backbone network output unit is used to process the biometric image through the feature map module of the backbone network to obtain a local branch feature map. The local branch feature map includes the number of local branch feature image batches, the number of local branch feature output channels, the width of the local branch feature map, and the height of the local branch feature map. The number of local branch feature image batches is the same as the number of image batches, and the number of local branch feature output channels is greater than the number of input channels.

[0032] A global data determination unit is configured to fuse the global branch feature map and the local branch feature map when the width of the local branch feature map is the same as the width of the global branch feature map and the height of the local branch feature map is the same as the height of the global branch feature map, thereby obtaining a fused feature map. The fused feature map includes the number of fused feature image batches, the number of fused feature output channels, the width of the fused feature map, and the height of the fused feature map. The number of fused feature image batches is the same as the number of image batches. The number of fused feature output channels is the sum of the number of global branch feature output channels and the number of local branch feature output channels. The width of the fused feature map is equal to the width of the local branch feature map, and the height of the fused feature map is equal to the height of the local branch feature map.

[0033] Optionally, the branch network output unit includes:

[0034] An embedding layer output unit is used to input the biometric image into the embedding layer of a pre-constructed global branch network and output an intermediate feature map. The intermediate feature map includes the number of intermediate image batches, the number of intermediate feature output channels, the width of the intermediate feature map, and the height of the intermediate feature map. The number of intermediate image batches is the same as the number of image batches, the number of intermediate feature output channels is greater than the number of input channels, the width of the intermediate feature map is smaller than the width of the image, and the height of the intermediate feature map is smaller than the height of the image.

[0035] The generation module output unit is used to convert the intermediate feature map into a global branch feature map through the generation module of the global branch network. The height of the global branch feature map is smaller than the height of the intermediate feature map, and the width of the global branch feature map is smaller than the width of the intermediate feature map.

[0036] Optionally, the embedding layer includes a first convolutional layer, a second convolutional layer, a BatchNorm network layer, and an activation function layer. The number of input channels of the first convolutional layer is a preset number of channels, the number of output channels of the first convolutional layer is half of the preset number of channels, the number of input channels of the second convolutional layer is half of the preset number of channels, and the number of output channels of the second convolutional layer is the preset number of channels.

[0037] The embedded layer output unit includes:

[0038] The first embedding layer output subunit is used to input the biometric image into the first convolutional layer to obtain a first intermediate result;

[0039] The second embedding layer output subunit is used to input the first intermediate result into the BatchNorm network layer to obtain the second intermediate result;

[0040] The third embedding layer output subunit is used to input the second intermediate result into the activation function layer to obtain the third intermediate result;

[0041] The fourth embedding layer output subunit is used to input the third intermediate result into the second convolutional layer to obtain the fourth intermediate result;

[0042] The fifth embedding layer output subunit is used to input the fourth intermediate result into the BatchNorm network layer to obtain an intermediate feature map.

[0043] Optionally, the output unit of the generation module includes:

[0044] The first generation module output subunit is used to deform the intermediate feature map through the generation module of the global branch network to obtain a three-dimensional tensor.

[0045] The second generation module output subunit is used to compress the three-dimensional tensor to obtain a compressed three-dimensional tensor.

[0046] The third generation module output subunit is used to convert the compressed 3D tensor into a global branch feature map.

[0047] Optionally, the device may also include:

[0048] The feature dimension extraction unit is used to input the fused feature map into the feature recognition module of the backbone network when the width of the local branch feature map is the same as the width of the global branch feature map and the height of the local branch feature map is the same as the height of the global branch feature map, and output the biometric features of the biometric image. The number of feature dimensions of the biometric features is the number of channels after convolving the number of output channels of the global branch features with the number of output channels of the local branch features.

[0049] A device for globally identifying biometric images, characterized in that it includes a memory and a processor;

[0050] The memory is used to store programs;

[0051] The processor is used to execute the program to implement the various steps of the method for globally recognizing biometric images as described above.

[0052] A storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for globally recognizing biometric images as described above.

[0053] Using the above technical solution, this application obtains biometric images in a four-dimensional tensor format. The biometric image includes the number of image batches, the number of input channels, the image width, and the image height. The biometric image is input into a pre-constructed global branch network, outputting a global branch feature map. The global branch feature map includes the number of global branch image batches, the number of global branch feature output channels, the global branch feature map width, and the global branch feature map height. Specifically, the number of global branch feature images is equal to the number of image batches, the number of global branch feature output channels is greater than the number of input channels, the global branch feature map width is less than the image width, and the global branch feature map height is less than the image height. The biometric image is processed by the feature map module of the backbone network to obtain a local branch feature map. The local branch feature map includes the number of local branch feature image batches, the number of local branch feature output channels, and the local branch feature map height. The global branch feature map is fused with the local branch feature map. The local branch feature map has the same width and height as the global branch feature map. When the width and height of the local branch feature map are the same as the global branch feature map, the global branch feature map and the local branch feature map are fused to obtain a fused feature map. This fused feature map includes the number of fused feature map batches, the number of fused feature output channels, the fused feature map width, and the fused feature map height. The number of fused feature map batches is the same as the number of image batches. The number of fused feature output channels is the sum of the number of global branch feature output channels and the number of local branch feature output channels. The fused feature map width and the fused feature map height are equal to the local branch feature map width and height. Therefore, while the backbone network performs high-precision analysis of local features in biometric images, the global branch network extracts global features from the biometric images, enabling the biometric features to integrate both global and local features, thereby enhancing the biometric recognition capability. Attached Figure Description

[0054] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0055] Figure 1 A schematic diagram illustrating a global biometric image recognition process provided in this application embodiment;

[0056] Figure 2 This application provides a schematic diagram of an embedding layer for processing four-dimensional tensor data in an embodiment of the present application.

[0057] Figure 3 A schematic diagram of a device for globally recognizing biometric images provided in an embodiment of this application;

[0058] Figure 4 This is a schematic diagram of the structure of a device for globally recognizing biometric images, provided in an embodiment of this application. Detailed Implementation

[0059] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0060] The proposed solution can be implemented based on a terminal with data processing capabilities, such as a computer, server, or cloud platform.

[0061] Next, combined Figure 1 The method for globally identifying biometric images according to this application may include the following steps:

[0062] Step S110: Obtain biometric images in four-dimensional tensor format.

[0063] The biometric images include the number of image batches, the number of input channels, the image width, and the image height. The number of image batches can represent the total number of biometric images in a single batch. Biometric images can be facial images or palm images; implementation of this invention constitutes legal use.

[0064] For example, a biometric image can be represented as [B, C, H, W], where B represents the batch number of images and can be 128, C represents the number of input channels and can be 1 or 3, H represents the image width and can be 112 pixels, and W represents the image height and can be 112 pixels.

[0065] Step S120: Input the biometric image into the pre-constructed global branch network and output the global branch feature map.

[0066] Specifically, the global branch feature map can be four-dimensional tensor feature data, which may include the number of global branch image batches, the number of global branch feature output channels, the width of the global branch feature map, and the height of the global branch feature map.

[0067] Among them, the number of global branch feature images is equal to the number of image batches, the number of global branch feature output channels is greater than the number of input channels, the width of the global branch feature map is less than the width of the image, and the height of the global branch feature map is less than the height of the image.

[0068] For example, when the biometric image input to the global branch network is [128, 3, 112, 112], the output global branch feature map can be [128, 64, 7, 7], indicating that the number of branch feature images in the global branch feature map is 128, the number of branch feature output channels in the global branch feature map is 64, and the width and height of the global branch feature map are both 7.

[0069] Step S130: Process the biometric image through the feature map module of the backbone network to obtain the local branch feature map.

[0070] Specifically, the local branch feature map can include the number of local branch feature image batches, the number of local branch feature output channels, the width of the local branch feature map, and the height of the local branch feature map. The biometric recognition backbone network can be a MobileFaceNet backbone network or a ResNet backbone network.

[0071] In this case, the number of local branch feature image batches is the same as the number of image batches, and the number of local branch feature output channels is greater than the number of input channels.

[0072] Furthermore, the number of output channels for local branch features can be greater than the number of output channels for global branch features.

[0073] For example, when the biometric image input to the feature map module of the biometric recognition backbone network is [128, 3, 112, 112], the local branch feature map can be [128, 96, 7, 7], indicating that the batch size of the local branch feature map is 128, the number of local branch feature output channels is 96, and the width and height of the local branch feature map are both 7.

[0074] Step S140: When the width of the local branch feature map is the same as the width of the global branch feature map, and the height of the local branch feature map is the same as the height of the global branch feature map, the global branch feature map and the local branch feature map are fused to obtain a fused feature map.

[0075] Specifically, the fused feature map can include the number of batches of fused feature images, the number of fused feature output channels, the width of the fused feature map, and the height of the fused feature map.

[0076] Among them, the number of batches of fused feature images is the same as the number of batches of images, the number of fused feature output channels is the sum of the number of global branch feature output channels and the number of local branch feature output channels, the width of the fused feature map is equal to the width of the local branch feature map, and the height of the fused feature map is equal to the height of the local branch feature map.

[0077] For example, when the four-dimensional tensor data of the local branch feature map is [128, 96, 7, 7] and the four-dimensional tensor data of the global branch feature map is [128, 64, 7, 7], the fused feature map can be obtained as [128, 160, 7, 7]. This means that the batch size of the fused feature map is 128, the number of fused feature output channels is 160, and the width and height of the fused feature map are both 7.

[0078] This embodiment provides a method for globally recognizing biometric images. It acquires biometric images in a four-dimensional tensor format, where each image includes the number of image batches, the number of input channels, the image width, and the image height. The biometric images are then input into a pre-constructed global branch network, outputting a global branch feature map. This global branch feature map includes the number of global branch image batches, the number of global branch feature output channels, the global branch feature map width, and the global branch feature map height. Specifically, the number of global branch feature images equals the number of image batches, the number of global branch feature output channels is greater than the number of input channels, the global branch feature map width is less than the image width, and the global branch feature map height is less than the image height. The biometric images are then processed by a feature map module in the backbone network to obtain local branch feature maps, which include the number of local branch feature image batches and the number of local branch feature output channels. The system comprises a number of local branch feature maps, a width of local branch feature maps, and a height of local branch feature maps. The number of local branch feature map batches is the same as the number of image batches. The number of local branch feature output channels is greater than the number of input channels. When the width of the local branch feature map is the same as the width of the global branch feature map, and the height of the local branch feature map is the same as the height of the global branch feature map, the global branch feature map and the local branch feature map are fused to obtain a fused feature map. The fused feature map includes the number of fused feature map batches, the number of fused feature output channels, the width of the fused feature map, and the height of the fused feature map. The number of fused feature map batches is the same as the number of image batches. The number of fused feature output channels is the sum of the number of global branch feature output channels and the number of local branch feature output channels. The width of the fused feature map is equal to the width of the local branch feature map, and the height of the fused feature map is equal to the height of the local branch feature map. This demonstrates that while the backbone network performs high-precision analysis of local features in biometric images, the global branch network extracts global features from the biometric images, thus integrating global and local features and enhancing the biometric recognition capability.

[0079] In some embodiments of this application, the process of step S120 above, which involves inputting the biometric image into a pre-constructed global branch network and outputting a global branch feature map, is described. This process may include:

[0080] S1. Input the biometric image into the embedding layer of the pre-built global branch network and output the intermediate feature map.

[0081] Specifically, the intermediate feature map can be four-dimensional tensor data, which may include the number of intermediate image batches, the number of intermediate feature output channels, the width of the intermediate feature map, and the height of the intermediate feature map.

[0082] Among them, the number of intermediate image batches is the same as the number of image batches, the number of intermediate feature output channels is greater than the number of input channels, the width of the intermediate feature map is smaller than the width of the image, and the height of the intermediate feature map is smaller than the height of the image.

[0083] For example, when the biometric image input to the feature map module of the biometric recognition backbone network is [128, 3, 112, 112], the intermediate feature map can be [128, 64, 28, 28], indicating that the number of intermediate image batches of the intermediate feature map is 128, the number of intermediate feature output channels is 64, and the width and height of the intermediate feature map are both 28.

[0084] S2. The intermediate feature map is converted into a global branch feature map through the global branch network generation module.

[0085] The global branch feature map can be a four-dimensional tensor data. The height of the global branch feature map is smaller than the height of the intermediate feature map, and the width of the global branch feature map is smaller than the width of the intermediate feature map. The generation module of the global branch network can include a multi-head attention module and a multilayer perceptron (MLP) module.

[0086] Understandably, the module structures of the multi-head attention module and the MLP module are similar to those of the corresponding modules in the ViT network. The difference between the generator module and the corresponding module in the ViT network is that the generator module ultimately outputs a four-dimensional tensor, while the module corresponding to the generator module in the ViT network ultimately outputs a three-dimensional tensor.

[0087] Specifically, the process of converting intermediate feature maps into global branch feature maps through the global branch network generation module can include:

[0088] S21. The intermediate feature map is deformed by the generation module of the global branch network to obtain a three-dimensional tensor.

[0089] S22. Compress the three-dimensional tensor to obtain a compressed three-dimensional tensor.

[0090] S23. Convert the compressed 3D tensor into a global branch feature map.

[0091] For example, when the intermediate feature map input to the generator module of the global branch network is [128, 64, 28, 28], the intermediate feature map becomes a three-dimensional tensor [128, 28*28, 64] after shape transformation. The height and width are multiplied together to form a dimension. This dimension is swapped with the dimension of the number of channels. Then, the three-dimensional tensor is compressed to obtain a compressed three-dimensional tensor [128, 7*7, 64] with a smaller size than the original three-dimensional tensor. This compressed three-dimensional tensor can be converted into the global branch feature map [128, 64, 7, 7].

[0092] In some embodiments of this application, the embedding layer mentioned in the above embodiments is described. The embedding layer may include a first convolutional layer, a second convolutional layer, a BatchNorm network layer, and an activation function layer.

[0093] In this design, the number of input channels in the first convolutional layer can be a preset number, and the number of output channels in the first convolutional layer can be half of the preset number. The number of input channels in the second convolutional layer can also be half of the preset number, and the number of output channels in the second convolutional layer can be a preset number. The activation function layer can use either the GELU function or the ReLU function.

[0094] Specifically, the preset number of channels can be customized.

[0095] Based on this, the process mentioned in the above embodiments, which involves inputting biometric images into the embedding layer of a pre-constructed global branch network and outputting intermediate feature maps, will be described, such as... Figure 2 As shown, the process may include:

[0096] Step S210: Input the biometric image into the first convolutional layer to obtain the first intermediate result.

[0097] Step S220: Input the first intermediate result into the BatchNorm network layer to obtain the second intermediate result.

[0098] Step S230: Input the second intermediate result into the activation function layer to obtain the third intermediate result.

[0099] Step S240: Input the third intermediate result into the second convolutional layer to obtain the fourth intermediate result.

[0100] Step S250: Input the fourth intermediate result into the BatchNorm network layer to obtain the intermediate feature map.

[0101] It is understandable that the first convolutional layer, the second convolutional layer, the BatchNorm network layer, and the activation function layer can form a convolutional group to integrate information and reduce the dimensionality of biometric images. By passing the biometric image through the first convolutional layer, the BatchNorm network layer, the activation function layer, the second convolutional layer, and then back to the BatchNorm network layer for processing, the original space of the biometric image can be mapped to a new multidimensional space.

[0102] In some embodiments of this application, when the width of the local branch feature map is the same as the width of the global branch feature map, and the height of the local branch feature map is the same as the height of the global branch feature map, as mentioned in the above embodiments, after fusing the global branch feature map and the local branch feature map to obtain the fused feature map, the process may further include extracting biometric features. This process may include:

[0103] The fused feature map is input into the feature recognition module of the backbone network, and the biometric features of the biometric image are output.

[0104] The number of feature dimensions for biometric features is the number of channels obtained by convolving the number of output channels of global branch features with the number of output channels of local branch features.

[0105] Specifically, the fused feature map is input into the feature recognition module of the biometric recognition backbone network. The feature recognition module can output a two-dimensional tensor, which can include image batch information and feature dimension information.

[0106] For example, when the fused feature map is [128, 64+96, 7, 7], the fused feature map is input into the feature recognition module of the biometrics backbone network. The feature recognition module can output a two-dimensional tensor of [128, X], which means that X feature dimensions are extracted from 128 biometric images. X can be equal to 64+96, less than 64+96, or greater than 64+96.

[0107] It is understandable that since the number of feature dimensions is obtained by convolving the number of global branch feature output channels with the number of local branch feature output channels, the feature dimensions contain information about both global and local feature dimensions. The number of global feature dimensions can correspond to the number of global feature output channels, and the number of local feature dimensions can correspond to the number of basic feature output channels. Therefore, the extracted biometric features can enhance globality while providing high-precision analysis of local features.

[0108] The apparatus for realizing global recognition of biometric images provided in the embodiments of this application will be described below. The apparatus for realizing global recognition of biometric images described below can be referred to in correspondence with the method for realizing global recognition of biometric images described above.

[0109] See Figure 3 , Figure 3 This is a schematic diagram of a device structure for globally recognizing biometric images, as disclosed in an embodiment of this application.

[0110] like Figure 3 As shown, the device may include:

[0111] Tensor data acquisition unit 11 is used to acquire biometric images in four-dimensional tensor format, wherein the biometric images include image batch number, input channel number, image width and image height;

[0112] The branch network output unit 12 is used to input the biometric image into a pre-constructed global branch network and output a global branch feature map. The global branch feature map includes the number of global branch image batches, the number of global branch feature output channels, the width of the global branch feature map, and the height of the global branch feature map. The number of global branch feature images is equal to the number of image batches, the number of global branch feature output channels is greater than the number of input channels, the width of the global branch feature map is less than the width of the image, and the height of the global branch feature map is less than the height of the image.

[0113] The backbone network output unit 13 is used to process the biometric image through the feature map module of the backbone network to obtain a local branch feature map. The local branch feature map includes the number of local branch feature image batches, the number of local branch feature output channels, the width of the local branch feature map, and the height of the local branch feature map. The number of local branch feature image batches is the same as the number of image batches, and the number of local branch feature output channels is greater than the number of input channels.

[0114] The global data determination unit 14 is used to fuse the global branch feature map and the local branch feature map when the width of the local branch feature map is the same as the width of the global branch feature map and the height of the local branch feature map is the same as the height of the global branch feature map, to obtain a fused feature map. The fused feature map includes the number of fused feature image batches, the number of fused feature output channels, the width of the fused feature map, and the height of the fused feature map. The number of fused feature image batches is the same as the number of image batches. The number of fused feature output channels is the sum of the number of global branch feature output channels and the number of local branch feature output channels. The width of the fused feature map is equal to the width of the local branch feature map, and the height of the fused feature map is equal to the height of the local branch feature map.

[0115] Optionally, the branch network output unit includes:

[0116] An embedding layer output unit is used to input the biometric image into the embedding layer of a pre-constructed global branch network and output an intermediate feature map. The intermediate feature map includes the number of intermediate image batches, the number of intermediate feature output channels, the width of the intermediate feature map, and the height of the intermediate feature map. The number of intermediate image batches is the same as the number of image batches, the number of intermediate feature output channels is greater than the number of input channels, the width of the intermediate feature map is smaller than the width of the image, and the height of the intermediate feature map is smaller than the height of the image.

[0117] The generation module output unit is used to convert the intermediate feature map into a global branch feature map through the generation module of the global branch network. The height of the global branch feature map is smaller than the height of the intermediate feature map, and the width of the global branch feature map is smaller than the width of the intermediate feature map.

[0118] Optionally, the embedding layer includes a first convolutional layer, a second convolutional layer, a BatchNorm network layer, and an activation function layer. The number of input channels of the first convolutional layer is a preset number of channels, the number of output channels of the first convolutional layer is half of the preset number of channels, the number of input channels of the second convolutional layer is half of the preset number of channels, and the number of output channels of the second convolutional layer is the preset number of channels.

[0119] The embedded layer output unit includes:

[0120] The first embedding layer output subunit is used to input the biometric image into the first convolutional layer to obtain a first intermediate result;

[0121] The second embedding layer output subunit is used to input the first intermediate result into the BatchNorm network layer to obtain the second intermediate result;

[0122] The third embedding layer output subunit is used to input the second intermediate result into the activation function layer to obtain the third intermediate result;

[0123] The fourth embedding layer output subunit is used to input the third intermediate result into the second convolutional layer to obtain the fourth intermediate result;

[0124] The fifth embedding layer output subunit is used to input the fourth intermediate result into the BatchNorm network layer to obtain an intermediate feature map.

[0125] Optionally, the output unit of the generation module includes:

[0126] The first generation module output subunit is used to deform the intermediate feature map through the generation module of the global branch network to obtain a three-dimensional tensor.

[0127] The second generation module output subunit is used to compress the three-dimensional tensor to obtain a compressed three-dimensional tensor.

[0128] The third generation module output subunit is used to convert the compressed 3D tensor into a global branch feature map.

[0129] Optionally, the device may also include:

[0130] The feature dimension extraction unit is used to input the fused feature map into the feature recognition module of the backbone network when the width of the local branch feature map is the same as the width of the global branch feature map and the height of the local branch feature map is the same as the height of the global branch feature map, and output the biometric features of the biometric image. The number of feature dimensions of the biometric features is the number of channels after convolving the number of output channels of the global branch features with the number of output channels of the local branch features.

[0131] The device for globally recognizing biometric images provided in this application embodiment can be applied to devices for globally recognizing biometric images, such as terminals: mobile phones, computers, etc. Optionally, Figure 4 A hardware block diagram of a device for globally recognizing biometric images is shown, with reference to... Figure 4 The hardware structure for globally identifying biometric images may include: at least one processor 1, at least one communication interface 2, at least one memory 3, and at least one communication bus 4.

[0132] In this embodiment of the application, the number of processor 1, communication interface 2, memory 3, and communication bus 4 is at least one, and processor 1, communication interface 2, and memory 3 communicate with each other through communication bus 4;

[0133] Processor 1 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0134] Memory 3 may include high-speed RAM, and may also include non-volatile memory, such as at least one disk storage device;

[0135] The memory stores a program, which the processor can call. The program is used for:

[0136] Acquire biometric images in a four-dimensional tensor format, wherein the biometric images include the number of image batches, the number of input channels, the image width, and the image height;

[0137] The biometric image is input into a pre-constructed global branch network, and a global branch feature map is output. The global branch feature map includes the number of global branch image batches, the number of global branch feature output channels, the width of the global branch feature map, and the height of the global branch feature map. The number of global branch feature images is equal to the number of image batches, the number of global branch feature output channels is greater than the number of input channels, the width of the global branch feature map is less than the width of the image, and the height of the global branch feature map is less than the height of the image.

[0138] The biometric image is processed by the feature map module of the backbone network to obtain a local branch feature map. The local branch feature map includes the number of local branch feature image batches, the number of local branch feature output channels, the width of the local branch feature map, and the height of the local branch feature map. The number of local branch feature image batches is the same as the number of image batches, and the number of local branch feature output channels is greater than the number of input channels.

[0139] When the width of the local branch feature map is the same as the width of the global branch feature map, and the height of the local branch feature map is the same as the height of the global branch feature map, the global branch feature map and the local branch feature map are fused to obtain a fused feature map. The fused feature map includes the number of fused feature image batches, the number of fused feature output channels, the width of the fused feature map, and the height of the fused feature map. The number of fused feature image batches is the same as the number of image batches. The number of fused feature output channels is the sum of the number of global branch feature output channels and the number of local branch feature output channels. The width of the fused feature map is equal to the width of the local branch feature map, and the height of the fused feature map is equal to the height of the local branch feature map.

[0140] Optionally, the refined and extended functions of the program can be found in the description above.

[0141] This application embodiment also provides a storage medium that can store a program suitable for execution by a processor, the program being used for:

[0142] Acquire biometric images in a four-dimensional tensor format, wherein the biometric images include the number of image batches, the number of input channels, the image width, and the image height;

[0143] The biometric image is input into a pre-constructed global branch network, and a global branch feature map is output. The global branch feature map includes the number of global branch image batches, the number of global branch feature output channels, the width of the global branch feature map, and the height of the global branch feature map. The number of global branch feature images is equal to the number of image batches, the number of global branch feature output channels is greater than the number of input channels, the width of the global branch feature map is less than the width of the image, and the height of the global branch feature map is less than the height of the image.

[0144] The biometric image is processed by the feature map module of the backbone network to obtain a local branch feature map. The local branch feature map includes the number of local branch feature image batches, the number of local branch feature output channels, the width of the local branch feature map, and the height of the local branch feature map. The number of local branch feature image batches is the same as the number of image batches, and the number of local branch feature output channels is greater than the number of input channels.

[0145] When the width of the local branch feature map is the same as the width of the global branch feature map, and the height of the local branch feature map is the same as the height of the global branch feature map, the global branch feature map and the local branch feature map are fused to obtain a fused feature map. The fused feature map includes the number of fused feature image batches, the number of fused feature output channels, the width of the fused feature map, and the height of the fused feature map. The number of fused feature image batches is the same as the number of image batches. The number of fused feature output channels is the sum of the number of global branch feature output channels and the number of local branch feature output channels. The width of the fused feature map is equal to the width of the local branch feature map, and the height of the fused feature map is equal to the height of the local branch feature map.

[0146] Optionally, the refined and extended functions of the program can be found in the description above.

[0147] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0148] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0149] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for globally recognizing biometric images, characterized in that, include: Acquire biometric images in a four-dimensional tensor format, wherein the biometric images include the number of image batches, the number of input channels, the image width, and the image height; The biometric image is input into a pre-constructed global branch network, and a global branch feature map is output. The global branch feature map includes the number of global branch image batches, the number of global branch feature output channels, the width of the global branch feature map, and the height of the global branch feature map. The number of global branch feature images is equal to the number of image batches, the number of global branch feature output channels is greater than the number of input channels, the width of the global branch feature map is less than the width of the image, and the height of the global branch feature map is less than the height of the image. The biometric image is processed by the feature map module of the backbone network to obtain a local branch feature map. The local branch feature map includes the number of local branch feature image batches, the number of local branch feature output channels, the width of the local branch feature map, and the height of the local branch feature map. The number of local branch feature image batches is the same as the number of image batches, and the number of local branch feature output channels is greater than the number of input channels. When the width of the local branch feature map is the same as the width of the global branch feature map, and the height of the local branch feature map is the same as the height of the global branch feature map, the global branch feature map and the local branch feature map are fused to obtain a fused feature map. The fused feature map includes the number of fused feature image batches, the number of fused feature output channels, the width of the fused feature map, and the height of the fused feature map. The number of fused feature image batches is the same as the number of image batches. The number of fused feature output channels is the sum of the number of global branch feature output channels and the number of local branch feature output channels. The width of the fused feature map is equal to the width of the local branch feature map, and the height of the fused feature map is equal to the height of the local branch feature map. The biometric image is input into a pre-constructed global branch network, and the output is a global branch feature map, including: The biometric image is input into the embedding layer of a pre-constructed global branch network, and an intermediate feature map is output. The intermediate feature map includes the number of intermediate image batches, the number of intermediate feature output channels, the width of the intermediate feature map, and the height of the intermediate feature map. The number of intermediate image batches is the same as the number of image batches, the number of intermediate feature output channels is greater than the number of input channels, the width of the intermediate feature map is less than the width of the image, and the height of the intermediate feature map is less than the height of the image. The intermediate feature map is converted into a global branch feature map through the global branch network generation module. The height of the global branch feature map is smaller than the height of the intermediate feature map, and the width of the global branch feature map is smaller than the width of the intermediate feature map. The intermediate feature map is converted into a global branch feature map through the generation module of the global branch network, including: The intermediate feature map is deformed by the generation module of the global branch network to obtain a three-dimensional tensor; The three-dimensional tensor is compressed to obtain a compressed three-dimensional tensor; The compressed 3D tensor is converted into a global branch feature map.

2. The method according to claim 1, characterized in that, The embedding layer includes a first convolutional layer, a second convolutional layer, a BatchNorm network layer, and an activation function layer. The number of input channels of the first convolutional layer is a preset number of channels, and the number of output channels of the first convolutional layer is half of the preset number of channels. The number of input channels of the second convolutional layer is half of the preset number of channels, and the number of output channels of the second convolutional layer is the preset number of channels. The biometric image is input into the embedding layer of a pre-constructed global branch network, and intermediate feature maps are output, including: The biometric image is input into the first convolutional layer to obtain the first intermediate result; The first intermediate result is input into the BatchNorm network layer to obtain the second intermediate result; The second intermediate result is input into the activation function layer to obtain the third intermediate result; The third intermediate result is input into the second convolutional layer to obtain the fourth intermediate result; The fourth intermediate result is input into the BatchNorm network layer to obtain an intermediate feature map.

3. The method according to claim 1 or 2, characterized in that, When the width of the local branch feature map is the same as the width of the global branch feature map, and the height of the local branch feature map is the same as the height of the global branch feature map, after fusing the global branch feature map and the local branch feature map to obtain a fused feature map, the method further includes: The fused feature map is input into the feature recognition module of the backbone network, and the biometric features of the biometric image are output. The number of feature dimensions of the biometric features is the number of channels after convolving the number of output channels of the global branch features with the number of output channels of the local branch features.

4. A device for globally recognizing biometric images, characterized in that, The apparatus, applied to the method for globally identifying biometric images as described in claim 1, comprises: Tensor data acquisition unit is used to acquire biofeature images in four-dimensional tensor format, wherein the biofeature images include image batch size, number of input channels, image width and image height; The branch network output unit is used to input the biometric image into a pre-constructed global branch network and output a global branch feature map. The global branch feature map includes the number of global branch image batches, the number of global branch feature output channels, the width of the global branch feature map, and the height of the global branch feature map. The number of global branch feature images is equal to the number of image batches, the number of global branch feature output channels is greater than the number of input channels, the width of the global branch feature map is less than the width of the image, and the height of the global branch feature map is less than the height of the image. The backbone network output unit is used to process the biometric image through the feature map module of the backbone network to obtain a local branch feature map. The local branch feature map includes the number of local branch feature image batches, the number of local branch feature output channels, the width of the local branch feature map, and the height of the local branch feature map. The number of local branch feature image batches is the same as the number of image batches, and the number of local branch feature output channels is greater than the number of input channels. A global data determination unit is configured to fuse the global branch feature map and the local branch feature map when the width of the local branch feature map is the same as the width of the global branch feature map and the height of the local branch feature map is the same as the height of the global branch feature map, thereby obtaining a fused feature map. The fused feature map includes the number of fused feature image batches, the number of fused feature output channels, the width of the fused feature map, and the height of the fused feature map. The number of fused feature image batches is the same as the number of image batches. The number of fused feature output channels is the sum of the number of global branch feature output channels and the number of local branch feature output channels. The width of the fused feature map is equal to the width of the local branch feature map, and the height of the fused feature map is equal to the height of the local branch feature map.

5. The apparatus according to claim 4, characterized in that, The branch network output unit includes: An embedding layer output unit is used to input the biometric image into the embedding layer of a pre-constructed global branch network and output an intermediate feature map. The intermediate feature map includes the number of intermediate image batches, the number of intermediate feature output channels, the width of the intermediate feature map, and the height of the intermediate feature map. The number of intermediate image batches is the same as the number of image batches, the number of intermediate feature output channels is greater than the number of input channels, the width of the intermediate feature map is smaller than the width of the image, and the height of the intermediate feature map is smaller than the height of the image. The generation module output unit is used to convert the intermediate feature map into a global branch feature map through the generation module of the global branch network. The height of the global branch feature map is smaller than the height of the intermediate feature map, and the width of the global branch feature map is smaller than the width of the intermediate feature map.

6. The apparatus according to claim 5, characterized in that, The embedding layer includes a first convolutional layer, a second convolutional layer, a BatchNorm network layer, and an activation function layer. The number of input channels of the first convolutional layer is a preset number of channels, and the number of output channels of the first convolutional layer is half of the preset number of channels. The number of input channels of the second convolutional layer is half of the preset number of channels, and the number of output channels of the second convolutional layer is the preset number of channels. The embedded layer output unit includes: The first embedding layer output subunit is used to input the biometric image into the first convolutional layer to obtain a first intermediate result; The second embedding layer output subunit is used to input the first intermediate result into the BatchNorm network layer to obtain the second intermediate result; The third embedding layer output subunit is used to input the second intermediate result into the activation function layer to obtain the third intermediate result; The fourth embedding layer output subunit is used to input the third intermediate result into the second convolutional layer to obtain the fourth intermediate result; The fifth embedding layer output subunit is used to input the fourth intermediate result into the BatchNorm network layer to obtain an intermediate feature map.

7. A device for globally recognizing biometric images, characterized in that, Including memory and processor; The memory is used to store programs; The processor is configured to execute the program to implement the steps of the method for globally recognizing biometric images as described in any one of claims 1-3.

8. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method for global identification of biometric images as described in any one of claims 1-3.

Images