Search

CN-116883687-B - Method, device, equipment and storage medium for globally identifying biological characteristic image

CN116883687BCN 116883687 BCN116883687 BCN 116883687BCN-116883687-B

Abstract

The application discloses a method, a device, equipment and a storage medium for globally identifying a biological characteristic image, wherein the method comprises the steps of acquiring the biological characteristic image with a data format of four-dimensional tensor, inputting the biological characteristic image into a pre-constructed global branch network, outputting a global branch characteristic image, processing the biological characteristic image through a characteristic image module of a main network to obtain a local branch characteristic image, and fusing the global branch characteristic image and the local branch characteristic image to obtain a fused characteristic image when the width of the local branch characteristic image is the same as the width of the global branch characteristic image and the height of the local branch characteristic image is the same as the height of the global branch characteristic image. Therefore, the local features of the biological feature image are analyzed with high precision by the backbone network, and the global features of the biological feature image are extracted by the global branch network, so that the biological recognition features are fused with the global features and the local features, and the biological recognition feature recognition capability is enhanced.

Inventors

  • YANG QI
  • CHEN SHUKAI

Assignees

  • 熵基科技股份有限公司

Dates

Publication Date
20260505
Application Date
20230725

Claims (8)

  1. 1. A method for globally identifying a biometric image, comprising: acquiring a biological characteristic image with a data format of four-dimensional tensor, wherein the biological characteristic image comprises the number of image batches, the number of input channels, the image width and the image height; inputting the biological characteristic images into a pre-constructed global branch network, and outputting a global branch characteristic image, wherein the global branch characteristic image comprises a global branch image batch number, a global branch characteristic output channel number, a global branch characteristic image width and a global branch characteristic image height, the global branch characteristic image number is equal to the image batch number, the global branch characteristic output channel number is larger than the input channel number, the global branch characteristic image width is smaller than the image width, and the global branch characteristic image height is smaller than the image height; Processing the biological characteristic images through a characteristic image module of a backbone network to obtain local branch characteristic images, wherein the local branch characteristic images comprise local branch characteristic image batch numbers, local branch characteristic output channel numbers, local branch characteristic image width and local branch characteristic image heights, the local branch characteristic image batch numbers are the same as the image batch numbers, and the local branch characteristic output channel numbers are larger than the input channel numbers; 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, wherein the fused feature map comprises the quantity of fused feature image batches, the quantity of fused feature output channels, the width of the fused feature map and the height of the fused feature map, the quantity of the fused feature image batches is the same as the quantity of the image batches, the quantity of the fused feature output channels is the sum of the quantity of the global branch feature output channels and the quantity of the local branch feature output channels, the width of the fused feature map is the same as the width of the local branch feature map, and the height of the fused feature map is the same as the height of the local branch feature map; inputting the biological characteristic image into a pre-constructed global branch network, and outputting a global branch characteristic image, wherein the method comprises the following steps of: Inputting the biological characteristic images into an embedded layer of a pre-built global branch network, and outputting an intermediate characteristic image, wherein the intermediate characteristic image comprises an intermediate image batch number, an intermediate characteristic output channel number, an intermediate characteristic image width and an intermediate characteristic image height, the intermediate image batch number is the same as the image batch number, the intermediate characteristic output channel number is larger than the input channel number, the intermediate characteristic image width is smaller than the image width, and the intermediate characteristic image height is smaller than the image height; Converting the intermediate feature map into a global branch feature map through a generation module of the global branch network, wherein the global branch feature height of the global branch feature map is smaller than that of the intermediate feature map, and the global branch feature width of the global branch feature map is smaller than that of the intermediate feature map; Converting, by the generation module of the global branch network, the intermediate feature map into a global branch feature map, including: the intermediate feature map is deformed through a generation module of the global branch network to obtain a three-dimensional tensor; compressing the three-dimensional tensor to obtain a compressed three-dimensional tensor; And converting the compressed three-dimensional tensor into a global branch characteristic diagram.
  2. 2. The method of claim 1, wherein the embedded layer comprises a first convolution layer, a second convolution layer, a BatchNorm network layer, and an activation function layer, the first convolution layer having a predetermined number of channels, the first convolution layer having a predetermined number of output channels that is half the predetermined number of channels, the second convolution layer having a predetermined number of input channels that is half the predetermined number of channels, the second convolution layer having a predetermined number of output channels; inputting the biological characteristic image into an embedded layer of a pre-built global branch network, and outputting an intermediate characteristic image, wherein the method comprises the following steps of: Inputting the biological characteristic image into the first convolution layer to obtain a first intermediate result; Inputting the first intermediate result to the BatchNorm network layer to obtain a second intermediate result; Inputting the second intermediate result to the activation function layer to obtain a third intermediate result; inputting the third intermediate result into the second convolution layer to obtain a fourth intermediate result; and inputting the fourth intermediate result to the BatchNorm network layer to obtain an intermediate feature map.
  3. 3. The method according to claim 1 or 2, wherein when the local branch feature map width is the same as the global branch feature map width and the local branch feature map height is the same as the global branch feature map height, fusing the global branch feature map and the local branch feature map to obtain a fused feature map, further comprising: Inputting the fusion feature map to a feature recognition module of the backbone network, and outputting the biological recognition features of the biological feature images, wherein the feature dimension number of the biological recognition features is the number of channels after the number of global branch feature output channels and the number of local branch feature output channels are convolved.
  4. 4. An apparatus for globally identifying a biometric image, applied to the method for globally identifying a biometric image as claimed in claim 1, the apparatus comprising: The tensor data acquisition unit is used for acquiring a biological characteristic image with a data format of four-dimensional tensor, wherein the biological characteristic image comprises the number of image batches, the number of input channels, the image width and the image height; The branch network output unit is used for inputting the biological characteristic images into a pre-constructed global branch network and outputting global branch characteristic images, wherein the global branch characteristic images comprise global branch image batch quantity, global branch characteristic output channel quantity, global branch characteristic image width and global branch characteristic image height, the global branch characteristic image quantity is equal to the image batch quantity, the global branch characteristic output channel quantity is larger than the input channel quantity, the global branch characteristic image width is smaller than the image width, and the global branch characteristic image height is smaller than the image height; the main network output unit is used for processing the biological characteristic images through a characteristic image module of a main network to obtain local branch characteristic images, wherein the local branch characteristic images comprise local branch characteristic image batch numbers, local branch characteristic output channel numbers, local branch characteristic image width and local branch characteristic image height, the local branch characteristic image batch numbers are the same as the image batch numbers, and the local branch characteristic output channel numbers are larger than the input channel numbers; And the global data determining unit is used for fusing the global branch feature map with the local branch feature map to obtain a fused feature map when the local branch feature map is the same as the global branch feature map in width and the local branch feature map is the same as the global branch feature map in height, wherein the fused feature map comprises a fused feature image batch number, a fused feature output channel number, a fused feature map width and a fused feature map height, the fused feature image batch number is the same as the image batch number, the fused feature output channel number is the sum of the global branch feature output channel number and the local branch feature output channel number, the fused feature map is the same as the local branch feature map in width, and the fused feature map is the same as the local branch feature map in height.
  5. 5. The apparatus of claim 4, wherein the branch network output unit comprises: The embedding layer output unit is used for inputting the biological characteristic images to an embedding layer of a pre-built global branch network and outputting an intermediate characteristic image, wherein the intermediate characteristic image comprises an intermediate image batch number, an intermediate characteristic output channel number, an intermediate characteristic image width and an intermediate characteristic image height, the intermediate image batch number is the same as the image batch number, the intermediate characteristic output channel number is larger than the input channel number, the intermediate characteristic image width is smaller than the image width, and the intermediate characteristic image height is smaller than the image height; The generation module output unit is used for converting the intermediate feature map into a global branch feature map through the generation module of the global branch network, the global branch feature height of the global branch feature map is smaller than that of the intermediate feature map, and the global branch feature width of the global branch feature map is smaller than that of the intermediate feature map.
  6. 6. The apparatus of claim 5, wherein the embedded layer comprises a first convolutional layer, a second convolutional layer, a BatchNorm network layer, and an activation function layer, the first convolutional layer having a number of input channels that is half of the number of preset channels, the first convolutional layer having a number of output channels that is half of the number of preset channels, the second convolutional layer having a number of input channels that is half of the number of preset channels, the second convolutional layer having a number of output channels that is the number of preset channels; the embedded layer output unit includes: The first embedding layer output subunit is used for inputting the biological characteristic image into the first convolution layer to obtain a first intermediate result; a second embedded layer output subunit, configured to input the first intermediate result to the BatchNorm network layer to obtain a second intermediate result; a third embedded layer output subunit, configured to input the second intermediate result to the activation function layer, to obtain a third intermediate result; A fourth embedded layer output subunit, configured to input the third intermediate result to the second convolution layer, to obtain a fourth intermediate result; And a fifth embedded layer output subunit, configured to input the fourth intermediate result to the BatchNorm network layer, to obtain an intermediate feature map.
  7. 7. A device for globally identifying a biometric image, comprising a memory and a processor; the memory is used for storing programs; the processor for executing the program to perform the steps of the method of globally identifying a biometric image as claimed in any one of claims 1 to 3.
  8. 8. A storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the method of globally identifying a biometric image of a biometric as claimed in any one of claims 1 to 3.

Description

Method, device, equipment and storage medium for globally identifying biological characteristic image Technical Field The application relates to the technical field of intelligent recognition, in particular to a method, a device, equipment and a storage medium for globally recognizing a biological characteristic image. Background With the continuous development of information technology, information mining is interesting for a plurality of researchers, and useful information is mined from massive information. For different objects, the corresponding unique features of the object, such as fingerprints, palmprints, faces and the like of the person, and researchers can determine a specific object by extracting feature information from a face image or complex information of the face image. For the recognition extraction method for extracting the features from the image, researchers can design an algorithm or train a model of the features, so that the algorithm and the model implement the feature extraction function on the image, and feature extraction is realized. Typically, the image needs to be processed prior to being input into the algorithm/model so that it conforms to the data input format of the algorithm/model, such as the format of four-dimensional tensor data. In the current biological feature recognition model, the local part of the biological feature image can be recognized with high precision, but the discrimination of the global feature is not strong. How to enhance global feature recognition of biometric images is a matter of concern. Disclosure of Invention In view of the above problems, the present application provides a method, apparatus, device and storage medium for globally identifying a biometric image, so as to enhance global feature identification of the biometric image, and combine high precision of local features to make a biometric identification model more characterization and characterization capable. In order to achieve the above object, the following specific solutions are proposed: a method of globally identifying a biometric image, comprising: acquiring a biological characteristic image with a data format of four-dimensional tensor, wherein the biological characteristic image comprises the number of image batches, the number of input channels, the image width and the image height; inputting the biological characteristic images into a pre-constructed global branch network, and outputting a global branch characteristic image, wherein the global branch characteristic image comprises a global branch image batch number, a global branch characteristic output channel number, a global branch characteristic image width and a global branch characteristic image height, the global branch characteristic image number is equal to the image batch number, the global branch characteristic output channel number is larger than the input channel number, the global branch characteristic image width is smaller than the image width, and the global branch characteristic image height is smaller than the image height; Processing the biological characteristic images through a characteristic image module of a backbone network to obtain local branch characteristic images, wherein the local branch characteristic images comprise local branch characteristic image batch numbers, local branch characteristic output channel numbers, local branch characteristic image width and local branch characteristic image heights, the local branch characteristic image batch numbers are the same as the image batch numbers, and the local branch characteristic output channel numbers are larger than the input channel numbers; 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 comprises the quantity of fused feature image batches, the quantity of fused feature output channels, the width of the fused feature map and the height of the fused feature map, wherein the quantity of fused feature image batches is the same as the quantity of image batches, the quantity of fused feature output channels is the sum of the quantity of global branch feature output channels and the quantity of local branch feature output channels, the width of the fused feature map is the same as the width of the local branch feature map, and the height of the fused feature map is the same as the height of the local branch feature map. Optionally, inputting the biometric image into a pre-constructed global branch network, and outputting a global branch feature map, including: Inputting the biological characteristic images into an embedded layer of a pre-built global branch network, and outputting an intermediate characteristic image, wherein the intermediate characteristic image co