CN-116311402-B - Palm vein living body identification method, palm vein living body identification device and readable storage medium

CN116311402BCN 116311402 BCN116311402 BCN 116311402BCN-116311402-B

Abstract

The application discloses a palm vein living body identification method which comprises the steps of extracting a palm vein effective area of a palm vein image, carrying out azimuth rotation correction on the palm vein effective area, carrying out palm vein texture feature extraction on the palm vein effective area subjected to azimuth rotation correction by utilizing a first feature extraction layer of a main network to obtain a texture feature image f 1 , carrying out texture enhancement processing on the texture feature image f 1 to obtain a texture feature image T, carrying out feature extraction on the texture feature image f 1 by utilizing a second feature extraction layer of the main network to obtain a feature image f 2 , inputting the feature image f 2 into an attention module of a palm vein texture local area to obtain an attention feature image D, carrying out average pooling on the obtained feature image f 2 and the attention feature image D to obtain a texture feature matrix P, and inputting the texture feature matrix P into a full-connection layer of the main network to carry out palm vein living body true and false judgment. The application also provides a storage medium and a recognition device applying the palm vein living body recognition method.

Inventors

CHEN HAITAO
OUYANG YICUN
LI XI
FU LEI
LI SHANLU
LI XIAOKAI
Mo jiayuan
LI QIU

Assignees

盛视科技股份有限公司

Dates

Publication Date: 20260505
Application Date: 20230223

Claims (6)

1. A method for identifying a palm vein living body, comprising: Extracting palm vein effective area of palm vein image, carrying out azimuth rotation correction on the palm vein effective area to make point between index finger root and middle finger root and point between middle finger root and ring finger root on same horizontal line parallel to X axis, positioning four finger root points in the palm vein effective area, wherein the four finger root points comprise point P 1 between thumb root and index finger root, point P 2 between index finger root and middle finger root, point P 3 between middle finger root and ring finger root and point P 4 between ring finger root and little finger root, judging whether the palm vein effective area belongs to left or right hand according to the four finger root points, adjusting the palm vein effective area to upward direction, carrying out azimuth rotation correction on the palm vein effective area according to structural characteristics of left and right hands; The step of judging whether the detected palm vein effective area belongs to left and right or right hands according to four finger root points comprises the steps of establishing a two-dimensional rectangular coordinate system by taking a key point P 1 as an origin, calculating deflection angles alpha n1 of connecting lines of P 2 、P 3 、P 4 and P 1 relative to the X-axis direction, wherein n is E (2, 3, 4), judging whether P 2 、P 3 and P 4 are positioned in the same quadrant of the two-dimensional rectangular coordinate system by taking the key point P 1 as the origin, judging whether the palm vein effective area belongs to left and right or right hands according to deflection angles alpha 21 、α 31 and alpha 41 corresponding to the P 2 、P 3 and P 4 , selecting two key points P i and P j positioned in the same quadrant, and judging whether the palm vein effective area belongs to left and right or right hands according to deflection angles alpha i1 and alpha j1 corresponding to the P i and P j , wherein i is E (2, 3, 4), j is E (2, 3, 4) and i < j; The palm vein effective area is adjusted to be upward in the finger direction, and the palm vein effective area is corrected in an azimuth rotation mode according to the structural characteristics of the left hand and the right hand, wherein key points P 2 and P 4 are selected, an azimuth rotation angle beta=arctan ((y 4 '-y 2 ')/ (x 4 '-x 2 ') is calculated for the palm veins belonging to the right hand, an azimuth rotation angle beta=arctan ((y 2 '-y 4 ')/ (x 2 '-x 4 ') is calculated for the palm veins belonging to the left hand), if the azimuth rotation angle beta is larger than 0, the palm vein effective area is rotated anticlockwise around the central point of an image by an angle beta, and if the azimuth rotation angle beta is smaller than 0, the palm vein effective area is rotated clockwise around the central point of the image by an angle beta, wherein x 2 'and y 2 ' are respectively the abscissa and the ordinate of a 2 nd key point P 2 , and x 4 'and y 4 ' are respectively the abscissa and the ordinate of a 4 th key point P 4 ; extracting palm vein texture features of the palm vein effective area subjected to rotation correction by using a first feature extraction layer of a main network to obtain a texture feature map f 1 ; performing texture enhancement processing on the texture feature map f 1 to obtain a texture feature map T; Performing feature extraction on the texture feature map f 1 by using a second feature extraction layer of the backbone network to obtain a feature map f 2 ; Inputting the feature map f 2 into an attention module of the palm vein texture local area to obtain an attention feature map D; carrying out average pooling on the obtained texture feature map T and the attention feature map D to obtain a texture feature matrix P, and inputting the texture feature matrix P into a full-connection layer of a main network to carry out palm vein living body true and false judgment; The first feature extraction layer and the second feature extraction layer are all a plurality of convolution layers and are distributed according to the front-back sequence of the backbone network.
2. The palm vein living body recognition method according to claim 1, wherein the step of performing texture enhancement processing on the texture feature map f 1 to obtain a texture feature map T includes: Downsampling the texture feature map f 1 to obtain an average pooled feature map t; Upsampling the feature map t to the same dimension as the texture feature map f 1 ; Aggregating the texture feature map f 1 and the feature map t after expansion processing; and (3) carrying out dense convolution on the aggregation result to obtain a texture feature map T after texture enhancement.
3. The palm vein vital identification method according to claim 1, wherein the palm vein vital identification method further comprises: Extracting features of the feature map f 2 by using a third feature extraction layer of the backbone network to obtain a deep feature map f 3 ; Carrying out average pooling on the deep feature map f 3 to obtain a global feature map G; the texture feature matrix P and the global feature graph G are input into a full-connection layer of a main network together to judge the palm vein living body true and false; The third feature extraction layer is a plurality of convolution layers, and the second feature extraction layer and the third feature extraction layer are all a plurality of convolution layers and distributed according to the front-back sequence of the backbone network.
4. The method of claim 3, wherein the backbone network is resnet network, the first feature extraction layer is a convolution layer of the first through seventh layers of the resnet network, the second feature extraction layer is a convolution layer of the eighth through fifteenth layers of the resnet network, and the third feature extraction layer is a convolution layer of the sixteenth through forty-ninth layers of the resnet network.
5. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a program code which, when run on a computer, causes the computer to execute the metacarpal vein living body recognition method as claimed in any one of claims 1 to 4.
6. A palm vein living body recognition apparatus, characterized by comprising a processor and a memory connected to the processor, the memory for storing executable code and the processor for executing the executable code of the memory, the palm vein living body recognition method according to any one of claims 1 to 4 being implemented when the processor tries to execute the code.

Description

Palm vein living body identification method, palm vein living body identification device and readable storage medium Technical Field The present application relates to the field of image processing, and more particularly, to a palm vein living body recognition method, a palm vein living body recognition device, and a readable storage medium. Background The two common identity recognition technologies of fingerprint recognition and face recognition are mature, and are widely applied to the aspects of identity recognition, security check, entrance guard, medical research and the like. Palm vein is used as a biological feature with higher safety, has been studied for decades, and the palm vein identification has the characteristics of non-contact and the like, and has a very good application prospect in the aspect of the identity identification of personnel in the current public place. In order to ensure the safety and accuracy of palm vein identification, the palm vein identification process needs to resist spoofing attack of non-living palm veins, for example, the palm vein image printed is used for identity verification, so that whether the palm veins are living or not needs to be identified in the palm vein identification process. Disclosure of Invention Aiming at the prior art, the technical problem solved by the application is to provide a palm vein living body identification method, an identification device and a computer readable storage medium capable of identifying the palm vein of a true or false living body. To solve the above technical problem, in a first aspect, the present application provides a method for identifying a palm vein living body, including: Extracting a palm vein effective area of a palm vein image, and carrying out azimuth rotation correction on the palm vein effective area to enable points between the index finger root and the middle finger root and points between the middle finger root and the ring finger root to be on the same horizontal line parallel to an X axis; extracting palm vein texture features of the palm vein effective area subjected to rotation correction by using a first feature extraction layer of a main network to obtain a texture feature map f 1; performing texture enhancement processing on the texture feature map f 1 to obtain a texture feature map T; Performing feature extraction on the texture feature map f 1 by using a second feature extraction layer of the backbone network to obtain a feature map f 2; Inputting the feature map f 2 into an attention module of the palm vein texture local region of interest to obtain an attention feature map D, and Carrying out average pooling on the obtained feature map f 2 and the attention feature map D to obtain a texture feature matrix P, and inputting the texture feature matrix P into a full-connection layer of a main network to carry out palm vein living body true and false judgment; The first feature extraction layer and the second feature extraction layer are all a plurality of convolution layers and are distributed according to the front-back sequence of the backbone network. In one possible implementation, the step of performing texture enhancement processing on the texture map f 1 to obtain the texture map T includes: Downsampling the texture feature map f 1 to obtain an average pooled feature map t; Upsampling the feature map t to the same dimension as the texture feature map f 1; Aggregating the texture feature map f 1 and the feature map t after expansion processing; and (3) carrying out dense convolution on the aggregation result to obtain a texture feature map T after texture enhancement. In one possible implementation manner, the palm vein living body identification method further includes: Extracting features of the feature map f 2 by using a third feature extraction layer of the backbone network to obtain a deep feature map f 3; Carrying out average pooling on the deep feature map f 3 to obtain a global feature map G; the texture feature matrix P and the global feature graph G are input into a full-connection layer of a main network together to judge the palm vein living body true and false; The third feature extraction layer is a plurality of convolution layers, and the second feature extraction layer and the third feature extraction layer are all a plurality of convolution layers and distributed according to the front-back sequence of the backbone network. In one possible implementation, the backbone network is a resnet network, the first feature extraction layer is a convolution layer of the first through seventh layers of the resnet network, the second feature extraction layer is a convolution layer of the eighth through fifteenth layers of the resnet network, and the third feature extraction layer is a convolution layer of the sixteenth through forty-ninth layers of the resnet network. In one possible implementation, the step of performing azimuthal rotation correction on the palm vein effective area suc