Research On Dorsal-hand Vein Recognition Under Weak Constraint Conditions

In the current era of increasing information interaction,identity recognition under weak constraints of different acquisition equipment,test environments,time periods,and postures of hand will inevitably become the development trend of biometric technology.However,under weak constraints,there may be five problems of differences in illumination contrast,position shift,angular rotation,scale scaling,and affine deformation in the collected dorsal-hand vein images,resulting in poor image recognition accuracy.Therefore,how to achieve high-robustness feature extraction and high-precision identity recognition under weak constraints is our research and forward direction.The main work and innovations of this paper are summarized as follows:1?Propose an improved threshold segmentation algorithm based on the ratio of the maximum connected domain area and connected domain quantity to adjust coefficient ? in binary vein images so as to solve the problems of vein rupture and poor connectivity.Based on the principle of intra-class pattern similarities,the optimal segmentation coefficient ? of the binary vein image was determined to avoid severe over-segmentation or under-segmentation.The improved segmented binary vein image database was verified using the Scale-invariant feature transform(SIFT)algorithm under cross-device conditions,and the recognition rate was increased from 73.48% to 77.4%.2?Propose a multi-directional detail component feature key point extraction method based on wavelet decomposition.With the combination of vertical & diagonal feature key points,the texture information of the binary vein image in the vertical direction,inclined and curved regions can be detected.Distinctive Efficient Robust Feature(DERF)was employed to describe the feature key points,and the results were more detailed and accurate.When the initial radius 1r of the sampling points in the algorithm was set to 4,the recognition result of the dorsal-hand vein using the DERF descriptor under cross-device conditions reached 93.4%.3?Propose a recognition method of fine-tuning the segmentation coefficient ? and coarse matching.Based on dorsal-hand contour and vein connection of the binary vein images collected by device A,the contour correction and segmentation threshold coefficient fine-tuning of the test binary vein images collected by device B were performed to further ensure the similarities of images in the same class under across-device conditions.Set coarse matching filtering criterion before feature extraction to reduce the number of candidate sets and the computational complexity in the matching stage.Under cross-device conditions,the experimental verification improved the recognition rate to 97.86%.The research results show that the dorsal-hand vein recognition rate under weak constraints has increased from 73.48% to 97.86%,which verifies the innovativeness of the thesis work.