Research On Feature Extraction Algorithm Of Low-quality Finger Vein Images

With the advent of the information era,people's demands for information security protection are increasing day by day.Finger vein recognition technology has become a necessary means of identification and authentication due to its advantages of non-contact,living body collection and high security level.However,finger vein recognition technology has not been widely marketed due to the limitation of finger vein image quality.The low-contrast,uneven gray,offset and fuzzy low-quality images caused by these factors that disadvantages of the acquisition device,different finger positions and individual physiological differences directly affect the difficulty of finger vein feature extraction and reduce the recognition performance.Therefore,this paper mainly studies the feature extraction algorithm of finger vein.The main research contents are as follows:(1)Finger vein image preprocessing.Perform edge detection and morphological operation and the HU moment on finger vein images.The image is subjected to translational rotation correction with normalized finger region of interest.(2)An eigenvalue ratio enhancement algorithm based on Hessian matrix is proposed.In this paper,the Frangi filter enhancement algorithm is applied to the image enhancement of finger veins,but the enhancement effect is poor.The main principle of Frangi filter is analyzed and improved,and the enhancement algorithm of eigenvalue ratio based on Hessian matrix is proposed.The proposed algorithm can not only enhance the detail of the low-quality finger vein image but the thick vein.Moreover,it overcomes the inhibition of Frangi filtering enhancement algorithm on venous bending and intersections,and eliminates the cavitation phenomenon in venous structure.(3)An image enhancement algorithm based on difference Gabor filtering is proposed.For low quality finger vein with low contrast and fuzzy image,it is easy to enhance the noise during image enhancement.Aiming at the shortcomings of insufficient feature extraction of low-quality finger vein images,wavelet fusion algorithm is used to fuse differential Gabor filtering,matched filtering enhancement and eigenvalue ratio enhancement algorithms.The advantages are complementary,the vein feature are effectively enhanced,and the anti-noise performance is better.(4)The maximum entropy threshold image segmentation algorithm based on geneticalgorithm is adopted.For threshold segmentation of low-quality finger vein images,simple threshold segmentation cannot effectively segment image feature information.The OTSU segmentation method and the maximum entropy threshold segmentation algorithm have the disadvantages of large computational complexity and time-consuming.The maximum entropy threshold image segmentation algorithm based on genetic algorithm is considered.Adaptive global optimization segmentation can shorten the running time,improve the anti-noise performance and meet the real-time requirements.The template matching method is used to verify the performance of the proposed algorithm.Under the authentication mode,the EER of the extraction algorithm based on the eigenvalue ratio enhancement was 0.71%,and the EER of the extraction algorithm based on the fusion enhancement was 0.19%.In the recognition mode,the recognition rate of the extraction algorithm based on the eigenvalue ratio enhancement is 98.91%,and the recognition rate of the extraction algorithm based on the fusion enhancement is99.13%.In conclusion,the recognition performance of this algorithm is improved compared with other finger vein extraction algorithms.