Research On Segmentation And Feature Extraction Method For Finger-vein Image

As a new biometric recognition technology,finger-vein recognition has a wide application prospect in the field of machine vision and pattern recognition.Compared with the traditional identification technology,it has many significant advantages,such as high security,convenient collection,friendly interaction,and so on.Finger-vein recognition mainly includes the following four key steps: image acquisition,region of interest(ROI)segmentation,image restoration and enhancement,feature extraction and matching.Finger-vein images are collected by taking advantage of the trait that hemoglobin in human blood can absorb near-infrared light.Therefore,the acquisition process is easily affected by the scattering effect of the tissue inside the finger,which seriously degrades the quality of acquired images.In addition,changes in finger posture could cause translation,rotation,and even deformation of acquired images.The above factors bring great difficulties and challenges to finger-vein recognition.Although some achievements have been made in research on finger-vein recognition,there are still some issues that need to be solved.This dissertation focuses on the key points of finger-vein recognition and carry out the research work.The specific research methods and contributions are summarized as follows:(1)A method for ROI segmentation of finger-vein image based on the fusion of finger contour and gradient distribution is proposed.To solve the problem of finger image rotation caused by the variation of finger posture,a correction method based on finger contour information is proposed.Based on the corrected finger image,a method for locating finger joint positions based on gradient distribution is proposed.The ROI image is finally segmented reliably by detecting joint positions of the finger.The experimental results show that the proposed method can reliably segment the effective finger-vein ROI image,and compared with existing methods,it can retain richer venous information.(2)An adaptive finger-vein image restoration method based on atmospheric scattering theory is proposed.To deal with the issues of low image quality and poor contrast caused by scattering of finger internal tissues,a restoration method that is suitable for finger-vein images is proposed by analyzing the correlation and difference between the atmospheric scattering phenomenon and the imaging process of finger-vein images.Since related parameters in the restoration method are difficult to be figured out directly,corresponding approximation estimation methods are proposed for the absorptivity ?(x)and atmospheric light A,respectively.The experimental results reveal that the proposed method is able to improve the quality of finger-vein images effectively.Compared with existing methods,the restored finger-vein image has higher contrast and is more beneficial for improving the performance of the finger-vein recognition system.(3)A method for segmenting finger-vein network features based on the active contour is proposed.To solve the problem that vein networks are probable discontinuous or over-segmented during the extraction of finger-vein network features,the active contour-based method is used for the first time to extract finger-vein network features.By analyzing the characteristic of finger-vein images,an active contour model for the segmentation of finger-vein networks is established by integrating an effective edge fitting item.To deal with the problem that the processing result of active contour-based methods is susceptible to the initial contour,an initial contour localization method based on the fuzzy kernel clustering algorithm(KFCM)is proposed to improve the stability of its processing results.The experimental results show that the proposed method can reliably extract effective vein network features.Compared with existing methods,the segmented vein networks are more continuous and smooth,which is more beneficial to improve the accuracy of the finger-vein recognition system.(4)A method for extracting semantic features of finger-vein images based on generative adversarial network(GAN)is proposed.To solve the issue of the unstable recognition results of finger-vein semantic feature extraction methods on different datasets,a generative network FCGAN(fully convolutional GAN)is established,which utilizes the fully convolutional structure and strictly constrained loss function to synthesize high-quality finger-vein samples and augment the dataset for the first time.Based on samples augmented by FCGAN,a semantic feature extraction method FCGAN-CNN is proposed to obtain highly distinguished and reliable finger-vein features.The experimental results show that FCGAN can generate more diverse and high-quality finger-vein samples.Compared with existing methods,FCGAN-CNN can extract finger-vein semantic features with higher discrimination and more robustness.