Research Of Hand Dorsal Vein Recognition The Key Algorithms Based On Multi Features Fusion

In the field of public safety, the original identity authentication methods are exposing more and more defects, biometric recognition technologies with reliability and security appeared, and vein recognition is considered as the biometric recognition method with better confidentiality, convenience and reliability than other methods because it has many advantages such as living identification, internal feature, non-contact acquisition and specific acquisition environment. In recent years, domestic and foreign researchers had achieved some achievements in vein recognition field, but there are many key techniques needed to be solved.Having reviewed a large number of domestic and foreign literatures and the existing vein recognition algorithms are summarized, the key problems in vein recognition process are discussed and investigated deeply, and new ideas and algorithms are proposed. In this thesis the main innovative points are reflected in the following respects.Firstly, since the defects the acquired near infrared vein images have bad clarity and quality, a new near infrared vein image acquisition algorithm based on quality assessment is proposed, in the proposed algorithm the acquired vein image quality is assessed objectively in real time, the near infrared light intensity is adjusted according to assess the result, and the vein image with good quality is acquired.Secondly, because the dorsal skin thicknesses in different regions are different, a vein information segmentation algorithm based on function optimization in the regions of interest, in the proposed algorithm the regions with vein information are determined, and convex function optimization is applied in every interest regions to separate vein information, then segmentation results of all the interest regions are fused and the integral vein information is acquired. Thirdly, according to the analysis to vein image from different perspectives, four vein recognition algorithms based on different vein features are proposed which are shown as follows.1) According to the characteristic of shape invariants among the vein curves, a vein recognition algorithm based on regional shape is proposed, and the regional edges are approached by a lot of line segments, the angles sequence of these line segments are regarded as the regional features, and recognize one’s identity through fusing all regional features in the vein images;2) Due to the limitations of wavelet describing curves information, a vein recognition algorithm based on ridgelet transform after dividing image into blocks is proposed, in the proposed algorithm the directional features of all local vein curves are fused to recognize one’s identity;3) In order to improve the robustness to multiple noncooperation factors in acquiring images, an identity recognition algorithm based on multiple features sparse representation of dorsal hand vein image is proposed, the global and local features of gray vein image are analyzed in the algorithm, and the unknown identity is determined through judging that whether or not the samples in vein database can represent the unknown vein image sparsely;4) For the defection of losing some vein features under single light intensity, a recognition algorithm base on fusing multi HMMs with Contourlet subband energy observations is proposed, in the proposed algorithm every vein object is regarded as a model, the Contourlet subband energy is regarded as feature observation, and the identity is recognized by the probability of model generating observation.Then all the proposed recognition algorithms are used to analyze vein image features from a single view, such that a multi recognition algorithms fusion strategy based on Bayesian minimum risk is used to judge one’s identity.Experiment shows that for the vein database with more samples, there are some incooperative factors in image acquisition process, after fusing multi recognition algorithms, the correct recognition rate can reach a high level, at the same time, the false rejection rate and false acceptance rate are lower than some other algorithms, and the recognition algorithm after fusing results has good robustness. Finally, the obtained innovative achievements are summarized, and the further research works are prospectived.