In our increasingly digitalized and connected society, personal impact expands from several meters to millions of miles. We all have a high probability of making conversation, communicating, sending email or even making business with people who connect with us from remote places via the Internet or wireless devices. In such an environment, identity recognition is becoming more and more difficult. Biometrics has emerged as a novel, effective, and promising method to establish identity. As the most natural tool for human beings to perceive and reconstruct surrounding environment, hand is also popular in the field of Biometrics. Exploring different aspects of hand, hand geometry, palmprint and palm vein all fall into this category. Although higher and higher recognition rates are reported in the literature, the success of most hand based modalities relies on a contact device with pegs for image acquisition, which may bring hygiene concern and reluctance of use. Recently, a growing trend towards relieving the users from contact device has emerged and the idea of peg-free or further contact-free hand biometrics is proposed. However, the accuracy of hand based biometrics system degrades along with the removal of peg or/and contact plane. Scale changes, in-depth rotation, non-linear skin deformation originating from pose change, are most commonly encountered image variations in touchless environments. In this thesis, we focus on the performance degradation challenge of contact-free palmprint recognition and investigate the problem of image cquisition, pre-processing and fusion. The main contributions of this thesis include: 1. We propose the idea of compensating image deformation originating from contact-free image acquisition condition via image warping. Since the palm region lacks of stable and universal landmarks, we characterize hand pose by utilizing hand contour landmarks and further eliminate image deformations (including in-plane/in-depth rotation, scale, location and skin deformation) via image warping. By analyzing the matching performance of palm region warping, whole hand warping and fusion of hand regions based warping and hand shape recognition, we finally draw to the conclusion that image warping based method improves the performance of touch-less hand image recognition. 2. We design a contact-free multi-spectral palm image acquisition equipment which features flexible wavelengths selection/combination. We collect a multi-spectral palm image da...
修改评论