With the development of information technology and computer networks, the information security becomes more and more important. In many situations, we need to verify personal identities before accessing private information. Biometrics emerges to provide a novel and effective solution to this demand. It captures and recognizes characteristics of human bodies or behaviors in order to achieve personal identification. Palmprint image recognition is a novel member in biometrics. It evaluates similarity of line-like textures inside central regions of palms between different individuals to recognize identities. In recent ten years, based on much research progress, the performance of palmprint recognition systems has been enhanced to a large degree. In this dissertation, we focus on the challenge of contactless palmprint recognition and investigate the problem of image acquisition, preprocessing, palmprint image representation and recognition. The main contributions of this thesis include: (1) We propose an overall framework of contactless palmprint recognition. Following this framework, we firstly investigate the problem of palmprint acquisition in a contactless way under complex backgrounds, through which we can get clear hand images. (2) We design prototype systems to achieve contactless palmprint image under cluttered scenes. In the systems, we combine characteristics of skin color and shape of hands to achieve automatic and efficient hand detection in complex scenes. Based on results of hand detection, we correct and normalize rotation, translation and scales of subjects’ hands by the hand principal lines then obtain regions of interests for real-time palmprint recognition. (3) To describe the palmprint more accurately, we propose a palmprint representation based on contrast context histogram. This method can describe the palmprint more clearly and remove the unsteady point when matching the palmprint. So the result of this representation is much better than the competitive code method.
修改评论