Research On Vein Feature Extraction Based On Fractal And Multi-wavelets Theory

Biological feature recognition is a kind of identity authentication technology; it is based on the inherent human biological characteristics or behavioral characteristics. With a combination of computer technology, the biometric technology has been widely used in many fields. As a new method of biometric identification, vein recognition, although with a late start, has become the leading biometric identification method after several years of development.Using biological features to authenticate identity, the key issue is the research of feature extraction and matching algorithms. Taking vein of human back of the hand as the subject, a research of feature extraction methods of vein based on fractal and multi-wavelets theory is put forward in this thesis.In order to make feature extraction and matching go well, vein images must be preprocessed. Methods such as contrast-limited adaptive histogram equalization, threshold image are applied in preprocessingVein feature extraction is the most critical part of vein recognition. The previous methods used in vein recognition are summarized in this paper, meanwhile, the advantages and disadvantages are briefly analyzed, and then, vein feature extraction methods are raised. The global and local features which are separately used in classification and matching are extracted form vein images. Global features are extracted by using box counting method while local ones by multi-wavelets decomposition method. The fractal dimension which could stand for the coarseness of images can be figured out by box counting method; after one-story and two-story multi-wavelets decomposition, one-dimension coefficients and multi-scale coefficients can be coded. Ultimately vein eigenvectors are obtained.Experiments have been done in this paper:the coarseness of vein images carried out on box counting in vein feature extraction is used in classification; the results prove that the method is effective. After feature extraction based on one-dimension coefficients and multi-scale coefficients coding, the eigenvectors are matched by cosine distance and Hamming distance method.Finally, achievement of thesis is summarized, ideas about vein recognition methods are raised, and vein recognition research forecast is proposed as well.