Fingerprint Image Segmentation And The Reference Point Extraction Algorithm

In recent years, automatic fingerprint identification system, which consists of fingerprint image enhancement, fingerprint classification, feature extraction and fingerprint matching, has drawn a substantial attention in the literature. Until now, automatic fingerprint identification system is still not satisfying for the following two problems: (i) To segment the valid fingerprint area from background under some non-perfect situation; (ii) To match two fingerprint images in that (a) features may have undergone elastic deformation due to skin elasticity and (b) features may be scaled, translated, and/or rotated.In this paper, we deeply analyze and investigate the above two key problems of automatic fingerprint identification system. The main contents are as follows:For fingerprint segmentation, aiming at the deficiency of existing method, we first bring forward a new thought of fingerprint segmentation based on wavelet packet decomposition and carry it out. The fingerprint image is decomposed into sub-images using wavelet packet decomposition at some specified level firstly. Each sub-image consists of pixels in a specified range of frequency of the original fingerprint image; then an appropriate sub-image is selected and reconstructed and the edges of the valid fingerprint image are detected in this sub-image; finally the fingerprint image is segmented according to the edges we detected. The experiments on some extremely low quality fingerprint images confirm us the effectiveness of our approach.Delauany based reference points extraction method is presented. Minutia based matching is the most well-known and widely used method for fingerprint matching as one of graph-based matching methods. Two kinds of minutiae are adopted in matching: ridge ending and ridge bifurcation. Thus, fingerprint matching can be regarded as a point-pattern matching problem. Some pairs of reference points are first located to align the template image and the query image before matching. Reference points could be obtained using Delaunay triangulation net based on triangulation matching.