Research On Fingerprint Identification And Fast Matching Algorithms

With the development of the computer and information technologies, identity identification technology by computer makes quite great progress. Because of the invariability and uniqueness of the fingerprints, the fingerprint verification and identification have been become of the widest used identity identification technologies.This thesis does deep research mainly on fingerprint identification technology's essential parts, including fingerprint pretreatment, feature extraction, feature match and the method of setting up fingerprint feature database. In the fingerprint pretreatment part, Combining the strong denoising character of wavelet transform with the less processing time of two-dimensional decomposed Cabor filter, A fingerprint enhancement algorithm based on decompositing Gabor function in wavelet domain is proposed. This algorithm uses wavelet transform to decompose the gray-scale fingerprint image into sub-image at one level, only doing the ridge direction estimation of the low-frequency sub-graph. This algorithm can not only undermine the effects of the high-frequency information of the fingerprint image to direction estimation, but also reduce the time of direction estimation. At the same time, the number of multiplication of two-dimensional decomposed Gabor filter is significantly less than the number of two-dimensional Gabor filter. Finally, we use inverse wavelet transform for reconstructing the enhancement image of the four sub-graphs. This thesis also does deep analysis on binarization and thinning algorithm. In the feature extraction part, a kind of 8-neighbour extraction algorithm was proposed and distributing regulation of various pseudo minutiae in fingerprint image were analyzed. Special algorithms were designed with respect to various noises to delete pseudo minutiae. We also design a reasonable person information table to store fingerprint feature, this can improve the performance of the database. In the feature match part, the thesis proposes a new segmented point pattern matching algorithm based on vector-based triangular. First of all, according to the relationship between two subsets consisted of three neighboring minutiae derived from the two minutiae sets to acquire a pair of reference point and matching score for fingerprint matching. Then according to the reference point, the feature point will be converted to polar coordinates to portray. At the same time of calibrating the coordinates of the input images, we introduce the technique of variable bounding box to eliminate local deformation. Then, the difference of minutiae to the reference point will be computing as the ruling condition to re-calculation the matching score. Then we statistics the matching points, when it is greater than a pre-set threshold, that we consume the two fingerprint image matching. Finally, an automated fingerprint identification system was accomplished in the integrated software developing tool Visual C++ 6.0, and its capability was tested to be effective.