Study On Some Solutions For Automatic Fingerprint Recognition

People pay more and more attention to the biometric authentication because of such features as superior security, convenience and easy operation, which includes recognition of fingerprint, voice, face, iris, palm print and so on. Compared with other biometric authentication technologies, fingerprint recognition is advantageous in that it has high efficiency, convenient collecting and low cost. Particularly, semiconductor fingerprint sense chip emerges in recent years, so the fingerprint recognition has begun to be used in embedded equipment.This paper summarizes fingerprint verification methods and technologies and made intensive study of fingerprint recognition solution, including segmentation of fingerprint image, measure of fingerprint image quality, singularity detection, fingerprint matching and how to develop test software of fingerprint recognition research. The algorithm of fingerprint image segmentation uses three block features— block mean gray value, block variance and block clusters degree—for segmentation, in which definition of block mean gray value is improved. A SVM classifier, which can solve small-sample learning problem with good generalization ability, is trained to classify each block. And solution for computing speed problem brought by SVM is proposed. Secondly, a complete quality assessing algorithm is presented by analyzing size of fingerprint area, fingerprint location, number and location of singularity and direction field of fingerprint and so on. By this means, we can estimate the quality of the fingerprint image well. In this paper, we detect singularity based on poincare index which is improved, and adopt k means to remove spurious singularity. At last, an efficient algorithm of Hough transform-based fingerprint matching is developed. The algorithm can simplify the complexity of parameter space significantly and solve the problem that too much calculation and memory space are needed for the regular approach. Moreover, it is characteristic of a very strong robustness and can deal with low-quality fingerprint image in a satisfactory way.In order to test our algorithm, a test system of fingerprint recognition research is designed and a lot of experiments are performed in it. All these results show that our algorithm achieve good effect.