Study On Some Problems In Fingerprint Identification

Fingerprint identification as an important and full of challenge research topic of pattern recognition, has significant meaning both in theory and application aspects as to improving its performance quality. For this respect, this thesis has discussed the problems of fingerprint image pre-processing, denoising and features extraction, using the knowledge of digital image processing, pattern recognition and computational simulation. The major contributions of this thesis are listed as follows:Firstly, a few important parts of fingerprint image pre-processing are analyzed and a fingerprint quality evaluation method is improved based on previous tasks, which can judge the fingerprints too dry or too wet even those located deflectively. It can avoid inputting those images with low quality or non-standard and reduce workload and identification time in the future work.Secondly, a self-adaptive greyscale transformation module is proposed based on uncomplete Beta function about fingerprint image denoising, and there are two adjusting parameters in this module .A simple simulated annealing and genetic algorithms(SAGA) is proposed based on adaptive stretching module simultaneously. This algorithm is proved to be feasible and effective seen from the Experimental results by optimizing a dinoising fingerprint image.Thirdly, a fingerprint image feature extraction method is introduced based on fingerprint ridge distribution modal according to ridges distribution sequences in the fingerprint, In contrast with traditional extraction method based on space domain or minutiae, this method can extract most of the fingerprint features which can completely stand for the correlative fingerprint to a certain extent, and improve identification efficiency.The theory and technology of automatic fingerprint identification is systematically studied in this thesis. All algorithms are implemented on PC under VC++6. 0 and Matlab. The results demonstrate that the processing effects of fingerprint images can be greatly enhanced and the identification efficiency can be improved evidently.