Research On Fingerprint Image Processing And Fingerprint Matching Algorithm

With the development of information technology and network technology, information security has drawn increasing attention. In order to protect their own information, data and property, many occasions need to identify their visitors. The identification system, based on fingerprint recognition technology, with its unique technical advantages and cost benefits, are widely applied to various situations.However, the current fingerprint identification system has severval problems: such as feature matching less efficient; easily be caused by influenced by false feature points matching, on the blurred images can not be accurately and so on. This paper will do a systematic study of most aspects of the fingerprint image processing system,by compared to existing algorithms, improvement and innovation.Fingerprint recognition system by processing and image processing purposes was roughly divided into:preprocessing, feature extraction and feature matching.A fingerprint image preprocessing including normalization, image segmentation, image enhancement, binarization and thinning. This paper focuses on image enhancement algorithm to study the direction-based filtering of image enhancement.In the fingerprint feature extraction stage, this paper's algorithm based on extracted feature points the neighborhood of the thined image, and then carried out to deal with pseudo-feature points. Before removing the pseudo-feature points, the first carried out to the edge treatment; remove the fingerprint image around the feature points. Then according to different types of characteristics of pseudo-feature points, using the corresponding algorithm removed.In the matching phase, this paper used the twice feature matching algorithm which based on the feature point. Initial matching stage, by building local feature vector, and innovative manner, according to various characteristics with the amount of deformation of other factors by the different weights into a different match, "scoring" system, to achieve the benchmark feature points and matching points to be set strike; second matching stage, the use of reference feature points to achieve the characteristics of fingerprints to be matched with the template fingerprint feature points between the coordinates of calibration points. And through the introduction of variable bounding box, discriminant whether the two feature points match. Last obtain all the feature points matched the total number, and compare it with the threshold,if it's more than the threshold, then they are the same fingerprint. or not.Finally, this paper did thousands times of Matlab simulation,which is use the FVC2002DB₁ and DB₂ standard fingerprint databases, to test the whole fingerprint recognition algorithm.The result show that the FAR10 were 0.85% and 5.5%, the FAR100 were 1.8% and 9.5%, the EER were 1.7% and 7.0%. So we found that the matching precision (refer to the FAR10 and FAR100)of this algorithm is better than other similar test results and comprehensive performance (refer to the EER) should be improved a little more. This results met expectations.