Research On Feature Extraction And Matching Algorithm For Small Area And Low Quality Fingerprint Image

In recent years,fingerprint identification technology has gradually become popular in residents’ lives,and is used for authentication of some software and hardware devices such as smart phones.However,the recognition accuracy of small-area and low-quality fingerprint images has not yet reached the actual demand,especially in the field of criminal investigation,the dependence on high-precision fingerprint identification technology is increasing.Feature extraction and matching of small-area and low-quality fingerprint images are studied in this thesis in order to improve the effect of fingerprint recognition.In order to extract higher quality features from low-quality fingerprint images,a contrast enhancement feature extraction algorithm based on SIFT is proposed in this thesis.In the same scale space,two images with different scales are superimposed with different energy coefficients to enhance the contrast of low-scale image.Then,when the local features of the image are described,it ensures that the direction of the gradient component is unchanged and enlarges the amplitude at the same time,so as to improve the robustness of feature information.Finally,to reduce the possibility of error preemption in feature matching,each energy column of descriptor is contracted non-linearly to improve the discrimination of descriptor vectors and thus improve the sensitivity of feature matching.The experimental results show that the number of matches in the proposed algorithm can reach more than 2times of SIFT results on average,and the number of bidirectional optimal matches has increased by 120% on average.With regard to feature matching,in order to reduce the computational complexity of existing matching algorithms,this thesis proposes a non-iterative feature point matching algorithm according to the characteristics of fingerprint images.Firstly,the set of bidirectional optimal matches of two images to be registered is calculated by cosine similarity of feature descriptor vector.Then,the adjacency matrix and connected sub-graph of set are obtained by analyzing the connectivity between bidirectional optimal pairs through multiple constraints,and the homography matrix is calculated after filtering the matches with larger errors in the connected sub-graph.Finally,the feature points are mapped by homography matrix,and the optimal matching pairs are found in the mapped local area to complete accurate matching.The experimental results show that the matching time of proposed algorithm is only 10～20milliseconds,which is several times to dozens of times shorter than the improved RANSAC algorithm,and reaches state-of-the-art performance on matching accuracy.