Fingerprint Feature Extraction Based On The Subspace Analysis Methods

Fingerprints are applied to identification has a long history. It has become the most widely used biometric identification technology because of the characteristics of universality, uniqueness and portability of fingerprints. Although the theoretical and applied research about finger-print recognition has acquired significant development, the accuracy and efficiency of fingerprint recognition algorithms still need to be improved. This paper analyzed the achievements of fingerprint recognition algori-thms both at home and abroad in recent years. On this basis, the improved algorithm was proposed for fingerprint segmentation, extraction of region of interest and feature extraction respectively:1. An adaptive fingerprint image segmentation algorithm based on multiple features was put forward to avoid the appearance of the blocking effect of the foreground image edges. It made use of three local features (mean, variance and total variance) to determine the local block threshold adaptively without the experience. Meanwhile, it combined the point level segmentation with the block level segmentation in order that it could be capable of avoiding the appearance of the blocking effect of the foreground image edges at a low computational cost.2. An extracting method of the fingerprint region of interest based on curvature measurement and Poincare Index was proposed. The designed method firstly located the reference point in the trip points set of orientation field. Then it took advantage of the method based on the Poincare Index and that based on the curvature to detect one reference points set among the extracted trip points set respectively. Finally, the centroid of the intersection of the two reference points sets was considered as the extracted reference point. 3. In this paper, the region of interest was utilized to extract features instead of the whole fingerprint information so as to reduce the computational cost. In addition, the two typical methods of the subspace analysis methods:two-dimensional principal component analysis(2DPCA) and two-dimensional linear discriminate analysis(2DLDA) were applied to extract fingerprint image features. And the performances of the two methods were analyzed according to the experiment results.The algorithm performances are evaluated on FVC2004 database. Experiment results show that:1. The proposed segmenting algorithm could avoid the appearance of the blocking effect effectively. And it also could extract the foreground area to the more complex background fingerprint image more accurately;2. The designed extracting method of the fingerprint region of interest decreased the computational cost to a great extent because it located the reference point in the trip points set of orientation field. Besides, it could detect the reference point of all type fingerprints more effectively and precisely and was free to resolution and rotation.3. Both 2DPCA and 2DLDA used to extract the fingerprint image feature could obtain a good recognition performance.