Automatic fingerprint recognition technique performs personal identification based on the intrinsic physiological feature of fingerprint and has wide application prospect. Automatic fingerprint recognition technique has received extensive attention from researchers all over the world. Fingerprint matching plays significant role on the performance of recognition system. This thesis focuses on fingerprint matching, in which some key problems are studied deeply. And the main contributions of this thesis are as follows: 1. A segmentation method based on active contour model is proposed. Firstly, descriptors of fingerprint foreground are constructed, then active contour model is performed on the descriptor image, therefore, the foreground border is located automatically. Due to the block effect, the envelope of fingerprint foreground is coarse in the sense of pixel level. A curve filter method as a postprocessing is proposed to smooth the border of foreground. Experimental results show that borders of foreground can be located with single-pixel precision. 2. Base on the gray scale gradient, the estimation of fingerprint orientation filed is improved through direction composition in Fourier domain. A new detection method is proposed by combining the measure of local orientation disorder with the curvature field which calculated from orientation field. According to singular points, the searching range of minutia pairing is bounded, only those minutiae which fall into a certain scope are considered when match two fingerprints, thus, the matching speed can be improved greatly. 3. Proposes a fingerprint alignment method based on local ridge similarity. We firstly sample the thinning ridges which associate with corresponding minutiae. The horizontal and vertical shift are estimated directly from minutiae, while the rotation between two ridges is estimated through curve fitting. Final, the similarity between two ridges is represented by a related measure with fitting error. The minutia pair corresponding to the peak in the three dimension histogram of local ridge similarity is assigned as the reference points. Experimental results show that the proposed method can effectively accomplish fingerprint alignment which provides the basis to measure the similarity between two fingerprints. 4. A robust approach of measuring the similarity between two fingerprints is proposed. Firstly, we model the orientation difference and apply Bayesian rule to obtain a probability which indicates the likelihood weather two fingerprints came from the same finger. Secondly, we select absolute reliable minutiae by minutiae scoring and double the bounding box. Therefore, the less such minutiae failing to match means the more similarity between two fingerprints. And finally, we fuse the two similarity measures with the matching minutia score to give a comprehensive matching score. Experimental results demonstrate that the fused score performs better than the single one.
修改评论