Segmentation of fingerprint image based on gradient magnitude and coherence

Abstract

Fingerprint image segmentation is an important pre-processing step in automatic fingerprint recognition system. A well-designed fingerprint segmentation technique can improve the accuracy in collecting clear fingerprint area and mark noise areas. The traditional grey variance segmentation method is widely and easily used, but it can hardly segment fingerprints with low contrast of high noise. To overcome the low image contrast, combining two-block feature; mean of gradient magnitude and coherence, where the fingerprint image is segmented into background, foreground or noisy regions, has been done. Except for the noisy regions in the foreground, there are still such noises existed in the background whose coherences are low, and are mistakenly assigned as foreground. A novel segmentation method based on combination local mean of grey-scale and local variance of gradient magnitude is presented in this paper. The proposed extraction begins with normalization of the fingerprint. Then, it is followed by foreground region separation from the background. Finally, the gradient coherence approach is used to detect the noise regions existed in the foreground. Experimental results on NIST-Database14 fingerprint images indicate that the proposed method gives the impressive results.