A novel approach of K-means based fingerprint segmentation algorithm

To overcome the shortcomings of the existed algorithms, in this paper, a novel approach of K-means based fingerprint segmentation algorithm is proposed. Firstly, the fingerprint image is divided into non-overlapping pieces, and then the feature vector for each piece is represented by its variance, direction and energy spectrum. Then, K-means, the un-supervised clustering algorithm, is utilized to label those vectors of pieces. Finally, a series of post-processing are done to remove the residual isolated blocks in prospect or background area. Experimental results show that the proposed algorithm, being of comparable segmentation speed to the state-of-the-art ones, is capable of adapting distinguished fingerprint image acquisition devices and fingerprint images of varied resolution and quality.