Fingerprint Skeleton Extraction Based on Improved Principal Curve

In the fingerprint recognition system, skeleton extraction for low quality fingerprint images is an emphasis and difficulty task. Traditional methods are, however, susceptible to noise. In view of this, we propose a principal curves-based approach to alleviate this difficulty. In the paper, according to some characteristics of the fingerprint dataset, we improve the original principal graph algorithm proposed by Kegl to obtain principal curves, which can be served as the skeleton of a fingerprint. Experimental results show that our improved principal curve algorithm is better in efficiency and quality than the original algorithm. Our algorithm contains more information quantity and is proved to be more accurate and anti-noisy than thinning algorithm.