Minutiae are the two most prominent and well-accepted classes of fingerprint features arising from local ridge discontinuities: ridge endings and ridge bifurcations. However, the preprocessing stage doesn't eliminate all possible defects in the original gray-level image and the orientation estimation in a poor image is extremely unreliable. In order to further eliminate the false minutiae caused by low quality, a minutiae verification mechanism is proposed to improve the identification performance. The minutiae verification mechanism uses the concept of eigen-codebook to find the optimal projection bases for true minutiae regions and false minutiae regions. Experimental results show that the performance is improved efficiently even less training data.
Relation:
第十六屆電腦視覺、圖學暨影像處理研討會論文摘要集=Proceedings of 16th Conference on Computer Vision, Graphics, and Image Processing，頁182-187
