Rotational invariant fingerprint matching using local directional descriptors

Fingerprint matching techniques have been extensively investigated in the literature. Minutiae-based and image-based approaches are two major classes of fingerprint matching. Minutiae-based approaches are still not able to achieve high matching accuracy, particularly in low quality images. Now-a-days, image-based methods are given more attention as they extract more discriminatory information in comparison to minutiae-based methods. In this paper, we have proposed an image-based method, which exploits the rotation invariant local directional fixed length descriptors. This method involves preprocessing, region of interest (ROI) computation and extraction of proposed local directional fixed length descriptors. The proposed descriptors are tested on public benchmark databases FVC2002 and FVC2004 using Euclidian distance, Chi-square distance, histogram intersection, and least square support vector machine (LS-SVM). The effectiveness of the proposed method is demonstrated by comparing the experimental results with those of other image-based approaches proposed in the literature.