Abstract
Pore-based high-resolution fingerprint recognition has been researched for many years, with studies demonstrating that pores can improve the accuracy of fingerprint identification. However, existing methods that rely solely on pores to measure the similarity between two fingerprints face challenges, particularly when dealing with partial fingerprints with limited overlapping areas. This article presents a novel high-resolution fingerprint recognition method that integrates multiple features. The proposed approach consists of three main steps. First, pixelwise correspondences are established using a dense matching algorithm, facilitating one-to-one matched pixels for overall similarity estimation and image alignment. On the basis of the alignment, pores outside the overlapping regions are removed. Second, pores within the common areas are matched using a partial graph matching algorithm, which reduces the impact of outliers and noise on the matching results. Since most of the outliers are already eliminated during the alignment step, the accuracy of pore matching is further enhanced. Finally, the overlapping areas of the two fingerprint images are used to calculate the overall similarity using a vision transformer (ViT) network. The final similarity between the two fingerprints is computed by integrating the pore matching results with the overall similarity of the overlapping areas. Experimental evaluations are finally conducted to assess the performance of the proposed method.