Performance Evaluation of Latent Fingerprint Enhancement and its perspectives

Abstract

This paper analyzes the impact of existing image enhancement techniques on latent fingerprint recognition, scores different datasets and their enhanced recognition results with an evaluation tool, summarizes and reviews existing datasets and related techniques, and analyzes the reasons for their mixed results. It evaluates the effects of different enhancement models, such as FingerGAN, on both public fingerprint datasets and newly compiled latent datasets, namely the MUST and LFIW databases. Using metrics like GMean, GSTD, AUC, and EER, the paper compares the recognition results before and after enhancement to determine the effectiveness of these techniques. The findings suggest that FingerGAN significantly improves recognition rates for latent fingerprints of poorer quality, while it has a mixed or negative impact on higher quality datasets. The analysis highlights the potential and challenges of enhancing latent fingerprints, especially in complex real-world scenarios.