Towards Latent Fingerprint Recognition: Evaluating the Performance of Advanced Algorithms

This paper presents an in-depth evaluation of various fingerprint recognition algorithms using the Latent Finger-print in the Wild (LFIW) dataset. By analyzing ROC curves, score distributions, FMR, FNMR, and DET curves, we assess the performance of five algorithms: LatentMinuComp_v0, NBIS, NeuroTech, MarkIDNet1, and MarkIDNet2. Our findings indi-cate that NeuroTech outperforms the other algorithms with the lowest error rates, followed by LatentMinuComp_v0. Conversely, NBIS, MarkIDNet1, and MarkIDNet2 exhibit higher error rates and face challenges in distinguishing between genuine and impostor samples. Future research will focus on optimizing these algorithms, enhancing their robustness, conducting real-world scenario testing, exploring fusion techniques, implementing continuous learning, and improving quality assessment to advance fingerprint recognition technology.