LFIQ: Latent fingerprint image quality
Soweon Yoon, Kai Cao, Eryun Liu, Anil Kumar Jain · 2013
Latent fingerprint images are typically obtained under non-ideal acquisition conditions, resulting in incomplete or distorted impression of a finger, and ridge structure corrupted by background noise. This necessitates involving latent experts in latent fingerprint examination, including assessing the value of a latent print as forensic evidence. However, it is now generally agreed that human factors (e.g., human visual perception, expertise of latent examiners, workload, etc.) can significantly affect the reliability and consistency of the value determinations made by latent examiners. We propose an objective quality measure for latent fingerprints, called Latent Fingerprint Image Quality (LFIQ), that can be effectively used to distinguish latent fingerprints of good quality, which do not require any human intervention, and to compensate for the subjective nature of value determination by latent examiners. We investigate several factors that determine the latent quality: (i) ridge quality based on ridge clarity and connectivity of good ridge structures, (ii) minutiae reliability based on a minutiae dictionary learnt from high quality minutia patches, and (iii) position of the finger by detecting a reference point. The proposed LFIQ metric is based on triangulation of minutiae incorporating the above three factors. Experimental results show that (i) the proposed LFIQ is a good predictor of the latent matching performance by AFIS and (ii) it is also correlated with value determination by latent examiners.