Latent Fingerprint Image Quality Assessment Using Deep Learning

Abstract

Latent fingerprints are fingerprint impressions unintentionally left on surfaces at a crime scene. They are crucial in crime scene investigations for making identifications or exclusions of suspects. Determining the quality of latent fingerprint images is crucial to the effectiveness and reliability of matching algorithms. To alleviate the inconsistency and subjectivity inherent in feature markups by latent fingerprint examiners, automatic processing of latent fingerprints is imperative. We propose a deep neural network that predicts the quality of image patches extracted from a latent fingerprint and knits them together to predict the quality of a given latent fingerprint. The proposed approach eliminates the need for manual ROI markup and manual feature markup by latent examiners. Experimental results on NIST SD27 show the effectiveness of our technique in latent fingerprint quality prediction.