LAFIN: A Convolutional Neural Network-based Technique for Singular Point Extraction and Classification of Latent Fingerprints

Abstract

Latent fingerprints left at crime scenes are useful evidence in the court of law. Law enforcement agencies have been using latent fingerprints for many years as a reliable forensic evidence for crime scene analysis. However, due to poor quality images and complex image background, current state-of-the-art for automatic latent fingerprint processing is not as reliable as rolled or live scan fingerprints. In this paper, we propose a Convolutional Neural Network (CNN) based model (LAFIN) to classify latent fingerprints into five different classes. The classification result along with fingerprint image is then fed to a mini classifier to extract the singular points (if any) present in the latent fingerprint. The CNN model for latent fingerprint classification is trained on different fingerprint images from IIIT-D latent database and NIST Special database 4. The results on publicly available IIIT-D latent fingerprint database demonstrate the efficiency of the proposed approach.