Robust CBFR for Forensic Application using CNN
Sampada S. Bulbule, Mukul Sutaone, Vibha Vyas · 2019
Face recognition for forensic applications is a very challenging task. The challenges arise due to the unavailability of the mug shot of the culprit from a crime scene. The available face is either occluded or blurred. In both the cases, Component-based face recognition is capable of recognizing an individual or the culprit through a part or component of the face, which is the major focus of the paper. Convolutional Neural Network is the base of the proposed framework for face recognition via components: nose, mouth and, eyes. The novelty of this framework is that it can single-handedly recognize an individual by full face or by anyone or all the components of the face using Deep Learning with high recognition rate. In a practical scenario, the collected images are corrupted with different noises that reduce the recognition rate. Robustness of the system is tested for different noises viz. White Gaussian noise, Speckle noise, and Salt Pepper noise. Results show the descending trend in the accuracy of the recognition as the intensity of the noise is increased. Faces94, which is the standard database of face, is used for the experiment.