A Robust Face Recognition System for One Sample Problem

Most of the practical applications have limited number of image samples of individuals for face verification and recognition process such as passport, driving licenses, photo ID etc. So use of computer system becomes challenging task, when image samples available per person for training and testing of system are limited. We are proposing a robust face recognition system based on Tetrolet, Local Directional Pattern (LDP) and Cat Swam Optimization (CSO) to solve this problem. Initially, the input image is pre-processed to extract region of interest using filtering method. This image is then given to the proposed descriptor, namely Tetrolet-LDP to extract the features of the image. The features are subjected to classification using the proposed classification module, called Cat Swarm Optimization based 2-Dimensional Hidden Markov Model (CSO-based 2DHMM) in which the CSO trains the 2D-HMM. The performance is analyzed using the metrics, such as accuracy, False Rejection Rate (FRR), & False Acceptance Rate (FAR) and the system achieves high accuracy of 99.65%, and less FRR and FAR of 0.0033 and 0.003 for training percentage variation and 99.65%, 0.0035 and 0.004 for k-Fold Validation.