Using only two states of discrete HMM for high-speed face recognition

This paper presents a simple and fast face recognition system based on two states of discrete Hidden Markov Model (HMM). The minimization in the number of states leads to high processing speed and less memory occupation. Median filter is applied to each image under process, where it is the most suitable filter used to eliminate the effect of noise on images, and thereby enhancing the performance of the system. The features are extracted from reduced size images using a combination of maximum variance and Singular Value Decomposition (SVD). More reduction in processing data is achieved by assigning a single value to each feature vector. This process greatly speeds up the training and classification steps, where a discrete type of left-to-right HMM is used in this work. Experimental results verify that the proposed work is superior to previous approaches of HMM face recognition, where it achieves 100% recognition rate, high speed, and extremely low memory usage.