Face recognition system based on continuous one-state model

Abstract

Face recognition (FR) has received an important concern in our contemporary life, especially in control and security applications. This work proposes a novel method for tackling the problem of FR using the minimum number of states (i.e. one state) of continuous Hidden Markov Model (HMM). The contribution of the proposed method can be viewed in the use of continuous one-state HMM, which has not yet been used in other approaches. Furthermore, the concept that early adopted, regarding the relationship between facial regions and number of states, is invalidated, accordingly. The implementation of this method can be summarized as follows: i) filtering the face image using a median filter, ii) reduction of additional noise and image size using multiple level of discrete wavelet transform (DWT), iii) training the outcomes of the DWT using one-state HMM of continuous output density with one Gaussian mixture coefficient, and iv) recognizing images for validation. The advantage of using continuous output densities appears in conducting noisy images, such that the proposed method highly reduces the effect of noise. The simulation results present that the recognition rate is about 100% despite the presence of 25% and 50% of impulsive noise densities on the images of the ORL and Yale databases, respectively.