Hidden Markov Models for face recognition

Biometric recognition and authentication can be used in many fields such as; surveillance, security systems, access control, and many more. The main objectives of this paper is to implement a face recognition system using one-dimension Hidden Markov Models (HMMs), where a model is trained for each user. The idea behind using HMMs in face recognition is that the face structure may be viewed as a sequence of distinct parts or regions, and that the order of this sequence is always preserved (e.g., forehead, eyes, nose, mouth, and chin). Features extracted from each region acts as a stream of frames, or vectors describing the structure of the face. The image of a face may be modeled using a one-dimension HMM by assigning each of these region to a state. In this paper, three proposed methods are used to extract the feature vectors namely, the Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT), and Principal Component Analysis (PCA). Several HMMs topologies were used in this paper, the results reported are obtained using the Olivetti Research Ltd (ORL) database of faces. Finally a fusion of the three systems is used to improve the recognition performance.