Recognizing faces under varying poses with three states Hidden Markov Model

In this paper, we propose an approach for face recognition under varying poses through a three states Hidden Markov Model (3s-HMM). We use discrete cosine transform for feature extraction. The aim of this paper is to evaluate the performance of 3sHMM approach for different face databases that contains ample number of images with varying poses (for instance we consider 0° to ± 60° orientations in yaw). 3 images per subject are used for training and rest all the images for recognition. The sequences of overlapping window are extracted from each facial image, computing the DCT coefficients for each of them. The whole sequence is then modelled by using 3sHMM. The method is compared for different face databases showing comparable results.