Face Recognition Using Probabilistic Model

Abstract

The effectiveness of real-time facial recognition systems is of great importance. In this work, hidden Markov model is used applying three states for face recognition with further reduction of feature vector coefficients. A sequence of overlapping blocks have been used for dividing the face images. The model is trained using an observation sequence containing the eigenvectors and coefficients of these blocks, and each subject is given a separate HMM. Applying DWT during the preprocessing stage has reduced the computational complexity of the proposed model and maximum noise filtration has been achieved. Moreover, singular value decomposition has been applied on face images and test images are rejected and accepted based on threshold singular value determined empirically. For feature extraction, principal component analysis is employed. Recognized test images are identified dependent on majority criteria utilizing different observation vectors of image. Experimental results of Yale and ORL directory in noisy environments such as Salt and Pepper and noise free environments with show that the recognition authenticity of the presented model is comparative to the existing methods with reduced computational cost.