Non-overlapped sampling based Hidden Markov model for face recognition

In this paper, a novel method for face recognition, based on Hidden Markov Model using non-overlapped sampling, is proposed. Conventional Hidden Markov Model (HMM) approaches always model a face using the observation vectors generated by overlapped technique, leading to low efficiency and redundant information. The singular value vector and 2D discrete cosine transform (2D-DCT) coefficients of no-overlapping sub-images are fused in feature level by the canonical correlation analysis (CCA) to construct an efficient set of observation vectors. Experiments to evaluate the proposed approach are carried out on the Georgia Tech (GT) face databases and the Olivetti Research Laboratory (ORL) databases. The results show that the proposed method has reduced the training and recognition time obviously by using non-overlapped technique with equal or better performance than previous methods.