Abstract:

The EHMM relies on the output maximum likelihood probability to determine the face image. However, because of the similarity between face images, this method can lead to recognition errors. We propose a face recognition method based on the EHMM-SVM. The two-dimensional discrete cosine transform (2D-DCT) is used to extract face features so as to obtain the observation vector sequence. The output probability of each face image corresponding to the EHMM model is obtained by the double nested Viterbi algorithm. The output probability is input into the SVM for classification training and recognition tests, and the results of face recognition are obtained. ORL and YALE face databases are used in  the experiments. Experimental results show the feasibility and effectiveness of the method.

Key words: EHMM, SVM, face recognition, 2D-DCT, Viterbi