Face verification using D-HMM and adaptive K-means clustering

In this paper, we propose a pseudo 2 Dimension Discrete HMM (P2D-DHMM) for face verification. Each face image is scanned for frontal face in two ways. One way from top to bottom and one way from right to left by a sliding window and two set features are extracted. 2D-DCT coefficients as features are extracted. K-means clustering is used for generation two codebook and then by the vector quantization (VQ) two code words for each face image are generated. These code words are used as observation vectors in training and recognition phase. Two separate Discrete HMM (each HMM for each way) is trained by Baum Welch algorithm for each set of containing image of the same face (λ^v_c, λ^h_c). A test face image is recognized by finding the best match (likelihood) between the image and all of the HMMs (λ^v_c + λ^h_c) face models using forward algorithm. Experimental results show the advantages of using P2D-DHMM recognizer engine instead of conventional continues HMM.