Lip feature extraction and reduction for HMM-based visual speech recognition systems

摘要: Lipreading is a main part of audio-visual speech recognition systems which are mostly faced with redundancy extracted features. In this paper, new approach has been proposed to increase the lipreading performance by extraction discriminant way, first, faces detected; then, lip key points in four cubic curves characterize contours. Next, visual features from contours for each frame. To discriminate unit (word) others, that frames arranged feature vector. Moreover, differences frame k previous used construct more informative vectors. solve small sample size problem, direct linear analysis (D-LDA) employed reduce size. classify these transformed features, hidden Markov model (HMM) recognize units. The algorithm was applied on M2VTS database. Results show applying D-LDA reduction provides better classification accuracy compare employ HMM without reduction.