Word Recognition based on Hidden Markov Models with Three-Dimensional Lip Shape Information

摘要: We propose a new method of word recognition which is based on Hidden Markov Models (HMMs) and uses three-dimensional image information. Conventional methods extract features from two-dimensional However, since such vary according to the relative position between camera face, rate declines when speaker not directly facing [11]. The proposed extracts three-dimensionalimagesgenerated using high-speed restoration moir´ analysis. extracted are modeled by Gaussian distributions their temporal dependencies HMMs. show experimentally, real sequential images, that achieves higher than subspace constrained face.