Recognizing multiple persons' facial expressions using HMM based on automatic extraction of significant frames from image sequences

摘要: A method that can be used for recognizing facial expressions of multiple persons is proposed. In this method, the condition muscles assigned to a hidden state HMM each expression. Then, probability updated according feature vector obtained from image processing. Image processing performed in two steps. First, velocity estimated every successive frames by using an optical flow algorithm. two-dimensional Fourier transform applied field at regions around eye and mouth. The coefficients lower frequencies are selected form vector. mixture density approximating output so as represent variation among persons. To cope with case when displayed continuously, computation modified such peak motion detected, new sequence assumed start previous frame minimal motion. Experiments show effective because recognition accuracy improves number mixtures increases. addition, correctly recognizes expression continuously follows another one.