| Recently,as an auxiliary method for Automatic Speech Recognition(ASR),computer lip-reading technology has received more and more concern.Some key technologic problems in computer lip-reading system,such as lip detection,feature extraction and lip-reading recognition were researched in this paper.The method employing support vector machine to segment mouth from face image was presented,and the characteristic parameters were extracted according to the lip template and moving lip feature.Based on these,a dynamic clustering algorithm to classify the lip-shape based on the criteria of the minimum square error sum was proposed.Time series of state that described the variation of moving lip were trained according to improved ant colony algorithm,and meaning of pattern matching were proceeded by DTW algorithm.Experiments show the recognition algorithm has satisfactory result in classifying isolated pronunciation. |
PDF Full Text Request