Research On The Technology Of Lip-reading Recognition Based On HMM

With the development of human-computer interaction technology, lip-reading technology as an auxiliary means of speech recognition has been morewidely used in the human computer interaction system. In this paper, regardingcomputer webpage browsing basic platform, human-computer interaction modebased on lip-reading as the research object, the study focuses on lip-readingrecognition. A complete lip reading recognition system consists of four elements:video image acquisition module, positioning module, lip feature extractionmodule and the recognition module. Through the detail analysis on the keyproblems in the above elements, we put forward the corresponding solution inthis paper, and the main research contents are as follows:According to face detection in lip-reading system, using the YCbCr colorspace, we presents a novel light intensity self adaptive algorithm of facedetection to locate the face region, and rough segment lip based on geometricfeatures of human face. The experimental results show that this method improvesthe stability in face detection and orientation under non uniform light intensit y.Based on the clustering effect of lip color in YIQ color space, we designed aself-adaptive chromatic filter to separat lip color, this method overcomes theindividual differences in lip color and precise position the lip.By comparing with different feature extraction methods, this paper proposedan approach that combin the shape feature and gray information of lip as the lipmovement information for subsequent identification. Both the shape feature oflip contour and pixel feature based on gray statistics of lip use the scalar toreduce individual differences and retain all the lip movement information as faras possible.Based on the theory of hidden Markov model and the operation feature ofwebpage browsing, this paper designs corresponding model for the webpageoperation instruction.And the hidden Markov model was further trained andtested by the the lip movement information features. Final recognition result ismore ideal, and proved the design and improved the lip feature extraction in thispaper is effective.