Speaker Identification Research Based On Lip Motion Features

With continuous development of application of the information processing technology, there are more requirements of authentication on its speed and efficiency, although using static biometric already has many research in depth, some has been in practical use in daily life, in recent years, much attentions has been paid to researches using dynamic biometric features to identification which can improve the identification rate and recognition speed.Based on the analysis and study of general technology of speech recognition and the advantages and disadvantages of traditional lip-reading recognition technology, this paper designed and constructed a speaker identification system based on lip motion features by combining static and dynamic biometric feature come from the speaker, meanwhile deep research has done to the composition of the proposed system. The main work and achievements are as follow.(1) Based on the study and analysis of research achievements of lip motion features, this paper designed and constructed a speaker identification system model based on lip motion features, which composed by three parts:video preprocessing, lip feature information extraction, identification. In addition, the system model introduced a self-learning control mechanism, so that identification system has adaptability and scalability.(2) In the stage of video pre-process, in order to eliminate the harmful effect on extraction of lip motion features caused by head movement, add acquire pure information of lip motion, this paper use technique of image rotation to successive face images, in order to compensate the head movement, thus to improve signal to noise ratio of lip characteristic information.(3) In the process of lip feature information extraction, using the pattern which integrates a speaker's lip geometric and dynamic feature as the speaker's characteristic feature. We designed a model of lip feature information extraction, and designed algorithm for key points track and location, algorithm for key points selection, algorithm for motion vector extraction, algorithm for fitting key points trajectory curve, the simulation results show that the feasibility of the algorithm.(4) In the phase of speaker recognition, using a two-phase feature matching model, on the basis of algorithm design of geometric feature matching and motion feature matching algorithm, this paper gives a complete identification process, at last we make simulations of the identification process from two aspects of its feasibility and efficiency, and the experiments verified the effectiveness, high recognition rate and accuracy of the proposed method.