Lip information extraction based on the fusion of geometry and motion features.

Lip feature information extraction decides the recognition effect in speaker identification based on the shape of lip. To solve the problems of less useful information and low recognition rate, this paper proposed a new extraction method of lip information which is based on the fusion of geometry and motion features. Firstly, key points in lip sequence images were tracked and located. Then on this basis, geometry and motion characteristics were extracted. Finally lip feature information of speaker was got by the fusion of geometry and motion features, and the recognition result was chosen by similarity comparison and the speaker was identified. The experiment indicates that the extraction method of lip information based on the fusion of geometry and motion features has more comprehensive information and higher recognition rate. © 2011 IEEE.