Lip Feature Extraction And Recognition With Head-camera

Lip-reading has been an important topic in computer vision and speech-recognition, and has a bright future for application in Intelligent Human-computer Interface and Multimedia Intelligent System. This paper focuses on the lip image capture, lip feature extraction and lip recognition with Head-Camera. Head-Camera can make user conveniently capture the lip image capture and extract the lip feature.Head-Camera collects lip image movably. PIG computes the initiate points of Snake automatically. Adopt CONVEF Snake Model to extract lip contour, which is able to track lip contour in video sequences with accuracy and robustness. CONVEF Snake Model uses virtual electric field as external power, regarded as a pure convolution operation to improve the calculation speed. The lip contour and the pixel information are combined according to their weights to generate the mix-feature as the descriptor of the lip. ISOMAP is employed to reduce the dimension of the lip feature and discover the intrinsic structure of the lip reading. CHMM is applied to model and recognize the lip reading based on the manifold feature. Experiments conducted on the real lip image illustrate that the proposed method could extract lip features and do recognition fast and accurately, especially in speaker-independent.