The Research Of Several Key Technologies In Mouth-shape Recognition

In this paper, a model of skin's color according to the theories of skin's color likelihood is established. Threshold used to separate the skin color from the images is selected adaptively. A method which can realize the face detection and orientation is developed. And a method of segmenting and extracting the mouth from the facial image based on color image is proposed. The classifier using R,G,B and U as the feature vectors to separate lip color from skin color is designed by the principle of Support Vector Machine. In order to describe the edge outlines of the mouth binary image, several curves are used to make up a mathematics formulary of the lips. Based on mathematics formulary of the lips, a energy function calculating the minimum value of the energy function to obtain the shape feature of lips is constructed. Locating the lip in video sequences is one of the primary steps of the automatic lip-reading system. The feature values of first image are used as the initialization values of figure curves in next image. This method realizes lip detecting and locating. Finally, a serial and useful feature values are selected to build up a small database of lips.