An Inner Contour Based Lip Moving Feature Extraction Method for Chinese Speech

To model the lip moving of Chinese speech with fewer features, this paper presents a feature extraction method that bases on the quadratic interpolation of inner lip contours. The most difficult task of applying inner contours to represent lips is the removing of teeth effects. To address this problem, this paper firstly maps the color space into chromatic color space, which can decrease the effect of teeth and paves the way for extracting the inner lip. In order to get correct lip contour, an approach for extracting precise lip contour feature points is then presented. According to the shape characteristics of lips when pronounce Chinese words, three quadratic curves are applied to represent the upper- and lower-inner contours of the lips and their parameters are computed by nearest-neighbor interpolation algorithm. Our experiments also show that, being compared with classical lip contour representation methods, the advantages of representing lip moving via inner lip contours are the fewer of features required and its robustness from various lip shapes among persons, and thus is more efficient for the lip reading recognition and lip moving synthesis applications