Lip contour segmentation is an important step in some applications such as in an automatic speech recognition system. Because of the associating with visual information, the recognition rate would be greatly increased, especially in noisy conditions. Here, we propose a new lip contour segmentation method with two stages. In the first stage, the lip image has been set in ROI (Region of Interesting) by a face detector which called AdaBoost to avoid the disturbance of complex backgrounds. In the second stage, the ROI has been transformed to the chromaticity color space, and the lip contour was extracted by a binary transformed image of ROI that using K-Means to find the threshold of lip contour pixel values for each frame. After extracting the lip contour, the lip contour feature points are located by histogram projecting with a binary transformed image in ROI. Extracting results are shown in the experiment results, we've compared our method to some conventional color space methods, and it shows good results of contour extraction in our method.
展开▼
