Using shape priors for improved lip segmentation

Lip segmentation is an important problem which is necessary to be solved in many applications, especially in audio-visual speech recognition. In this paper, a level-set based method that utilizes adaptive color distributions and shape priors for lip segmentation is introduced. More precisely, an implicit curve representation which learns the color information of lip and non-lip points and shape information of lip regions from a training set is employed. The model can adapt itself to the image of interest using a coarse elliptical region. Extracted lip contour provides detailed information about the lip shape. We show that using shape priors improve the segmentation performance, especially the recall rate.