Improved lip contour extraction using K-means clustering and ellipse fitting

A Visual Speech Recognition system is completely dependent on correct extraction of lip contours. The extracted lip contour can be useful in tracking the lip movements in Visual Speech Recognition. The accuracy of the system can be improved by correct extraction of the lip contours. In this paper, a computer vision approach is proposed to automatically extract lip contour from the face image. The proposed technique uses Viola Jones algorithm to localize the face and mouth in the image. Unlike existing methods merge thresholds used in Viola Jones are made iterative and adaptive which makes it invariant to the quality of input image. The colour space conversion and clustering method results in separation of lip from non lip pixels which are further subjected to morphological operations and ellipse fitting for an efficient lip contour extraction. The proposed method has been tested on 4000 images from the VidTimit database and results showed significant improvement in lip segmentation than some of the existing methods for lip contour extraction. The proposed method is computationally efficient and robust and can be used for real time applications.