Lip contour extraction scheme using morphological reconstruction based segmentation

摘要

In this paper, a two stage lip contour extraction scheme is proposed, where at the first stage, pixel intensity variation based information is used for obtaining a preliminary estimation of lip and in the second stage, morphological reconstruction based segmentation is employed to detect the final lip region. From a given video frame, first the mouth region is extracted by using block threshold based binary conversion and some morphological operation. Analysing the variation of RGB pixel intensity pattern, intensity ratio based lip region detection is performed, which provides an accurate estimate of lower lip region. However, because of critical shape of the upper lip, further processing is carried out by using morphological opening by reconstruction. Finally, polynomial curve fit is performed to obtain the lip contour. From extensive experimentation on several real-life images from audio-visual clips, it is found that the proposed method offers high level of accuracy in image.