Lip Segmentation Based on Improved Clustering of Spatial Patterns
Rong Chuan-zhe · Journal of Military Communications Technology · 2013
Lip segmentation is a very important preprocessing process for lip-reading system and authentication system based on lip. The extraction effects directly affect the performance of real application systems. In this paper, a new lip extraction method based on improved clustering of spatial patterns was proposed, with each pixel in a colorimage taken as a patternincluding not only its feature value but its spatial information. So, the clustering of all pixels of an image was converted to the clustering of the corresponding patterns. Then a component was used to represent the color information of the pixels. At the same time, geodesic distance took the place of euclidean distance to represent the distance between the pixels in the feature space. Experimental results show that the proposed approach can segment images of lips more effectively.