Robust statistical estimation applied to automatic lip segmentation
Automatic lip segmentation is an indispensable pre-requisite in face-video applications that make use of the mouth-region. Lip segmentation can be treated as a three-stage process: mouth-region detection, separation of the constituent clusters in this region and identification of the cluster containing the lip pixels. This paper describes a novel method of performing automatic, single-frame, chromaticity based lip segmentation with no prior model information or heuristic assumptions. It uses a robust statistical estimator to identify the different regions in the image and then performs post-processing based on cluster colour and shape to identify the lip region.