Lip segmentation and tracking based on Chan-Vese model

Lip reading has wide spread application, e.g. audio-visual Automatic Speech Recognition (AV-ASR), silent speech interface and person identification. Lip segmentation is one of important step in lip reading, because it provides basic information to be processed in subsequent steps. Lip tracking is a process of locating lip to associate lip in consecutive video frames. Chan-Vese model is a region-based segmentation algorithm, which also can be used as tracking method. This algorithm can detect boundary of object which not defined by gradient, while classical active contour can't be applied. This method also can detect object by any initial curve in the image, not necessarily surround the object. This paper investigates about lip segmentation and tracking based on Chan-Vese model, preceded by the color segmentation.