Videophones and videoconferencing have received much attention in telecommunication applications recently and it is believed that there will be great demand in such technology for business and personal use in the near. future. In order to transmit large amount of video signals across a low-speed network, we usually apply some video compression algorithms which may, however, cause the problem of degradation in picture quality. Model-based image coding method was proposed to solve this problem. No matter how this method has been implemented, the very basic requirement which most researchers have seldom addressed is that the system should be able to automatically locate both eyes and the mouth and segment the facial area in a single-frame head-and-shoulder image with noisy background. This...[ Read more ]

Videophones and videoconferencing have received much attention in telecommunication applications recently and it is believed that there will be great demand in such technology for business and personal use in the near. future. In order to transmit large amount of video signals across a low-speed network, we usually apply some video compression algorithms which may, however, cause the problem of degradation in picture quality. Model-based image coding method was proposed to solve this problem. No matter how this method has been implemented, the very basic requirement which most researchers have seldom addressed is that the system should be able to automatically locate both eyes and the mouth and segment the facial area in a single-frame head-and-shoulder image with noisy background. This problem is known as a preprocessing of many applications such as face recognition and videophone, and is being addressed in this research.

We propose a system that consists of three steps. Facial features such as eyes, sides of face, shoulders and the mouth are detected individually in the first step. All detected features as well as their false alarms are input to the second step, which is an uncertainty reasoning model that can handle incomplete information and resolve conflicts among the input. A face model based on spatial relations of facial features is built in the reasoning model so that the best match of each facial feature is found. In the last step, the system takes those matched facial features to estimate the facial region, and finally segment the facial area by outlining the face contour using the active contour model.