A Robust Technique For Mouth And Eyes Extraction

Model-based coding is a compression technique that has been recently proposed for low bandwidth transmission such as real life video conferencing. Implementation of such a technique requires robust algorithms for extraction of facial features. This paper describes some algorithms we developed for the extraction of mouth and eyes from a sequence of video frames. To demonstrate the robustness of the proposed solution, we conducted a series of tests with various skin colors, illuminations, face orientations, eye colors, and shirt colors. The experiments showed that our algorithms achieve 93 % correct extraction of eyes and mouth, when varying skin color, illumination, and face orientation and 100 % correct extraction of eyes and mouth when varying eye color, shirt color