A Study for Facial Feature Localization and Extraction

• Main Authors: Chin-Yuan Kuo, 郭晉源
• Other Authors: Yih-Ming Su
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/06281341685571701743

碩士 === 義守大學 === 電子工程學系碩士班 === 98 === This research is mainly to develop a system for face detection and facial feature extraction. The algorithm consists of face detection, mouth detection, skin color detection and eye detection. Firstly, a face image with complex background is extracted to use skin color detection and remove the background. Then, rectangle feature extraction and AdaBoost classification approaches are used to localize the face region. Furthermore, the eye and lip candidate region are divided using facial geometrize relationship. In the eye candidate region, a template matching approach is used to obtain the position, and then the range of the lip candidate region is narrowed to achieve better result. The best lip region is extracted using edge projection and color difference between the lip and skin color after detecting lip candidate region. The result shows that the use of skin color detection can not only reduce the computation time for face detection, but also have high level of fault tolerant in the complex environment. The experiment results used self-made videos to do the test of facial feature position and selection. The system performed 15 frames/s, and also achieved the overall detection rate of 75 %.