Automatic Facial Expression Analysis System

• Main Authors: Ming-Wey Wu, 吳明衛
• Other Authors: Shu-Mei Guo
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/40832107848444450601

碩士 === 國立成功大學 === 資訊工程學系碩博士班 === 91 === Most facial expression recognition systems restrict that the color of background must be simple, the face in an image have been identified and localized, or the facial feature must be manually extracted. Differing from previous systems, we propose a fully automatic facial expression analysis system that automatically detects faces with different sizes in complex background, extracts fourteen feature points, and recognizes four kinds of facial expressions — happy, angry, surprised and neutral. Three parts in the system are developed to extract facial expression information for the fully automatic recognition. The first part is face detection with evolutionary computation, which can find the best-fit ellipse to cover the face contour and include the greater area of skin color. The second part is facial feature extraction using the preceding ellipse information to get fourteen feature points, two points in each eyebrow, three points in each eye, and four points in the mouth. The third part is facial expression recognition with the back-propagation neural network to recognize four motions (i.e. happiness, anger, surprise and neutral). Our proposed system has been tested on the famous JAFFE (Japanese Female Facial Expression) database and the self-shoot database. The experimental results with the overall recognition accuracy of 86% for the JAFFE database and 74% for the self -shoot database demonstrate the effectiveness of the proposed system.