人脸表情自动识别技术研究进展

Research Advances in Automatic Facial Expression Recognition

作为生物特征识别与情感计算领域的一个极富挑战性的交叉课题,人脸表情自动识别技术在各种应用的推动下发展很快,但鲁棒的自动人脸表情识别系统至今尚未建立.人脸表情识别的3个关键环节是人脸检测、人脸特征定位与提取和人脸表情的情感分类.在上述关键环节上均取得了重要进展.需要解决的问题包括提高人脸检测算法的鲁棒性、人脸特征提取的针对性和准确性、人脸刚性运动分离和三维表情识别.人脸表情数据库建设和应用研究,也是今后研究的重点.提出了人脸表情自动识别系统评价指标,包括技术的实用性、成果的可比性、系统的专用性和实时性.

As a challenging interdiscipline of biologic feature recognition and affection calculation, the technique of automatic recognition of facial expression （ facial expression automatic recognition system, FEARS） develops quickly driven by demands of various applications. Nevertheless, fttUy automatic facial expression recognition systems with acceptable robustness have not yet come forth due to the great difficulties. The three key procedures for automatic recognition of facial expression are facial image detection, location and extraction of facial features and emotion classification, and great advances have been achieved on the three fields. The remaining problems include improving the robustness of facial recognition algorithms, precise and pertinence of identification of facial features, separation of rigid facial actions and recognition of three-dimensional facial expressions. To build the data bases of facial expressions and application research are also important. Indexes, such as applicability, comparability and appropriative and real-time performances, are proposed for evaluating automatic recognition systems of facial expressions.