Abstract: Extracting illumination invariants from face image is an effective way to improve the robust-illumination performance of incomplete training sample face recognition. The previous algorithms extracting the high frequency characteristics of face image from different angles as illumination invariants can’t get integral essential feature of face and have some limitations. In this paper, from the perspective of feature level and decision level fusion, a complex illumination face recognition method based on multi-feature fusion is proposed. The proposed algorithm can utilize the different advantages of several illumination invariants and improve the robust-illumination performance of face recognition under complex lighting environment. The experimental results on the Yale B + and the uncontrolled lighting face database show that the proposed algorithm is effective.

Key words: illumination invariants, multi-feature fusion, principal component analysis, face recognition

摘要： 提取人脸图像光照不变量是提高不完备训练样本人脸识别光照鲁棒性的一个有效途径。以往算法分别从不同角度提取人脸图像的高频特征作为光照不变量不能提取完整的人脸本征，具有一定的局限性。从特征级和决策级融合的角度提出了一种基于多特征融合的复杂光照人脸识别方法。所提算法能发挥不同光照不变量的自身优势，明显提高复杂光照人脸识别的光照鲁棒性。Yale B+和非控光照人脸库的实验结果表明所提算法的有效性。

关键词: 光照不变量, 多特征融合, 主成分分析, 人脸识别