Abstract: Illumination invariants extraction based on illumination estimation is an effective method to improve the performance of face recognition under complex illumination conditions. The past algorithms, estimating illumination from a digital image based on the slowly changing characteristic of light, are unable to obtain accurate illumination and illumination invariants. In this paper, based on the principle of imaging, the slowly changing characteristic of light, the complex illumination environment, image fusion and smoothing filtering, an effective method for illumination estimation and illumination invariant extraction from a face image is proposed. The proposed algorithm can deal with the shadow edge well, and extract the illumination invariants containing many facial details, which is closer to the original facial features. The experimental results on the Yale B+ and the CAS-PEAL-R1 face databases under complex illumination environment show that the proposed algorithm is effective.

Key words: illumination estimation, illumination invariants, face recognition

摘要： 基于光照估计的光照不变量提取是提高复杂光照人脸识别性能的一种有效方法。以往算法仅考虑光照缓慢变化特性从人脸图像中估计光照，无法获取准确的光照和光照不变量。综合考虑图像的成像原理、光照缓慢变化特性和复杂照明环境，结合图像融合和平滑滤波，提出一种有效的人脸图像光照估计、光照不变量提取方法。所提算法能较好地处理阴影边缘问题，提取含有丰富面部细节特征、更接近于人脸本征的光照不变量。复杂光照Yale B+和CAS-PEAL-R1人脸库上的实验结果表明所提算法具有高效性。

关键词: 光照估计, 光照不变量, 人脸识别