基于光照补偿空间的鲁棒人脸识别

Illumination compensation subspace based robust face recognition

如何应付光照条件变化是人脸识别的主要研究课题,文中提出了一种基于单幅注册图像的光照不变人脸识别方法.文中首先分析证明了同一个主体的不同光照图像之间的差别,可以通过一个与身份和光照独立的子空间进行线性描述,我们称之为光照补偿空间;然后介绍了光照补偿空间的构建,并且提出了一种简单有效的基于光照补偿的人脸识别算法,将图像光照差异补偿与人脸识别过程统一在一起,通过一次1优化就可以完成人脸身份的识别.文中方法在没有明显增加优化复杂度的前提下,有效降低了光照差异对于人脸识别算法精度的影响.文中提出的方法在Extended Yale B图像库上进行了验证.

How to deal with illumination variation is an important research topic of face recognition. In this paper, an illumination invariant face recognition strategy is put forward to deal with single gallery recognition. We give a proof that the illumination distinction among face images of s specific people could be linearly described through a carefully constructed subspace that is independent of both appearance and illumination, which is called the illumination compensation subspace. And then the subspace construction approach is introduced. Finally, a straightforward yet effective illumination compensation based face recognition algorithm is proposed which unifies both the calculation of illumination compensation and face recognition process into once l1 optimization. In a word, our approach effectively reduces the influence of illumination diversity on face recognition accuracy without incurring evident increase in optimization complexity. The proposed method has been evaluated on the Extended Yale B dataset.