DCT域上人脸光照正则化及识别  

Lighting normalization for face recognition in DCT domain

作　　者：许嘉伟[1] 冯国灿[1] 

机构地区：[1]中山大学数学与计算科学学院,广州510275

出　　处：《计算机应用》2009年第1期47-50,共4页journal of Computer Applications

基　　金：国家自然科学基金资助项目(60675007);教育部重点基金资助项目(104145)

摘　　要：现有的光照正则化处理算法,都是在空间域中进行的,为避免海量图像解压缩的时间消耗,在JPEG图像上直接进行光照正则化处理,提高人脸识别效率,在DCT域上,基于三维辐照度方程,把差图像法推广到了DCT域上,并在DCT域上提出了分量图像法。实验表明:差图像法与分量图像法均能在DCT域中有效地削弱光照方向对人脸识别的负面影响。All the existing lighting normalization algorithms are implemented in spatial domain. In order to save the time in images decompression and increase the speed in face recognition, we carried out lighting normalization for JPEG hnages directly. In this paper, we applied the Differential-image algorithm in the Discrete Cosine Transform （DCT） domain based on the 3-dimensional irradiance equation, and proposed Component-image method in DCT domain. Experiments show that the Differential-image algorithm and Component-image method in DCT domain can weaken the negative impact of face recognition from light directions effectively, improve the efficiency for face recognition in JPEG images.

关 键 词：离散余弦变换 人脸识别 光照正则化 差图像 分量图像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]