Abstract: The performance of face recognition is seriously impacted by illumination, expression, posture and occlusion variations, for which low frequency Discrete Cosine Transform（DCT） coefficients retransforming based on Contrast Limiting Adaptive Histogram Equalization（CLAHE） is proposed. Original images are divided into some non-overlapping patches and CLAHE is used to do local contrast stretching so as to reduce noise. Illustration variation of face image is removed by reducing suit numbers of low frequency DCT coefficients. Kernel principle component analysis is used to extract features. Nearest neighbor classifier is used to finish classification and recognition. The effectiveness and reliability of proposed algorithm have been verified by experiments on ORL,extended YaleB and AR face database. Experimental results show that proposed algorithm has higher recognition accuracy than several advanced standardized technologies.

Key words: robust face recognition, adaptive histogram equalization, Discrete Cosine Transform（DCT）, coefficients retransforming, Kernel Principal Component Analysis（KPCA）

摘要： 针对光照、表情、姿态、遮挡等变化显著影响人脸识别系统性能的问题，提出了基于限制对比度自适应直方图均衡化（CLAHE）的低频离散余弦变换（DCT）系数重变换算法。将图像划分成多个互不重叠的局部小块，使用CLAHE对每个局部小块进行局部对比拉伸以实现去噪，通过缩减适当数目的低频DCT系数来消除人脸图像中的光照变化；利用核主成分分析进行特征提取，采用K-最近邻分类器以完成最终的人脸识别。在ORL、扩展YaleB和AR人脸数据库上的实验验证了所提算法的有效性和鲁棒性，实验结果表明，相比其他几种较为先进的人脸识别技术，所提算法取得了更高的识别率，同时大大降低了识别所用时间。

关键词: 鲁棒人脸识别, 自适应直方图均衡化, 离散余弦变换, 系数重变换, 核主成分分析