受限直方图均衡化的低频DCT系数重变换算法  

Low Frequency DCT Coefficients Retransformed Algorithm of Contrast Limited Adaptive Histogram Equalization

作　　者：魏锐[1] 徐亮[1] 

机构地区：[1]黄淮学院信息工程学院,河南驻马店463000

出　　处：《实验室研究与探索》2015年第5期16-20,35,共6页Research and Exploration In Laboratory

基　　金：国家科技支撑计划项目(2012BAH12B01);河南省教育厅重点科技攻关项目(13A520786)

摘　　要：针对现实人脸识别中由于光照、表情、姿态或其他物体引起的面部遮挡而严重影响识别率的问题,提出了受限直方图均衡化的低频DCT系数重变换算法。首先,将图像划分成多个互不重叠的局部小块,使用受限直方图均衡化对局部子块进行局部对比拉伸以实现去噪;然后,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;最后,利用核主成分分析进行特征提取,最近邻分类器完成最终的人脸识别。在ORL、扩展Yale B及1个户外人脸数据库上的实验验证了所提算法的有效性及鲁棒性,表明相比几种线性表示算法,本文算法在处理鲁棒人脸识别时取得了更高的识别率。In view of the reality in face recognition due to illumination, expression, pose or other objects caused by facial shade seriously affects the recognition rate of the problem, this paper puts forward a method for some heavy low frequencies by using discrete cosine transform （DCT） coefficients retransformed algorithm of contrast limited adaptive histogram equalization. Firstly, the image is divided into several non-overlapping locally small blocks, and denoising is achieved by using the contrast limited adaptive histogram equalization about local sub-block for local contrast stretching; Secondly, using the appropriate number of low frequency DCT coefficients eliminates the illumination change in face image; Finally, by using kernel principal component analysis for feature extraction, the nearest neighbor classifier completes the final face recognition. By ORL, extended Yale B and an outdoor experiment on face database the effectiveness of the proposed algorithm and robustness is verified. The experimental results show that compared with several kinds of linear algorithm, this algorithm in dealing with a robust face recognition has achieved higher recognition rate.

关 键 词：人脸识别 自适应直方图均衡化 低频离散余弦变换 系数重变换 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]