CLAHE融合低频DCT系数重变换的人脸识别算法

Face Recognition Algorithm Based on Fusion CLAHE and Low Frequency DCT Coefficients Retransformation

针对人脸识别中由于伪装(如围巾、太阳镜和头发)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CLAHE融合低频DCT系数重变换的人脸识别算法。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。在扩展Yale B、FRGC V2.0及一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。实验结果表明,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。

Towards the problem that serious affect the recognition rate of face, such as disguise （e.g. , scarves, sunglasses and hair） and facial shade by other objects, the face recognition algorithm based on fusion CLAHE and low frequency DCT coefficients transformation was proposed. Firstly, divided the image into several non-overlapping local patches, used the histogram equalization to do local contrast stretching on local sub-block and realized denoising. Secondly, with the appropriate number of low frequency DCT coefficients to eliminate the illumination change in face image. Finally, with kernel principal component analysis, to carry out feature extraction, and with the nearest neighbor classifier, completed the final face recognition. The effectiveness and robustness of the proposed algorithm were experimentally verified in extended Yale B, FRGC V2.0 and an outdoor face database. The experimental results showed the proposed algorithm achieved higher recognition rate when dealing with robust face recognition, compared with several kinds of linear algorithms.