基于小波的人脸去光照识别算法研究

Using Wavelet Reconstructed Face for Recognition

研究了面部图像的多级小波分解与重构,分析了光照、表情等变化对小波分解低频近似系数的影响,指出了光照变化对小波低频近似系数影响最大,表情和个体差异的影响次之。在此基础上提出应用PCA变换的方法来重建光照影响小的小波低频近似分量,以此来替换原图的小波低频近似系数,再通过小波重构得到去光照人脸。在识别时,运用特征脸的方法对重构脸进行了特征提取与识别研究,在AR人脸库上进行了测试,并对不同小波基进行了比较,结果表明该方法能有效地去除光照等因素影响,识别效果得到了较大地提高。

The wavelet decomposition and reconstruction on face image were analyzed, and the fact that illumination had more affections than expression on the low frequency approximation coefficients was given, followed by expression. Based on this, a novel reconstruction method was proposed, which replaced the low frequency approximation coefficients of illumination and expression faces by the PCA rebuilt vector, then a new illumination unrelated face was reconstructed. And the followed recognition system, based on the classic eigenface method shows its good performance on AR face database.