基于二次多项式模型的人脸光照归一化

 根据人脸光照变化非线性的特性,本文提出用二次多项式模型来描述非正面光照条件下的人脸图像与正面光照条件下的人脸图像在对应象素点灰度之间的关系,进而提出一种人脸光照归一化方法.本方法的一个重要特点是独立于先验的物理模型而通过建立统计回归模型来学习不同光照图像之间的关系.此外,提出用基于PCA的方法对光照归一化后的人脸图像进行加权补偿,进一步改善图像的视觉效果.在Extended YaleB和CMU-PIE人脸库上的实验结果表明,新方法在改善人脸视觉效果同时能大大提高人脸识别准确率.

Illumination Normalization of Human Face Based on the Second-Order Polynomial Model

We proposed to describe the relationship between the pixel grey values of the face images under frontal and non-frontal illumination conditions by using a second order polynomial model.Correspondingly,an illumination normalization method based on such nonlinear model was formed.The proposed method learns the illumination variations in a statistical manner by using regression model without any prior physical knowledge.Furthermore,in order to improve the visual quality,a PCA-based weighting compensation for the normalized face image was proposed.The experimental results on the Extended Yale B and CMU-PIE face databases show that the proposed method can attain good visualization for face images,and significantly improve the face recognition performance.