一个改进的球面谐波模型及其在光照人脸识别中的应用

An improved spherical harmonics model for face recognition under varying lighting

从根本上解决人脸识别中的光照问题,仅从人脸的平面图像考虑是不够的,需更多地利用人脸的纹理和形状等本质特征。提出用中性人脸的形状来表示特定人脸的形状,并通过计算球面谐波基图像进行光照估计,实现光照补偿。对真实照片去光照的处理和在Yale B人脸库上的实验结果表明,该方法有效地利用了人脸的形状特征,能比较准确地估计出人脸的光照,明显地去掉光照不均的影响,识别率得到较大提高。

IN order to improve the accuracy of face recognition under varying lighting, the essential features of human faces, such as texture and shape are more important than that of 2D face images. An average face was used to describe the shape of special ones. Illumination estimation achieved by calculating spherical harmonies basic image was used for lighting normalization. Experiments were done on both real photographs and the Yale B face database. The results show that this approach could achieve more efficiency of using facial shape characteristics, better accuracy of the face illuminations estimation. and lower influence of illuminations. It greatly improves the performance of face recognition under varying lighting.