提出了一种新的光照不变人脸识别的图像预处理算法称为分段局部归一化方法(SLN)。其思想是对图像像素分段，使得每段中各像素对应的物体表面点具有相近的表面法向量分布，因而对光源具有相似的灰度响应，然后局部归一化在各段中进行以削弱光照影响。该算法首先建立物体的朗伯（Lambert）表面反射模型，用奇异值分解方法估计出人脸形状的平均表面法向量分布矩阵，根据法向量方向利用聚类算法对像素进行分段，然后在各段中进行局部的像素归一化处理，最后传统的人脸识别算法如PCA在归一化后的图像中进行。在Harvard和YaleB人脸图像库中的识别试验表明，该算法能有效地提高在非均匀光照条件下的人脸识别率。

A new preprocessing algorithm for illumination invariant face recognition called Segmented Local Normalization (SLN) is proposed. The main idea is to produce image segmentation so that in each segment, pixel points have similar surface normal distribution and then have similar intensity responses to the light source. Then the local pixel normalization is processed in each segment in order to eliminate illumination. The algorithm firstly establishes Lambert object surface reflection model and secondly a general face surface normal matrix is estimated using SVD. Then the clustering algorithm based on the surface normal directions is used to obtain the image segments, and a local normalization is applied in each image segment. Finally, the traditional face recognition algorithm like PCA is applied on the normalized images. Experimental results based on the Harvard and YaleB face database show that under uneven illumination conditions, the algorithm can increase the face recognition rate efficiently.