Abstract: Illumination variation causes the uneven grayscale distribution of the facial image, thus leading a big contract difference in part, then the descent of face recognition. Therefore, in the first place, based on the homomorphic filtering, the paper presents a new method of Gaussian filtering by changing the filtering function. In the following step, it enlarges the dynamic grayscale range of the image through histogram equalization for the sake of extracting the Gabor wavelet feature from the facial image. At last, it recognizes the facial image through nearest neighbor method. The Yale and CMU and PIE database, featuring significant illumination variation, can provide the best result by reducing dimension of facial image and shortening the extracting time.

Key words: illumination variation, Gaussian filter, histogram equalization, Gabor wavelet

摘要： 光照变化易使人脸图像的灰度分布不均，造成局部对比度差别较大，会引起人脸识别正确率下降。为此在同态滤波的基础上，改变滤波函数，提出了高斯滤波的人脸识别方法，接着对滤波后的图像直方图均衡化，来增加图像的灰度动态范围，然后对人脸图像提取Gabor小波特征，最后利用最近邻法识别人脸图像。在光照变换明显的Yale B和CMU PIE数据库识别效果最好，降低了人脸图像的特征维数，缩短了特征提取时间，有效地提高了人脸识别率。

关键词: 光照变化, 高斯滤波, 直方图均衡化, Gabor小波