人脸识别的光照预处理算法

Illumination pretreatment algorithm of face recognition

针对光照变化对人脸图像的改变,降低人脸识别系统的识别率的问题,提出改进的光照预处理的方法,将图像用小波变换取低频成分,再进行双直方图均衡化处理,提高图像的光照对比度;然后对同一张图像,用高斯高通滤波器取高频成分,对图像进行信号增强;再对两种处理后的图像进行一定比例融合,用空域锐化方法再进行图像特征增强。用主成分分析(principal component analysis,PCA)方法进行降维,线性鉴别分析(linear discriminant analysis,LDA)方法进行特征提取。实验结果表明,在小训练样本情况下,较经典PCA方法错误率可下降25%左右。

Change of illumination would alter face image recognition and decrease the recognition rate. An improved illumination pretreatment method is put forward by using wavelet for low frequency component of the image and double histogram equalization processing to improve the image illumination contrast. Then the high frequency components of the same image are gained with Gaussian high- pass filter,and the image signal is enhanced. Then two kinds of processed image are fused to a certain proportion,and the image features are enhanced with the airspace image sharpening method. The Principal Component Analysis is used for dimension reduction and Linear Discriminant Analysis is used for feature extraction. Experimental results show that in the case of small training samples,the error rate can decrease 25% compared with classical PCA method.