基于Gabor特征的快速贝叶斯人脸识别  

Rapid Bayesian Face Recognition Using Gabor Transform

作　　者：牛丽平[1] 闫林[1] 郑延斌[1] 薛占鳌[1] 

机构地区：[1]河南师范大学计算机与信息技术学院,河南新乡453007

出　　处：《控制工程》2012年第6期1055-1058,共4页Control Engineering of China

摘　　要：对贝叶斯分类中两种计算相似度的公式进行简化,设计了基于Gabor特征的快速贝叶斯人脸识别方法。首先对人脸图像进行Gabor小波变换,把同一个人的Gabor特征与本人的类内Gabor均值图像之差作为类内差异图像,对类内差异图像进行小波分解取其低频图像;然后采用2DPCA方法建立类内特征子空间。对测试图像进行Gabor特征提取,并与目标图像的类内均值Gabor特征作差形成差异图像,对差异图像进行小波分解,提取低频子图信息投影到类内特征子空间,利用简化的贝叶斯准则进行分类。实验结果表明,该方法降低了算法复杂度,提高了识别率,并对光照和表情变化具有较强的鲁棒性。In Bayesian classifier, the probabilities are small enough to be well beyond the minimum value. So, to simplify the similarity computation in the Bayesian algorithm, Ta rapid Bayesian approach based on Gabor filter is proposed. Firstly, it utilizes the convolution of the images and the Gabor filters to extract features and different images formed by Gabor features of images and the average of within class. And then, difference images are decomposed by applying wavelet transform. Bayesian recognition is processed using low frequen- cy sub-band in 2DPCA space. Experimental result shows that the method has the advantages of simple computation and high recognition rate under expression and illumination changes.

关 键 词：人脸识别 贝叶斯分类 GABOR变换 小波变换 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]