Gabor特征均值的子空间人脸识别算法的改进

针对人脸图像局部特征提取不充分的问题,在基于子空间分析的人脸识别算法中,提出了在线性和非线性空间中实现基于2DGabor均值的子空间人脸识别算法.首先,根据人脸图像的5个特殊区域,对人脸图像进行分块处理,分别对每一块进行2DGabor运算,并把每个训练样本相应像素点得到的特征矢量取均值,得到图像的特征向量,然后在线性和非线性空间中利用2DPCA(two-dimensional principle component analysis)和KDA(kernel fisher discriminant analysis)对特征向量进行降维处理,最后利用最近邻分类器和支持向量机分类器SVM(support vector machine)进行特征分类与识别,通过对ORL和FERET标准人脸库图像进行的实验仿真即对比结果表明,基于2DGabor均值的方法不仅提高识别率,而且对于人脸光照、姿态和表情变换均具有良好的鲁棒性.

Face Recognition Using 2DGabor Mean Values in Subspace

According to the defect of the local feature extraction in face recognition,a novelty subspace method of face recognition based on the local 2DGabor mean value implementation in linear and nolinear space was proposed.Firstly,each facial image in the training sample set is divided according to the five special face regions and then the feature of five key regions was extracted through 2DGabor,mean value was calculated from feature vectors gained from corresponding pixel of every test sample,secondly,2DPCA（two-dimensional principle component analysis） and KDA（kernel fisher discriminant analysis） are used to decrease the dimension of the gained eigenvector in linear and nolinear space,finally the nearest neighbor classification and SVW（support vector machine）are adopted to recognize the face images.The numerical experiments on facial database of ORL and FERET show this method achieves better effect of face recognition than other methods and shows stronger robustness to changes of illumination,expression,poses and so on.