改进的基于Gabor-LDA的人脸识别方法

鉴于Gabor特征对光照、表情等变化比较鲁棒,并已在人脸识别领域取得成功应用,提出了一种改进的Gabor-LDA算法.首先对人脸图像进行多方向、多尺度Gabor小渡滤波,然后对得到的特征向量使用改进的主成分分析方法(PCA)变换降维,采用自适应加权原理重建类内散布矩阵和类间散布矩阵,从而改进了最佳鉴别分析(LDA)判别函数,有效地解决了训练样本类均值与类中心的偏离问题.对Yale人脸库的数值试验表明,该算法比传统算法有更好的性能.

Improved face recognition method based on Gabor-LDA

Since Gabor feature is robust to illumination and expression variations and has been successfully used in face recognition. A novelty method of face recognition based on Gabor-LDA is presented. First, the proposed method decomposes the face image by con-volving the face image with multi-orientation and multi-scale Gabor filters to extract the eigenvectors. And then an improved algorithm of principal component analysis (PCA) is proposed. This algorithm is used to decrease the dimension of the eigenvector. The method of adaptively weight is used to reconstruct the within-class scatter matrix and between-class scatter, and then the linear diseriminant analy-sis (LDA) function is improved. The problem of the class mean of training samples deviates from the center of this class is resolved by this improved LDA discriminate function. The numerical experiments on facial database of YALE show this method achieves better per-formance of face recognition than traditional methods.