对半偶图像作LDA变换的人脸识别方法

对于人脸图像而言，偶图像相当于原始图像与它的镜像的和。为了增大训练样本，许多人脸识别算法把镜像图像加入到训练集中。与这些方法不同，本文把原始训练集中的每一幅图像都变换为它的半偶图像得到新的训练集。对于新的训练集，采用LDA(Linear discriminant analysis)方法提取特征后作分类。在ORL和IIS人脸库上的实验结果表明，对半偶图像作LDA的人脸识别方法有较高的识别率。

Applying LDA on Half-Even Images for Face Recognition

Any function can be decomposed into an odd function and an even function. For a face image, its even image can be obtained by adding its mirror image to its original image. To enlarge the training set, many face recognition algorithms are used to add mirror images to the training set. Different from those methods, the proposed method in this paper substitutes each training image with its half even image. Using the new training set, linear discriminant analysis (LDA) is employed for the feature extraction. Experimental results on ORL and IIS face database show that the proposed method can achieve better performances.