基于DCT域内拉普拉斯值排序的人脸识别方法

摘要

基于DCT域内的人脸识别方法的关键是如何选择有效的DCT系数，提出了一种基于DCT域内拉普拉斯值排序的人脸识别方法。首先将图像划分为若干个大小相同的子块，对每一个子块进行DCT变换，得到分块DCT系数，然后利用拉普拉斯值作为局部保持能力判据选择那些能够很好保持样本流形结构的分块DCT系数，最后对选定的DCT系数执行LPP算法提取识别特征，在ORL和Yale人脸数据库上的实验结果证明了该方法的有效性。%The key of face recognition in DCT domain is how to select effective DCT coefficient.For this purpose,a method for face recognition in DCT domain with Laplacian Score ranking is proposed.Firstly,the image is divided into several blocks with the same size.For each block,DCT is used to obtain block DCT coefficient.Then effective block DCT coefficient is selected according to locality preserving power criterion with Laplacian Score.Ultimately,LPP is performed on the selected block DCT coefficients to extract recognition features.The experiments on ORL and Yale face database shows that the improved method is effective.