一种基于近邻稀疏表示的人脸识别新方法

A New Method Based on the Nearest Neighbor Sparse Representation in Face Recognition

稀疏表示近些年来被广泛用于人脸识别。由于在现实中,同类图像之间往往不可避免存在光照、姿态、甚至遮挡等差异,如果使用这些有各种差异的图像样本去表示某一特定状态下的图像,则表示的效果势必会受到影响。为进一步提高稀疏表示在人脸识别中的性能,基于原始协同分类(CRC)算法,引入近邻思想,即在各类训练样本中分别寻找与测试样本相近的若干样本,以构建新的近邻样本集;在此基础上进行协同表示,并利用每类样本系数分别重构待测样本,最后基于重构样本集再次协同表示。这种基于近邻样本的二次稀疏重构表示法,使识别更精确,并在一定程度上提升了运行效率。在ORL,YALE,FERET及AR人脸数据库上通过仿真验证了该方法的有效性。

Sparse representation has been widely used in human face recognition in recent years.Because in reality,illumination,pose,even occlusion and the other differences often inevitably exit in the various images,if using these image samples that have various differences to represent the images in a particular state,the effect of representation is bound to be affected.In order to further improve the performance of sparse representation in human face recognition,the article introduce the nearest neighbor thought based on the original collaborative representation classification（CRC）algorithm,namely choosing some training samples that are similar to the testing sample,so as to construct the new sample set for collaboratively representing,and the coefficients for each kind of samples are used to reconstruct the tested sample respectively,finally,using all the reconstructed samples for collaboratively representing again.This secondary sparse reconstruction method based on the nearest neighbor samples makes the recognition more accurate,and to a certain extent,improves the running efficiency.In the ORL,YALE,FERET and AR face database,simulation experiments are carried out,and the results verify the validity of the proposed method.