基于全局不相关的多流形学习

为提升人脸识别算法的鲁棒性,减少判别信息的冗余度,提出基于全局不相关的多流形判别学习算法(UFDML)。使用特征空间到特征空间的距离,学习样本局部判别信息,提出全局不相关约束,使提取的判别特征是统计不相关的。在Yale,AR,ORL人脸库上的实验结果表明,与LPP(局部保持投影)、LDA(线性判别分析)、UDP(非监督判别投影)等人脸识别算法相比,所提算法的平均识别率高于其它算法,验证了其有效性。

Multi manifold learning based on global uncorrelation

To improve the robustness of face recognition algorithm and reduce the redundancy of identifying information,a discriminant learning algorithm based on global uncorrelated manifold(UFDML)was proposed.The distance of the feature space to feature space and global uncorrelated constraint were used,local discriminant information of samples was learnt to make the extracted discrimination feature statistically uncorrelated.The effectiveness of the proposed algorithm was verified through the experimental results of Yale,AR,ORL face database,which shows that the average recognition rate of the proposed algorithm is higher than that of the other face recognition algorithms such as LPP(locally preserving projection),LDA(linear discriminant analysis)and UDP(unsupervised discriminant projection).