基于双重局部保持的不完整多视角嵌入学习方法

现有的多视角降维方法多数假设数据是完整的，但该假设在实际应用中难以实现。为解决不完整多视角数据降维问题，提出一种新的不完整多视角嵌入学习方法。基于多视角数据的一致性与同一视角下样本间的线性相关性学习一组重构系数，对缺失样本进行线性重构，通过学习所有视角的公共低维嵌入，保持原始空间的局部几何结构。在此基础上，设计一种惩罚参数来度量重构样本的可靠度，从而权衡缺失样本对学习结果的负面影响。实验结果表明，该方法在Yale、ORL和COIL-20数据集上NMI值分别达到65.63%、73.23%和78.27%，较MVL-IV算法分别提升8.37%、16.71%和20.24%。

Incomplete Multi-view Embedded Learning Method Based on Double Locality Preserving

Most of the existing multi-view dimensionality reduction methods assume that the data is complete, but it is unrealistic for practical applications.In order to solve the problems in the dimensionality reduction of incomplete multi-view data, this paper proposes a new method for incomplete multi-view embedded learning.Based on the consistency of multi-view data and the linear correlation between samples under the same view, a set of reconstruction coefficients is learnt for the reconstruction of the missing samples.Then the common low-dimensional embedding of all views is learnt to maintain the local geometric structure of the original space.Additionally, a penalty parameter is designed to measure the reliability of the reconstructed samples, so as to balance the negative influence of missing samples on the learning results.Experimental results show that the NMI value of the proposed method reaches 65.63%, 73.23% and 78.27% on the datasets of Yale, ORL and COIL-20, increased by 8.37%, 16.71% and 20.24% compared with the MVL-IV algorithm.