A Novel Semi-Supervised Dimensionality Reduction Framework
Xin Guo, Yun Tie, Lin Qi, Ling Guan · IEEE Multimedia · 2016
In pattern recognition, when a dataset contains multiple classes, and the structures of the classes are different, single manifold assumption can hardly guarantee the best classification performance. It is more reasonable to assume each class lies on a separate manifold. Here, the authors propose a novel framework of semisupervised dimensionality reduction for multimanifold learning. To address the issue of label insufficiency under the multimanifold assumption, they propose solving three challenging problems: clustering unlabeled samples into different manifolds using sparse manifold clustering, even when they are close to each other; predicting the label of image sets instead of a single sample by calculating the manifold-to-manifold distance; and constructing three kinds of graphs for each manifold to exploit more information from unlabeled samples. During the investigation, they demonstrated that most existing dimension-reduction methods based on manifold learning can be viewed as special cases of the proposed framework. Experimental results verify the advantages and effectiveness of this new framework.