基于半监督典型相关分析的多视图维数约简

为了有效地在半监督多视图情景下进行维数约简,提出了使用非负低秩图进行标签传播的半监督典型相关分析方法。非负低秩图捕获的全局线性近邻可以利用直接邻居和间接可达邻居的信息维持全局簇结构,同时,低秩的性质可以保持图的压缩表示。当无标签样本通过标签传播算法获得估计的标签信息后,在每个视图上构建软标签矩阵和概率类内散度矩阵。然后,通过最大化不同视图同类样本间相关性的同时最小化每个视图低维特征空间类内变化来提升特征鉴别能力。实验表明所提方法比已有相关方法能够取得更好的识别性能且更鲁棒。

Semi-supervised canonical correlation analysis based multi-view dimensionality reduction

In order to efficiently reduce dimensionality in multi-view semi-supervised scenarios, semi-supervised canonical correlation analysis methods which using nonnegative low-rank graph to propagate labels are proposed. Global linear neighborhoods captured by nonnegative low-rank graph can utilize information from both direct and reachable indirect neighbors to preserve the global cluster structures, while the low-rank property retains a compressed representation of the graph. After label information of unlabeled samples are estimated by label propagation algorithm, soft label matrices of all samples and probabilistic within-class scatter matrices in each view are constructed. Then, by maximizing the correlations between samples of the same class from cross views and minimizing within-class variations in the low-dimensional feature space of each view simultaneously, discriminative power of features are enhanced. Experimental results demonstrate that the proposed methods can achieve better recognition performances and robustness than existing related methods.