随着高维数特点的非结构化数据大量涌现，降维在数据处理过程中越来越重要。为了充分挖掘数据中潜在的信息，我们提出了一种基于实例级和约束块级信息的判别性半监督降维方法IDSDRC。该算法同时利用成对约束的实例级以及约束块级信息，并综合考虑数据样本的局部几何结构。基于Yale、ORL和AR标准人脸数据库进行人脸聚类的实验，实验结果显示IDSDRC算法比一些现有的半监督降维算法效果好，验证了该算法的有效性。%With high-dimensional unstructured data available in great numbers,dimensionality reduction is more and more important in data processing. In this paper,an instance-level based discriminative semi-supervised dimensionality reduction method with chunklets named IDSDRC is also proposed, which aims to simultaneously use both instance-level and chunklet-level information together with unlabeled data for dimensionality reduction. Experimental results on standard face databases for face clustering show that IDSDRC is efficient and superior to several established semi-supervised dimensionality reduction methods.
展开▼
