摘要
近年来多视图学习引起了研究者的广泛关注。在多视图学习中,数据主要来自于多个视图(或特征集)。多视图数据的最大优点是可以从不同视图之间提取互补信息。传统多视图学习方法是在不同视图上单独地训练分类器。这些方法利用了视图之间的互补信息,但是忽略了去除不同视图之间的冗余信息。为了解决上述问题,提出一种基于多视图核鉴别分析的识别方法。该方法通过基于核判别分析从各个视图中提取出相互正交的投影矩阵,从而能够提取出兼具互补和无冗余的特征。在AR和Oxford Flowers17公共数据库上的实验结果验证了所提算法的有效性。
Multi-view learning has caused wide public concern of researchers in recent years. In multi-view learning, data is mainly from many views (or feature set). The biggest advantage of multi-view data is that it can extract complementary information from different views. The traditional multi-view learning method learns classifiers in different views independently. These methods utilize the complementary information between views,but ignore the redundant information between different views. In order to solve the above problem, a recognition method based on multi view kernel discrimiuant analysis is proposed. It uses kernel discriminant analysis to extract projection matrix from each view and makes the transformations orthogonal, so that it can extract both complementary and non-redundant features. Experimental results on public database like AR and Oxford Flowers17 verify the effectiveness of the algorithm proposed.
出处
《计算机技术与发展》
2016年第12期92-95,共4页
Computer Technology and Development
基金
国家自然科学基金资助项目(61272273)
关键词
多视图学习
互补信息
冗余信息
核鉴别分析
multi-view learning
complementary information
redundant features
kernel discriminant analysis