网络图像标注中多视图半监督稀疏特征选择算法研究

项目名称： 网络图像标注中多视图半监督稀疏特征选择算法研究

项目编号： No.61502143

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 计算机科学学科

项目作者： 史彩娟

作者单位： 华北理工大学

项目金额： 21万元

中文摘要： 面对爆炸性增长的网络图像，自动图像标注技术成为对网络图像进行有效检索、组织和管理的一个重要途径。本项目的研究目标是通过多视图半监督稀疏特征选择算法研究来提升网络图像标注的准确度和效率。具体研究内容：(1) 研究多视图数据间的一致和互补属性，同时充分利用多视图间的一致性和互补性构建多视图半监督稀疏特征选择方法；(2) 研究多视图结构化稀疏表示，同时考虑不同视图，以及同一视图中不同特征的重要性，构建多视图结构化稀疏特征选择算法；(3) 为了避免构建图拉普拉斯矩阵的高计算代价，研究一种自适应半监督学习方法，从而实现快速的自适应半监督稀疏特征选择。本项目的特色是基于多视图学习、结构化稀疏表示和半监督学习等，对多视图半监督稀疏特征选择算法进行深入研究，通过提升特征选择的性能来提升网络图像标注的准确度和效率。

中文关键词： 网络图像标注；稀疏特征选择；多视图学习；结构化稀疏表示；半监督学习

英文摘要： Facing the explosive growth of the web images, automatic image annotation has become an important way for effectively retrieving, organizing, and managing web images. This project aims to enhance the accuracy and efficiency of web image annotation by researching on multi-view semi-supervised sparse feature selection. The specific research contents include: (1) With research on consistent and complementary properties of multi-view data, the consistency and the complementarity between different views will be made full use simultaneously to construct multi-view semi-supervised sparse feature selection algorithms; (2) With research on multi-view structured sparse representation, a multi-view structured sparse feature selection algorithm will be constructed by taking into account the importance of different view and the importance of different features in the same view; (3) In order to avoid the high computational cost of constructing the graph Lapalacian matrix, an adaptive semi-supervised learning method will be studied, and than a adaptive semi-supervised sparse feature selection algorithm will be proposed. The characteristic of this project is to deeply research the multi-view semi-supervised sparse feature selection algorithms based on multi-view learning, structured sparse representation, and semi-supervised learning to enhance the feature selection performance, and then to enhance web image annotation accuracy and efficiency.

英文关键词： Web Image Annotation;Sparse Feature Selection;Multi-view Learning;Structural Sparse Representation;Semi-supervised Learning