葛宏伟

个人信息Personal Information

教授

博士生导师

硕士生导师

主要任职:计算机科学与技术学院党委书记

性别:男

毕业院校:吉林大学

学位:博士

所在单位:计算机科学与技术学院

学科:计算机应用技术

办公地点:创新园大厦A832

联系方式:hwge@dlut.edu.cn

电子邮箱:gehw@dlut.edu.cn

扫描关注

论文成果

当前位置: 中文主页 >> 科学研究 >> 论文成果

Support vector description of clusters for image annotation

点击次数:

论文类型:会议论文

发表时间:2011-11-19

收录刊物:EI、Scopus

摘要:Image annotation is a challenging problem due to the rapid growing of real world image archives. In this paper, we propose a novel approach to the solving of this problem based on a variant of the support vector clustering (SVC) algorithm, i.e., the support vector description of clusters. The system has two major components, the training process and the annotating process. In the training process, clusters of image manually annotated by descriptive words are used as training instances. Each cluster is described by a one-cluster SVC model. The proposed model can exploit the advantage of SVC for its ability to delineate cluster boundaries of arbitrary shape. Moreover, the training process of the one-cluster SVC model is formulated as the process of building density estimator for underlying distribution of the cluster. In the annotating process, for a test image, the probability of this instance being generated by each model is computed. And then the relevant words are selected based on the obtained probabilities. Simulated experiments were conducted on the Corel60k data set. The results demonstrate the performance of the proposed algorithm, compared with the performance of other algorithms.