图像数据库检索中的信息过滤反馈方法

User’s Relevance Feedback Method Based on Information Filtering in Content-based Image Retrieval

利用系统积累的反馈历史数据来改善图像检索的效果引起了越来越多的关注 .该文在分析用户相关反馈记录的基础上 ,结合相关反馈记录中的用户评价数据和其对应的检索样本的图像内容两方面信息 ,提出了一种基于反馈记录的模糊聚类的反馈记录信息过滤分析方法来改进检索性能 .实验显示 ,与现有方法相比 ,该文方法在图像检索的效果和反馈记录的利用效率方面都有明显改善 .

An image retrieval method using the accumulated user's relevance feedback records is presented to improve the performance of image retrieval. The method is based on the semi-supervision fuzzy clustering of the feedback records with information filtering, that is both the user's relevance evaluation and the corresponding query images of the records are used to predict the semantic correlation of the databases images and current retrieval. The merits of the method are as follows: (l) more semantic correlation information can be obtained using the information filtering; (2) with the clustering of feedback records, the efficiency of the analysis of feedback records can be improved. Experiments on the data set of 11,000 Coral images show that the method outperforms the traditional ones.