【摘要】 相关反馈技术是基于内容的图像检索领域近期研究的热点之一。现有方法通过用户给出的相关反馈图像来寻求唯一的全局最优检索点与权重。由于图像内在的复杂性,唯一的全局检索点通常难以准确覆盖用户的检索目标子空间,使得反馈对检索效果的改进表现出不稳定性。为解决这一问题,本文使用模糊聚类的方法对反馈图像集进行分析,并在此基础上给出了一个基于反馈图像的多聚类中心的相关反馈方法。实验对比显示,该方法能明显改善图像检索的效果且具有较强的稳定性。更多还原

【Abstract】 The user’ s relevance feedback technique is t he most promising approach to improve the effect of content-based image retrieval. However, the performance of relevance feedback technique in image retrieval is not stable in many settings, and more feedback information does not mean more improving definitely. To address this problem, an unsupervised fuzzy clustering method is employed to classify the user’s relevance feedback images into several sub-classes, and each sub-class center point will be used to modify the original query. The score of each image is obtained according to the fuzzy set rules. A prototype system has been implemented to evaluate our method. Experiment results show that the efficiency of user’s relevance feedback can be improved apparently.更多还原