一种基于区域检测的图像检索相关反馈方法

摘要

This paper proposed a relevance feedback approach in image retrieve system .The approach is based on the detection of interest region of the query user .This approach makes use of the positively relevant images which are labeled by user and learn the user intention of the query so that the return images satisfy the user better .This approach can be divided into 3 steps:1 ) match the features of positively relevant images to those of the query image;2) verify the matched points using RANSAC model;3) select the interest region according to the verified points .As the interest region has been selected , one can use this region or put more weight on the features in this region as a new query image .The results of the experiments show that this approach can improve the precise of the retrieval and return images that satisfy the user better .%提出一种基于用户兴趣区域检测的图像检索相关反馈学习方法，利用用户反馈为正相关的图像进行学习，进而猜测用户意图，获得更令用户满意的检索结果。该反馈学习方法的框架如下：1）对正相关的反馈图像与查询图像进行特征匹配；2）对匹配的特征使用RANSAC模型进行校准；3）进行区域选取。在选取了兴趣区域以后，可以将该区域直接抠取或将其中的特征点权值增强作为新的待查图像进行检索，提高检索的精度。实验表明本文方法可以返回更符合用户心意的检索结果。