一种基于视觉词典优化和查询扩展的图像检索方法

Image Retrieval with Enhanced Visual Dictionary and Query Expansion

视觉词典方法 (Bag of visual words,Bo VW)是当前图像检索领域的主流方法,然而,传统的视觉词典方法存在计算量大、词典区分性不强以及抗干扰能力差等问题,难以适应大数据环境.针对这些问题,本文提出了一种基于视觉词典优化和查询扩展的图像检索方法.首先,利用基于密度的聚类方法对SIFT特征进行聚类生成视觉词典,提高视觉词典的生成效率和质量;然后,通过卡方模型分析视觉单词与图像目标的相关性,去除不包含目标信息的视觉单词,增强视觉词典的分辨能力;最后,采用基于图结构的查询扩展方法对初始检索结果进行重排序.在Oxford5K和Paris6K图像集上的实验结果表明,新方法在一定程度上提高了视觉词典的质量和语义分辨能力,性能优于当前主流方法.

The most popular approach in image retrieval is based on the bag of visual-words （BoVW） model. However, there are several fundamental problems that restrict the performance of this method, such as low time efficiency, weak discrimination of visual words and less robustness. So, an image retrieval method with enhanced visual dictionary and query expansion is proposed. Firstly, clustering by fast search and finding density peaks are used to generate a group of visual words. Secondly, non-information words in the dictionary are eliminated by Chi-square model to improve the distinguishing ability of the visual dictionary. Finally, an efficient graph-based visual reranking method is introduced to refine the initial search results. Experimental results of Oxford5K and Paris6K datasets indicate that the expression ability of visual dictionary is effectively improved and the method is superior to the state-of-the-art image retrieval methods in performance.