Image Re-ranking based on Topic Diversity

Abstract

Social media sharing sites allow users to annotate images with free tags, which greatly contribute to the development of web image retrieval. Searching for tag-based images is an important way to find the images users share on social networks. However, how to make the highest result a relevant arrangement and diversity is difficult. In this paper, we suggest a multi-order approach to image-based image retrieval with a view to enhancing coverage performance. First, we build a graph of the sign based on the similarity between each sign. The community detection method is then performed to remove the community theme for each tag. After that, an arrangement is made between the community and within the community to obtain the final results retrieved. In the process of arrangement between communities, the adaptive random walk model is used to arrange the community based on the multiple information of each subject society. In addition, we build an inverted index structure for images to speed up the search process. Experimental results on the Flickr data set and NUS-scale data sets show the effectiveness of the proposed approach.