Tag Based Image Search by Social Re-ranking

Abstract

Social networking sites such as Flickr allow users to annotate images using free tags, which contribute significantly to the development and organization of web image retrieval. Searching for tag-based images is an important way to find images that social users have contributed to such social sites. However, how to make the highest result a relevant arrangement and diversity is difficult. In this paper, we propose a social system to rearrange the image retrieval based on the marks taking into account the appropriateness and diversity of the image. We aim to rearrange images according to their visual information, semantic information and social directories. Preliminary results include images contributed by different social users. Each user typically contributes to several images. First we classify these images by rearranging the users. Users who have a higher input to the specified query rank higher. We then serialize the user within the set of user-ordered images, and only the most appropriate image is selected from each user's photo collection. These specific images form the final results retrieved. We are building an inverted index structure for social image data sets to speed up the search process. Experimental results on the Flickr data set show that our social reclassification method is effective and effective.