Improving image retrieval precision using combination of circular reranking and time-based reranking

Abstract

Search reranking is regarded as a common way to boost image retrieval precision. The problem is not simple especially when there are multiple features to be considered for search, which often happens in image retrieval. This paper proposes the combination of Circular reranking and Time-based reranking methods for improving the precision of image retrieval. Circular reranking utilises multiple features of an image, usually textual and visual descriptions, for reranking. Initially, it will conduct multiple runs of random walk for obtaining initial search results. Secondly, two features of an image are exchanged for better mutual reinforcement which makes multiple keyword search possible. Lastly, reranked results are attained through exchanging the ranking scores among different features in a cyclic manner. Time-based reranking is based on the count of Time, View and Download, of an image. Time count is the time duration between opening and closing of an image. View and Download counts are the total number of views and downloads respectively for an image. In our approach, Time-based reranking is performed on the Circular reranked list for improving precision, appropriately combining features of both reranking methods, while retrieving images during search.