Diversifying Image search results

Image re-ranking is the process which reorders the original text based search results by identifying the intrinsic relationship between the images and query term to generate a new order list which would better meet the user needs. The Image search result diversification aims at providing a set of results which covers a wide range topics related to the user query. An efficient information retrieval system should return relevant as well as diversified results to the user. In real world applications, the images returned for a certain query are naturally in hierarchical organization, rather than a simple parallel relation. By identifying all the topics that are related for the given query and returning the images from all these topics can achieve the diversity in the top ranked images from the search engine. This paper propose an approach which identifies all the related terms for the given user query term by the Tag Frequency-Inverse Document Frequency algorithm. In addition, the Google knowledge graph Application Programming Interface is also used to identify the related terms for a given query. The images obtained for those related terms are refined further to obtain only the relevant images. Then the images are selected by means of Hierarchical Topic Aware Re-Ranking model for returning the diverse image search results to the user. Such a diversified result can improve the efficiency of the web image search engines.