Enhancing the web image search results through direct and indirect relevance model
Saravanan Arumugam, S. Sathya Bama · AIP conference proceedings · 2022
Due to the Internet revolution, search engines have become an inevitable part of our everyday life. However, retrieving required information in the form of text and images is still a challenging task. Unlike extracting text content from the web, extracting the images and ranking them based on the query is even more difficult as the number of images on the web is increasing day by day due to social media sites. Several methods exist in fetching the relevant images from the web. However, most of the methods are ineffective in retrieving relevant images as it involves outliers due to polysemy problem. This paper presents the framework for improving the performance of the search engine concerning the web image search results. The method utilizes direct as well as indirect relevance score computation for the images using text based and context based information. The direct relevance score computation is carried out based on the given query and the text representation of the images whereas, indirect relevance score computation uses the topmost relevant topics exist in the text documents and the relevant topics exist in the contextual content of an image thereby eliminating the duplicate images. Based on the experiments performed, the proposed method improves the results of the image search significantly with proper re-ranking of images. The overall increase in the precision for the proposed model than other models under comparison is around 5% to 10%.