There are many photos sharing websites like Flickr which allow users to annotate pictures with descriptive keywords called tags. Tag based image search helps to find images contributed by users in such social media sharing websites, which significantly support to the development of the web image retrieval and organization. How to make top positioned ranked result relevant is the challenging problem. For that we re-rank images according to visual data, semantic data and social clues. The inter-user re-ranking and intra-user re-ranking methods are used to achieve a good trade-off between the diversity and relevance performance. First we sort the images contributed by different users using inter-user ranking. Then we apply intra-user ranking method on ranked image set. Here, the most relevant image of each user is chosen from the image set. And these images form the final result. We construct an inverted index structure for image dataset to improve the searching process. Keywords: image Search; relevance; re-ranking; social media; tag based image retrieval
I. INTRODUCTIONSocial media has taken the world by storm through dozens of websites and other forms of technologies improving the way people communicate with each other. There are many social media sharing websites that have millions of members allowing them to make connections, share images, search and more on a regular basis. Sites like Flickr allow users to create and share media information as well as describe the created content with tags. Figure 1 shows an example of a social image associated with user providing tags. Tags play as a major character in searching of images. Tag related image search is an effective method than content based image search. It provides a development in web image retrieval. However tags contributed by users may incomplete or ambiguous. But this is not unexpected because of the uncontrolled influence of tagging, diversity of knowledge and cultural background of the users. Applying the text based retrieval method may form unsatisfied results. Therefore, a ranking method that can explore tags and the images may provide a better social image search result. The ranking issues in tag based image retrieval have gained wide alertness among researchers in recent years. However, the ranking innovations face following difficulties in its development. First issue is tag mismatch. In social tagging all users have to label their uploaded images with their own tag in social network and share with others. Every user has their own habit to tag photos. Therefore, even for same image there will be several different tags. Thus many irrelevant tags are introduced with uploaded image .Second one is query ambiguity. Users cannot describe perfectly with their single keywords, so add little information to a user's contribution. In addition, equivalent words and polysemy are alternate reasons for the query uncertainty. Images taken at the same interval and fixed spot are fairly similar. How to tackle these issues in re-ranking is the fundamental ...