A novel approach to link semantic gap between images and tags via probabilistic ranking

Nowadays there is tremendous fame of social networking sites that leads to a broad investigation in tag-based social image search. Both visual features and tags plays important role in the research field. Existing approach uses tags and visual features consecutively or independently to estimate the relevance of images. In this paper we tackle the problems of Automatic Image Annotation (AIA) for image retrieval that delivers the global image search engine that covers different applications such as Image to Image Retrieval, Image to Tag suggestions, Tag to Image Retrieval and Tag to Tag suggestion which is based on user contributed tags on the websites such as flicker. The proposed approach comprises of global feature extractions of images. Then we predicate relationship between image and tag by using asymmetric graph as a probabilistic view. Once the relationship is found ranking is employed with the help of t-step random walk model. Finally pseudo relevance feedback is applied on the ranked images. Additionally experimental analysis of above model is conducted on Corel image dataset to illustrate the effectiveness of this approach.