Effective image object retrieval and tag refinement by augmenting unsupervised semantic features

Image Retrieval is an evolving technique because of the rapid growth of images and videos in Social Medias such as Flickr and Facebook. To retrieve images user gives query as image or keyword. But these techniques provide low recall rates because of the varying lighting conditions, viewpoints and unambiguous tags by the user. In this paper we have tackled the problem by leveraging both visual and textual features in order to provide image retrieval effectively. We introduced a framework to augment each image with semantic features (Visual and Textual) by using graphs among images. Images are retrieved by propagation and selection on visually and textually similar graphs. Tag of images are also refined using semantic features. Experimental results confirm that proposed framework effectively retrieve images comparing to previous methods.