Image Search Using Features and Social Ranking

Researchers are gaining interest in matching the textual query with visual pictures and their encompassing texts or tags for quick image search. The system proposes a social re-ranking feature for tag-based image retrieval with the thought of images relevance and diversity. The system extracts relevant images based on given image input or tags input. For image retrieval visual features like SURF GIST and hierarchical wavelet packet descriptor are used. The system applies re-ranking of pictures according to their visual information semantic info and social clues. The initial results include images contributed by completely different social users. Typically every user contributes many images. Initially system sorts these images by inter-user re-ranking. Users that have higher contribution to the given query rank higher. Then it consecutively implements intra-user re-ranking on the ranked users image set. The system works in two phases: Training and testing. In training phase image visual information semantic info and social clues are extracted and saved. In testing phase test image or tags are matched with training data and re-ranking algorithm is applied to extract top k matched images. The proposed system is enforced and tested using java Experimental results on Flickr dataset show that our social re-ranking methodology is effective and economical.