Social images tag ranking based on visual words in compressed domain

Abstract

With the introduction of many image compression standards, the social images are stored and transmitted in compressed formats such as JPEG. For large-scale image database, tag ranking must fully decompress the compressed data to predict tag relevance based on visual content. In order to improve the accuracy of tag ranking and further reduce the ranking time, social images tag ranking based on visual words in compressed domain is proposed in this paper, which includes three steps: (1) low-resolution social images are constructed from the compressed image data; (2) visual words are created according to extracted SIFT descriptors in low-resolution social image; (3) the neighbor voting model is utilized to rank the image tags after matching the similarity based on visual words of an image. In order to evaluate the performance of the proposed method, average NDCG (normalized discounted cumulative gain) and tag ranking time are compared. Experimental results show that the proposed method can significantly reduce the time of image tag ranking under ensuring the ranking accuracy of social image tags.