Automatic image annotation is the process by which keywords are automatically assigned to a digital image. However, results of the state-of-the-art image annotation methods are still far from satisfactory. To tackle this issue, this paper proposes a novel approach to automatically refine
the original annotations of images by salient object detecting and neighbor voting. For a given image with original annotations, we segment it to non-overlapping block-based structures and then execute the refined structure clustering process to find salient objects from training dataset.
Afterwards, terms of salient objects together with candidate annotations are submitted to social image community, and then returned images with user-supplied tags are used as neighbors to vote for each candidate annotation. Hence, final annotations are constructed by terms of salient objects
and the candidate annotations with higher voting score. Experimental results conducted on Corel5K dataset and Flickr photos demonstrate that our method is efficient to improve the performance of image annotations.
ADVANCED SCIENCE LETTERS is an international peer-reviewed journal with a very wide-ranging coverage, consolidates research activities in all areas of (1) Physical Sciences, (2) Biological Sciences, (3) Mathematical Sciences, (4) Engineering, (5) Computer and Information Sciences, and (6) Geosciences to publish original short communications, full research papers and timely brief (mini) reviews with authors photo and biography encompassing the basic and applied research and current developments in educational aspects of these scientific areas.