Automatic image annotation is significance for image understanding and retrieve of web image, so it becomes the new hot research topic in recent years. This paper proposes an effective annotation method based on the Diverse Density (DD) algorithm and keywords correction. The method includes two sub-processes, basic image annotation and annotation refinement. In the basic label process, we use the improved DD algorithm to find the visual feature vector for some semantic concept. The general DD algorithm uses all instances in the positive bags as start points of the optimization process, which will greatly increase the computing time. We cluster the same visual feature regions in the positive bags and use the clustering centers as start points instead of all instances. Moreover, the negative instances have been used to guild the selection of start point. Then, we integrate the improved DD algorithm into the Bayesian framework to realize the initial image annotation. In the annotation refinement, the correlations between keywords are added to refine those candidate annotations from the prior process. Finally, experimental results and comparisons on the Corel image set are given to illustrate the performance of the new algorithm.

Date Published: 8 July 2011
PDF: 6 pages
Proc. SPIE 8009, Third International Conference on Digital Image Processing (ICDIP 2011), 80091L (8 July 2011); doi: 10.1117/12.896094

Published in SPIE Proceedings Vol. 8009:
Third International Conference on Digital Image Processing (ICDIP 2011)
Ting Zhang, Editor(s)