FBCC: An Image Similarity Algorithm Based on Regions

摘要

Region-based image retrieval has been an active research area for the past few years. A good similarity measure that combines information from all regions is very important for region-based retrieval systems. In this paper, we propose FBCC (Foreground-Background Corresponding Comparison), a novel image similarity measure based on region comparison. The basic idea is comparing query foreground regions with database foreground regions and query background regions with database background regions. Three factors have been considered in the algorithm: the comparable credit between two regions, the significance of each region and the difference of total number of regions between two images. Experimental results on a testbed of 10.000 general-purpose images show that this approach is effective for center-surround images.