Color Image Segmentation for Region Queries in Image Databases

We present a novel color image segmentation framework, dedicated to region queries in content-based image retrieval (i.e. for queries such as “find me more images containing similar regions”). The goal is to segment the image into a few regions of interest. The novelty of our technique comes from the unification of the feature-space and the imagespace segmentation in a common framework. The method uses no prior modeling of the image nor of the region, focusing on local feature distributions and their spatial stability in a multi-feature, multi-resolution approach. Experiments are presented on real-world imagery, demonstrating the power of the method for segmentation and region queries in image databases.