An information theoretic approach to image segmentation [content based image retrieval applications]

An ever increasing amount of image data presents both the need and challenges for content based image retrieval. Image segmentation is required to support querying at the level of objects. In this paper, we present image segmentation as a problem of maximizing the information by segmentation. The segmentation is done by classifying the image features in the feature space and measuring the information gained by the classification by an evaluation function. An important feature of the evaluation function is that it provides a direct way to incorporate the limitation of our perception ability. Experimental results on general real world images demonstrate the effectiveness of our algorithm.