On hierarchical segmentation for image compression

The present paper describes a hierarchical image segmentation scheme keeping in mind its use in image compression. At each level of hierarchy, the segmentation provides a sub-image consisting of compact, homogeneous regions. A number of thresholds based on conditional entropy of the image guides the entire process. Small regions are merged. Objective measures based on correlation and contrast have been proposed for evaluation of the segmentation technique, and the result of the proposed algorithm has been compared with those of three different existing multi-level thresholding algorithms. (C) 2000 Elsevier Science B.V. All rights reserved.