In this paper, we present a novel region-based imag e analysis platform. With a human in the loop, it provides an efficient method for object extraction, image segmentation, and filtering. The region space is organised into a Binary Partition Tree represen tation where small homogenous regions are successively merged. The hindsight of the merging process is th erefore stored within the tree structure and the evolutions from the seed to root nodes are studied. By analysing these evolutions, discontinuities caused by the merging of heterogeneous regions are identified and the corresponding tree nodes short listed. These nodes indicate salient details within the image and from them tree simplification and region based filtering becomes possible. Given a simpler version of the original tree, manually assisted object extraction and image segmentation can be easily achiev ed and the experimental results show that our method facilitates the extraction of semantic content from the image.