CONTROLLING KNOWLEDGE-BASED IMAGE SEGMENTATION USING AN ITERATIVE SPATIAL DATA STRUCTURE CONSTRUCTION ALGORITHM

In this paper, we present a control algorithm for knowledge-based segmentation of multidimensional image data. Based on the iterative expansion of spatial data structures, the algorithm provides a controlled environment during segmentation for accessing and manipulating image data at different spatial resolution scales. The control algorithm (1) facilitates the delivery and application of information and knowledge of diverse sources to the image classification decision making processes; and (2) helps to coordinate and balance the progress of image segmentation in order to reduce processing-order dependence of the final results. We prove that by choosing a sequence of consistent and progressively more stringent ''stopping criteria'', the control algorithm, once established, will work with any arbitrary image of a given type. The effectiveness of this control algorithm has been demonstrated through test results from a prototype system with 3D biomedical image data.