Knowledge-based image segmentation
Chih-Chen Hung · 1990
Computer vision systems have numerous military, space, industrial and medical applications. In general, computer vision systems understand, or try to understand, a three-dimensional scene image. Basically, image analysis consists of four stages: segmentation, region description, relational description and matching. Image segmentation is the process of partitioning an image into meaningful regions. More than twenty years of research have not yet produced a truly reliable image segmentation system which can handle various imaging conditions and scene contents. Since the conventional approach has not produced the desired results, a new approach is needed. A knowledge-based approach to image segmentation and design of a system based on this approach is presented in this dissertation. The system consists of five major components: (1) knowledge base, (2) image classifier, (3) segmentation controller, (4) segmentation evaluation module and (5) library of procedures. The classifier simplifies the segmentation process by mapping the input image to one of the predefined conceptual views. The knowledge base supports dynamic segmentation strategy. With the help of the evaluation module, the controller is capable of recovering from errors. The blackboard architecture allows efficient inter-module communication. The system is suitable for software as well as hardware implementation. Simulation results strongly support the theory and prove that the knowledge-based approach can significantly improve segmentation.