| A major problem in implementing an image understanding system (such as a photo-interpretation system or a robot vision system) is to segment images into disjoint regions of uniform gray tone or texture. This dissertation presents a new image segmentation expert system (ISES) approach for solving this problem based on the use of both edge and texture features.; The ISES is composed of two parts: (1) image processing tools which process images by localized brightness compensation (LBC), heuristic edge extraction (HEE), and region formation (RF), and (2) a knowledge-based controller which is composed of an inference engine, rule base, and hypothesis base. The LBC is a preprocessing operation for an input image for improving the image quality. The HEE is used to obtain spatially accurate, one-pixel-wide boundaries of uniform regions. The RF operation contains a new version of the A* algorithm (minimum-cost search) and a modified Fisher's pattern classification scheme based on the simultaneous use of both edge and texture features. The combined simultaneous use of the edge detection and texture classification, under the knowledge-based control of the ISES, makes the segmented-region boundaries one pixel wide, spatially accurate, without edge gaps, and without noisy micro-edges inside segmented regions. Thus it achieves high-quality region formation (i.e. image segmentation).; The software of the ISES was implemented on the PRIME 750 Computer and DeAnza image display system in the RPI Image Processing Laboratory. It was also implemented on a VAX/VMS 785 Computer and Raster Technologies image display system in the Image Science Laboratory, Kodak Research Laboratories, Rochester, N.Y. A number of test sample images are given in the dissertation. |
