A boundary-based image segmentation and representation method for binary images

Image processing systems have proven very useful in many space, industrial, military, and medical applications. The goal of any image processing system is to interpret a two-dimensional projection of a three-dimensional scene. Most imaging systems are composed of four stages: segmentation, region description, relational description, and interpretation. A boundary oriented segmentation method for binary images is presented in this dissertation. A curve extraction network that is capable of identifying curve segments present in binary images is described. A line detector based on two state machines that processes an extracted curve segment to output a polyline description of the input curve is also described. The curve extraction network and the state machine based line detector are suitable for hardware implementation. A graph based data structure that is compatible with the segmentation process is developed to represent objects of a digital image. The data structure used for object representation is independent of the specific object to be described. The knowledge representation scheme supports coarse/fine object matching. Scale and rotation change between input object and stored template can be easily resolved using the knowledge representation scheme. A process for identifying a complex object in an input digital image using the object template stored in the knowledge base is demonstrated.