Syntactic Algorithms for Image Segmentation and a Special Computer Architecture for Image Processing

Abstract

Abstract : Several efficient algorithms for image recognition and segmentation and a new computer architecture for image processing are proposed. The algorithms are 'Syntactic' in that they perform structural or spatial analysis rather than statistical analysis, and a 'grammer' is inferred for describing the structures of patterns in an image. Depending on the requirements of the problem, an appropriate grammatical approach is used by the syntactic algorithm. A finite-state string grammar is applied to the image recognition of highways, rivers, bridges, and commercial/industrial areas from LANDSAT images. There are two major methods in the string grammar approach for image recognition; namely, the syntax-directed method and syntax-controlled method. For the syntax directed method, syntactic analysis is performed by a template matching which is directed by the syntactic rules. For the syntax - controlled method an automation which is directly controlled by the syntactic rules is used for the syntactic analysis. A tree grammar is applied to the image segmentation of terrain and tactical targets from LANDSAT and infrared images respectively. The tree grammar approach utilizes a tree automation to extract the boundaries of the homogeneous region segments of the image. The homogeneity of the region segment is obtained through texture measurements of the image. The computer architecture proposed is a special purpose system in that it can perform an image processing task on several picture-points of an image at the same time, and thus takes advantage of the fact that image processing tasks usually exhibit 'parallelism'. This architecture uses a distributed computing approach.