Theory and design of an on-line minicomputer-based image processing system

Abstract

Current reserach activity at Rensselear Polytechnic Institute in varied and unique applicatiosn of image process techniques has prompted the design of a flexible and efficient facility dedicated to these tasks. This flexibility and efficienct facility dedicated to these tasks. This flexibility and efficiency have been achieved largely through the development of special purpose, modular processors and associated on line, interactive software packages.

Photographs document the operation of the system. Schematics and theory of design are provided for the many hardware subsystems. Examples of system software development are also included. Conclusions recommend further areas of research in image processor design, both in terms of general goals and terms of in specific augmentations of the present system.

The integrated conceptual development and hardware/software design of the Rensselaer On-Line Image Processing System (ROLIPS) are dcoumented in this Thesis. The goal is twofold: the exposition of a theory of system design together with a detailed description of its actual implementation in hardware and in software. Major innovations of this design include the incorporation of a special purpose scan processor for the formatting of data by oriented line segments and other structural elements, the design and implementation of alternate image data management schemes (file-oriented and record-sequential), and the augmentation, through hardware, of the CPU instruction set to include image I/O macro-operations. The design parameters for each of these innovations are developed from theoretical considerations of intrinsic image properties. The use of these properties in the design of possible new arithmetic/logic units is also discussed.

Image processing systems based on general purpose processors and raster-scan image I/O devices are limited in efficiency and flexibility by te nature of their component devices. These limitations may be traced to three aspects of the system: the nature of the scan operation, the complements of I/O and arithmetic/logic instructions, and the management of data storage resources. A fixed, raster scan is often undesirable in that it imposes a directionality on image data which is not a property of the image itself. It is also inefficient from the viewpoint of maximizing the rate of convergence of sequential estimation schemes. A general purpose instruction set is poorly suited to image processing in that basic I/O and artihmetic macro-operations such as scan line access or digital filter operations require long stirngs of machine code. Finally, data management schemes in most operating systems are poorly suited to the establishment and access of large, two-dimensional data files.