In image processing and analysis new parallel computer architectures have proved necessary since the classical sequential general propose computer often is to slow. The very large data volumes and the complexity of the processing make great demands on storage capability as well as processing power. The PICAP system of Linköping University has been designed to meet the demands in an interactive image processing environment. Being a multi-processor system it achieves its speed through global but also local parallelism. A virtual image memory alleviates the problem of storage space. An extensive menu-based command language makes interaction with the system a simple matter also for the less experienced user.

The enormously large amount of data in image processing obviously calls for special processors as well as for effective algorithms. This paper describes a new pipe-lined processor, the real-time relational processor, ROP, that implements neighborhood relational processing in full video-speed. The processor use relations between pixels to form the lowest level, the microstructure, of the relational hierarchy. The main parts of the processor are present and examples on algorithms, that runs in real-time by the processor, from different fields of image analysis is introduced. The new algorithms make template matching on graylevel images, direction-labelled edge detection and selective noise-filtering respectively.

Image processing puts heavy demands on computational speed and storage space. The PICAP system of Linkoping University has been designed to meet these demands in an interactive image processing environment, utilizing state-of-the-art hardware and software components. PICAP is a multiuser system, and computational speed is achieved through parallelism at both global and local levels. A virtual image memory alleviates the problem of storage space. The users can invoke the PICAP facilities by means of a menu-based language or a general-purpose high-level language. The modular and open-ended architecture of PICAP makes it highly adaptive to future advances in both hardware and software technologies.

A parallel picture processing machine (PPM) is presented. The proposed machine can perform local operations of both logical and arithmetical character on three by three neighborhoods of digitized pictures. It is essentially a two-dimensional machine but in a restricted sense it can be regarded as three-dimensional. The instruction repertoire is thoroughly discussed and several program examples are presented. In thelast section the implementation of the machine is described and v various tradeoffs between speed and economy are discussed. A dynamically programmable microprogram store is shown to give great freedom in program design.