In recent years much effort has been devoted to the study of picture processing algorithms and pictorial pattern recognition. The efforts have resulted in both theoretical understanding of and practical approaches to some of the basic problems involved.
Common to the areas of picture processing is the great need for processing power. One picture may contain as much as several million picture elements (pixels). A scene viewed through a common television camera for example, represents approximately two megabits of data renewed 25 times per second. To produce or analyze data at such rates is completely beyond the capacity of the general purpose computer. Even if the speed is increased by a factor of ten, wich may be anticipated in the future, the GPC will fall short when faced with real-time applications. The problems stern from the fact that pictures represent a new type of data, orders of magnitude richer in information than that for which the GPC's architecture was designed.
Many attempts have been made to design a computer architecture that is more suited for picture processing [3-6] [9-11] and in some cases the machines have also been constructed.
In the following we will restrict ourselves to an analysis of the computational problems of pictorial pattern recognition. Starting with a survey in this chapter, the principles and motivations behind the PICAP processor will be described in chapter 2 followed by the PICAP implementation, examples of PICAP operation and evaluation in chapters 3, 4 and 5.