Neighborhood Operators: An Outlook

Abstract

The goal of a great part of the research activity in digital image processing is twofold: 1) to design and efficiently implement algorithms for image analysis and description and 2) to improve the understanding of the human visual and perception system. An old discussion about the usefulness of these two aspects -and of their complementarity -is still alive since for some people, the ways in which Nature has established to perform vision should have no bearing on the ways in which an artificial vision system should be designed. Conversely, the group having the opposite view claims that evolution is a significant process which has optimized the solution of a computational problem such as the one posed by vision (in 3-D, with colour and, in some instances, with motion). As examples of these two different ways of thinking, the first one may be represented by the work done by M.L. Minsky, especially on computational geometry(1) and, another example is the work on mathematical morphology by (2) J. Serra , whilst the second one, namely the biologically based approach is well (3) depicted by the work of B. McCormick starting in the early days of the ILLIAC computers up to the interesting proposal of a highly parallel structure for real time image processing(4). The three cultural areas which have been mentioned reflect three important viewpoints: that geometry in the discrete needs special mathematics, that set theory can be used to define a picture algebra and that a good knowledge of the physiology of the human vision system can be used for redesigning computer architectures which will perform computation at levels which are close to the ones in natural systems. The reason for mentioning these approaches is that they all share one common point: that the traditional Von Neumann architecture is not adequate to perform artificial vision operations neither at a reasonable efficiency nor at a level of flexibility similar to the human system.