Computer vision is firmly tied to one representation for images: an array of samples measured on a rectangular grid. For some work, this restriction matters little; but for much low-level processing, the representation is crucial in determining the operations that can readily be carried out. The author suggests that the exploration of alternatives may be worth some effort, and to support this he investigated one possibility: the logarithmically sampled image. This representation has interesting properties and can support a number of useful operations. The author shows in particular how straight line detection and a fairly general form of two- dimensional matching may be performed using it, and I suggest ways in which the work may be developed.
This paper is not available online