Mapping algorithms on to platforms: an approach to algorithm and hardware co-design

The aim of this paper is to bring together aspects of our recent work in machine vision. We discuss the relative merits of various platforms for the implementation of machine vision algorithms with particular emphasis on the DIP chip, a custom device designed following an in- vestigation into the computational requirements of a range of image manipulation algorithms. We will show: (i) how certain operations may be made easily avail- able in hardware; (ii) that algorithms may be redesigned to take this into account; (iii) that high performance processors other than work- stations are the natural platforms for these algorithms; and that (iv) that such platforms will be necessary to convincingly demonstrate the potential scalability of these algorithms.