Circular pictures, their digitisation and processing

Abstract

Nearly all of the previous work in visual pattern recognition has been based upon Cartesian images, that is those in which addressing is performed using Cartesian co-ordinates. In many instances circular patterns are evident and, in such cases, polar-co-ordinate addressing is both more natural and more convenient to use. The paper lists various applications areas which generate such pictures and chief among these is theautomatic visual inspection of industrial artefacts. Several techniques for digitising polar images are presented, including flying-spot scanners, rotating-image methods, and spiral-scan television. A variety of `polar? picture-processing techniques are discussed, including: (a) tangential lowpass and highpass filtering of polar images (b) radial lowpass and highpass filtering of polar images (c) `circular? filtering of Cartesian images (d) rectangular filtering of polar images (e) chain-coding of polar images (f) correlation in polar images (g) frequency and sequency analysis (h) gradient operators (i) integration along circular or radial paths (j) finding the centre of a circular picture