Contour coding through stretching of discrete circular arcs by affine transformation

This paper presents a method for coding binary contour images. Coding is lossy in nature. A contour is first decomposed into line and are segments through an extraction of some of its dominant pixels (called the key pixels including the pixels of inflexion). Detection of inflexion points (pixels) is made through the Gaussian circle and its image. Each contour are is then approximated by suitably stretching a locally defined discrete circular are through affine transformation(s). Stretching can be performed iteratively. To get discrete circular arcs, a generation algorithm has been presented along with their properties in conjunction with rings and discs. Performance of the coding algorithm has been compared to that of a method which reconstructs the contour image approximately. Comparison with the performance of exact or lossless methods has also been carried out to show how far it is away from them. (C) 2000 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.