Polygonal approximation of outlines by scale-based dominant point detection

This paper improves on previous methods for representing contours of images in scale-based polygonal approximations. The method adds more outline shape to the lower scale levels than seen elsewhere. Other methods (such as Teh and Chin (1989), Pei and Lin (see Pattern Recognition, vol.25, no.11, p.1307, 1992)) place polygon vertices only at the outline dominant feature points, whereas here we sample outlines at a spacing that depends upon the scale-level as well as the dominant points. This allows the outline sample density to be more uniform and better related to the scale-level and prevents the approximating polygons from losing details of the outline shape too quickly. The first stage of the process converts a raster scanned image into a vectorised representation of its object outlines. Then a scale-space map of the dominant turning points in the outlines is constructed, using scale-space filtering. Next, back-tracing and point-adding techniques are used to produce more accurate and better scale-related approximations. Finally, the contour points are output, from higher to lower scale-level, to provide a hierarchy of coarse-to-fine polygonal approximations of the outline at different scales