In an earlier section we concentrated on the problem of determining boundaries between regions by detecting local discontinuities and grouping them to form object boundaries. In this section, we consider the dual problem of mapping individual pixels into sets of pixels of similar properties (intensity, colour, texture etc.) which define the regions within a continuous image. Region growing is probably less common than edge detection as a low level processing operation of computer vision systems, but it is applicable in environments which are highly textured or coloured, for example outdoor scenery viewed by a mobile vehicle. (The results of a Canny edge detector applied to a tree or forest is difficult to interpret, even if applied at multiple scales.)
Several approaches to regional segmentation have been developed; here we consider a local and global approach, by analogy with the different approaches to contour definition. First, we examine an approach that starts from the assumption that all pixels are separate regions, and attempts to merge pixels into regions or groups of similar pixels.
[ Contents |
Region growing: a recursive approach ]
Comments to: Sarah Price at ICBL.