Enhanced contour chain-coding and topological hierarchy analysis in a single-pass raster scan

In this paper we present an enhancement of Cederberg´s raster-scan approach for generating chain-encoded representations of contours of segmented regions and their inherent topological hierarchy tree within a binary image. The presented algorithm uses a 3 × 3 neighborhood window to build chain-code of all possible contour pixel arrangements of objects and holes within objects in a single pass through the image. The amount of template comparisons for each pixel window has been drastically reduced by taking basic interdependencies between a line-wise scanning beam and borders of binary objects into account. Furthermore, the complete topological hierarchy tree is established by inferring the global surroundness relations among the region borders from the previous obtained local information. With these main characteristics, the algorithm is well suited for efficient parallel processing implementations in streaming applications (e.g. in FPGAs) in that it avoids unpredictable memory accesses to a video frame buffer as can be found with the border following technique. Moreover - on the basis of the hierarchy tree - an effective pre-selection of contours can be established in that only objects or holes down to a defined child-level will be further processed, enabling significant bandwidth-savings for subsequent storage or transmission purposes.