Hybrid thinning through reconstruction

One difficulty with many pixel-wise thinning algorithms is that they produce unacceptable results at junction points, or in the presence of contour noise. The authors present a novel approach to detecting ambiguous regions in a thinned image. The method uses the reconstructability properties of appropriate thinning algorithms to reverse the thinning process and automatically detect those pixels which may have resulted from more then one stroke in the image. The ambiguous regions are then interpreted and reconstructed using domain specific or derived contextual information. The approach has the advantage of using local methods to rapidly identify strokes (or regions) which have been thinned correctly and allowing more detailed analysis based on non-local methods in the remaining regions.