Multiscale hierarchical segmentation

In this article three different methods for building a hierarchical graph from scale space images are discussed. The first method has been suggested by Lifshitz and Pizer [8]. It creates graph structures for image description by defining a linking relationship between pixels in successively blurred versions of the initial image. The second method has been proposed by Kuijper [6]. It uses isointensity surfaces through scale space critical points to find a scale space hierarchy and a ”pre-segmentation” of the initial image. The third method is a method with similarities to the ”scale space primal sketch” described by Lindeberg [9]. This algorithm is based on so called attraction areas and it combines the strong points from the two other previously mentioned methods.