(3,4)-weighted skeleton decomposition for pattern representation and description

摘要

A digital pattern, perceived as the superposition of elongated regions, is decomposed into simple regions through the decomposition of its (3,4)-weighted skeleton. The skeleton is interpreted as a curve in 3D space, where the three coordinates of any pixel are its planar coordinates and the distance label. The 3D curve is divided into rectilinear segments, which constitute the spines of elementary regions, i.e. regions with linearly changing width and orientation. Then, the spines are analysed to simplify the skeleton decomposition and avoid redundancy. Spines identifying regions unnecessary for the description of the pattern are annihilated, while contiguous spines, corresponding to sufficiently similar regions, are merged. The resulting skeleton components are used to represent and describe the simple regions into which the pattern is decomposed. Decomposition at different resolution levels can be obtained by selecting different threshold values during the polygonal approximation, performed to divide the skeleton into rectilinear pieces, and/or the successive merging step.