Object-oriented Hierarchical Image Vectorization

We present proximity graphs based approach to hierarchical image segmentation and vectorization. Our method produces an irregular pyramid that contains a stack of vectorized images of successively reduced levels of detail. We are jumping off from the over-segmented image represented by polygonal patches, which are attributed with spectral information. We employ constrained Delaunay triangulation combined with the proximity and closure principles known from the visual perception to extract the initial polygonal patches. They are built upon a triangular mesh composed of irregular sized triangles. We then represent the image as a graph with vertices corresponding to the polygons and edges reflecting polygon relations. This is followed by the iterative graph contraction based on Minimum Spanning Tree. The graph contractions merge the polygonal patches based on spectral differences between neighboring polygonal patches. The approach can be generalized to the multi-criteria MST to integrate other factors important for polygon agglomeration, in addition to spectral proximity considered in this investigation. An important characteristic of the approach is that initial and agglomerated polygonal patches are built in a way to retain spatial relationships among spectral discontinuities present in the original image. * Corresponding author.