Graph cut based optimization
Many problems of early vision and computer graphics can be formulated in terms of energy minimization. In this thesis, we focus on improvement of an algorithm, that solves the energy minimization for two labels via graph cuts and whose implementation is avaliable as GridCut library. We extend the GridCut by expansion move algorithm, which solves the energy minimization for multiple labels, and apply it to 2D grid-like graphs with 4 and 8 connected neighboring system and 3D grid-like graphs with 6 and 26 connected neighboring system. We compare the resulting implementation with the GCO library.