Approximate Translational Building Blocks for Image Decomposition and Synthesis

Abstract

We introduce approximate translational building blocks for unsupervised image parsing. Such building blocks are frequently appearing copies of image patches that are mapped coherently under translations. We exploit the coherency assumption to find approximate building blocks in noisy and ambiguous image data, using a spectral embedding of co-occurrence patterns. We quantitatively evaluate our method on a large benchmark data set and obtain clear improvements over state-of-the-art methods. We apply our method to texture synthesis by integrating building blocks constraints and their offset statistics into a conventional Markov Random Field model. A user study shows improved retargeting results even if the images are only partially described by a few classes of building blocks.