Cartoon-Texture-Noise Decomposition with Transport Norms

Christoph Brauer, Dirk Lorenz

Abstract: We investigate the problem of decomposing an image into three parts, namely a cartoon part, a texture part and a noise part. We argue that norms originating in the theory of optimal transport should have the ability to distinguish certain noise types from textures. Hence, we present a brief introduction to optimal transport metrics and show their relation to previously proposed texture norms. We propose different variational models and investigate their performance.