The performance of the atlas term in MRF-MAP based DfD algorithm

Depth from Defocus (DfD) is an effective method to infer the depth information of a scene. This technique requires a focused and defocused image of the same scene. The state of the art DfD literature uses an energy function optimization method based on the Markov Random Field model with Maximum a Posteriori statistics (MRF-MAP). Typically this energy function consists of two terms: first, the absolute difference term and second, the neighborhood term. This paper describes the algorithm modified with a new term, which we call the atlas term. In this research, we explore how this term behaves in the DfD algorithm to improve accuracy and running time.