Counter-Examples for Bayesian MAP Restoration

摘要

Bayesian MAP is most widely used to solve various inverse problems such as denoising and deblurring, zooming, reconstruction. The reason is that it provides a coherent statistical framework to combine observed (noisy) data with prior information on the unknown signal or image. However, this paper exhibits a major contradiction since the MAP solutions substantially deviate from both the data-acquisition model and the prior model. This is illustrated using experiments and explained based on some known analytical properties of the MAP solutions.