We show how the Expected Patch Log Likelihood (EPLL) proposed by Zoran and Weiss can be reinterpreted as the law of the global image conditioned by a global coherence between overlapping patches. This helps to better understand and to rewrite the patch aggregation step common to all patch-based image processing models, with applications both in image restoration and image editing.

This is poster number 33 in Poster Session

Authors:

Alexandre Saint-Dizier (Universite Paris Descartes)

Keywords:

bayesian methods, epll, image deblurring, image enhancement, image reconstruction, image representation, patch