A Non-Local Pairwise Energy Based Model For The HDR Image Compression Problem

Max Mignotte
DIRO, Département d'Informatique et de Recherche Opérationnelle, CP 6128,
Succ. Centre-Ville, P.O. 6128, Montréal (Québec), H3C 3J7.


ABSTRACT

In this work, a new Gibbs energy based compression model for the display of high-dynamic-range images. The proposed tone mapping method tends to exploit the biologically-inspired dynamic retina concept which is herein mathematically expressed via an image representation based on the specification of the statistical distributions of the non-local gradient magnitude. In this framework, the contrast reduction problem is expressed by a Gibbs energy with non-local pairwise pixel interactions whose cost function is minimized by a conjugate gradient descent procedure.

comparison with the bilateral filter


Images
The following images are courtesy of Raanan Fattal

belgium.hdr
synagogue.hdr

The following images are courtesy of Paul Debevec

memorial.hdr

The following images are courtesy of Jack Tumblin

bigFogMap.hdr


The following images are courtesy of Fredo Durand

smalldesignCenter.hdr
smallOffice.hdr

The following images are courtesy of http://hdri.cs.nott.ac.uk/v1/download.php
     
     

The following images are courtesy of Rafal Mantiuk

nancy_church_2.hdr
nancy_church_3.hdr


mpi_atrium_1.hdr
mpi_atrium_3.hdr

The following images are courtesy of ?

forest_path.hdr