Compressive Rendering: an application of Compressive Sensing to global illumination

Written by Luis Paulo Santos

Luis Paulo Santos

Universidade do Minho, Braga, Portugal
http://gec.di.uminho.pt/psantos/

Seminar in English
December 5, 2013 at 10:00
University of West Bohemia, UV115

Abstract

  Compressive Sensing (CS) is a relatively new sampling theory that has been exciting the scientific community with its promise, and demonstrated results, that a vast class of discrete signals can be represented and perfectly reconstructed from a number of samples well below the classical Nyquist rate. These impressive results are based both on

• the realization that if a discrete signal admits a sparse representation on a given basis, then sampling requirements can be significantly loosened

• the development of strong guarantees of perfect reconstruction from the undersampled data.

  Physically based rendering is a very demanding sampling process, requiring the numerical evaluation of huge numbers of light paths. Sensing, understood as the evaluation of a tree of light paths, is achieved through a computationally expensive numerical simulation process, each measurement entailing tracing an arbitrarily large number of rays. Applying CS to rendering (coined Compressive Rendering (CR)) constitutes thus an obvious approach, which has indeed been tackled with very promising results. Throughout this talk I will present some of these results and will propose new applications and grounds of research.