Selective tone mapper

Abstract

Recently human visual attention has been increasingly exploited in computer graphics in order to reduce the high cost of computing high-fidelity images. By only computing in high quality those areas of a scene which are perceptually important, significant computation time can be saved without the user being aware of the quality differences within the image. Tone mapping (TM) is a key part of the high-fidelity image synthesis process, allowing high dynamic range images to be best displayed on low dynamic range computer monitors. Although it is possible to achieve, by means of graphics hardware support, interactive tone mapping using lower quality global tone mappers, to-date interactive performance for the superior quality, but computationally significantly more expensive, local tone mapping has been possible but only at a much reduced quality. This report presents the novel concept of Selective Tone Mapper, that it is a general framework, for reducing the computational costs of the existing local TM operators. We propose a GPU implementation of this framework which overcomes many of the problems modern graphics hardware have, in order to fully exploit the advantages of the Selective Tone Mapper concept.