Cornell University Program of Computer Graphics
Real-time hardware based tone reproduction.John Crane Mollis.
Master's thesis, Cornell University, January 2004.
The human visual system is exposed to a vast range of illumination conditions, far greater than any display device can reproduce, and its response to these conditions varies greatly. To create an immersive impression and accurate portrayal of a scene on a computer monitor requires complex modeling of visual response through tone reproduction algorithms and the simulation of adaptation effects in real-time. However, all current tone reproduction operators are off-line and address only a portion of the visual phenomena necessary for completeness, thereby limiting their applicability. A mostly unexplored problem is how perceptually accurate and full featured tone reproduction can be incorporated into interactive applications where visual effects will be dynamic and often very dramatic. Previous work in this area has been constrained with respect to the generality of tone reproduction models used, the scope of available input and the hardware output performance.
The aim of this thesis is two-fold. First, we create a real-time tone reproduction operator that includes as many phenomena as possible and is based upon psychophysical data. This requires a combination and extension of the best operators for predictive tone mapping and significant acceleration using current commodity graphics hardware. Care is taken to not restrict available input or compromise the predictive nature of the operator through artificial approximations. The result is a widely applicable, fast and comprehensive operator with applications in lighting engineering, architectural walkthroughs, flight and car simulators, entertainment and due to it's predictive nature, low vision simulation.
The second aim of this thesis is to use our perceptually based operator to construct a low vision simulation tool for evaluating real or simulated environments for suitability with older individuals.
This paper is available as a PDF file Mol04.pdf (2.6M).