Psychophysics for Comparing Real and Synthetic Scenes (invited lecture)

Advances in image synthesis techniques allow us to simulate the distribution of light energy in a scene with great precision. Unfortunately, this does not ensure that the displayed image will have authentic visual appearance. Reasons for this include the limited dynamic range of displays, and any residual shortcomings of the rendering process. Furthermore, it is unclear to what extend human vision will encode such departures from perfect physical realism. This leads to a need for including the human observer in any process which attempts to evaluate the perceptual significance of any errors in reproduction. Our psychophysical studies address this need.

This talk provides an introduction to the application of psychophysics to the evaluation and advancement of computer graphics with respect to the real scenes they are intended to depict. It covers the fundamentals of the design and organisation of psychophysical experiments, data collection and analysis and the application of results to rendering algorithms. The emphasis of this seminar is on the practical issues which must be addressed so that human subjects may easily make perceptual evaluations between the real and synthetic scenes. Case studies, involving comparing a test environment consisting of a small room containing complex objects to its rendered counterpart, will also be discussed.