The visibility and perception of contrast depends strongly on the state of luminance adaptation in early vision mechanisms. Naka and Rushton (1966) modeled luminance adaptation of individual photoreceptors or as pooled across a small retinal region. However, most adaptation models used in computer graphics and vision assume a larger ad-hoc pooling radius. In this work we propose an empirical model for the local luminance adaptation, based on new psychophysical experiments. A custom high-dynamic-range display was build to produce contrasts in excess of 100000:1 and luminance above 5000 cd/m2. The display was used to measure baseline adaptation due to full-field luminance stimuli, and the adaptation due to various patterns of disks and rings. We discuss the predictive power of several candidate models, ranging from simple Gaussian-weighted luminance averaging to general averaging kernels in the log-luminance domain. We found that the pooling radius is smaller than the ad-hoc radius used in many applications, but larger than the extent of the laterally interconnecting retinal neurons. This suggests that luminance adaptation is also pooled in receptive fields in LGN or the visual cortex. Our predictive model can be used as an adaptation function in existing applications in vision and graphics.
|Publication status||Published - 21 Aug 2015|
|Event||38th European Conference on Visual Perception - Liverpool, United Kingdom|
Duration: 24 Aug 2015 → 27 Aug 2015
|Conference||38th European Conference on Visual Perception|
|Abbreviated title||ECVP 2015|
|Period||24/08/15 → 27/08/15|