Abstract
Human observers can estimate the mean hue of a hue ensemble. We studied the spatial sampling characteristics of ensemble perception of hue by systematically varying the external noise and the amount of information available in a discrimination task. Presented for 500 ms against a gray background, our stimuli consisted of 1, 4, 16, or 64 1-degree square elements, each with a uniform hue. The hues ranged from yellow to blue on a hue circle in CIELAB color space. In different conditions, the hues were drawn from a von Mises distribution with one of three levels of external noise (no-, low-, or high-noise). Observers responded whether they perceived the comparison stimulus as yellower or bluer than a standard in a 2IFC task. The average comparison hue varied around the average standard hue. Discrimination thresholds were estimated by fitting psychometric functions to the data. The number of elements utilized in averaging was estimated through equivalent noise modeling. The relative importance of edge and surface information was tested by two 16-element conditions with the elements either abutting or separated by a 1/3-degree gap. Discrimination thresholds increased with increasing external noise, but decreased as the number of elements increased, the improvement being greater with higher noise. A gap separating the stimulus elements had no effect on performance. Modeling the number of samples used by the observer as a fixed power of the samples available gave an excellent fit to the data. For the 64-element stimulus, estimated effective number of samples ranged from 16 to 41. Control experiments confirmed that performance improved with the number of elements, not stimulus area. Observers sample and average stimulus elements to estimate mean hue, similarly to ensemble perception in other domains. The observed sampling clearly surpasses earlier estimates, and performance is most affected by surface, not edge information.
Acknowledgement: This project was supported by the Academy of Finland (grant number 319404).