Computational-observer analysis of illuminationdiscrimination

Journal of Vision

article

Xiaomao Ding¹  Ana Radonjić²  Nicolas P. Cottaris²  Haomiao Jiang³  Brian A. Wandell⁵  David H. Brainard² 

[1] Neuroscience Graduate Group, University of Pennsylvania;Department of Psychology, University of Pennsylvania;Department of Electrical Engineering, Stanford University;Current address: Google Research;Department of Psychology, Stanford University

关键词: color vision;    illumination discrimination;    ideal observer;    computational observer;   

DOI  :  10.1167/19.7.11

来源: Association for Research in Vision and Ophthalmology

【 摘 要 】

The spectral properties of the ambient illuminationprovide useful information about time of day andweather. We study the perceptual representation ofillumination by analyzing measurements of how wellpeople discriminate between illuminations across sceneconfigurations. More specifically, we compare humanperformance to a computational-observer analysis thatevaluates the information available in theisomerizations of cone photopigment in a model humanphotoreceptor mosaic. The performance of such anobserver is limited by the Poisson variability of thenumber of isomerizations in each cone. The overalllevel of Poisson-limited computational-observersensitivity exceeded that of human observers. This wasmodeled by increasing the amount of noise in thenumber of isomerizations of each cone. The additionalnoise brought the overall level of performance of thecomputational observer into the same range as that ofhuman observers, allowing us to compare the patternof sensitivity across stimulus manipulations. Keypatterns of human performance were not accounted forby the computational observer. In particular, neitherthe elevation of illumination-discrimination thresholdsfor illuminant changes in a blue color direction (whenthresholds are expressed in CIELUV DE units), nor theeffects of varying the ensemble of surfaces in thescenes being viewed, could be accounted for byvariation in the information available in the coneisomerizations.

【 授权许可】

CC BY|CC BY-NC-ND