Abstract
When observers match the color of natural objects they base their judgments on the brightest parts of the objects. In the present work we test whether this behavior is an effective strategy for lightness estimation. The luminance of diffusely reflecting surfaces is proportional to the cosine of the angle between the surface normal and the direction of the incident light. However, the complex interaction between object geometry and natural light fields can hardly be treated analytically. We resorted to a physical-based rendering simulation to find the most robust estimate of the objects’ reflectance. Using the software RADIANCE interfaced with a MATLAB toolbox, we rendered a set of virtual objects under a large variety of different viewing and illumination conditions using simulated natural light fields. Each view was rendered with different values of reflectance. For each rendered object, we calculated the percentiles of the radiance distribution as potential lightness estimates. We found that the distribution of the standard deviations for each estimate has approximately an inverted U-shape with minima for the darkest and the brightest object regions. Given that the dependency of the luminance on the incident light direction has to be maximal for the most illuminated parts of the objects, the sharp drop in variability for the extremely bright portions of the objects can only be due to a reduction in the variability of their orientation within the light field. Intuitively, this means that for most natural objects there is nearly always a region that is close to perpendicular to the direction of the light source. Results of an ROC-analysis show that reflectance discriminability increases with the luminance of the object region which is compared. Results of both analyses indicate the most illuminated regions are most diagnostic of the object´s reflectance.
Meeting abstract presented at VSS 2012