Lighting quality research using rendered images of offices

Forty participants viewed a series of high-quality, computer-rendered colour images of a typical open-plan partitioned office, and rated them for attractiveness. The images were projected at realistic luminances and 33% of full size. The images were geometrically identical, but the outputs of four lighting circuits depicted in the renderings were independently manipulated. Initially, the lighting circuit outputs were random, but a genetic algorithm was used to generate new images that retained features of prior, highly-rated, images. As a result, the images converged on an individual’s preferred scene. Luminances in the preferred image were similar to preferred luminances chosen by people in real settings. A sub-set of images was rated on Brightness, Non-Uniformity and Attraction scales. Ratings were significantly related to simple photometric descriptors of the images. In particular, around 50% of the variance in Attraction ratings was predicted by average image luminance and its square, or by average image luminance and a measure of luminance variability.