Image-based perceptual analysis of lit environments

Abstract

Modern lighting systems typically provide a number of control parameters, e.g. colour and intensity, which allows nearly infinite possible configurations in one single room. The growing complexity of control makes it increasingly challenging for the user to configure the system and fully utilize the additional degrees of freedom. In contrast, an intuitive control interface takes into account the perceptual meaning of the lighting configurations. Unfortunately, collecting user ratings to construct perceptual models for lit spaces is cumbersome. To address this problem, we introduce an image-based mapping method that can rapidly evaluate the perceptual impression of lighting scenes using photographs or renderings. We discuss potential applications, guidelines and limitations of this method. In summary, we were able to closely approximate ratings-based mapping (normalized dissimilarity value <0.04). Among three dimensionality reduction methods, principal component analysis achieved the lowest dissimilarity and required the least images with a resolution as low as six by six pixels. Furthermore, simulations revealed that one perceptual model might suffice for the same type of offices. Offices of different types, on the other hand, require new mapping.