Filters & Lumination: Creating Multi-Illuminant Images for Computational Color Constancy

White-balancing is a very important process for anyone who deals with photography. It is present in every digital camera, and it has a significant influence on how an image will look. It removes the chromatic effect of scene illumination so that the final image looks as though it is illuminated by a perfectly white light. This needs to be done so that images look natural. The Human Visual System does this and without white-balancing digital images look odd. Many methods have been developed, and it was shown that the best results are obtained using learning-based methods. Learning-based methods rely on large, diverse datasets for proper training and evaluation. While there are many datasets with images affected by a single uniform illuminant, thorough research on images affected by multiple illuminants has only recently started. To help with such research, in this paper we propose a new way to create multi-illuminant color constancy images. With our approach, images of a diverse set of scenes with a variable number of illuminants can be created. Our approach also includes an automatic way to create a per-pixel illumination mask for each image. We used around 100 different scenes to evaluate our dataset creation approach. We also evaluate several methods from the literature on our images.