Finding Images with Similar Lighting Conditions in Large Photo Collections

When we look at images taken from outdoor scenes, much of the information perceived is due to the ligthing conditions. In these scenes, the solar beams interact with the atmosphere and create a global illumination that determines the way we perceive objets in the world. Lately, exploration of the sky like the main illuminance component has began to be explored in Computer Vision. Some of these studies could be classified like color-based algorithms while some others fall in the physicsbased category. However most of them assume that the photometric and geometric camera parameters are constant, or at least, that they could be determined. This work presents a simple and effective method in order to find images with similar lighting conditions. This method is based on a Gaussian mixture model of sky pixels represented by a 3D histogram in the La∗b∗ color space.