Segmenting an image can yield large objects whose generic color may be classified as memory colors. Typically sky, skin, foliage, and/or water are the dominant memory colors and can be found in the majority of personal photographs. An algorithm identifies potential memory color objects and uses their chromatic signature to predict the scene illuminant from a limited class of common light sources.
Ted Cooper, "Memory Colors as Illuminant Predictors" in Proc. IS&T 9th Color and Imaging Conf., 2001, pp 146 - 152, https://doi.org/10.2352/CIC.2001.9.1.art00027