Multi-Scale Block-Based Color Constancy

Oguzhan Ulucan, Diclehan Ulucan, Marc Ebner

Abstract: Removing the effects of the light source from a color-casted image is a key step for numerous computer vision applications. Alongside introducing new color constancy algorithms, researchers also aim at improving the existing methods by utilizing different approaches. Using strategies that have not been investigated in the field of color constancy in detail might help us both to design cost-efficient simple yet effective algorithms and to improve the existing methods. Thereupon, we propose a color constancy method relying on surface orientations which utilizes block-based operations in scale space by only considering the salient regions of the image. According to the evaluations, our multi-scale color constancy method achieves state-of-the-art performance. Moreover, we demonstrate that the performance of several existing color constancy algorithms can be improved by benefiting from scale space.