Bag of Color Features For Color Constancy

Abstract

In this paper, we propose a novel color constancy approach, called Bag of Color Features (BoCF), building upon the Bag-of-Feature pooling, which allows for reducing the number of parameters needed for illumination estimation up to 95%. At the same time, the proposed method is consistent with the color constancy assumption stating that global spatial information is not relevant for illumination estimation and that the local information (edges,etc) is enough. Furthermore, BoCF is consistent with color constancy statistical approaches and can be interpreted as a learning-based generalization of many statistical-based approaches. To further improve the accuracy of illumination estimation, we propose a novel attention mechanism for the BoCF model. We propose two variants based on self-attention. BoCF approach and its variants achieve competitive, to the state of the art, results while requiring much fewer parameters on three benchmark dataset: ColorChecker RECommended, INTEL-TUT version 2, and NUS8 datasets.