Color constancy by combining low-mid-high level image cues

In general, computational methods to estimate the color of the light source are based on single, low-level image cues such as pixel values and edges. Only a few methods are proposed exploiting multiple cues for color constancy by incorporating pixel values, edge information and higher-order image statistics. However, expanding color constancy beyond these low-level image statistics (pixels, edges and n-jets) to include high-level cues and integrate all these cues together into a unified framework has not been explored.

In this paper, the color of the light source is estimated using (low-level) image statistics, (intermediate-level) regions, and (high-level) scene characteristics. A Bayesian framework is proposed combining the different cues in a principled way.

Our experiments show that the proposed algorithm outperforms the original Bayesian method. The mean error is reduced by 33.3% with respect to the original Bayesian method and the median error is reduced by 37.1% on the re-processed version of the Gehler color constancy dataset. Our method outperforms most of the state-of-the-art color constancy algorithms in mean angular error and obtains the highest accuracy in terms of median angular error.