Image Decolorization Based on Information Theory
In this paper we introduce a novel non-linear mapping technique to effectively decolorize images. Designed in a multi-scale fusion fashion, we first derive three input images represented by the color channels R, G and B. In order to transfer to the decolorized image only the relevant features of the derived inputs, we define two weight maps based on information theoretic approaches. The first weight map extracts visually salient regions based on a information maximization strategy while the second weight map filters the amount of local variation of each derived input computing local entropy per patch. Finally, to reduce the local distortions that might be introduced by the weight maps discontinuities, our decolorization strategy is designed in a multi-scale fusion. We also introduce a blind measure to accurately evaluate image decolorization methods. Our comprehensive qualitative and quantitative validation demonstrates that our method yields very competitive results ; Part of this work has was supported by 2020 European Union Research and Innovation Horizon 2020 under the grant agreement Marie Sklodowska-Curie No 712949 (TECNIOspring PLUS), as well as the Agency for the Competitiveness of the Company of the Generalitat de Catalunya - ACCIO: TECSPR17-1-0054. Part of this work was supported by research grant GNaC2018 - ARUT, no. 1361-01.02.2019, financed by Politehnica University of Timisoara and also by Spanish project from Ministery of Economy and Competitivity, MINECO, TIN2016-75866- C3-3-R