Invertible Image Decolorization With CFEH and Reversible Data Hiding

Abstract

In the field of invertible image decolorization, how to reduce artifacts in the smooth grayscale regions and prevent color distortion at the boundaries of the reconstructed color image is a crucial issue. In this paper, we propose an invertible deep learning network with extraction and hiding of color information. Our approach separates the original color image into the luminance and chromaticity planes by using orthogonal transformation, which enhances the independence and completeness of color and luminance information. Then, the color feature extraction module is developed to minimize color information distortion, while the color hiding module is adopted to hide color information invisibly. Compared with existing deep-learning-based methods, the proposed network can preserve more color information while ensuring the quality of grayscale images by processing color and grayscale information separately. Furthermore, we propose a reversible data hiding strategy that enhances the performance of the reconstructed color images. Our method outperforms learned invertible image decolorization methods, as demonstrated through experiments on the VOC2012, Kodak24, and NCD datasets.