Colorization-based image coding based on superpixel

This paper proposes a novel colorization-based color image coding method. This method segments an input image into small segments called superpixels and restores the chrominance of each superpixel as the linear transformation of its luminance. A lot of color image coding methods using colorization have been proposed, but no methods utilize a local correlation between the luminance component and the chrominance component effectively. Therefore conventional methods cannot recover the information of local variation and the output image becomes excessively smoothed. We introduce the region segmentation to suppress large fluctuations in the segmented area. This segmentation makes possible to restore the chrominance from the luminance using a linear transformation. The simulation with some standard images shows that the framework actually performs better result than the conventional algorithms based on JPEG.