Interactive Colorization via Multi-Cues Manipulation

Related Work

Generally speaking, colorization could be roughly classified into two categories--global colorization and local colorization. Both of the two are essential tools for various visual editing task. Global colorization conveys the whole color characteristics of a reference image to a user-specified target image. Reinhard et al. (2001) proposed a global color mapping method mapping a reference image’s color characteristics to another target image using statistical analysis. The color of the reference image is propagated corresponding to the mean and standard deviation of the colors in the target image, keeping the image content naturally. As effective as it is, the quality of colorization relies on the similarity between target and reference images.

Subsequently, lots of complementary methods are proposed to improve Reinhard’s statistical mapping method (Reinhard et al., 2001) to achieve better color stylization results. Chang et al. (2005) proposed a basic color categories-based method that was applied within each pair of convex hull of the same category. Similarly, some approaches of mapping are presented, e.g. Tai et al.’s probabilistic segmentation color transfer (Tai et al., 2005), Wang et al.’s data-driven image color theme enhancement (Wan et al., 2010), Dong et al.’s dominant colors mapping based method (Dong et al., 2010) and Wang et al.’s training based method (Wang et al. 2011).