Visualization of color transfer between images in RGB color space

Abstract

This paper proposes a novel approach for visualizing the process of color transfer between images. The color transfer is a technique by enforcing an image having similar color characteristics with another. Considering the pixels of reference image and target image as point clouds in RGB color space, then the color transfer becomes geometrical transformation to make the target image have the same distribution with that of the reference image. The geometrical transformation matrices are computed by decomposing the covariance matrix of the two images. For visualizing the process of color transfer, the transform matrices are interpolated. Since rotation matrix cannot be interpolated directly, a quaternion interpolation is proposed for completing this task. Moreover, a curved interpolation path is designed to avoid saturation shortness during visualization. The experiment results demonstrate that the proposed algorithm performs well.