Multi-View Image Color Correction Using 3D Point Set Registration

Abstract

In this paper, a novel color correction method for multi-view images is proposed to improve color consistency successfully. Contrast to the conventional methods trying to find a transfer function per color channel, the proposed method deforms the color distribution of the image non-rigidly into that of the reference image by exploiting a 3D point set registration algorithm. In order to alleviate the computational complexity of the 3D registration process, we approximate the image color distribution consisting of all the image pixels by extracting a much smaller number of representative color points to be used for 3D registration. Extensive experiments confirm that the proposed method achieves excellent performance in color consistency enhancement both subjectively and objectively, and the proposed representative color point set approximation enhances the computational efficiency successfully.