Fast Colorization Using Edge and Gradient Constrains

Abstract

This paper proposes a fast but effective user-guided colorization algorithm. The main difficulty with colorization is its intensive computational cost. And the color sometimes diffuses from one region to others. We make full use of the information given by the gray-scale images, including the edge, gradient and gradient direction, to propagate color over regions from the user’s scribbles. We introduce a novel local distance definition which reduces the color confusion between two regions obviously. Two-Dimensional Programming is used to get the minimum distance from the scribbles to every pixel, and every pixel is blended by the chrominance with top three minimum distances. This speeds up the colorization process. Our algorithm can also be extended to recolorization and movie colorization. Experimental results show that our algorithm outperforms many state-ofthe- art algorithms from small amount of user scribbles. Besides, the implementation require less than 1 second on the image with 320*240 pixels.