Novel colorization method based on correlation neighborhood similarity pixels priori

Colorization is a computer-aided process of automatically adding colors to grayscale images or videos according to the colors scribbled on by the user. In this paper, we propose a novel colorization method which based on correlation neighborhood similarity pixels priori. First, according to a key observation that similar intensities neighboring pixels have similar colors, our method searches the similarity neighborhood pixels group. Then, we compute the weighted coefficients of the neighborhood pixels in the luminance image and transmit the weighted coefficients to the chrominance. Finally, we obtain the chrominance values by solving a quadratic optimization problem which uses the colors scribbled on by the user as the linear constraints. The experimental results show that our approach is quite effective especially in the boundary parts. Moreover, our method can give better visually results when only a few colors scribbled on.