扩散张量加权梯度域图像彩色化方法

Image Colorization Based on Diffusion Tensor in Weighted Gradient Domain

基于扩散张量的加权拉普拉斯核推广了图像彩色化的泊松解法,该彩色化过程是通过颜色在亮度值扩散张量加权的梯度场引导下自动传播完成的.首先在灰度图像上由用户手工地给定少量的颜色条带;然后计算每个像素的扩散张量,并利用这些扩散张量构造加权梯度场,从而导出基于散度的图像彩色化方程;最后求解方程,获得灰度图像着色结果.实验结果表明:该方法效果良好,比原泊松解法有显著改善.

Based on weighted Lapalcian kernel with diffusion tensors, a generalized Poisson solver for colorization is proposed in this paper. The colorization process is accomplished by propagating the colors automatically according to the weighted gradient fields with luminance diffusion tensors. Firstly, a user needs to provide the gray image with a few color scribbles manually. Secondly, the diffusion tensor field of the gray image at each pixel is calculated and applied to construct weighted gradient field, thus to create a divergence-based colorization equation. Finally, a fully colorized image is obtained by solving this equation. Experiments show that the proposed method can generate good colorization results and outperform significantly the original Poisson solver.