避免颜色越界的各向异性扩散图像彩色化

Image Colorization without Edge Crossover of Colors Based on Anisotropic Diffusion

图像彩色化是根据人工描绘的局部初始颜色用计算机为灰度图像着色。该文提出一种各向异性的非线性扩散图像彩色化方法,该方法能够克服颜色越界问题。首先建立基于偏微分的扩散方程,通过设置自适应的张量函数,使颜色在平滑区域快速均匀扩散,在颜色交界处按各向异性方式扩散,能够有效增强边缘颜色的连贯性。算法采用有限差分法实现扩散方程的数值求解,与当前流行的彩色化技术相比,该方法能得到更加清晰自然的彩色化图像和更高的图像质量。

Image colorization is to create a color picture from gray image automatically according to manually added scribbles. A colorization algorithm based on anisotropic diffusion is proposed to overcome the problem of edge crossover of colors. By defining an adaptive diffusion tensor, a nonlinear anisotropic diffusion function based on Partial Differential Equation （PDE） is established. Colors are diffused uniformly and quickly in smooth areas, while anisotropic diffusion takes place near edges, which effectively improves color transition by suppressing unwanted crossover. Numerical solution of the PDE based on a finite difference method gets satisfactory colorization effects and better quality in edge regions compared with the popular techniques.