基于自然彩色图像的彩色化权值函数建模

针对现有的权值函数采用人为指定的核函数,难以准确地反映图像复杂处像素间的色度相似关系,提出了利用自然彩色图像特性构建权值函数的方法。根据自然图像的像素在RGB空间的分布特征,提取出色度与灰度的局部线性关系,并利用其作为先验知识结合最小二乘法推广至整幅图像,进而获得一种新的权值函数。该权值函数能将相近的两像素间的灰度差异、位置差异及周围像素的灰度分布统一到色度相似程度的计算中。实验结果表明,该权值函数有助于获得更优的彩色化结果,特别是在复杂边界处效果提升明显。

Modeling colorization weight function based on natural color images

Some existing function is assigned artificially as one of some kernel functions. It is hard to describe the affinity accurately between pixels in complicated texture area. Aiming at this problem, this paper employed a feature of natural color images in modeling weight function. It extrcted a local linear chrominance-luminance relation firstly from the distribution of pixels in RGB color space, and then served as prior assumption , generalized to the whole image with least square method, and finally provided a novel chrominance weight function model which could integrate luminance difference, space distance and luminance distribution of their neighborhoods into estimation of their chrominance affinity. Experiment results indicate that the weight function can benefit for better colorization, especially in complicated edges.