高分辨率灰度图像的快速多分辨率着色

Fast Multi-Resolution Colorization of High-Resolution Gray Images

灰度图像的计算机快速着色有着广泛的应用前景,尤其是对老照片或者旧影片的彩色化处理.基于最优化的着色方法利用少量的人工标色,就能够得到好的着色效果,但是该方法在处理较高分辨率图像时需要消耗大量的存储和计算资源,有时甚至不能得到最后的结果.在最优化着色方法的基础上,文中利用图像着色的特征,提出了使用最优化着色的金字塔模型进行多分辨率着色处理的新方法,始终控制矩阵计算的规模,能够很快地得到视觉效果一致甚至更好的着色效果.实验结果也证明新方法无论在计算时间、存储需求还是最后着色的质量上,尤其是针对高分辨率图像,比原方法有了很大的改进.

Colorization could be used to increase the visual appeal of old black and white photos and movies. By means of colorization using optimization, one can add colors automatically to a monochrome image or movie after annotating the image with a few color scribbles, based on a simple premise neighboring pixels in space that have similar intensities should have similar colors. However, the previous method may fail to deal with high-resolution gray images well because of its quite large size of computation. In this paper, the authors utilize features of the image colorization as an optimization problem to overcome this limitation. An image pyramid model with the edge detection is presented to reduce the computation size for the colorization. As a result, the new method can colorize high-resolution gray images to advantage in computing speed and visual effects.