一种基于彩色化的单幅彩色图像超分辨率重建

Colorization-based Single Frame Color Image Super-resolution

对单幅彩色图像进行超分辨率(Super-resolution,SR)重建,一般是将原始RGB图像转换为YUV图像,对亮度分量Y进行SR,而对色度分量U、V只进行简单的插值。因为图像插值很容易产生边缘模糊和锯齿,本文用彩色化算法对低分辨率的U、V分量进行处理,并提出一种基于彩色化的迭代反投影(Iterative Back-projection,IBP)方法,改善了传统基于插值的IBP算法对边缘图像的处理效果。实验结果表明,文中算法得到图像的客观质量和主观质量既优于传统的基于双立方插值处理的SR彩色图像,也优于其它以彩色化为基础进行色度图像处理的SR图像。

For single-frame color image super-resolution, most techniques use super-resolution reconstruc- tion only on the Y-channel. Usually directly use interpolation algorithms for the chrominance channels （U,V） which decide the color. Because interpolation algorithm often produces ＂jaggy＂ and ＂ringing＂ ar- tifacts, this paper integrated colorization into color image SR using the colorization algorithm to enhance the low-resolution （LR） chrominance channels and the iterative back-projection（IBP） method to improve the accuracy of the LR chrominance information. This paper also proposed a eolorization-based IBP （CIBP） algorithm to enhance the traditional interpolation-based IBP algorithm. Experimental results dem- onstrated that compared with the bicubic-interpolation algorithm and the latest colorization-based chromi- nance process algorithm, the proposed algorithm can reconstruct the high-resolution （HR） chrominance channels with higher Peak Signal Noise Ratio（PSNR） and Structural Similarity （SSIM）.