摘要
为了提高单帧降质图像的分辨率,利用迭代反投影(iterative back projection,IBP)的方法改进了基于稀疏表示的图像超分辨率重建算法。该算法将高分辨率图像减去IBP的重建结果后的差值用于学习高分辨率字典,并提出一种基于非局部相似性的全局后处理过程,降低了稀疏表示求解的计算量同时提高了图像重建效果。实验结果表明,与现有的其它算法相比,改善了重建图像的主观质量,PSNR和SSIM也得到提高。
To deal with the single-image scaleup problem, an improved algorithm via sparse representation based on IBP is pre- sented. The difference between high-resolution images and reconstruction results of IBP is made use of to train high-resolution dictionary, which leads to numerical shortcuts. A global post-processing stage based on nonlocal similarity is also proposed in the new algorithm to improve reconstructed images. Extensive experiments validates that our algorithm achieves both visual and ob- jective （i. e. PSNR and SSIM） improvements over previous state-of-the-art methods.
出处
《计算机工程与设计》
CSCD
北大核心
2014年第3期934-938,共5页
Computer Engineering and Design
基金
国家自然科学基金项目(61363069
11271388)
广西自然科学基金项目(2013GXNSFDA019030
2013GXNSFAA019331
2012GXNSFBA053014
2012GXNSFAA053231
2011GXNSFA018158)
广西教育厅重点基金项目(201202ZD040
201202ZD044)
关键词
图像超分辨率
迭代反投影
稀疏表示
非局部
全局后处理
image super-resolution（SR）
IBP
sparse representation
nonlocal（NL）
global post-processing
