摘要
为了提高图像超分辨率重建的效果,提出一种基于自适应匹配追踪稀疏表示的图像超分辨重建算法.该方法采用k次奇异值分解算法联合训练适用于高、低分辨率图像块的联合字典对;然后,寻找输入的低分辨率图像块在低分辨率字典下的稀疏表示;最后,利用字典间的相似性,通过低分辨率稀疏系数和高分辨率字典来生成清晰的高分辨率图像.在稀疏表示过程中,求解稀疏表示系数的优化算法大多使用正交匹配追踪算法.为了提高重构精度,缩短算法时间,采用自适应匹配追踪算法进行求解.实验表明,该算法的重构精度明显优于其他算法,对边缘和细节具有更好的重构能力,并且能够缩短字典训练的时间.
In order to improve the effect of image super-resolution reconstruction, we propose a new sparse representation optimization method based on sparse representation reconstruction method proposed by Yang et al. The algorithm uses k-means singular value decomposition algorithm to jointly train two dictionaries of low-resolution ( LR )and high-resolution( HR) image patches. Then, we seek a sparse representation for each patch of the input LR image under the LR dictionary. According to the similarity of the dictionaries, the sparse representation of the LR image patch can be applied with the HR image patch dictionary to generate a HR image patch. In sparse representation, most of the optimization algorithms for sparse representation coefficients use Orthogonal Matching Pursuit (OMP) algorithm. In order to improve the reconstruction accuracy and shorten the time of OMP algorithm,this paper uses Adaptive Matching Pursuit (AMP) algorithm to solve sparse representation coefficients. Experiments show that the reconstruction accuracy of this algorithm is obviously better than other algorithms, it has better reconstruction ability for edges and details, and it can shorten the time of dictionary training.
作者
华臻
张海程
李晋江
HUA Zhen;ZHANG Hai-cheng;LI Jin-jiang(College of Computer Science and Technology,Shandong Technology and Business University,Yantai 264000,Chin;Shandong Co-Innovation Center of Future Intelligent Computing,Yantai 264000,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2018年第10期2339-2344,共6页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61472227,61772319,61602277)资助.
关键词
超分辨率重建
稀疏表示
自适应匹配追踪算法
字典学习
super-resolution reconstruction
sparse representation
adaptive matching pursuit algorithm
dictionary learning