基于非局部相似性和分类半耦合字典学习的超分辨率重建

Super-Resolution Reconstruction Based on Non-Local Similarity and Clustered Semi-Coupled Dictionary Learning

为了提升单幅彩色图像的超分辨率重建质量,提出了一种改进的基于学习的超分辨率方法.针对半耦合字典学习超分辨率算法训练精度不高的缺陷,采用稀疏域分类与半耦合字典学习交替进行的启发式策略.在训练阶段引入稀疏域非局部相似性约束项,使用改进了的非局部约束l1范数优化问题求解算法,训练得到多组高、低分辨率字典和映射矩阵.在重建阶段利用分类稀疏表示、非局部相似性并结合残差补偿进一步提高重建精度.实验结果表明,该方法在主观和客观评价标准下均取得了较好的重建效果,显著提升了超分辨率重建质量.

In order to improve the super-resolution reconstruction quality of single color image,a modified learning based super-resolution approach was proposed. To resolve the problem of low training accuracy of semi-coupled dictionary learning（SCDL）super-resolution algorithm,this paper presented a novel heuristic strategy in which clustering in sparse domain and semi-coupled dictionary learning were performed in turn. In training phase,the non-local similarity constrain term in sparse domain was introduced into the optimization function,and then several groups of highlow resolution dictionaries and mapping functions were trained by a modified non-locally constrained l1-norm optimization algorithm. In reconstruction phase,combined with residual compensation,clustered sparse representation and non-local similarity were used to further improve the super-resolution reconstruction quality. Experimental results show that the proposed approach can achieve better reconstruction effect in both subjective and objective evaluation criteria and significantly improve the quality of super-resolution.