基于双稀疏模型和非局部自相似约束的超分辨率算法研究  

Image super-resolution algorithm based on double-sparsity model and non-local self-similarity

作　　者：朱晨 杨欣[1] 谢堂鑫 周大可[1] ZHU Chen;YANG Xin;XIE Tang-xin;ZHOU Da-ke(College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China)

机构地区：[1]南京航空航天大学自动化学院,江苏南京210000

出　　处：《云南民族大学学报（自然科学版）》2020年第1期59-64,共6页Journal of Yunnan Minzu University:Natural Sciences Edition

基　　金：国家自然科学基金(61573182).

摘　　要：在基于稀疏表示的图像超分辨率方法中,字典的选择对最终重建质量具有重要影响.目前K-SVD作为基于外部样本学习的过冗余字典在图像重建领域取得广泛的成功,但同时也限制信号输入维度,带来信号降维过程的信息损失.针对这一问题,提出引入一种双稀疏模型,结合结构化字典和非结构化字典优点,避免降维过程信息损失同时保证训练精度;重建阶段引入非局部自相似性约束,迭代求解稀疏系数,降低编码噪声,最终重建高分辨率图像.实验结果表明,该算法在图像质量客观评价指标上优于对比算法,并且在主观视觉效果上获得更清晰的边缘等细节信息.Among sparse coding-based super-resolution algorithms,the choice of dictionary has a significant impact on the quality of image reconstruction.Nowadays,K-SVD dictionary as the over-redundant dictionary based on external sample learning achieves much success in the field of image reconstruction,but it also limits the dimension of input signal,resulting in information loss during signal dimensionality reduction.To solve this problem,this paper proposes a double sparse model through combining the advantages of structured dictionary and unstructured dictionary to avoid information loss in the dimension reduction process while ensuring training accuracy.Non-local self-similarity constraint is introduced at the reconstruction phase and coding noise is reduced by iteratively solving sparse coefficients,and finally a high-resolution image is reconstructed.The experimental results show that the proposed algorithm outperforms the previous algorithm in the objective evaluation of image quality,and obtains a clearer edge and more details on subjective visual effects.

关 键 词：超分辨率 双稀疏模型 非局部自相似性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]