摘要
针对稀疏表示超分辨率重建算法中稀疏表示系数正则化效果不明显、字典完备性弱以及重建图像存在虚边缘等问题,提出了一种改进的稀疏表示超分辨率重建算法。首先对正则化正交匹配追踪(regularized orthogonal matching pursuit,ROMP)稀疏表示系数求解算法进行了改进,通过引入局部约束加权来提高稀疏表示系数的精度、增强图像的纹理特性;然后,将Huber影响函数用于提取图像的先验特征信息,以增强图像特征、提升高分辨率字典的表示能力;最后,提出了基于学习的迭代反投影方法,提高了图像后处理阶段预测误差的准确性,进一步改善了高分辨率重建图像效果。实验结果表明,该方法在峰值信噪比和视觉效果上都有所提高,重建图像的纹理特性和质量得到了有效增强。
This paper proposes an improved algorithm of super-resolution to solve the problems that the sparse representation coefficient regularization is not effect, the high-resolution dictionary＇ s completeness is weak and the reconstruction image has false edges. Firstly, the ROMP algorithm is improved by introducing weighted partial constraint to improve the effect of regularization and enhance the texture features of images. Seeondly, the prior itfformation of image is extracted by Huber influenee funetion, so that image features are enhanced and the expression eapaeity of high resolution dietionary is improved. Finally, an iterative back projection method based on study is proposed, which can improve accuracy of the prediction error in image post-processing stage and achieve the high quality resolution reconstruction effectively. The simulation and analysis show that the proposed method has certain improvement on the peak signal-to-noise ratio and visual effect, and it can improve the quality of reconstruction images.
出处
《重庆邮电大学学报（自然科学版）》
CSCD
北大核心
2016年第3期400-405,共6页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
基金
陕西省自然科学基金资助(2013JM8025)
航空科学基金资助(20141996018)~~
关键词
超分辨率重建
稀疏表示
字典训练
图像特征
迭代反投影
super-resolution reconstruction
sparse representation
dictionary training
image features
iterative back projeetion
