摘要
图像具有大量的局部结构相似区域,并且这种相似性可以在多个尺度上保持。基于这一特征,利用结构相似指标进行相似性匹配生成相似的低分辨率图像序列,从而把单幅图像的超分辨问题转化为图像序列超分辨问题来解决。文中提出了一种新的自适应的正则化方法,正则参数的选取使得目标函数存在全局最优解。最后证明了算法的收敛性。实验表明,该方法具有很好的复原效果。
An image contains many local structure similar areas and this similarity holds across scales.Based on the feature,we achieve similarity matching using the structure similarity index,then generate low resolution image sequences,consequently solve single image super-resolution by transforming to image sequence super-resolution.A new adaptive regularization method is proposed,and the choice of the regularization parameter can make cost function have a global optimal solution.In the end the convergence of the algorithm is proved.Our experimental results show that this algorithm has a good effect.
出处
《计算机工程与应用》
CSCD
北大核心
2007年第22期34-37,共4页
Computer Engineering and Applications
基金
国家自然科学基金(the National Natural Science Foundation of China under Grant No.60672135)。
关键词
图像复原
超分辨
结构相似度
正则化
image restoration
super-resolution
structure similarity
regularization
