一种结合稀疏表示和纹理分块的遥感影像超分辨率方法

提出了一种基于稀疏表示和纹理分块的单幅遥感影像超分辨率方法,主要利用先验知识及影像自身的纹理信息重构遥感图像。首先,提取用于字典学习的图像块,从高、低分辨率遥感图像块中训练出冗余字典,采用正交匹配追踪方法更新字典,用迭代的方法直到算法收敛;然后,将训练的字典应用于遥感影像超分辨率重构。重构时将图像块分成平滑块和非平滑块两种类型,平滑块采用双三次卷积方法重构,非平滑块采用低分辨率遥感图像块的稀疏表示系数及高分辨率图像块冗余字典重构。实验结果表明,此方法重构速度较快,并在视觉及客观评价指标上有较好的超分辨率效果。

Remote Sensing Image Super-Resolution Method Using Sparse Representation and Classified Texture Patches

A super-resolution method based on sparse representation and classified texture patches wasproposed,mainly using the priori knowledge and texture to reconstruct remote sensing images.First,extract image blocks for dictionary learning,the over-complete dictionary was learned from the highand low resolution remote sensing image blocks.Orthogonal match pursuit was used to calculate thesparse conefficients,then the coefficients were fixed,iterative method was used to update the diction-ary until the algorithm converges.Then,the training dictionary was used to reconstruct the remotesensing images.In this step,the image was divided into smooth patches and non-smooth patches,bicubic interpolation was used for smooth patches while sparse conefficients and over-complete diction-ary were used for non-smooth patches.Experiment shows that this method has a faster reconstructionspeed and can achieve satisfied super-resolution results in the visual effects and objective evaluation in-dicators.