Abstract: In order to enhance the spatial resolution of remote sensing images by software, this paper proposes a remote sensing image super-resolution reconstruction algorithm based on double-sparse K-SVD dictionary learning. Based on the sparse representation theory, the K-SVD dictionary learning algorithm is used to solve the low resolution dictionary and its sparse coefficient. The sparse coefficient passes to the high-resolution dictionary learning space to form high and low resolution dictionary pairs and reconstruct high-resolution remote sensing images. In the dictionary learning and sparse reconstruction stages, different degrees of sparseness are set. The experiments are respectively carried out, using TM5 image, resource No.3 image and remote sensing image in USC_SIPI image library to reconstruct. The results show that the regardless of whether the reconstructed image is noisy, the peak signal-to-noise ratio and similarity of the algorithm are higher than those of Bicubic and Zeyde. The introduction of K-SVD and double-sparse parameters not only reduces dictionary learning time, but also has high enhancement capability of spatial resolution.

Key words: super-resolution, remote sensing image, dictionary learning, sparse reconstruction

摘要： 为了通过软件方式增强遥感影像的空间分辨率，提出了一种基于双稀疏度K-SVD字典学习的遥感影像超分辨率重建算法。基于稀疏表示理论，利用K-SVD字典学习算法求解低分辨率字典及其稀疏系数，将稀疏系数传递至高分辨率字典学习空间，形成高、低分辨率字典对，重建得到高分辨率遥感影像，并在字典学习和稀疏重建两个阶段设置了不同的稀疏度。实验分别采用TM5影像、资源三号影像以及USC_SIPI图像库中的遥感影像进行重建，结果表明，不论重建影像有无噪声，所提算法的峰值信噪比和结构相似指标均高于Bicubic法以及Zeyde的算法。K-SVD和双稀疏度参数的引入，不仅减少了字典学习时间，且具有高的空间分辨率提升能力。

关键词: 超分辨率, 遥感影像, 字典学习, 稀疏重建