Abstract: In order to improve the spatial resolution of remote sensing images, this paper proposes improved joint dictionary learning algorithm. Method of optimal directions is exploited as an updating dictionary algorithm to learn coupled dictionary, and introduces sparse coefficient acquired by learning low resolution dictionary into the high resolution dictionary learning space. Exploiting sparse reconstruction method eventually generates a high resolution remote sensing image. At the same time, this algorithm is optimized, and training samples are automatically intercepted. By experiments, the results show that the proposed approach can achieve better reconstruction quality than existing algorithm in the subjective evaluation criteria. It also demonstrates effectively that the method is much faster than some classic algorithms in the process of learning dictionary, the reconstructed image is more clear and texture structure is more obvious.

Key words: coupled dictionary, method of optimal directions, super-resolution reconstruction, remote sensing imagery, sparse representation

摘要： 针对遥感影像超分辨率重建问题，提出了一种改进联合字典学习的超分辨率重建模型。利用最优方向字典更新算法进行耦合字典对的学习，将由低分辨率字典学习得到的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。该算法通过优化，实现训练样本自动截取，通过验证实验表明：与已有的经典算法相比，提出的算法定量评价指标有明显改善，同时，在字典学习过程中所需时间远少于现有经典算法，大大提高了遥感影像重建的效率，其重建影像更加清晰，几何纹理结构更加明显，证明了该算法的高效性。

关键词: 耦合字典, 最优方向法, 超分辨率重建, 遥感影像, 稀疏表示