基于稀疏表示和字典学习的全色与多光谱影像融合

Remote Sensing Image Fusion Based on the Sparse Representation and Dictionary Learning

Pansharpening方法通过融合多光谱影像的光谱信息和全色影像的空间细节信息来得到高分辨多光谱影像。然而传统的Pansharpening方法易导致产生光谱扭曲和空间信息丢失现象。受到影像稀疏表示超分重建理论启发,本文提出了一种新的基于稀疏表示和字典学习的Pansharpening方法。该方法以影像的高频特征作为训练样本,通过字典学习的方法来获取高低分辨率影像字典,使用正交匹配追踪算法求解出影像的稀疏表示系数,最终通过高分辨影像字典与稀疏系数相乘得到融合影像。实验结果表明:本文提出的方法能很好地保持遥感影像的光谱信息和空间细节信息。

Pansharpeningmethods canfuse the spectral information of muhispectral image and the spatial information of panchromatic image to obtain high -resolution muhispectral image. However, the traditional pansharpening methods usually suffer spectral distortions and spatial detail loss. Inspired by super resolution theory based on image sparse representation, this paper proposes a new pansharpening method based on sparse representation and dictionary learning. The proposed method uses high frequency feature of image as training samplesto obtain image dictionary through dictionary learning method. Sparse coefficientis solved by orthogonal matching pursuit algorithm. Then, the fused image is calculated by multiplyingthe obtained sparse coefficients and the dictionary of the highreso- lution image. Experiment shows that the proposed method can maintain spatial and spectral information of remote sensing image well.