基于训练字典的遥感图像融合

为了提高多光谱图像与全色图像的融合质量，利用稀疏表示理论，提出了一种基于训练字典的融合算法。该算法对多光谱图像的亮度分量进行亮度平滑滤波（SFIM）得到新的亮度分量，利用图像块随机采样学习得到的训练字典对全色图像和新的亮度分量进行稀疏表示，采用空间频率取大的融合规则对稀疏系数进行融合，通过重构和IHS逆变换得到融合结果。对不同场景、不同卫星的多光谱图像和全色图像进行实验，结果表明，该方法能在提高空间分辨率的同时更好地保持光谱特性。

Remote sensing image fusion based on training dictionary

An image fusion algorithm is presented based on sparse representation and training dictionary in order to improve the fusion quality of multi-spectral image and panchromatic image. A new intensity component is firstly obtained by taking the method of smoothing filter-based intensity modulation on the original intensity component of the multi-spectral image. The panchromatic image and intensity component are obtained for their sparse coefficient through training dictionary which generated by randomly sampling raw patches. Then, the fused coefficient is obtained by the fusion rule which choosing spatial frequency maximum. Finally, the fused results are obtained through reconstruction and IHS inverse transformation. The improved algorithm has been tested on various multi-spectral and panchromatic satellite remote sensing images for different areas. The experimental results prove that the proposed method can improve the spatial resolution and better maintain the spectral characteristics.