In this article, we proposed a novel image fusion method based on multiscale convolution sparse decomposition (MCSD). A unified framework based on MCSD is first utilized to decompose panchromatic (PAN) image and the spatial component of upsampled low spatial resolution multispectral (LR MS) images, which can produce the corresponding low frequencies and feature maps. By combining convolution sparse decomposition with multiscale analysis, MCSD can efficiently approximate the spatial and spectral information in images. Next, a binary map generated from gradient information is utilized to integrate the low frequencies of LR MS and PAN images. For feature maps, the fusion gain for each pixel is calculated according to the similarity between the local patches from them. Finally, the fused image is reconstructed by the sum of fused low frequency and feature maps. Some experiments are conducted on QuickBird and GeoEye-1 satellite datasets. Compared with other methods, the proposed method performs better in visual and numerical evaluations.

中文翻译：

---

使用多尺度卷积稀疏分解融合全色和多光谱图像

在本文中，我们提出了一种基于多尺度卷积稀疏分解（MCSD）的新型图像融合方法。首先利用基于 MCSD 的统一框架来分解全色 (PAN) 图像和上采样的低空间分辨率多光谱 (LR MS) 图像的空间分量，可以产生相应的低频和特征图。通过将卷积稀疏分解与多尺度分析相结合，MCSD 可以有效地逼近图像中的空间和光谱信息。接下来，利用从梯度信息生成的二值图来整合 LR MS 和 PAN 图像的低频。对于特征图，每个像素的融合增益是根据它们的局部补丁之间的相似性计算的。最后，融合图像由融合低频和特征图之和重建。一些实验是在 QuickBird 和 GeoEye-1 卫星数据集上进行的。与其他方法相比，所提出的方法在视觉和数值评估方面表现更好。