Based on image gradients, an alternative image multi-scale decomposition approach for remote sensing image processing is derived through solving a Poisson equation in this paper. First, the image gradients are obtained. Second, the gradients with given magnitudes are attenuated using some operators. Finally, the modified gradients are used to solve a Poisson equation to obtain the smooth base layer and thus the detail layer. This multi-scale decomposition approach is used for remote sensing image enhancement and remote sensing image smoothing, which outperforms the other methods such as bilateral filter (BLF) and weighted least squares based multi-scale decomposition (WLSMD) in edge-preserving image decomposition via experiments.