We systematically analyze the parameters of rolling guidance filtering and propose a hybrid multi-scale decomposition method based on rolling guidance filtering, utilizing the law of parameter variation. First, the infrared and visible images are decomposed into base, small-scale, and large-scale layers using this method. Second, a fusion rule that combines the pixel and gradient energies is applied to the base layer, and another fusion rule based on an adaptive dual-channel pulse-coupled neural network (DAPCNN) is used to combine the large- and small-scale layers. Finally, the fused image is obtained via inverse hybrid multi-scale decomposition. Compared with other common image decomposition methods, the proposed method can not only extract the image's texture details and preserve its edge features but also prevent the halo phenomenon at the edges. The experimental results show that the proposed method can extract the target information from the infrared image and fuse it into the visible image. Compared with existing fusion methods, the proposed method shows clear advantages not only in a subjective human visual evaluation, but also in several objective evaluation metrics, namely the mutual information, information entropy, standard deviation, non-linear correlation information entropy, and Chen-Varshney indexes.