Infrared and visible image fusion via gradientlet filter and salience-combined map

In this study, we innovatively propose salience-combined map and gradientlet filter for infrared and visible image fusion. It can enhance the infrared image of the target and also retain more detailed textures. First, our method is based on a multi-scale decomposition framework and gradientlet filter to decompose the source graph into approximate layers and residual layers. The approximate layers preserve smooth areas of the source images without edge blurring. The residual layers reflect the small gradients and noise of the source image. Since the texture part of the residual layer is weak, we introduce a Gamma-enhanced gradient map to complement the texture. The initial fusion image can be obtained by fusing the approximate layers and the residual layers. The salience-combined map directly extracts salient objects from infrared images according to pixel threshold segmentation, and extracts background information other than objects from visible images. Then the salience-combined map is used to guide the initial fusion image to get the final image. In our qualitative analysis, we compared our method against 5 traditional methods and deep learning-based methods. In the quantitative assessment, utilizing 29 pairs of randomly selected source images, our algorithm distinctly showcased absolute superiority across various metrics, including EN, SF, AG, and FD. The aforementioned results affirm that our method ensures the generation of fused images with clear targets and rich details.