The fusion of infrared and visible images is difficult because of their different modalities. Current fusion methods are difficult to maintain both complementary information and good visual effects, such as methods of region discrimination based on visual saliency and methods based on total variation (TV). Among them, methods of region discrimination based on visual saliency for fusion have better complementary information but poor visual consistency, while methods based on total variation for fusion have good visual consistency, but there is no proper regularization to ensure sufficient selection and fusion of complementary information. In this paper, an improved infrared and visible image fusion method via visual saliency and fractional-order total variation is proposed. First, the infrared and visible images are fused through the saliency map to obtain a fused image, and then the fused image and an selected original image are fused by the fractional-order total variation to obtain the final fused image. In this paper, visual saliency map-based fusion makes the fused image contain as much complementary information as possible from the source image, while fractional-order total variation-based fusion makes the fused image have better visual effects. Compared with the state-of-the-art image fusion algorithm, the experimental results show that the proposed method is more competitive in retaining image texture details and having visual effects.

中文翻译：

---

基于显着图改进图像融合的分数阶总变异

红外和可见光图像的融合是困难的，因为它们的模态不同。目前的融合方法难以同时保持互补信息和良好的视觉效果，例如基于视觉显着性的区域区分方法和基于总变异（TV）的方法。其中，基于视觉显着性进行融合的区域判别方法具有较好的互补信息，但视觉一致性较差，而基于全变异的融合方法具有较好的视觉一致性，但没有适当的正则化来保证互补信息的充分选择和融合。 . 本文提出了一种基于视觉显着性和分数阶总变异的改进的红外和可见光图像融合方法。第一的，红外和可见光图像通过显着图融合得到融合图像，然后融合图像和选定的原始图像通过分数阶总变差融合得到最终的融合图像。在本文中，基于视觉显着图的融合使得融合图像包含尽可能多的来自源图像的互补信息，而基于分数阶总变异的融合使得融合图像具有更好的视觉效果。与最先进的图像融合算法相比，实验结果表明，该方法在保留图像纹理细节和视觉效果方面更具竞争力。基于视觉显着图的融合使融合图像包含尽可能多的来自源图像的互补信息，而基于分数阶总变异的融合使融合图像具有更好的视觉效果。与最先进的图像融合算法相比，实验结果表明，该方法在保留图像纹理细节和视觉效果方面更具竞争力。基于视觉显着图的融合使融合图像包含尽可能多的来自源图像的互补信息，而基于分数阶总变异的融合使融合图像具有更好的视觉效果。与最先进的图像融合算法相比，实验结果表明，该方法在保留图像纹理细节和视觉效果方面更具竞争力。