DSG-Fusion: Infrared and visible image fusion via generative adversarial networks and guided filter

Highlights：

• A novel GAN based image fusion method is proposed (DSG-Fusion). By integrating the guided filter into an end-to-end deep learning framework, the DSGFusion can not only extract more background and detail information, but also avoid the manual design of complex fusion rules.

• Features of two source images are extracted by two independent data flows, and their base layers and detail layers participate in feature extraction and fusion process respectively. The double-stream architecture of the designed network can extract more representative information from source images.

• The DSG-Fusion is compared with five existing methods on two public datasets. The extensive experimental results illustrate that DSG-Fusion preserves more intensity and texture information from source images, and the background and details are clearer.