Infrared and visible image fusion method based on saliency detection and multi-scale texture detail feature extraction

A single Visible (VIS) image or Infrared (IR) image cannot clearly present the texture details and heat source information of the scene. The fusion task of IR images and VIS can combine the respective characteristics of the two types of modalities to form a new image with rich details, which can better serve advanced vision tasks. Firstly, Kmeans and connected component analysis are used to obtain the IR saliency map. Then, we use Gaussian pyramid to perform multi-scale decomposition of VIS images. Use Gabor filter to extract texture to generate VIS detail map. Finally we use the adaptive fusion rule to obtain high-quality fused images. By conducting a large number of experiments on TNO and LLVIP, and compared them with existing image fusion methods. The experimental results show that our proposed algorithm has potential performance in both quantitative and qualitative evaluations, and has real-time performance during the fusion process.