Semantic Injection Infrared and Visible Image Fusion Via a Disentangled Representation Network

Image fusion is the task of generating images from the same scene with different sensors. Images obtained by different sensors have different expressions under the image target semantics. Good semantic aware images can better serve the subsequent advanced image processing tasks. Most of the existing fusion methods are based on the underlying logic of the image, without considering the high-level semantic fusion. We propose a semantic injection disentangled representation network SDR. The source image is decomposed into structural and modal features by the corresponding encoder, and a generator based on semantic injection is constructed to generate the fused image under Semantic injection. On the open data set, our fusion network is compared with eight most advanced methods. Qualitative and quantitative experiments show that our method is superior to the current most advanced fusion methods. The fusion result preserves the structural information of the infrared image and the texture details of the visible image. The model proposed in this paper combines the advantages of feature decoupling and semantic injection, and has a better effect on fused images.