Infrared and visible fusion imaging via double-layer fusion denoising neural network

We propose an infrared and visible fusion imaging method with a double-layer fusion denoising neural network (DFDNN). The DFDNN is designed in an encoder-fusion-decoder architecture and reconstructs the high-quality image from the infrared and visible images captured by the corresponding imaging device. A nest connection architecture is developed to avoid the semantic gap between the encoder and decoder. A noise estimation map is added to DFDNN to achieve the denoising function of the network. An infrared and visible fusion imaging system is built to verify the effectiveness and performance of the proposed method. Experimental results on a public dataset and the practical dataset obtained from the experimental system show that the proposed method performs favorably against the nine state-of-the-art fusion methods in terms of visual perception and quantitative evaluations. The proposed method may find applications in medical imaging and night monitoring.