基于图像退化模型的红外与可见光图像融合方法

基于深度学习的红外与可见光图像融合算法依赖人工设计的相似度函数衡量输入与输出的相似度，这种无监督学习方式不能有效利用神经网络提取深层特征的能力，导致融合结果不理想。针对该问题，该文首先提出一种新的红外与可见光图像融合退化模型，把红外和可见光图像视为理想融合图像通过不同退化过程后产生的退化图像。其次，提出模拟图像退化的数据增强方案，采用高清数据集生成大量模拟退化图像供训练网络。最后，基于提出的退化模型设计了简单高效的端到端网络模型及其网络训练框架。实验结果表明，该文所提方法不仅拥有良好视觉效果和性能指标，还能有效地抑制光照、烟雾和噪声等干扰。

Infrared and Visible Image Fusion Method Based on Degradation Model

Infrared and visible image fusion algorithms based on deep learning rely on artificially designed similarity functions to measure the similarity between input and output. The unsupervised learning method can not effectively utilize the ability of neural networks to extract deep features, resulting in unsatisfactory fusion results. Considering this problem, a new fusion degradation model of infrared and visible image is proposed in this paper, which regards infrared and visible images as the degraded images produced by ideal fusion images through mixed degradation processes. Secondly, a data enhancement scheme for simulating image degradation is proposed, and a large number of simulated degradation images are generated by using high-definition datasets for training the network. Finally, a simple and efficient end-to-end network model and its network training framework are designed based on the proposed degradation model. The experimental results show that the method proposed in this paper not only has good visual effects and performance indicators, but also can effectively suppress interferences such as illumination, smoke and noise.