基于差分融合与边缘增强的轻量级红外与可见光图像融合算法

A Lightweight Infrared and Visible Image Fusion Algorithm Based on Difference Fusion and Edge Enhancement

为了解决现有融合方法在融合性能与计算资源消耗上存在不平衡的问题,提出了一种基于差分融合与边缘增强的轻量级红外与可见光图像融合算法。该方法通过基于结构重参数化的编码器和深度编码块来提取图像特征信息,用差分融合模块(DFM)融合不同模态的特征信息,再将融合后的特征信息通过边缘增强模块(EEM)来强化特征的边缘信息。在训练完成后,用结构重参数化技术优化多分支结构和推理速度,在融合性能不变的情况下,降低网络的计算资源消耗。最后,在MSRS和TNO数据集上进行的实验表明,所提方法在视觉效果和定量指标方面具有优越性。

For purpose of solving the imbalance between the performance and calculating resource consumption of fusion methods existed,a lightweight infrared and visible image fusion algorithm based on difference fusion and edge enhancement was proposed,which has a structure reparameterization-based encoder and deep encoding blocks adopted to extract image feature information,make use of the difference fusion module(DFM)to fuse different modal feature information,and then enhance the edge information of the feature information fused through an edge enhancement module(EEM).After the training,having the structure reparameterization technique employed to optimize both multi-branch structure and inference speed was implemented,including reducing computational resource consumption of the network while maintaining the same fusion performance.Experiments on the MSRS and TNO datasets show that,the method proposed has superiority visual effects and quantitative metrics.