基于残差密集块和自编码网络的红外与可见光图像融合

红外与可见光图像融合是复杂环境中获得高质量目标图像的一种有效手段，在目标检测与跟踪、图像增强、遥感、医疗等领域有广泛应用前景.为解决目前基于深度学习的红外与可见光图像融合方法中存在的网络无法充分提取特征、特征信息利用不充分和融合图像清晰度低的问题，本文提出了一种基于残差密集块的端到端自编码图像融合网络结构，利用基于残差密集块的编码器网络将图像分解成背景特征图和细节特征图，然后将两种特征图进行融合，再通过解码器进行重构，还原出最终的融合图像.测试结果表明，本文的方法可以得到清晰度高、目标突出、轮廓明显的融合图像，在SF、AG、CC、SCD、Qabf、SSIM 6个融合质量评估指标上与目前代表性融合方法相比均有不同程度的提升，特别是在融合图像清晰度上优势明显，且对于模糊、遮挡、逆光、烟雾等复杂环境图像有较好的融合效果. 

Infrared and Visible Image Fusion Based on Residual Dense Block and Auto-Encoder Network

Infrared and visible image fusion is an effective means to obtain high-quality target images in complex environments. It has broad application prospects in the fields of target detection and tracking, image enhancement, remote sensing, and medical treatment. In order to solve the problems of the current deep learning-based infrared and visible image fusion methods that the network cannot fully extract featuresd, cannot fully utilize the feature information, and the clarity of fusion image is low, this paper proposes an end-to-end image fusion network based on residual dense block and auto-encoder, which uses an encoder network based on residual dense block to decompose the image into a background feature map and a detailed feature map, after that the two feature maps will be fused, and then reconstructed by the decoder to restore the final fusion image. The test results show that the method in this paper can obtain a fused image with high definition, prominent target and obvious outline, compared with the current representative fusion methods, the six fusion quality evaluation indicators of SF, AG, CC, SCD, Qabf, and SSIM have been improved in different degrees, especially has a huge advantage in the clarity of the fusion image. And for complex environmental images such as blur, occlusion, backlighting, and smoke, there is a good fusion effect.