基于多尺度链接的红外与可见光图像融合

Infrared and Visible Image Fusion Based on Multi-scale Link

摘要:

提出了一种红外和可见光图像的融合方法。该方法采用深度学习中的生成对抗网络(generative adversarial network,GAN)来融合两种模态图像,融合过程主要通过网络架构中生成器和鉴别器之间的对抗实现。生成器采用多尺度链接架构,有效提取并利用源图像的深层与浅层特征信息。同时鉴别器采用了与传统全局鉴别器不同的局部鉴别器,确保融合后的图像充分包含源图像的信息与特征分布。经实验验证,采用该方法融合后的图像能有效包含两种源图像各自的特征。

Abstract:

A fusion method for infrared and visible image was presented. The methodused generative adversarial network (GAN) of deep learning to fuse two modal images. Thefusion process was mainly achieved through adversarial interactions between generators anddiscriminators of the network architecture. The generator employed a multi-scale linkarchitecture to allow effective extraction and utilization of deep and shallow-level featuresfrom the source images.Moreover, the local discriminator which was distinct fromtraditional global discriminator was used to ensure comprehensive incorporation of theinformation and feature distributions from the source images in the fused output.Experimental results demonstrate the effectiveness of the proposed method in preserving thedistinctive characteristics of both source images in the fused output.