Infrared and Visible Image Fusion Based on Blur Suppression GAN

Article Introduction

The key to multi-sensor image fusion is the integration of infrared and visible light images. Generative Adversarial Networks (GANs) have significant advantages in automatic feature extraction and subjective visual enhancement for fusing infrared and visible light images. However, the different principles of infrared imaging and visible light imaging lead to edge and texture blurring in the GAN fusion results.

This study proposes an end-to-end generative adversarial network for infrared and visible light image fusion, effectively suppressing edge and texture blurring in the fused images. A residual-in-residual dense block (RRDB+SN) with switchable normalization is used in the generator as the feature extraction module, preserving the main infrared intensity information and texture details, while avoiding artifacts caused by batch normalization. Based on this, an anti-blur loss function based on the Weber local descriptor (WLD) is proposed to mitigate edge and texture blurring caused by the fusion of infrared and visible light images. Experiments validate the effectiveness of the generator structure and the WLD loss function. Qualitative and quantitative experiments conducted on public datasets demonstrate the advantages of this method compared to other state-of-the-art fusion methods.

Author Introduction

Yi Shi (First Author, Corresponding Author), Associate Professor at the School of Mechanical and Electrical Engineering, Chengdu University of Technology. He is responsible for several important scientific projects, including provincial and ministerial key laboratory open funds and national major science and technology projects, and has made a series of innovative academic achievements with significant international impact. His main research directions include deep learning image processing, infrared image processing, drone image processing, and intelligent information devices and system technologies. He has published over 20 papers in important domestic and international academic journals, including Pattern Recognition, Neurocomputing, Infrared Physics & Technology, Measurement, Chinese Journal of Electronics, and holds 6 domestic and international invention patents.

Yi Shi is a high-level paraplegic teacher working on the front lines of teaching and research. He was awarded the title of Most Beautiful Teacher in Sichuan Province in 2022, is a member of the Sichuan Disabled Inspirational Report Group, and has received other honors such as Model of Self-improvement in Chenghua District, Chengdu.

Infrared and Visible Image Fusion Based on Blur Suppression Generative Adversarial Network

YI Shi, LIU Xi, LI Li, CHENG Xinghao, WANG Cheng