基于双分支Transformer与动态图增强的多曝光图像融合方法

A Multi-Exposure Image Fusion Method Based on Dual-Branch Transformer and Dynamic Graph Enhancement

摘要: 多曝光图像融合(MEF)的目的是将不同曝光度的图像融合为一张细节清晰且亮度均衡的结果图像。针对现有方法在亮度调节不足、结构保持有限及多尺度特征建模不充分等问题，本文提出一种基于双分支Transformer与动态图增强的多曝光图像融合方法。通过双分支Transformer架构实现局部与全局特征的互补建模，有效兼顾细节清晰度与结构一致性的同时引入动态图增强机制，自适应捕捉跨曝光图像间的动态依赖关系，克服了传统静态融合的局限，进一步结合亮度调节与边缘保持策略，使得结果图像在视觉自然性与感知层次方面显著提升。在SICE、MEFB等公开数据集上的实验结果表明，本方法在主观感知与客观指标方面均优于现有方法，展现出良好的鲁棒性与视觉表现。

Abstract: Multi-exposure image fusion (MEF) aims to merge images with different exposure levels into a single result image featuring clear details and balanced brightness. Addressing limitations in existing methods—such as inadequate brightness adjustment, limited structural preservation, and insufficient multi-scale feature modeling—this paper proposes a multi-exposure image fusion approach based on a dual-branch Transformer and dynamic image enhancement. The dual-branch Transformer architecture enables complementary modeling of local and global features, effectively balancing detail clarity and structural consistency. Simultaneously, the introduction of a dynamic image enhancement mechanism adaptively captures dynamic dependencies across exposure images, overcoming the limitations of traditional static fusion. Furthermore, the integration of brightness adjustment and edge preservation strategies significantly enhances the visual naturalness and perceptual quality of the resulting images. Experimental results on public datasets such as SICE and MEFB demonstrate that our method outperforms existing approaches in both subjective perception and objective metrics, exhibiting robust performance and superior visual quality.