DA-MEF: A Multi-exposure Fusion Network via Detail Enhancement and Global Feature Attention

Multi-exposure image fusion (MEF) can produce high dynamic range (HDR) images with enhanced information and details from improperly exposed images. In recent years, MEF algorithms based on deep learning have shown good performance, but most of them have some defects because of the complex network and the neglect of unique features. In this work, we introduce a lightweight MEF method that synthesizes HDR images with high quality and efficiency. Specifically, we propose a comprehensive feature extraction block to accurately capture global information and local details. In this stage, we design a novel attention mechanism named GFA, which extracts valid features by comparing the similarities and differences between two inputs. Additionally, we introduce a detail enhancement fusion module that helps create HDR results with clear edges and rich details. To demonstrate the superiority of our approach, we conduct extensive experiments and compare our method with several cutting-edge techniques. Furthermore, we validate the effectiveness of the proposed modules through ablation experiments.