HDR imaging aims to extract reliable information from multiple low dynamic range exposures to generate high quality high dynamic range (HDR) images. However, in degraded environments, factors such as noise, blur, and motion induced ghosting can significantly impair reconstruction. To address this, we propose dual attention transformer (DAT), a novel architecture for joint HDR imaging and multi-type degradation correction. To robustly integrate spatial structure information across exposures, DAT introduces a structural connectivity stream, enhancing inter-frame structural complementarity and modeling complex nonlinear relationships, which allows effective extraction of key structures while suppressing motion and degradation induced distortions. To obtain high fidelity features with both semantic understanding and detail restoration, DAT employs a cross dimensional interaction mechanism for joint global spatial and high dimensional channel attention, enabling accurate identification and separation of multiple degradation factors and overcoming limitations of existing methods in extracting degradation-related channel features. Additionally, a hybrid loss provides targeted supervision for degradations, while combined gamma correction moderately brightens underexposed frames to improve signal-to-noise ratio and amplify details in dark region. Extensive experiments show that DAT achieves significant performance improvements under various synthetic degradation scenarios while maintaining a favorable balance between reconstruction quality and computational cost. Our code is publicly available at https://github.com/gzzzzzz781/PINet.
Gong et al. (Thu,) studied this question.