Multiexposure Image Fusion for Dynamic Scenes

In this study, a multiexposure image fusion approach for dynamic scenes is proposed, which includes four stages, namely, detail enhancement and reference image selection, ghost artifacts removal, weighting map estimation and refinement, and image fusion. Weighted least squares (WLS) optimization is used to enhance the details of input low dynamic range (LDR) images. To remove ghost artifacts, median threshold bitmap (MTB), independent component analysis (ICA), edge detection, and cross-image median filtering are used to detect motion objects. Mertens et al.'s approach is used to perform weighting map estimation and refinement. Finally, the final motion-free HDR-like image is generated by fusing input LDR images with final weighting maps. Based on the experimental results obtained in this study, the quality of the final HDR-like images by the proposed approach is better than those by the five comparison approaches.