For the problem that the high dynamic range(HDR)reconstruction in dynamic scenar-ios will have artifacts,an HDR reconstruction algorithm based on a densely connected coder-de-coder network is proposed.The algorithm is based on the attention mechanism module in channel and spatial dimensions,which guides and extracts image features corresponding to low dynamic range images with different exposures according to the attention feature map,automatically sup-presses undesired components caused by misalignment and oversaturation,and enhances the de-sired fine details in the non-reference image.The multi-level details of the feature map are gradu-ally extracted in the coding part through the densely connected coder-decoder network module,and dense connection is adopted to enhance the information interaction,and the image detail fea-tures are restored using the decoder,which ultimately generates artifact-less and high-quality high-dynamic-range images.Experiment results show that the algorithm can effectively attenuate the artifact phenomenon and obtain high dynamic range images with rich details.
high dynamic range imageattention moduledense connectionmulti-exposure fu-sionremoving artifacts
NETL
NSTL
万方数据
