Attention Guided Network for Multi Exposure Image Fusion

Abstract

The lack of ground-truth fused images for supervised learning, and exiting multi-exposure image fusion suffer from loss of edge features and blurred detail. To address these problems, we propose an attention guided network for multi-exposure image fusion. First, a dual channel Unet with independent weight is established, the feature of the target in different exposure images is extracted, the high-dimensional multi-scale feature maps of different exposure images are generated; Then, through visual attention mechanism generated the logical mask of the target region of interest area and superimposed on the high-dimensional multi-scale feature maps to highlight the target features and suppress the nontarget area. Finally, we concat the filtered high-dimensional multi-scale features, and the edge detailed information of underexposed and overexposed regions is preserved by dilated residual dense block and the high-dynamic range image is generated by building a feature reconstruction module. Based on end-to-end network and using content loss and structure loss calculation strategy constrain the neural network to achieve unsupervised learning. The experiment results indicate that the proposed methods achieve better imaging performance not only reducing the interference of background brightness information but also preserve the texture of under- and over-exposure images.