A Multi Exposure Synthetic HDR Image Algorithm based on Cascaded Network

At present, the large challenge of synthesizing HDR (high dynamic range) images from multiple exposure images is the problem of artifacts caused by moving objects and camera shake in dynamic scenes. In this paper, a three-layer cascaded network structure of HDR image generation method is proposed. With the deepening of the network, images with different exposure levels are input for processing. Each sub network only converts and extracts the feature of one exposure level image. Finally, a gradually improved HDR image is obtained through the cascaded structure. Through this cascaded network processing, the mutual interference and confusion of dynamic regions between multiple images can be reduced. Compared with the existing HDR generation methods, this method can solve the problem of artifacts and keep the texture details of the generated HDR image well.