HDR Image Generator Focused on Saturated Region Restoration with Contextual Loss

In this paper, we propose a novel method for generating a tone mapped high dynamic range (HDR) image from a single low dynamic range (LDR) image. At this time, in order to concentrate on restoring the saturated region of 0 or 255 in the image, in the proposed method, the ratio of L1 loss and GAN loss is set differently according to the ratio of the saturated region. In addition, contextual loss is applied to solve the blur and stain artifacts when L1 loss and GAN loss are applied. In the performance evaluation, the peak signal-to-noise ratio and structural similarity of the proposed method are 29.09 and 0.869, respectively, which outperform the benchmark method.