Deep Chain HDRI: Reconstructing a High Dynamic Range Image from a Single Low Dynamic Range Image

In this paper, we propose a novel deep neural network model that reconstructsa high dynamic range (HDR) image from a single low dynamic range (LDR) image.The proposed model is based on a convolutional neural network composed ofdilated convolutional layers, and infers LDR images with various exposures andillumination from a single LDR image of the same scene. Then, the final HDRimage can be formed by merging these inference results. It is relatively easyfor the proposed method to find the mapping between the LDR and an HDR with adifferent bit depth because of the chaining structure inferring therelationship between the LDR images with brighter (or darker) exposures from agiven LDR image. The method not only extends the range, but also has theadvantage of restoring the light information of the actual physical world. Forthe HDR images obtained by the proposed method, the HDR-VDP2 Q score, which isthe most popular evaluation metric for HDR images, was 56.36 for a display witha 1920$\times$1200 resolution, which is an improvement of 6 compared with thescores of conventional algorithms. In addition, when comparing the peaksignal-to-noise ratio values for tone mapped HDR images generated by theproposed and conventional algorithms, the average value obtained by theproposed algorithm is 30.86 dB, which is 10 dB higher than those obtained bythe conventional algorithms.