CNN-based LDR-to-HDR conversion system

We propose a system to convert low-dynamic-range images to high-dynamic-range (HDR) images. The system is based on a convolutional neural network (CNN). We use the CNN to distinguish light sources from non-light sources, and use that information to identify regions that should be brightened. Then the contrast is stretched, and the brightness of the region is further enhanced to obtain an HDR image.