DeepHDR-GIF: Capturing Motion in High Dynamic Range Scenes

In this work, we have proposed a novel computational photography application to generate a Graphics Interchange Format (GIF) image corresponding to High Dynamic Range (HDR) scene involving motion. Though HDR image and GIF image are prevalent in the computational photography community for a long time, according to our literature survey, this is the maiden attempt to combine them in a single framework. Like most other HDR image generation algorithms, the first step in the proposed framework is to capture a sequence of multi-exposure (−2EV, 0EV, 2EV) low dynamic range (LDR) images. The decided exposures (−2EV, 0EV, 2EV) are varied in a round-robin fashion, and continuous frames are captured to get adequate information about the motion of the scene. The next step is to combine sets of three consecutive multi-exposure LDR images to generate HDR images. Further, we take two successive HDR images and produced three in-between frames in a binary-search manner. At last, generated HDR frames and interpolated frames are merged in to a GIF image, which depicts the motion in the scene without losing out on the dynamic range of the scene. The proposed framework works on different types of dynamic scenes, Object movement or Camera Movement, and the results are observed to be visually pleasing without any noticeable artifacts.