📝 Abstract:
The increased interest in consumer-grade high dynamic range (HDR) images and videos in recent years has caused a proliferation of HDR deghosting algorithms. Despite numerous proposals, a fast, memory-efficient, and robust algorithm has been difficult to achieve. In this thesis, we address this problem by first leveraging the power of attention and U-Net-based neural representations and using a conservative hybrid deghosting strategy to enable the deployment to hardware-constrained devices. Then, we explore the use of event data with the bracketed low dynamic range (LDR) RGB image data to gain temporal precision and HDR information on the dynamics of the scene during the capture of the LDR bracket to guide the minimization of ghosting artifacts further.