Video Deblurring with Pixel-wise Duty Cycle

Within certain current video deblurring techniques, it's commonly presumed that the camera's duty cycle is stable and unchanging. Indeed, when the camera is auto-exposed, its operational cycle remains uncertain and variable. A flawed camera duty cycle might reduce the precision of the blur kernel and the motion vector within a frame, affecting the efficiency of video deblurring. To address the aforementioned issue, we propose a technique for video deblurring that relies on calculating the pixel-wise duty cycle (PDC) to reinstate blur videos resulting from camera tremors or object motion. The proposed technique is segmented into two parts: the section estimating the space-variant blur kernel and the section for restoring sharp frames. The initial part introduces a PDC technique that merges optical flow with a projective motion blur model to calculate the ratio of length between motion vectors within and between frames for each pixel. The second part involves building a pyramid frame to alternate between the latent sharp frame, optical flow, and PDC, enhancing the accuracy of deblurring outcomes. We propose an energy-driven function, incorporating spatial and temporal regularization, to mitigate the ringing artifacts and noises in the deblurred outcome, thereby controlling the hidden sharp frame, optical flow, and PDC. The experimental findings indicate that the suggested technique is capable of accurately predicting and eliminating intricate, uneven velocity motion blur in videos with natural blurring.