Reconstructing Reality Unified Correction of Rolling Shutter and Motion Blur from a Single Image

Abstract

of rolling shutter (RS) distortions and motion blur (MB), especially under camera or object motion. Unlike CCD sensors with global shutters, CMOS sensors capture images row-by-row, leading to geometric distortions when motion occurs during exposure. These distortions are often further complicated by motion blur, resulting in visually degraded images. This paper addresses the challenging task of restoring a single image affected by rolling shutter motion blur (RSMB) a combination of both artifacts without requiring multiple frames or auxiliary sensors. We adopt a model-based approach that represents RSMB as a weighted integration of the sharp latent image transformed under varying camera poses. By discretizing the pose space and applying the Efficient Filter Flow (EFF) approximation, we enable fast and spatially accurate deblurring using only a single input image. The proposed framework not only recovers visually faithful reconstructions but also provides practical benefits for real-world applications where re-capturing the scene is infeasible.