Non-uniform Motion Deblurring for Camera Shakes

This paper presents a novel blind motion deblurring algorithm for dealing with non-uniform motion blurs caused by camera shakes. Conventional motion deblurring algorithms assume the blur can be described by a uniform point spread function (PSF). However, this assumption is typically invalid in practice. We exploit a new representation of motion blurs which models the blur effects of camera shakes using a set of planar perspective projections (i.e., homographies). This representation can fully describe the motions of camera shakes in 3D world which cause non-uniform motion blurs. Our main contribution is the blind motion deblurring algorithm associated with this representation. We solve the ill-posed PSF estimation problem by transforming it into a well-posed image registration problem that estimates homographies. To improve the robustness and to resolve the initialization problem of our algorithm, we use multiple input images for deblurring. Our algorithm is experimented with both synthetic and real world examples, and produces superior deblurring results compared to the previous algorithms using a uniform PSF.