Richardson-Lucy Deblurring for Scenes under Projective Motion Path

This paper addresses the problem of modeling and correcting image blur caused by camera motion that follows a projective motion path. We introduce a new Projective Motion Blur Model that treats the blurred image as an integration of a clear scene under a sequence of projective transformations that describe the camera’s path. The benefits of this motion blur model is that it compactly represents spatially varying motion blur without the need for explicit blurs kernels or having to segment the image into local regions with the same spatially invariant blur. We show how to modify the Richardson-Lucy (RL) algorithm to incorporate our Projective Motion Blur Model to estimate the original clear image. In addition, we will show that our Projective Motion RL algorithm can incorporate stateof-the-art regularization priors to improve the deblurred results. Our Projective Motion Blur Model along with the Projective Motion RL is detailed together with statistical analysis on the algorithm’s convergence properties, robustness to noise, and experimental results demonstrating its overall effectiveness for deblurring images.