Motion blur is a common artifact in hand-held photography. It has the detrimental effect of derailing the aesthetic value of the captured images; in addition, most computer vision tasks warrant blur-free inputs. Presently, consumer cameras have gone beyond the conventional cameras in order to have additional benefits and functionalities. Three important such imaging devices are rolling shutter camera, light field camera, and unconstrained dual-lens camera. Their increasing popularity has necessitated the need for tackling motion blur in these devices. In this talk, we discuss models and methods for these cameras aimed at “restoring” motion blurred photographs, where we have no particular information about the camera motion or the structure of the scene being photographed – a problem referred to as blind motion deblurring.



First, we consider rolling shutter (RS) sensors which are today present in almost all cameras and smartphones. However, existing deblurring works for RS cameras are constrained by heavy computational cost, need for precise sensor information, inability to handle wide-angle lenses (which most cell-phone and drone cameras have) and irregular camera motion. To this end, we propose an RS blurring model and, based on that, a method that mitigates these issues significantly. Second, we consider the case of light field (LF) cameras. For LFs, the state-of-the-art deblurring method is limited to handling only downsampled LF, and further warrants high-end GPUs which is seldom practical from a consumer-end. To this end, we introduce a divide and conquer strategy for LF motion deblurring. Our approach is CPU-efficient computationally and can effectively deblur full-resolution LFs. Third, we move to the case of unconstrained dual-lens cameras, which have been quite successfully deployed in present-day smartphones. Despite the far-reaching potential of these cameras, there is not a single deblurring method for such systems. To this end, we propose a generalized blur model that elegantly explains the intrinsically coupled image formation model for dual-lens set-up. Further, we reveal an intriguing challenge that naturally disrupts captured depth information of scenes. We address this issue by devising a judicious prior, and based on our model and prior propose a practical blind motion deblurring method for dual-lens cameras. Finally, we focus on motion blur caused by dynamic scenes in unconstrained dual-lens cameras. In practice, apart from camera-shake, motion blur happens due to object motion as well. We address this problem using Deep Learning based on Signal processing. Our signal processing formulation allows accommodation of lower image-scales in the same network without increasing the number of parameters, and further introduces an optimal filtering scheme. We experimentally show that all our proposed techniques have substantial practical merit.