Deblurring of Low Light Images using Light-streak and Dark Channel

Abstract

Images taken by smartphones or handheld cameras in night or dark places are often blurry. To solve this, the device should be steady and have long exposure time. There are many existing deblurring techniques, but they often fail when the image does not have enough salient features. Thus using state of the art approach may result in failure and needs specific methods for deblurring of low light images. Light streaks are common in most of the low light images and are formed from point light sources. It contain more information regarding camera motion and blur kernels. Also while low light images contain dark pixels (very low intensity pixels), these pixels are not very dark when averaged with neighboring high intensity pixels during the blur process. The deblurring approaches based on light steaks and dark pixels are rare. Here a method is developed to select a light streak for kernel estimation and introduces a non-linear blur model that explicitly takes dark channel prior into account for estimating the blur kernel in an optimization framework.