Deblurring a Camera-Shake image using a Thinning Kernel
Shwu-Huey Yen, Pin-An Chen, Hwei-Jen Lin · 2018
The task of blind deblurring usually consists ofestimation of interim images and blur kernels. Due to thelack of information in kernels compared to that in interimimages, when only a blurred image is available, most ofdeblurring methods emphasis the estimation of interimimages. However, the resulting kernel is often wider than itshould be, thus degrading the quality of the deconvolvedimage. To remedy the problem of wide kernels, we presenta thinning scheme to better estimate a kernel. In this way, aclear image can be recovered from a camera-shake blurredimage. To mitigate the insufficient information of blurkernels, we make simple inferences and assumptions forkernels based on the trajectory of the camera shake. Underthese inferences and assumptions, we use a three-stepapproach to estimate the blur kernel. Firstly, we relax thecondition to find the shape of the blur kernel. Next, we usea thinning algorithm to obtain the skeleton of the blurkernel. Thirdly, we reweight the blur kernel by Gaussiandistribution. By repeating these steps a few times we canget a more accurate blur kernel. Finally, we can reconstructa high quality deblurred image by using the blur kernel. Theproposed method is tested by a public database and ourresults outperform those of two similar methods.