Fast Image Deblurring Method Based On Guided Image

With the popularity of smart phone and other digital products,images become one of important medium for people to transmit information.However,capturing a high quality photo is not a trivial task as there exist camera shake or the motion of objects in the scene during the exposure time.The camera shake or the motion of objects will lead to blurred images.Therefore,image deblurring has become an urgent demand and an important research topic in image processing.Particularly,capturing blurred images under dim light conditions is also a common phenomenon.This is because a long exposure time is usual y required under dim light conditions.The captured images may end up to be blurry due to inevitable motion between the hand-held camera and the scene.Moreover,the saturated pixels or regions produced by bright objects in the scene usual y exist in this kind of blurred images.Removering this kind of blurred images has many chal enges.Saturated pixels and regions cannot be modeled well by linear convolution model that most existing deblurring approaches assume.The unnatural darkness and less reliable edges in this kind of blurred images also make those edge selection based deblurrng methods fail to obtain pleasing results.In this paper,we first make a comprehensive study about the guided filtering.The most important motivation of the guided filtering is that the filtering results depend on the guidance image.Motivated by guided filtering,we propose a fast image deblurring method based on a guided image.Although the flash images cannot record the ambient light,their clear structures still have a compensation of the blurred image captured in the same scene.The clear structures in corresponding no-flash image are used as guidance to help restoring a latent sharp image.In the optimization stage,the Iterative Re-weighted Least Squares(IRLS)method is employed to obtain the solutions of the proposed model.Different from most existing deblurring methods,the proposed method does not require the kernel estimation step,which avoids the alternating iteration strategy used in the existing deblurring methods,thus greatly decreasing the running time.Experimental results show that the proposed guided image based deblurring method is capable to generate clear images from low-light blurred images,and performs well on the blurred images with saturated pixels or regions.In addition,the proposed algorithm needs less computational time compared to the existing methods,which demonstrates its efficiency.