Blind Restoration Of Blurred Image Based On Camera Shake

In photography, shaking of camera or relative movement of objects usually cause image blurry. Usually, the image blurry is called motion blurring if it is caused by the relative movement of the imaging sensor to the scene. The camera-shake blur that is studied in this paper also belongs to this kind. Currently, as the digital technology has become more sophisticated, the weight of the hand-held digital camera becomes lighter and the size is smaller. Camera-shake will inevitably occur during pressing the camera shutter. Especially for mobile phone cameras, the probability of this happening is greater. Because camera-shake blur makes some valuable photos missing important information, people need to restore a clear image from the blurred image urgently, to obtain useful information. Therefore, it is useful and essential to study and develop an efficient and fast deblurring technology.By convention, camera-shake can be modeled as the sum of convolution of the pure image with a blur kernel and noise produced by the camera-shake. Therefore, deblur process is called image deconvolution. When the kernel is known, the image deconvolution is called non-blind deconvolution. If the kernel is unknown, it is called blind deconvolution. Generally, the path of camera-shake is random and unpredictable. In this case, the process to obtain a clear image is called blind image restoration, or blind deconvolution. This paper is focus on removing camera-shake from a single blurred image using blind restoration. First, we estimate kernel which is the difficulty during the process. Secondly, restoring a clear image based on the estimated kernel. The algorithm is divided into two steps in this article. The first step is estimating kernel. We perform estimation by varying image resolution in a coarse-to-fine manner. At one level, we use bilateral filter, shock filter and EMD to obtain an initial latent image. Then combining the blurred image, we can estimate the kernel. We use a spatial prior to guide the recovery of a coarse version of the latent image. The second step is to obtain the sharp image by the Richardson-Lucy Algorithm based on the estimated kernel. Blurred images are selected from Levin Photo Gallery. Through simulation, it is confirmed that the algorithm proposed in this paper can estimate the effective kernel and get a clear image.