Image Restoration Based On Adaptive Blur Kernel Extraction

When taking pictures, due to various factors such as camera shake or relative movement, the captured image may become blurred. Shot moments of taking pictures, especially the remote sensing image, cannot be reproduced, making the restoration technology essential. For different scenarios, in this paper we propose three methods to extract the blur kernel of blurred images and to suppress the ringing effects during the restoration to obtain restoration effects with improved resolution.Three novel extraction methods are as follows. First, for the space target image with a starry background, we use Wiener filter to remove noise, with the best threshold segmentation method extracting the star zone, and we choose the most suitable blur kernel based on the maximum energy. Second, for the imaging system with an Inertial Measurement Unit, at first we get the motion information with the help of inertial device. Then we get the blur kernel using coordinate transformation. Third, we propose a method to get the kernel by human-computer interaction(HCI), which is extracting an image block which can show us the blur information, and then use a series of morphological treatments such as expansion, corrosion and thinning to optimum the kernel.After that we analyze one problem of deconvolution--the ringing effects, and propose solutions to reduce the respective ringing. We eliminate the boundary ringing by expanding our image border, making a larger image. For edge ringing, as the fact that distribution of the smooth area between blurred image and sharp image is similar, we propose a modified RL algorithm using Gaussian distribution to constraint their image gradient distributions, which reduces ringing effects effectively during the deconvolution process.Our paper presents three methods of extracting blur kernel and one method to suppress the ringing, confirmed the methods effective by experiments of real shot images.