Research On Image Restoration Method For Motion Blur

With the development of computer technology, image restoration has become one of the hottest research topics in digital image processing both at home and abroad, which has been applied to astronomical observations, remote sensing, military, medical imaging, biological research, criminal case solving, transportation, industrial vision, video restoration, and many other aspects. Image restoration been presented as a deconvolution problem belongs to the inverse problem which has an ill-posed property in mathematical physics, especially that dealing with blurred image with highly noise pollution is a challenging problem. Although there are many mature image deblurring methods, there exist many technical problems to be resolved, i.e. blur kernel estimation and modeling of irregular blur situation are still at the primary research stage.In this paper, the principles of digital image restoration techniques and the dominant restoration algorithms are discussed. The analysis and comparison of the advantages and drawbacks of existing restoration methods are presented, and from the point of view of kernel estimation, two methods of kernel estimation and the corresponding restoration algorithm are proposed. And also the mechanism of ringing effect in image restoration and its elimination method are analyzed. And a comprehensive image quality assessment aiming at the image restoration effect is proposed according to the characteristic of noise and blur.First, we get the blur kernel from two same scene shot with a noise image under short time exposure and a blur image under normal exposure, and then obtain the different smoothed images restored on the residual maps, and finally get clear result combined with images of different level of details by the joint bilateral filter. Experiments show that this method can extract accurate blur kernels, and RL algorithm on the residual maps with a gain control can suppress the amplifying noise and the spreading ringing. In order to obtain a robust blind image restoration method that is applicable to different blurred images and without complicated parameter tuning, a robust blind image restoration method of extracting blur kernel from a single image and deblurring is proposed. In this paper, an approach of blur kernel estimation is presented according to the relationship of image edges between blurred images and non-blurred ones, and a robust blind image restoration method is constructed under the framework of multi-scale space with some adaptive parameters introduced in each sub-algorithm. A variety of no-reference image quality evaluation methods on the result show, compared with previous blind image restoration methods which usually adopt the iterative method of minimization of energy function of blur kernel and clear image alternately, this method has a better deblurring ability on the basis of suppression of noise and ringing with simpler parameter tuning, algorithm is stable and steady, and its computation speed is 3 times faster.In addition, an effective method with combination of noise and blur intensity is used in image quality evaluation of blind image restoration for the first time. Experimental results show, comparing with the traditional evaluation methods, this method can balance the noise or ringing spreading and deblurring level on restored images availably, and can match the level of human visual system to a certain degree.