Research Of Motion Deblurring Based On Regularization Constraint

Blurred images restoration has been a focus in the study of image processing problems, as the popularity of mobile digital cameras in people’s life, the demand of solve the image blur caused by the filming equipment jitter or shooting the moving target has become more and more urgent. The most difficult problem of motion deblurring lies in we need restore the original image details clearly just depend on the single blurred image, from the point of view of mathematics, it is a ill-posed problem seriously. With the focus from scientific research scholars on the issue, in recent years, many excellent works and algorithms have sprung up and make a possible to solve this practical problem, but most of the existing algorithms have their own shortcomings and deficiencies, often characterized by the fuzzy image edge extraction and the select of the clear image constraint, and they also suffered from the ringing effect during the recovery and the speed of algorithm.In this is, we start from the space invariant models of fuzzy image, we introduce and analyze the methods which can recover the blur images caused by the linear motion, and based on this, we consider how to achieve the no-parametric blur kernel, focus on the more applicability image blind deconvolution problem. In our work, we summarize the present works in the handling of related problems, aiming at single blurred image with a global kernel function, propose a new blind deconvolution algorithm. The work of this are mainly on the three points:First, we use the image smooth to achieve the effectively edge for the motion deblur. We analysis the image texture edge cannot effectively express the motion process, instead of make a inaccuracy direction of fuzzy core, the selection of image edge determine the iterative direction of fuzzy core, ensure the accuracy of the kernel estimate.Second, we use the LO minimize gradient regularization as the constraint function of clear image, improve the calculation method of recovery under the multi-scale image framework of different resolutions, so that we get a good middle results closer to the original clear image, that ensures the robustness and efficiency of the algorithm. Also, we use the LO regularization to provide a smooth image to suppress the ringing effect, ensure the recovery in the blind deconvolution image of high quality.Finally, in view of the principle of license plate images, we combined the document image restoration framework, a deblurring algorithm of license plate is proposed by us.According to the characteristics of image binarization and the transparency, we get the original clear image edge information, and based on this, we restored the blur license plate image.