Research On Deblur Algorithm Of Motion Blurred Image

Since images are related to human life,the demand for high-quality images is increasing.But the image is susceptible to various factors that causes the quality to fall,a motion blurred image is one of the degraded images need to restored.Deblur algorithm research not only is an important part of image restoration,but also is of great significance in the field of digital image processing.There are two important work in deblur algorithm,the correct estimation of blur kernel is an important prerequisite for deblurring,the deconvolution to abtain high-quality clear image is the ultimate goal.Although a lot of solutions are proposed in present,there are still some problems to be solved,such as multiple iterations lead to serious time consumption;alternating iterations to update the clear image and blur kernel,the cost function model designed is bad or includes too many deconvolutions,these lead to ringing effect;at the same time if the motion blurred image with noise or image is not motion blurred image,it will affect the quality of the final image recovery while deblurring directly.This thesis will innovate and improve these problems.First of all,in order to ensure the correctness of object,according to the natural image gradient obeys the characteristics of the heavy tail distribution and evaluates coefficient satisfies the set threshold to identifying whether the image is a blurred image before deblurring the motion blurred.After determining the blur image,then according to the gradient gradient of the image whether it contains parallel dark stripes to distinguish between motion blur and other types of blur.These blur identification and judgment of blur types are able to handle the correct blurred image,so that not affecting the quality of the final restoration image.Secondly,according to the characteristics of the sparse difference between the clear image and the motion blurred image after the second sparse,this thesis proposes a second sparse mixed a priori cost function model with good stability,to avoid the cost function at the minimum point of the solution is a blur image.At the same time,in order to reduce the time consumption caused by multiple iterations,the method of fast threshold shrinkage is used to speed up the iterative step in updating the clear image.To solve the ringing effect,this thesis designs a non-blind deblur algorithm combining the super-Laplacian priori and the L0 regularization gradient prior after obtaining the exact blur kernel in blind deblurring stage.Bilateral filtering eliminates the artifacts of obtained clear image of the two algorithms,resulting in a high quality clear image.The proposed method not only can get better quality images,but also increase the speed.Thirdly,The image is susceptible to noise pollution,and the deblur algorithm is also sensitive to noise,so it is necessary to study the motion blurred images with noise.In this thesis,a deblur method combining adaptive median filtering and guided filtering is proposed.The method uses filter denoising and edge enhancement,that has little effect on the deblur algorithm.Adaptive median filtering not only removes noise well,but also preserves the integrity of the image.The structure of the image is enhanced by the guided filter,which makes the estimation of the blur kernel more accurate.After obtaining the accurate blur kernel,the non-blind deblur algorithm with the exception value dealt is guaranteed to obtain a clearer image.Finally,in order to improve the image quality evaluation of the deblur algorithm without reference to the original clear image,this thesis improves the full reference image quality evaluation method based on blur coefficients,and proposes a new image quality evaluation method based on edge energy fidelity.This method calculates the ratio of the edge energy of the original motion blurred image to the recovered sharp image as the fidelity coefficient to evaluate the recovered image quality.and then evaluates the advantages and disadvantages of the algorithm.the problem of quality evaluation of deblur algorithm without original image is solved.