单幅运动模糊图像的盲复原研究

发布时间：2018-04-02 16:13

本文选题：运动图像的盲复原　切入点：边缘检测　出处：《北京交通大学》2017年硕士论文

【摘要】：运动模糊图像的复原问题一直是图像复原领域研究的热点。在照片的拍摄过程中,由于相机与物体之间产生相对位移,导致发生运动模糊,严重影响到图像的后续处理与应用。图像去模糊就是采用一定的手段从模糊图像中恢复出清晰的图像,而在模糊核未知的情况下的复原问题就属于图像的盲复原范畴,这属于一种严重的病态问题。随着研究的不断深入,研究者提出很多优秀的算法,解决了图像盲复原的难题,但是许多现有的算法存在着一些不足,主要就是由于图像边缘提取和模糊核估计不准确导致的振铃效应以及图像细节信息丢失的问题。针对这些问题,本文从边缘提取和模糊核估计方面提出了一些改进算法,来改善运动模糊图像的盲复原效果。首先,本文提出了改进的Canny算子边缘提取算法。通过研究快速盲去模糊的算法,利用边缘检测算子进行边缘提取,再根据图像边缘信息的先验知识来估计模糊核。通过对边缘提取算法的研究,提出自适应的阈值设置方法,来设置高低阈值,滤除虚假边缘以及孤立的边缘点。实验结果表明该算法能够滤除大量的伪边缘,使得到的边缘更清晰、更连续。其次,本文提出了对估计的初始模糊核的细化方法。该方法通过边缘检测,估计模糊核,然后采用迭代支持域检测算法对模糊核进行修正细化,抑制初始模糊核中的一些离散值,使得到的模糊核更加精细、更加准确,采用超拉普拉斯分布拟合图像梯度分布,并且利用图像金字塔进行多尺度恢复,解决了模糊核过大的问题。实验结果表明该算法可以有效抑制初始模糊核中的离散值,获得较理想的模糊核,能有效地减弱复原图像的振铃效应以及能够保留大量的细节信息。
[Abstract]:The problem of motion blur image restoration has been a hot topic in the field of image restoration.In the process of photography, the relative displacement between the camera and the object results in motion blur, which seriously affects the subsequent processing and application of the image.Image de-blurring is to restore the clear image from the blurred image by certain means, but the restoration problem in the case of unknown fuzzy kernel belongs to the category of blind image restoration, which belongs to a serious pathological problem.With the deepening of the research, researchers put forward many excellent algorithms to solve the problem of blind image restoration, but many existing algorithms have some shortcomings.It is mainly caused by ringing effect caused by image edge extraction and inaccuracy of fuzzy kernel estimation and the loss of image detail information.To solve these problems, this paper proposes some improved algorithms for edge extraction and fuzzy kernel estimation to improve the blind restoration of motion blur images.Firstly, an improved edge detection algorithm for Canny operator is proposed.By studying the fast blind de-blurring algorithm, the edge detection operator is used to extract the edge, and then the fuzzy kernel is estimated according to the prior knowledge of the image edge information.Based on the research of edge extraction algorithm, an adaptive threshold setting method is proposed to set high and low thresholds and filter false edges and isolated edge points.Experimental results show that the algorithm can filter a large number of pseudo-edges and make the edges clearer and more continuous.Secondly, a refinement method for the initial fuzzy kernel of the estimation is proposed.In this method, the fuzzy kernel is estimated by edge detection, then the fuzzy kernel is refined by iterative support domain detection algorithm, and some discrete values in the initial fuzzy kernel are suppressed, which makes the fuzzy kernel more precise and accurate.The super-Laplace distribution is used to fit the image gradient distribution, and the image pyramid is used for multi-scale restoration, which solves the problem that the fuzzy kernel is too large.The experimental results show that the proposed algorithm can effectively suppress the discrete values in the initial fuzzy kernel, obtain the ideal fuzzy kernel, effectively attenuate the ringing effect of the restored image and retain a large amount of detail information.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】