基于混合数据项的运动去模糊变分方法  

Mixed data term based variation method for motion deblurring

作　　者：王国栋[1] 潘振宽[1] 刘存良[1] 郑世秀[1] 

机构地区：[1]青岛大学信息工程学院,青岛266071

出　　处：《仪器仪表学报》2013年第7期1552-1558,共7页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(61170106);山东省自然科学基金(ZR2010FQ030)资助项目

摘　　要：运动去模糊是当今图像处理领域的研究热点和难点。鉴于变分能量方程在图像处理中的理论优势,提出了基于变分能量方程的运动去模糊算法。为了利用图像中的梯度信息并消除噪声的影响,采用混合数据项。引入归一化的总变分项,使得运动去模糊的真正解能够使得变分能量下降。由于引入的归一化总变分项求解复杂,引入线性分裂Bregman迭代简化求解过程。由于运动轨迹的复杂性,算法在多尺度框架下进行,从粗尺度到细尺度依次渐近执行,直到最终求得模糊核函数。粗尺度上估计的点扩展函数和恢复的清晰图像作为下一精确尺度变分能量模型求解的初始值。估计出点扩展函数后利用总变分能量方程求得清晰的图像。实验结果验证了算法的有效性。Motion deblurring is a hot and difficult spot in current image processing research field. Due to the theoretical superiority of variation energy equation in image processing, we propose a motion deblurring method based on variation energy equation. In order to use the gradient information and eliminate the influence of noise in the image, the mixed data term is adopted. The normalized total variation term is introduced ,which can lead to the fact that the true solution of the motion blurring makes variation energy decrease. Because the introduced normalized total variation term is difficult to solve, the linear split Bregman method is introduced to reduce the complexity of the equation solution process. Because of the complexity of the motion trajectory, we establish a multiscale framework to carry out the deblurring procedure from course scale to fine scale until the blur kernel function is finally obtained. The point spread function estimated in coarse scale and the restored clear image are used as the initial value for solving the variation energy model in the next fine scale. After the point spread function is estimated, the total variation equation is used to solve the final clear image. Experiment results verify the validity of the proposed method.

关 键 词：运动去模糊 归一化总变分项 混合数据项 变分方法 分裂Bregman方法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]