模糊图像的盲去卷积算法概要.docVIP

模糊图像的盲去卷积算法 肖宿 韩国强 沃焱 （华南理工大学计算机科学与工程学院 广州 510006 ） 摘 要 贝叶斯框架下图像统计模型的选取和先验模型的建立 ,对于图像盲去卷积算法结果 的优劣具有重要的影响 .总变分模型是目前使用较多的先验模型 ,其优点是突出的边缘保持 能力 .而总变分模型的缺点也是比较明显的 ,它会给恢复的图像带来阶梯效应 ,产生虚假的边 缘 ,边缘的去噪效果不理想 .同时 ,作为非线性模型 ,它还会降低算法的效率和提高算法的复杂 度 .为了解决以上问题 ,文中引入 Sobolev 图像空间模型用于图像的盲去卷积.该模型提高了算 法的速度 ,降低了算法的整体复杂度 ,避免了阶梯效应和虚假边缘的出现 .在用变分近似的方法估计最优解时 ,变分准则的选取很重要 ,而常用的基于 Kullback-Leibler 测度的准则有时会使计算变得复杂 .本文提出了一种新的变分准则 ,它有效地简化了计算 ,提高了算法的速度 .实 验结果证明了本算法的有效性 ,在算法简化速度提升的同时 ,图像的盲去卷积取得了令人满意的效果 . 关键词 图像盲去卷积 ; 贝叶斯方法 ; 变分准则 ; 先验模型 ; 参数估计 中图分类号 : TP751.1文献标识码 : A Blind Deconvolution Algorithm of Blurred Images Xiao Su Han Guoqiang Wo Yan (School of Computer Science and Engineering, South China University of Technology, Guangzhou 510006) Abstract The selection of image statistical models and the construction of image prior models have important effects on the results of blind image deconvolution algorithms under Bayesian framework. Currently, the total variation model is a commonly used prior model which has the outstanding virtue that it can preserve edges well. However, it has its obvious drawback that it brings restored images the staircase effect, fake edges and sometimes bad results of denoising at the image edges. Meanwhile, as a nonlinear model, the total variation model will reduce the efficiency and increase the complexity of the algorithms. In order to solve the problems mentioned above, the Sobolev image space model is introduced to the blind image deconvolution. This model can increase the speed and reduce the overall complexity of the proposed algorithm without the staircase effects and the fake edges. It ’s important to select the proper variational criterion, when the variational method is used to obtain the optimal estimation, while the commonly used Kullback-Leibler divergence based criterion will sometimes make the calculations more complicated. This paper presents a novelty variational criterion which simplifies the calculations in the proposed algorithm and increases