基于退化转换及可分离全变差模型的图像修复  

Image Restoration Based on Degradation Conversion and Separable Total Variation Model

作　　者：王斌[1] 胡辽林[1] 薛瑞洋 

机构地区：[1]西安理工大学机械与精密仪器工程学院,陕西西安710048

出　　处：《光学学报》2014年第10期91-97,共7页Acta Optica Sinica

基　　金：陕西省自然科学基金(2014JM7273)

摘　　要：空域重叠的模糊和噪声引起图像退化,修复通常比较复杂。为简化图像修复,提高修复质量,提出利用二阶逼近算子将传统图像修复中既含噪声又含模糊的双退化模型转化为只含"动态噪声"的单退化模型。在传统全变差模型的基础上,提出了一种利用低维差分投影的思想,建立元素可分离的全变差模型,利用一阶梯度下降算法解决"动态噪声"问题。仿真结果表明,该方法适用于多种退化模型,即使在强退化环境下,依然可以有效地去除噪声和模糊,保留图像边缘和细节信息,使退化图像恢复到较理想的状态。Image restoration is actually complicated because of the spatial-domain overlapping of noise and blur which can cause image degradation. In order to simplify restoration and improve recovery quality, it is proposed that the original double degradation model being translated into a single degradation model which contains only dynamic noise by utilizing proximal operator. Based on traditional total variation model, the seperable version is proposed by introducing low-dimension differential projection. The dynamic noise can be removed through first-order gradient descent algorithm. The results show that this method is applicable to a variety of degradation models, which effectively remove noise and blur with edges and details kept, even in a strong degradation environment. This makes degraded image recover to ideal status.

关 键 词：图像处理 退化模型 逼近算子 动态噪声 全变差模型 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP751.1[自动化与计算机技术—计算机科学与技术]