基于原_对偶混合梯度下降法的图像恢复算法.pdfVIP

第 29卷第 4期 计算机应用 V ol. 29 No. 4 2009年 4月 Journal of ComputerApplications Apr. 2009 : 1001- 9081( 2009) 04- 0987- 03 ) 张 慧, 成礼智, 赵在新 (, 410073) ( h. zhang1984@ 163. com ) : 对基于原) 对偶混合梯度下降法的图像恢复算法进行了改进在原算法中, 长参数的设计对其恢复 效果和收敛速度影响较大为了改善算法性能, 通过引入中间变量改变算法的形式, 然后分离对偶向量的分量, 用加 权矩阵取代原算法中的 长参数, 并对其分开设计数值实验表明, 在峰值信噪比 ( PSNR) 和视觉效果相当的前提 下, 和原算法相比, 降低了参数设计对算法性能的影响, 且计算时间减少 50% 左右 : 全变分; 图像恢复; 原) 对偶; 混合梯度 : P391. 41 : A Im age restoration algorithm based on prim al-dual hybrid gradien t descen t m ethod ZHANG Hui, CHENG Li-zh,i ZHAO Zai-xin (S chool of S cience, N ationa l Un ivers ity of D ef ense T echnology, C hang sha H unan 410073, Ch ina ) A bstract: An mi proved algorithm for mi age restoration w as proposed based on Prmi al-Dual Hybrid G radient Descent (PDHGD) m ethod. he preferences have a great mi pact on convergence rate of the known algorithm. he form w as changed by introducing new variable, and the elem ents of the dualvector of the prmi al-dual hybridm odelw ere separated, and then step- size w as rep laced by using param eterm atrices. he numerical expermi ents show that the mi proved algorithm has advantage on choosing param eters compared to the known algorithm, and the iterative number and CPUp tmi e nearly declined by 50% , and at the sam e tmi e, the mi proved algorithm has exactly the sam e effect on mi age restoration. K ey w ords: otalVariation ( V); mi age restoration; prmi al-dual; hybrid gradient PDHGD 0 引言 , [ 1]