基于改进总广义变分的单幅红外图像超分辨率算法  

Single infrared image super-resolution algorithm based on improved total generalized variation

作　　者：陈继光 苏冰山 CHEN Jiguang;SU Bingshan(School of Intelligent Engineering,Zhengzhou University of Aeronautics,Zhengzhou 450046,China)

机构地区：[1]郑州航空工业管理学院智能工程学院,河南郑州450046

出　　处：《轻工学报》2020年第4期103-108,共6页Journal of Light Industry

基　　金：河南省科技攻关计划(高新技术领域)项目(172102210529);河南省高等学校重点科研计划项目(17A520062)。

摘　　要：针对传统总广义变分(TGV)算法在红外图像超分辨率重建过程中难以有效抑制噪声的问题,提出了一种基于改进TGV的单幅红外图像超分辨率算法.该算法首先将二阶TGV模型与一阶梯度锐化算子相结合,在算法实现的梯度上升阶段加上一阶梯度锐化算子,在梯度下降阶段的系数中加上一阶梯度锐化算子的系数,得到一种新的红外图像超分辨率正则化模型;然后采用一阶主-对偶优化算法求得高分辨率红外图像.实验结果表明,该算法的主观视觉效果和客观评价指标均优于其他传统算法,可获得质量较高的高分辨率红外图像,能有效抑制噪声,降低硬件实现的复杂度,有较强的实用性.Aiming at the problem that the tranditional total generalized variation(TGV)algorithm could not restrain noise effectively in the process of infrared image super-resolution,a single infrared image super-resolution algorithm based on improved TGV was proposed.Firstly,the algorithm was built by second-order TGV regularization model and first-order graduate sharpening operator.First-order graduate sharpening operator was added during the process of gradient ascent,and the factor of first-order graduate sharpening operator was added during the process of gradient descent,so this algorithm acquired a new kind of infrared image super-resolution regularization model.Then it inferred the high-resolution infrared image with a first-order primal-dual optimization scheme.The experimental results showed that the algorithm was superior to other traditional algorithms in terms of subjective visual effect and objective evaluation index,and could obtain high-quality high-resolution infrared images,effectively suppress noise and reduce the complexity of hardware implementation,and had strong practicality.

关 键 词：红外图像超分辨率 总广义变分 梯度锐化算子 正则化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]