您的位置:首页  > 论文页面

基于自适应高斯函数的鲁棒超分辨率图像复原

发表时间:2018-01-15  浏览量:934  下载量:77
全部作者: 张红,褚文华,曾雪迎
作者单位: 中国海洋大学数学科学学院
摘 要: 为增强复原图像对模型误差的鲁棒性,在正则化框架下提出一种基于自适应数据保真项和改进的双边全变差(bilateral total variation,BTV)正则项的超分辨率(super-resolution,SR)复原算法。自适应数据保真项由高斯函数导出并根据每一帧低分辨率(low-resolution,LR)图像对应的降质模型的精度自适应地确定阈值,具有自动区分处理模型误差和成像噪声的优点。改进的BTV正则项根据图像的局部特征,自动地调整“平滑性”度量,在平滑噪声的同时具有更优的细节保持能力。随后,采用梯度下降法求解相应的最小能量泛函,并提出一种在迭代过程中自适应更新阈值的加速策略,该数值求解过程可以视为对经典的迭代反投影方法的改进。利用模拟图像序列和真实视频序列分别对本文方法进行验证,结果表明本文方法能有效抑制模型误差对复原结果的影响,重建图像的视觉效果和量化指标均优于其他方法。
关 键 词: 计算数学;超分辨率复原;鲁棒估计;正则化;双边全变差
Title: Construction of robust multiframe super-resolution image based on self-adaptive Gaussian function
Author: ZHANG Hong, CHU Wenhua, ZENG Xueying
Organization: School of Mathematical Sciences, Ocean University of China
Abstract: To enhance the robustness of the restored image to model errors, we propose a super-resolution (SR) reconstruction algorithm using an adaptive data fidelity term and an improved bilateral total variation (BTV) term in a regularization framework. Using Gaussian function, the fidelity term is formed by a self-adaptive strategy depending on the accuracies of the estimated low-resolution (LR) image observation models, which has the advantages of automatic distinguishing between model error and image noise. The improved BTV regularization can adaptively adjust its smoothing effects according to image’s local structures, and hence preserve sharp features quite well while still being effective in removing noise. Gradient descent algorithm is used in solving the minimum energy functional and an acceleration strategy is developed with self-adaptive update of threshold in iteration process. This numerical scheme can be seen as the improvement of the traditional iterative back projection method. We use both synthetic and real data to test the proposed methods, the numerical experiments demonstrate that our method can effectively reduce the influence of the model errors on the restored image. The restored images by our method are superior to that by other methods in terms of the quantitative criteria and visual effects.
Key words: computational mathematics; super-resolution reconstruction; robust estimation; regularization; bilateral total variation
发表期数: 2018年1月第1期
引用格式: 张红,褚文华,曾雪迎. 基于自适应高斯函数的鲁棒超分辨率图像复原[J]. 中国科技论文在线精品论文,2018,11(1):1-10.
 
0 评论数 0
暂无评论
友情链接