多特征的POCS图像超分辨率重建方法  

POCS image super-resolution reconstruction method based on multi-feature

作　　者：白旭 卜丽静 赵国忱[1] 张正鹏 涂丽莹 BAI Xu;BU Lijing;ZHAO Guochen;ZHANG Zhengpeng;TU Liying(School of Surveying,Mapping and Geography,Liaoning Technical University,Fuxin,Liaoning 123000,China;School of Automation and Electronic Information,Xiangtan University,Xiangtan,Hunan 411100,China)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000 [2]湘潭大学自动化与电子信息学院,湖南湘潭411100

出　　处：《测绘科学》2022年第12期174-183,共10页Science of Surveying and Mapping

基　　金：国家自然科学基金青年科学基金项目(41801294)

摘　　要：针对传统凸集投影(POCS)方法重建图像存在边缘保持能力不足、细节丢失的问题,该文提出了一种多特征的凸集投影超分辨率重建方法。首先,根据局部相似性特征和图像的梯度特征构建参考帧,以保持图像的边缘细节特征。然后,运用光流法对上采样后的序列低分辨率(LR)图像和参考帧进行配准,同时利用比值稀疏约束模型估计参考帧的点扩散函数(PSF),并获取点扩散函数作用窗口的中心坐标和点扩散函数作用窗口的范围。最后利用参考帧和点扩散函数生成重建图像的初始估计,根据重建图像残差对参考帧进行修正,最终得到重建图像。为验证本文算法的有效性和鲁棒性,分别与上采样、迭代反投影、凸集投影几种重建方法进行实验对比,实验表明本文算法重建结果在主观和客观评价方面明显优于其它方法,本文算法可行。Aiming at the problems of lack of edge retention ability and detail loss in traditional POCS image reconstruction,a multi-feature POCS super-resolution reconstruction method was proposed.Firstly,a reference frame was constructed according to the local similarity feature and the gradient feature of the image which to maintain the edge detail feature of the image.Then,the LK optical flow method was used to register the upsampled sequential LR images and reference frames.At the same time,the ratio sparse constraint model is used to estimate the PSF of the reference frame and obtain the central coordinates and range of PSF action window.Finally,the initial estimation of reconstructed image is generated by using the reference frame and PSF,and the reconstructed image is finally obtained by modifying the reference frame according to the residual of reconstructed image.In order to verify the effectiveness and robustness of the algorithm presented in this paper,experimental comparison was conducted with several reconstruction methods,such as upsampling,iterative back-projection and convex set projection.Experimental results showed that this algorithm reconstruction results in subjective and objective evaluation are superior to other methods,this algorithm is feasible.

关 键 词：超分辨率重建 凸集投影 局部相似性 图像梯度 比值稀疏约束模型 

分 类 号：P237[天文地球—摄影测量与遥感]