针对鬼成像的聚合型三维目标匹配滤波  

Convergent Three-dimensional Target Matching Filtering for Ghost Imaging

作　　者：张顺尧 桑爱军[1] 宋立军 王世刚[1] ZHANG Shun-yao;SANG Ai-jun;SONG Li-jun;WANG Shi-gang(College of Telecommunication Engineering,Jilin University,Changchun 130022,China;Jilin Engineering Normal University,Changchun 130052,China)

机构地区：[1]吉林大学通信工程学院,吉林长春130022 [2]吉林工程技术师范学院,吉林长春130052

出　　处：《东北大学学报（自然科学版）》2023年第2期186-191,共6页Journal of Northeastern University(Natural Science)

基　　金：吉林省自然科学基金资助项目(20200201295JC);吉林省科技发展计划国际合作项目(20200801042GH).

摘　　要：针对视频鬼成像重构采样次数少、清晰度低、噪声大等问题,提出了一种聚合型三维目标匹配滤波方法.首先将鬼成像视频中的每一帧图像与待优化帧进行块匹配,将块匹配后的每一帧图像按照时序排序组成三维矩阵.根据每一帧图片的帧序,赋予其不同的帧权值,对矩阵进行三维加权中值滤波.经过运动目标的欠采样仿真与实验对比后,结果表明本文方法相较于现存的三维滤波方法,不仅噪声系数更低、结构保留程度更好,同时获到了更好的主观评价(Brisque).对比原始实验恢复图,本文方法在噪声系数和边缘模糊度上分别下降34.25%,6.86%,在Brisque主观评价上提升了45.84%.Aiming at the problems of low sampling frequency,low resolution and high noise in video ghost imaging reconstruction,an aggregated three-dimensional target matching filtering method was proposed.Firstly,each frame of the ghost imaging video is block-matched with the frame to be optimized and each frame of images after block matching is arranged to form a three-dimensional matrix according to the time sequence.According to the frame sequence of each frame picture,different frame weights are assigned to it.Then the matrix is subjected to three-dimensional weighted median filtering.After under-sampling simulation and experimental comparison of moving targets,the method proposed has not only lower noise figure and better structure retention,but also a better subjective evaluation compared with the existing three-dimensional filtering methods.Compared with the original experimental restoration graph,the proposed method reduces the noise figure and edge ambiguity by 34.25%and 6.86%,respectively,and the Brisque subjective evaluation was improved by 45.84%.

关 键 词：计算鬼成像 块匹配算法 中值滤波 三维加权滤波 视频滤波 

分 类 号：TN919.81[电子电信—通信与信息系统]