低秩聚类被动压缩鬼成像  

Passive compressive ghost imaging with low rank clustering

作　　者：雷腾 张义民 马一哲 丁学专[1,2,3] 吴滢跃 王世勇[1,2,3] LEI Teng;ZHANG Yiming;MA Yizhe;DING Xuezhuan;WU Yingyue;WANG Shiyong(Shanghai Institute of Technical Physics,Chinese Academy of Sciences,Shanghai 200083,China;University of Chinese Academy of Sciences,Beijing 100049,China;Laboratory of Infrared Detection and Imaging Technology,Chinese Academy of Sciences,Shanghai 200083,China)

机构地区：[1]中国科学院上海技术物理研究所,上海200083 [2]中国科学院大学,北京100049 [3]中国科学院红外探测与成像技术实验室,上海200083

出　　处：《光学精密工程》2024年第1期12-23,共12页Optics and Precision Engineering

基　　金：国家十四五预研基金资助项目(No.514010405)。

摘　　要：低采样率下的高质量鬼成像(GI)对于科学研究和实际应用具有重要意义,为了在低采样率条件下重建高质量图像,提出了一种高质量的被动式压缩鬼成像重构算法(PCGI-LRC)。基于图像的非局域相似块堆叠而成的矩阵具有低秩和稀疏奇异值的假设,从理论和实验上证明了一种对最小二乘问题与非局域相似块低秩近似问题进行联合迭代求解的方法,能够在低采样率(6.25%~50%)条件下实现高质量鬼成像。实验结果表明:与基于稀疏基约束的GI(GI-SBC)和基于全变分约束的GI(GI-TVC)相比,PCGI-LRC在峰值信噪比、结构相似性系数和视觉观测等方面均更优,在抑制重构噪声的同时保持了目标的细节信息,其中PSNR提升效果优于1.1 dB,SSIM提升效果优于0.04。High-quality ghost imaging(GI)at low sampling rate is of great importance for scientific research and practical applications.Therefore,the reconstruction of high-quality images under low sampling rate con⁃ditions remains the focus of GI research.In this paper,a high-quality passive compressive ghost image recon⁃struction algorithm was proposed,called PCGI-LRC.Based on the assumption that the matrices stacked with nonlocal similar blocks of an image have low-rank and sparse singular values,a joint iterative solution of the least squares problem was demonstrated theoretically and experimentally and the low-rank approximation problem of the nonlocal similar blocks can achieve high-quality ghost images under low sampling rate condi⁃tions(6.25%-50%).Moreover,the experimental results show that the proposed algorithm outperforms the GI based on sparse basis constraints(GI-SBC)and GI based on full variational constraints(GI-TVC)algorithms regarding peak signal-to-noise ratio(PSNR),structural similarity coefficient(SSIM),and visual ob⁃servation.Information of the target is preserved while the reconstruction noise is suppressed;the PSNR is improved by more than 1.1 dB and the SSIM improvement is higher than 0.04 dB.

关 键 词：鬼成像 图像重构 图像压缩 单像素成像 

分 类 号：O431.2[机械工程—光学工程]