基于非局部先验的深度压缩感知图像重构网络仲元红*周宇杰张静张晨旭(重庆大学微电子与通信工程学院重庆400044)摘要:传统的基于迭代的压缩感知(CS)图像重构算法易于集成图像先验信息,但存在性能不足、计算复杂度高等缺点。基于深度学习的图像重构算法重构性能通常优于传统的重构算法,并且具有更低的重构计算成本。因此,为了设计出一种更有效利用先验信息的深度学习图像重构算法,该文提出基于非局部先验的深度压缩感知图像重构网络。首先,将稀疏性和非局部先验相结合建立压缩感知图像重构模型,然后通过半二次方分裂法将模型分解为3个子问题,每一个子问题的求解都在深度学习的框架下展开,最后联合建立端到端的可训练的图像重构模型。仿真实验表明,在测试的采样率与数据集下该文所提算法的峰值信噪比与当前主流的重构算法SCSNet相比平均提升了0.18dB,与CSNet算法相比平均提升了约1.59dB,与ISTA-Net+算法相比平均提升了约2.09dB。关键词:图像重构;压缩感知;深度学习;非局部先验;半二次方分裂中图分类号:TN911.73;TP391.41文献标识码:A文章编号:1009-5896(2023)02-0654-10DOI:10.11999/JEIT211506DeepCompressiveSensingImageReconstructionNetworkBasedonNon-LocalPriorZHONGYuanhongZHOUYujieZHANGJingZHANGChenxu(SchoolofMicroelectronicsandCommunicationEngineering,ChongqingUniversity,Chongqing400044,China)Abstract:Thetraditionaliterative-basedCompressiveSensing(CS)imagereconstructionalgorithmiseasytointegrateimagepriorinformation,butithasshortcomingssuchasinsufficientperformanceandhighcomputationalcomplexity.Theperformanceoftheimagereconstructionalgorithmbasedondeeplearningisbetterthanthetraditionalreconstructionalgorithmsignificantly,andithaslowertimecost.Therefore,inordertodesignadeeplearningimagereconstructionalgorithmthatusespriorinformationmoreeffectively,adeepcompressivesensingimagereconstructionnetworkbasedonnon-localpriorsisproposed.Firstly,thesparsenessandnon-localpriorarecombinedtoestablishacompressedsensingimagereconstructionmodel.Secondly,themodelisdecomposedintothreesub-problemsbythehalfquadraticsplittingmethod.Thesolutionofeachsub-problemiscarriedoutundertheframeworkofdeeplearning.Finally,anend-to-endtrainableimagereconstructi...