Seyed Alireza Hosseini, Tam Thuc Do, Gene Cheung, Yuichi Tanaka
{"title":"通过展开图拉普拉奇正则构建可解释的深度去噪器","authors":"Seyed Alireza Hosseini, Tam Thuc Do, Gene Cheung, Yuichi Tanaka","doi":"arxiv-2409.06676","DOIUrl":null,"url":null,"abstract":"An image denoiser can be used for a wide range of restoration problems via\nthe Plug-and-Play (PnP) architecture. In this paper, we propose a general\nframework to build an interpretable graph-based deep denoiser (GDD) by\nunrolling a solution to a maximum a posteriori (MAP) problem equipped with a\ngraph Laplacian regularizer (GLR) as signal prior. Leveraging a recent theorem\nshowing that any (pseudo-)linear denoiser $\\boldsymbol \\Psi$, under mild\nconditions, can be mapped to a solution of a MAP denoising problem regularized\nusing GLR, we first initialize a graph Laplacian matrix $\\mathbf L$ via\ntruncated Taylor Series Expansion (TSE) of $\\boldsymbol \\Psi^{-1}$. Then, we\ncompute the MAP linear system solution by unrolling iterations of the conjugate\ngradient (CG) algorithm into a sequence of neural layers as a feed-forward\nnetwork -- one that is amenable to parameter tuning. The resulting GDD network\nis \"graph-interpretable\", low in parameter count, and easy to initialize thanks\nto $\\mathbf L$ derived from a known well-performing denoiser $\\boldsymbol\n\\Psi$. Experimental results show that GDD achieves competitive image denoising\nperformance compared to competitors, but employing far fewer parameters, and is\nmore robust to covariate shift.","PeriodicalId":501034,"journal":{"name":"arXiv - EE - Signal Processing","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2024-09-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Constructing an Interpretable Deep Denoiser by Unrolling Graph Laplacian Regularizer\",\"authors\":\"Seyed Alireza Hosseini, Tam Thuc Do, Gene Cheung, Yuichi Tanaka\",\"doi\":\"arxiv-2409.06676\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"An image denoiser can be used for a wide range of restoration problems via\\nthe Plug-and-Play (PnP) architecture. In this paper, we propose a general\\nframework to build an interpretable graph-based deep denoiser (GDD) by\\nunrolling a solution to a maximum a posteriori (MAP) problem equipped with a\\ngraph Laplacian regularizer (GLR) as signal prior. Leveraging a recent theorem\\nshowing that any (pseudo-)linear denoiser $\\\\boldsymbol \\\\Psi$, under mild\\nconditions, can be mapped to a solution of a MAP denoising problem regularized\\nusing GLR, we first initialize a graph Laplacian matrix $\\\\mathbf L$ via\\ntruncated Taylor Series Expansion (TSE) of $\\\\boldsymbol \\\\Psi^{-1}$. Then, we\\ncompute the MAP linear system solution by unrolling iterations of the conjugate\\ngradient (CG) algorithm into a sequence of neural layers as a feed-forward\\nnetwork -- one that is amenable to parameter tuning. The resulting GDD network\\nis \\\"graph-interpretable\\\", low in parameter count, and easy to initialize thanks\\nto $\\\\mathbf L$ derived from a known well-performing denoiser $\\\\boldsymbol\\n\\\\Psi$. Experimental results show that GDD achieves competitive image denoising\\nperformance compared to competitors, but employing far fewer parameters, and is\\nmore robust to covariate shift.\",\"PeriodicalId\":501034,\"journal\":{\"name\":\"arXiv - EE - Signal Processing\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-09-10\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - EE - Signal Processing\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2409.06676\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - EE - Signal Processing","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.06676","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
摘要
图像去噪器可通过即插即用(PnP)架构用于各种修复问题。在本文中,我们提出了一个通用框架,通过对最大后验(MAP)问题的解进行滚动,并将图谱拉普拉奇正则化器(GLR)作为信号先验来构建可解释的基于图谱的深度去噪器(GDD)。最近的一个定理表明,在温和条件下,任何(伪)线性去噪器$\boldsymbol \Psi$都可以映射为使用GLR正则化的MAP去噪问题的解,利用该定理,我们首先初始化了$\boldsymbol \Psi^{-1}$的图拉普拉斯矩阵$\mathbf L$ viatruncated Taylor Series Expansion (TSE)。然后,我们通过将共轭梯度(CG)算法的迭代展开到神经层序列中来计算 MAP 线性系统解,将其作为一个前馈网络--一个可以进行参数调整的网络。由此产生的GDD网络是 "可解释图 "的,参数数量少,并且易于初始化,这要归功于从已知性能良好的去噪器$\boldsymbol\Psi$中提取的$\mathbf L$。实验结果表明,与竞争者相比,GDD能实现具有竞争力的图像去噪性能,但使用的参数要少得多,而且对协变量偏移具有更强的鲁棒性。
Constructing an Interpretable Deep Denoiser by Unrolling Graph Laplacian Regularizer
An image denoiser can be used for a wide range of restoration problems via
the Plug-and-Play (PnP) architecture. In this paper, we propose a general
framework to build an interpretable graph-based deep denoiser (GDD) by
unrolling a solution to a maximum a posteriori (MAP) problem equipped with a
graph Laplacian regularizer (GLR) as signal prior. Leveraging a recent theorem
showing that any (pseudo-)linear denoiser $\boldsymbol \Psi$, under mild
conditions, can be mapped to a solution of a MAP denoising problem regularized
using GLR, we first initialize a graph Laplacian matrix $\mathbf L$ via
truncated Taylor Series Expansion (TSE) of $\boldsymbol \Psi^{-1}$. Then, we
compute the MAP linear system solution by unrolling iterations of the conjugate
gradient (CG) algorithm into a sequence of neural layers as a feed-forward
network -- one that is amenable to parameter tuning. The resulting GDD network
is "graph-interpretable", low in parameter count, and easy to initialize thanks
to $\mathbf L$ derived from a known well-performing denoiser $\boldsymbol
\Psi$. Experimental results show that GDD achieves competitive image denoising
performance compared to competitors, but employing far fewer parameters, and is
more robust to covariate shift.