Selecting an appropriate prior to compensate for information loss due to the measurement operator is a fundamental challenge in imaging inverse problems. Implicit priors based on denoising neural networks have become central to widely-used frameworks such as Plug-and-Play (PnP) algorithms. In this work, we introduce Fixed-points of Restoration (FiRe) priors as a new framework for expanding the notion of priors in PnP to general restoration models beyond traditional denoising models. The key insight behind FiRe is that smooth images emerge as fixed points of the composition of a degradation operator with the corresponding restoration model. This enables us to derive an explicit formula for our implicit prior by quantifying invariance of images under this composite operation. Adopting this fixed-point perspective, we show how various restoration networks can effectively serve as priors for solving inverse problems. The FiRe framework further enables ensemble-like combinations of multiple restoration models as well as acquisition-informed restoration networks, all within a unified optimization approach. Experimental results validate the effectiveness of FiRe across various inverse problems, establishing a new paradigm for incorporating pretrained restoration models into PnP-like algorithms. Code available at https://github.com/matthieutrs/fire.

Plug-and-play (PnP) methods are extensively used for solving imaging inverse problems by integrating physical measurement models with pre-trained deep denoisers as priors. Score-based diffusion models (SBMs) have recently emerged as a powerful framework for image generation by training deep denoisers to represent the score of the image prior. While both PnP and SBMs use deep denoisers, the score-based nature of PnP is unexplored in the literature due to its distinct origins rooted in proximal optimization. This paper introduces a novel view of PnP as a score-based method, a perspective that enables the re-use of powerful SBMs within classical PnP algorithms without retraining. We present a set of mathematical relationships for adapting popular SBMs as priors within PnP. We show that this approach enables a direct comparison between PnP and SBM-based reconstruction methods using the same neural network as the prior. Code is available at https://github.com/wustl-cig/scorepnp.

Deep neural networks trained as image denoisers are widely used as priors for solving imaging inverse problems. We introduce Stochastic deep Restoration Priors (ShaRP), a novel framework that stochastically leverages an ensemble of deep restoration models beyond denoisers to regularize inverse problems. By using generalized restoration models trained on a broad range of degradations beyond simple Gaussian noise, ShaRP effectively addresses structured artifacts and enables self-supervised training without fully sampled data. We prove that ShaRP minimizes an objective function involving a regularizer derived from the score functions of minimum mean square error (MMSE) restoration operators. We also provide theoretical guarantees for learning restoration operators from incomplete measurements. ShaRP achieves state-of-the-art performance on tasks such as magnetic resonance imaging reconstruction and single-image super-resolution, surpassing both denoiser- and diffusion-model-based methods without requiring retraining.

Diffusion models show promise for image restoration, but existing methods often struggle with inconsistent fidelity and undesirable artifacts. To address this, we introduce Kernel Density Steering (KDS), a novel inference-time framework promoting robust, high-fidelity outputs through explicit local mode-seeking. KDS employs an $N$-particle ensemble of diffusion samples, computing patch-wise kernel density estimation gradients from their collective outputs. These gradients steer patches in each particle towards shared, higher-density regions identified within the ensemble. This collective local mode-seeking mechanism, acting as "collective wisdom", steers samples away from spurious modes prone to artifacts, arising from independent sampling or model imperfections, and towards more robust, high-fidelity structures. This allows us to obtain better quality samples at the expense of higher compute by simultaneously sampling multiple particles. As a plug-and-play framework, KDS requires no retraining or external verifiers, seamlessly integrating with various diffusion samplers. Extensive numerical validations demonstrate KDS substantially improves both quantitative and qualitative performance on challenging real-world super-resolution and image inpainting tasks.

Face video restoration (FVR) is a challenging but important problem where one seeks to recover a perceptually realistic face videos from a low-quality input. While diffusion probabilistic models (DPMs) have been shown to achieve remarkable performance for face image restoration, they often fail to preserve temporally coherent, high-quality videos, compromising the fidelity of reconstructed faces. We present a new conditional diffusion framework called FLAIR for FVR. FLAIR ensures improved temporal alignments across frames in a computationally efficient fashion by converting a traditional image DPM into a video DPM. The proposed conversion uses a recurrent video refinement layer and a temporal self-attention at different scales. FLAIR also uses a conditional iterative refinement process to balance the perceptual and distortion quality during inference. This process consists of two key components: a data-consistency module that analytically ensures that the generated video precisely matches its degraded observation and a coarse-to-fine image enhancement module specifically for facial regions. Our extensive experiments show superiority of FLAIR over the current state-of-the-art (SOTA) for video super-resolution, deblurring, JPEG restoration, and space-time frame interpolation on two high-quality face video datasets.

Seismic data frequently exhibit missing traces, substantially affecting subsequent seismic processing and interpretation. Deep learning-based approaches have demonstrated significant advancements in reconstructing irregularly missing seismic data through supervised and unsupervised methods. Nonetheless, substantial challenges remain, such as generalization capacity and computation time cost during the inference. This work introduces a reconstruction method that uses a pretrained generative diffusion model for image synthesis and incorporates deep image prior (DIP) to enforce data consistency when reconstructing missing traces in seismic data. The proposed method has demonstrated strong robustness and high reconstruction capability of poststack and prestack data with different levels of structural complexity, even in field and synthetic scenarios where test data were outside the training domain. This indicates that our method can handle the high geological variability of different exploration targets. Additionally, compared to other state-of-the-art seismic reconstruction methods using diffusion models, during inference, our approach reduces the number of sampling timesteps by up to$4\times $. Our implementation is available athttps://github.com/PAULGOYES/CDDIP.git.

Deep learning-based methods provide remarkable performance in a number of computational imaging problems. Examples include end-to-end trained networks that map measurements to unknown signals, plug-and-play (PnP) methods that use pretrained denoisers as image prior, and model-based unrolled networks that train artifact removal blocks. Many of these methods lack robustness and fail to generalize with distribution shifts in data, measurements, and noise. In this paper, we present a simple framework to perform domain adaptation as data and measurement distribution shifts. Our method learns a small number of factors to add in a pretrained model to bridge the gap in performance. We present a number of experiments on accelerated magnetic resonance imaging (MRI) reconstruction and image deblurring to demonstrate that our method requires a small amount of memory and parameter overhead to adapt to new domains.

Image denoisers have been shown to be powerful priors for solving inverse problems in imaging. In this work, we introduce a generalization of these methods that allows any image restoration network to be used as an implicit prior. The proposed method uses priors specified by deep neural networks pre-trained as general restoration operators. The method provides a principled approach for adapting state-of-the-art restoration models for other inverse problems. Our theoretical result analyzes its convergence to a stationary point of a global functional associated with the restoration operator. Numerical results show that the method using a super-resolution prior achieves state-of-the-art performance both quantitatively and qualitatively. Overall, this work offers a step forward for solving inverse problems by enabling the use of powerful pre-trained restoration models as priors.

Deformable image registration (DIR) is an active research topic in biomedical imaging. There is a growing interest in developing DIR methods based on deep learning (DL). A traditional DL approach to DIR is based on training a convolutional neural network (CNN) to estimate the registration field between two input images. While conceptually simple, this approach comes with a limitation that it exclusively relies on a pre-trained CNN without explicitly enforcing fidelity between the registered image and the reference. We present plug-and-play image registration network (PIRATE) as a new DIR method that addresses this issue by integrating an explicit data-fidelity penalty and a CNN prior. PIRATE pre-trains a CNN denoiser on the registration field and "plugs" it into an iterative method as a regularizer. We additionally present PIRATE+ that fine-tunes the CNN prior in PIRATE using deep equilibrium models (DEQ). PIRATE+ interprets the fixed-point iteration of PIRATE as a network with effectively infinite layers and then trains the resulting network end-to-end, enabling it to learn more task-specific information and boosting its performance. Our numerical results on OASIS and CANDI datasets show that our methods achieve state-of-the-art performance on DIR.

Posterior sampling has been shown to be a powerful Bayesian approach for solving imaging inverse problems. The recent plug-and-play unadjusted Langevin algorithm (PnP-ULA) has emerged as a promising method for Monte Carlo sampling and minimum mean squared error (MMSE) estimation by combining physical measurement models with deep-learning priors specified using image denoisers. However, the intricate relationship between the sampling distribution of PnP-ULA and the mismatched data-fidelity and denoiser has not been theoretically analyzed. We address this gap by proposing a posterior-$L_2$ pseudometric and using it to quantify an explicit error bound for PnP-ULA under mismatched posterior distribution. We numerically validate our theory on several inverse problems such as sampling from Gaussian mixture models and image deblurring. Our results suggest that the sensitivity of the sampling distribution of PnP-ULA to a mismatch in the measurement model and the denoiser can be precisely characterized.

Plug-and-Play (PnP) priors is a widely-used family of methods for solving imaging inverse problems by integrating physical measurement models with image priors specified using image denoisers. PnP methods have been shown to achieve state-of-the-art performance when the prior is obtained using powerful deep denoisers. Despite extensive work on PnP, the topic of distribution mismatch between the training and testing data has often been overlooked in the PnP literature. This paper presents a set of new theoretical and numerical results on the topic of prior distribution mismatch and domain adaptation for the alternating direction method of multipliers (ADMM) variant of PnP. Our theoretical result provides an explicit error bound for PnP-ADMM due to the mismatch between the desired denoiser and the one used for inference. Our analysis contributes to the work in the area by considering the mismatch under nonconvex data-fidelity terms and expansive denoisers. Our first set of numerical results quantifies the impact of the prior distribution mismatch on the performance of PnP-ADMM on the problem of image super-resolution. Our second set of numerical results considers a simple and effective domain adaption strategy that closes the performance gap due to the use of mismatched denoisers. Our results suggest the relative robustness of PnP-ADMM to prior distribution mismatch, while also showing that the performance gap can be significantly reduced with only a few training samples from the desired distribution.

Regularization by Denoising (RED) is a well-known method for solving image restoration problems by using learned image denoisers as priors. Since the regularization parameter in the traditional RED does not have any physical interpretation, it does not provide an approach for automatic parameter selection. This letter addresses this issue by introducing the Constrained Regularization by Denoising (CRED) method that reformulates RED as a constrained optimization problem where the regularization parameter corresponds directly to the amount of noise in the measurements. The solution to the constrained problem is solved by designing an efficient method based on alternating direction method of multipliers (ADMM). Our experiments show that CRED outperforms the competing methods in terms of stability and robustness, while also achieving competitive performances in terms of image quality.

Implicit neural representations (INR) have been recently proposed as deep learning (DL) based solutions for image compression. An image can be compressed by training an INR model with fewer weights than the number of image pixels to map the coordinates of the image to corresponding pixel values. While traditional training approaches for INRs are based on enforcing pixel-wise image consistency, we propose to further improve image quality by using a new structural regularizer. We present structural regularization for INR compression (SINCO) as a novel INR method for image compression. SINCO imposes structural consistency of the compressed images to the groundtruth by using a segmentation network to penalize the discrepancy of segmentation masks predicted from compressed images. We validate SINCO on brain MRI images by showing that it can achieve better performance than some recent INR methods.

Deep model-based architectures (DMBAs) are widely used in imaging inverse problems to integrate physical measurement models and learned image priors. Plug-and-play priors (PnP) and deep equilibrium models (DEQ) are two DMBA frameworks that have received significant attention. The key difference between the two is that the image prior in DEQ is trained by using a specific measurement model, while that in PnP is trained as a general image denoiser. This difference is behind a common assumption that PnP is more robust to changes in the measurement models compared to DEQ. This paper investigates the robustness of DEQ priors to changes in the measurement models. Our results on two imaging inverse problems suggest that DEQ priors trained under mismatched measurement models outperform image denoisers.

Three-dimensional fluorescence microscopy often suffers from anisotropy, where the resolution along the axial direction is lower than that within the lateral imaging plane. We address this issue by presenting Dual-Cycle, a new framework for joint deconvolution and fusion of dual-view fluorescence images. Inspired by the recent Neuroclear method, Dual-Cycle is designed as a cycle-consistent generative network trained in a self-supervised fashion by combining a dual-view generator and prior-guided degradation model. We validate Dual-Cycle on both synthetic and real data showing its state-of-the-art performance without any external training data.

Limited-Angle Computed Tomography (LACT) is a nondestructive 3D imaging technique used in a variety of applications ranging from security to medicine. The limited angle coverage in LACT is often a dominant source of severe artifacts in the reconstructed images, making it a challenging imaging inverse problem. Diffusion models are a recent class of deep generative models for synthesizing realistic images using image denoisers. In this work, we present DOLCE as the first framework for integrating conditionally-trained diffusion models and explicit physical measurement models for solving imaging inverse problems. DOLCE achieves the SOTA performance in highly ill-posed LACT by alternating between the data-fidelity and sampling updates of a diffusion model conditioned on the transformed sinogram. We show through extensive experimentation that unlike existing methods, DOLCE can synthesize high-quality and structurally coherent 3D volumes by using only 2D conditionally pre-trained diffusion models. We further show on several challenging real LACT datasets that the same pretrained DOLCE model achieves the SOTA performance on drastically different types of images.

Plug-and-play (PnP) prior is a well-known class of methods for solving imaging inverse problems by computing fixed-points of operators combining physical measurement models and learned image denoisers. While PnP methods have been extensively used for image recovery with known measurement operators, there is little work on PnP for solving blind inverse problems. We address this gap by presenting a new block-coordinate PnP (BC-PnP) method that efficiently solves this joint estimation problem by introducing learned denoisers as priors on both the unknown image and the unknown measurement operator. We present a new convergence theory for BC-PnP compatible with blind inverse problems by considering nonconvex data-fidelity terms and expansive denoisers. Our theory analyzes the convergence of BC-PnP to a stationary point of an implicit function associated with an approximate minimum mean-squared error (MMSE) denoiser. We numerically validate our method on two blind inverse problems: automatic coil sensitivity estimation in magnetic resonance imaging (MRI) and blind image deblurring. Our results show that BC-PnP provides an efficient and principled framework for using denoisers as PnP priors for jointly estimating measurement operators and images.

Plug-and-Play (PnP) methods are efficient iterative algorithms for solving ill-posed image inverse problems. PnP methods are obtained by using deep Gaussian denoisers instead of the proximal operator or the gradient-descent step within proximal algorithms. Current PnP schemes rely on data-fidelity terms that have either Lipschitz gradients or closed-form proximal operators, which is not applicable to Poisson inverse problems. Based on the observation that the Gaussian noise is not the adequate noise model in this setting, we propose to generalize PnP using the Bregman Proximal Gradient (BPG) method. BPG replaces the Euclidean distance with a Bregman divergence that can better capture the smoothness properties of the problem. We introduce the Bregman Score Denoiser specifically parametrized and trained for the new Bregman geometry and prove that it corresponds to the proximal operator of a nonconvex potential. We propose two PnP algorithms based on the Bregman Score Denoiser for solving Poisson inverse problems. Extending the convergence results of BPG in the nonconvex settings, we show that the proposed methods converge, targeting stationary points of an explicit global functional. Experimental evaluations conducted on various Poisson inverse problems validate the convergence results and showcase effective restoration performance.

Physical and budget constraints often result in irregular sampling, which complicates accurate subsurface imaging. Pre-processing approaches, such as missing trace or shot interpolation, are typically employed to enhance seismic data in such cases. Recently, deep learning has been used to address the trace interpolation problem at the expense of large amounts of training data to adequately represent typical seismic events. Nonetheless, most research in this area has focused on trace reconstruction, with little attention having been devoted to shot interpolation. Furthermore, existing methods assume regularly spaced receivers/sources failing in approximating seismic data from real (irregular) surveys. This work presents a novel shot gather interpolation approach which uses a continuous coordinate-based representation of the acquired seismic wavefield parameterized by a neural network. The proposed unsupervised approach, which we call coordinate-based seismic interpolation (CoBSI), enables the prediction of specific seismic characteristics in irregular land surveys without using external data during neural network training. Experimental results on real and synthetic 3D data validate the ability of the proposed method to estimate continuous smooth seismic events in the time-space and frequency-wavenumber domains, improving sparsity or low-rank-based interpolation methods.

The past few years have seen a surge of activity around integration of deep learning networks and optimization algorithms for solving inverse problems. Recent work on plug-and-play priors (PnP), regularization by denoising (RED), and deep unfolding has shown the state-of-the-art performance of such integration in a variety of applications. However, the current paradigm for designing such algorithms is inherently Euclidean, due to the usage of the quadratic norm within the projection and proximal operators. We propose to broaden this perspective by considering a non-Euclidean setting based on the more general Bregman distance. Our new Bregman Proximal Gradient Method variant of PnP (PnP-BPGM) and Bregman Steepest Descent variant of RED (RED-BSD) replace the traditional updates in PnP and RED from the quadratic norms to more general Bregman distance. We present a theoretical convergence result for PnP-BPGM and demonstrate the effectiveness of our algorithms on Poisson linear inverse problems.

Regularization by denoising (RED) is a widely-used framework for solving inverse problems by leveraging image de-noisers as image priors. Recent work has reported the state-of-the-art performance of RED in a number of imaging applications using pre-trained deep neural nets as denoisers. Despite the recent progress, the stable convergence of RED algorithms remains an open problem. The existing RED theory only guarantees stability for convex data-fidelity terms and nonexpansive denoisers. This work addresses this issue by developing a new monotone RED (MRED) algorithm, whose convergence does not require nonexpansiveness of the deep denoising prior. Simulations on image deblurring and compressive sensing recovery from random matrices show the stability of MRED even when the traditional RED diverges.

Plug-and-Play Priors (PnP) and Regularization by Denoising (RED) are widely-used frameworks for solving imaging inverse problems by computing fixed-points of operators combining physical measurement models and learned image priors. While traditional PnP/RED formulations have focused on priors specified using image denoisers, there is a growing interest in learning PnP/RED priors that are end-to-end optimal. The recent Deep Equilibrium Models (DEQ) framework has enabled memory-efficient end-to-end learning of PnP/RED priors by implicitly differentiating through the fixed-point equations without storing intermediate activation values. However, the dependence of the computational/memory complexity of the measurement models in PnP/RED on the total number of measurements leaves DEQ impractical for many imaging applications. We propose ODER as a new strategy for improving the efficiency of DEQ through stochastic approximations of the measurement models. We theoretically analyze ODER giving insights into its convergence and ability to approximate the traditional DEQ approach. Our numerical results suggest the potential improvements in training/testing complexity due to ODER on three distinct imaging applications.

Deep neural networks for medical image reconstruction are traditionally trained using high-quality ground-truth images as training targets. Recent work on Noise2Noise (N2N) has shown the potential of using multiple noisy measurements of the same object as an alternative to having a ground-truth. However, existing N2N-based methods are not suitable for learning from the measurements of an object undergoing nonrigid deformation. This paper addresses this issue by proposing the deformation-compensated learning (DeCoLearn) method for training deep reconstruction networks by compensating for object deformations. A key component of DeCoLearn is a deep registration module, which is jointly trained with the deep reconstruction network without any ground-truth supervision. We validate DeCoLearn on both simulated and experimentally collected magnetic resonance imaging (MRI) data and show that it significantly improves imaging quality.

Deep unfolding networks are rapidly gaining attention for solving imaging inverse problems. However, the computational and memory complexity of existing deep unfolding networks scales with the size of the full measurement set, limiting their applicability to certain large-scale imaging inverse problems. We propose SCRED-Net as a novel methodology that introduces a stochastic approximation to the unfolded regularization by denoising (RED) algorithm. Our method uses only a subset of measurements within each cascade block, making it scalable to a large number of measurements for efficient end-to-end training. We present numerical results showing the effectiveness of SCRED-Net on intensity diffraction tomography (IDT) and sparse-view computed tomography (CT). Our results show that SCRED-Net matches the performance of a batch deep unfolding network at a fraction of training and operational complexity.

The plug-and-play priors (PnP) and regularization by denoising (RED) methods have become widely used for solving inverse problems by leveraging pre-trained deep denoisers as image priors. While the empirical imaging performance and the theoretical convergence properties of these algorithms have been widely investigated, their recovery properties have not previously been theoretically analyzed. We address this gap by showing how to establish theoretical recovery guarantees for PnP/RED by assuming that the solution of these methods lies near the fixed-points of a deep neural network. We also present numerical results comparing the recovery performance of PnP/RED in compressive sensing against that of recent compressive sensing algorithms based on generative models. Our numerical results suggest that PnP with a pre-trained artifact removal network provides significantly better results compared to the existing state-of-the-art methods.

Recent work has shown the effectiveness of the plug-and-play priors (PnP) framework for regularized image reconstruction. However, the performance of PnP depends on the quality of the denoisers used as priors. In this letter, we design a novel PnP denoising prior, called multiple self-similarity net (MSSN), based on the recurrent neural network (RNN) with self-similarity matching using multi-head attention mechanism. Unlike traditional neural net denoisers, MSSN exploits different types of relationships among non-local and repeating features to remove the noise in the input image. We numerically evaluate the performance of MSSN as a module within PnP for solving magnetic resonance (MR) image reconstruction. Experimental results show the stable convergence and excellent performance of MSSN for reconstructing images from highly compressive Fourier measurements.

Plug-and-play priors (PnP) is a methodology for regularized image reconstruction that specifies the prior through an image denoiser. While PnP algorithms are well understood for denoisers performing maximum a posteriori probability (MAP) estimation, they have not been analyzed for the minimum mean squared error (MMSE) denoisers. This letter addresses this gap by establishing the first theoretical convergence result for the iterative shrinkage/thresholding algorithm (ISTA) variant of PnP for MMSE denoisers. We show that the iterates produced by PnP-ISTA with an MMSE denoiser converge to a stationary point of some global cost function. We validate our analysis on sparse signal recovery in compressive sensing by comparing two types of denoisers, namely the exact MMSE denoiser and the approximate MMSE denoiser obtained by training a deep neural net.

The plug-and-play priors (PnP) framework has been recently shown to achieve state-of-the-art results in regularized image reconstruction by leveraging a sophisticated denoiser within an iterative algorithm. In this paper, we propose a new online PnP algorithm for Fourier ptychographic microscopy (FPM) based on the accelerated proximal gradient method (APGM). Specifically, the proposed algorithm uses only a subset of measurements, which makes it scalable to a large set of measurements. We validate the algorithm by showing that it can lead to significant performance gains on both simulated and experimental data.

In the past decade, sparsity-driven regularization has led to significant improvements in image reconstruction. Traditional regularizers, such as total variation (TV), rely on analytical models of sparsity. However, increasingly the field is moving towards trainable models, inspired from deep learning. Deep image prior (DIP) is a recent regularization framework that uses a convolutional neural network (CNN) architecture without data-driven training. This paper extends the DIP framework by combining it with the traditional TV regularization. We show that the inclusion of TV leads to considerable performance gains when tested on several traditional restoration tasks such as image denoising and deblurring.

Stochastic gradient descent (SGD) is one of the most widely used optimization methods for parallel and distributed processing of large datasets. One of the key limitations of distributed SGD is the need to regularly communicate the gradients between different computation nodes. To reduce this communication bottleneck, recent work has considered a one-bit variant of SGD, where only the sign of each gradient element is used in optimization. In this paper, we extend this idea by proposing a stochastic variant of the proximal-gradient method that also uses one-bit per update element. We prove the theoretical convergence of the method for non-convex optimization under a set of explicit assumptions. Our results indicate that the compressed method can match the convergence rate of the uncompressed one, making the proposed method potentially appealing for distributed processing of large datasets.

We consider the problem of estimating a vector from its noisy measurements using a prior specified only through a denoising function. Recent work on plug-and-play priors (PnP) and regularization-by-denoising (RED) has shown the state-of-the-art performance of estimators under such priors in a range of imaging tasks. In this work, we develop a new block coordinate RED algorithm that decomposes a large-scale estimation problem into a sequence of updates over a small subset of the unknown variables. We theoretically analyze the convergence of the algorithm and discuss its relationship to the traditional proximal optimization. Our analysis complements and extends recent theoretical results for RED-based estimation methods. We numerically validate our method using several denoiser priors, including those based on convolutional neural network (CNN) denoisers.

The problem of reconstructing an object from the measurements of the light it scatters is common in numerous imaging applications. While the most popular formulations of the problem are based on linearizing the object-light relationship, there is an increased interest in considering nonlinear formulations that can account for multiple light scattering. In this paper, we propose an image reconstruction method, called CISOR, for nonlinear diffractive imaging, based on our new variant of fast iterative shrinkage/thresholding algorithm (FISTA) and total variation (TV) regularization. We prove that CISOR reliably converges for our nonconvex optimization problem, and systematically compare our method with other state-of-the-art methods on simulated as well as experimentally measured data.

The radar autofocus problem arises in situations where radar measurements are acquired of a scene using antennas that suffer from position ambiguity. Current techniques model the antenna ambiguity as a global phase error affecting the received radar measurement at every antenna. However, the phase error signal model is only valid in the far field regime where the position error can be approximated by a one dimensional shift in the down-range direction. We propose in this paper an alternate formulation where the antenna position error is modeled using a two-dimensional shift operator in the image-domain. The radar autofocus problem then becomes a multichannel two-dimensional blind deconvolution problem where the static radar image is convolved with a two dimensional shift kernel for each antenna measurement. We develop an alternating minimization framework that leverages the sparsity and piece-wise smoothness of the radar scene, as well as the one-sparse property of the two dimensional shift kernels.

Multi-spectral (MS) image super-resolution aims to reconstruct super-resolved multi-channel images from their low-resolution images by regularizing the image to be reconstructed. Recently data-driven regularization techniques based on sparse modeling and deep learning have achieved substantial improvements in single image reconstruction problems. Inspired by these data-driven methods, we develop a novel coupled analysis and synthesis dictionary (CASD) model for MS image super-resolution, by exploiting a regularizer that operates within, as well as across, multiple spectral channels using convolutional dictionaries. To learn the CASD model parameters, we propose a deep dictionary learning framework, named DeepCASD, by unfolding and training an end-to-end CASD based reconstruction network over an image data set. Experimental results show that the DeepCASD framework exhibits improved performance on multi-spectral image super-resolution compared to state-of-the-art learning based super-resolution algorithms.

In remote sensing systems, exact knowledge of the sensor locations is critical for generating focused images. In order to accurately locate misplaced or perturbed sensors from their received signal data, we proposed a robust sensor localization method based on low-rank Euclidean distance matrix (EDM) reconstruction. To this end, an EDM of sensors and objects under detection is defined and partially initialized by computing distances between the inaccurate sensor locations and distances from the sensors to the objects using signal coherence analysis. We then decompose the noisy EDM with missing entries into a low-rank EDM corresponding to true sensor locations and a sparse matrix of distance errors by solving a constrained optimization problem using the alternating direction method of multipliers (ADMM). We verify our method with simulations on a uniform linear array with unknown perturbations up to several wavelengths.

In the past decade, sparsity-driven regularization has led to the advancement of image reconstruction algorithms. Traditionally, such regularizers rely on analytical models of sparsity [e.g., total variation (TV)]. However, more recent methods are increasingly centered around data-driven arguments inspired by deep learning. In this letter, we propose to generalize TV regularization by replacing the 11 -penalty with an alternative prior that is trainable. Specifically, our method learns the prior via extending the recently proposed fast parallel proximal algorithm to incorporate data-adaptive proximal operators. The proposed framework does not require additional inner iterations for evaluating the proximal mappings of the corresponding learned prior. Moreover, our formalism ensures that the training and reconstruction processes share the same algorithmic structure, making the endto-end implementation intuitive. As an example, we demonstrate our algorithm on the problem of deconvolution in a fluorescence microscope.

We propose a new compressive imaging method for reconstructing 2D or 3D objects from their scattered wave-field measurements. Our method relies on a novel, nonlinear measurement model that can account for the multiple scattering phenomenon, which makes the method preferable in applications where linear measurement models are inaccurate. We construct the measurement model by expanding the scattered wave-field with an accelerated-gradient method, which is guaranteed to converge and is suitable for large-scale problems. We provide explicit formulas for computing the gradient of our measurement model with respect to the unknown image, which enables image formation with a sparsity-driven numerical optimization algorithm. We validate the method both analytically and with numerical simulations.

Learning Tomography (LT) is a nonlinear optimization algorithm for computationally imaging three-dimensional (3D) distribution of the refractive index in semi-transparent samples. Since the energy function in LT is generally non-convex, the solution it obtains is not guaranteed to be globally optimal. In this paper, we describe linear and nonlinear tomographic reconstruction methods and compare them numerically. We present a review of the LT and, in addition, we investigate the influence of the initialization and exemplify the effect of regularization on the convergence of the algorithm. In particular, we show that both are essential for high-quality imaging in strongly scattering scenarios.

Computational imaging methods that can exploit multiple modalities have the potential to enhance the capabilities of traditional sensing systems. In this paper, we propose a new method that reconstructs multimodal images from their linear measurements by exploiting redundancies across different modalities. Our method combines a convolutional group-sparse representation of images with total variation (TV) regularization for high-quality multimodal imaging. We develop an online algorithm that enables the unsupervised learning of convolutional dictionaries on large-scale datasets that are typical in such applications. We illustrate the benefit of our approach in the context of joint intensity-depth imaging.

We consider the problem of fusing multiple cloud-contaminated aerial images of a 3D scene to generate a cloud-free image, where the images are captured from multiple unknown view angles. In order to fuse these images, we propose an end-to-end framework incorporating epipolar geometry and low-rank matrix completion. In particular, we first warp the multi-angular images to single-angle ones based on the estimated fundamental matrices that relate the multi-angular images according to their projective relations to the 3D scene. Then we formulate the fusion process of the warpped images as a low-rank matrix completion problem where each column of the matrix corresponds to a vectorized image with missing entries corresponding to cloud or occluded areas. Results using DigitalGlobe high spatial resolution images demonstrate that our algorithm outperforms existing approaches.

In the past two decades, nonlinear image reconstruction methods have led to substantial improvements in the capabilities of numerous imaging systems. Such methods are traditionally formulated as optimization problems that are solved iteratively by simultaneously enforcing data consistency and incorporating prior models. Recently, the Plug-and-Play Priors (PPP) framework suggested that by using more sophisticated denoisers, not necessarily corresponding to an optimization objective, it is possible to improve the quality of reconstructed images. In this letter, we show that the PPP approach is applicable beyond linear inverse problems. In particular, we develop the fast iterative shrinkage/thresholding algorithm variant of PPP for model-based nonlinear inverse scattering. The key advantage of the proposed formulation over the original ADMM-based one is that it does not need to perform an inversion on the forward model. We show that the proposed method produces high quality images using both simulated and experimentally measured data.

Total variation (TV) is a one of the most popular regularizers for stabilizing the solution of ill-posed inverse problems. This paper proposes a novel proximal-gradient algorithm for minimizing TV regularized least-squares cost functionals. Unlike traditional methods that require nested iterations for computing the proximal step of TV, our algorithm approximates the latter with several simple proximals that have closed form solutions. We theoretically prove that the proposed parallel proximal method achieves the TV solution with arbitrarily high precision at a global rate of converge that is equivalent to the fast proximal-gradient methods. The results in this paper have the potential to enhance the applicability of TV for solving very large-scale imaging inverse problems.

Multi-modal sensing is increasingly becoming important in a number of applications, providing new capabilities and processing challenges. In this paper, we explore the benefit of combining a low-resolution depth sensor with a high-resolution optical video sensor, in order to provide a high-resolution depth map of the scene. We propose a new formulation that is able to incorporate temporal information and exploit the motion of objects in the video to significantly improve the results over existing methods. In particular, our approach exploits the space-time redundancy in the depth and intensity using motion-adaptive low-rank regularization. We provide experiments to validate our approach and confirm that the quality of the estimated high-resolution depth is improved substantially. Our approach can be a first component in systems using vision techniques that rely on high-resolution depth information.

Generalized linear models, where a random vector x is observed through a noisy, possibly nonlinear, function of a linear transform z = Ax, arise in a range of applications in nonlinear filtering and regression. Approximate message passing (AMP) methods, based on loopy belief propagation, are a promising class of approaches for approximate inference in these models. AMP methods are computationally simple, general, and admit precise analyses with testable conditions for optimality for large i.i.d. transforms A. However, the algorithms can diverge for general A. This paper presents a convergent approach to the generalized AMP (GAMP) algorithm based on direct minimization of a large-system limit approximation of the Bethe free energy (LSL-BFE). The proposed method uses a double-loop procedure, where the outer loop successively linearizes the LSL-BFE and the inner loop minimizes the linearized LSL-BFE using the alternating direction method of multipliers (ADMM). The proposed method, called ADMM-GAMP, is similar in structure to the original GAMP method, but with an additional least-squares minimization. It is shown that for strictly convex, smooth penalties, ADMM-GAMP is guaranteed to converge to a local minimum of the LSL-BFE, thus providing a convergent alternative to GAMP that is stable under arbitrary transforms. Simulations are also presented that demonstrate the robustness of the method for non-convex penalties as well.

We present a new method for joint automatic extrinsic calibration and sensor fusion for a multimodal sensor system comprising a LIDAR and an optical camera. Our approach exploits the natural alignment of depth and intensity edges when the calibration parameters are correct. Thus, in contrast to a number of existing approaches, we do not require the presence or identification of known alignment targets. On the other hand, the characteristics of each sensor modality, such as sampling pattern and information measured, are significantly different, making direct edge alignment difficult. To overcome this difficulty, we jointly fuse the data and estimate the calibration parameters. In particular, the joint processing evaluates and optimizes both the quality of edge alignment and the performance of the fusion algorithm using a common cost function on the output. We demonstrate accurate calibration in practical configurations in which depth measurements are sparse and contain no reflectivity information. Experiments on synthetic and real data obtained with a three-dimensional LIDAR sensor demonstrate the effectiveness of our approach.

Total variation (TV) is a widely used regularizer for stabilizing the solution of ill-posed inverse problems. In this paper, we propose a novel proximal-gradient algorithm for minimizing TV regularized least-squares cost functional. Our method replaces the standard proximal step of TV by a simpler alternative that computes several independent proximals. We prove that the proposed parallel proximal method converges to the TV solution, while requiring no sub-iterations. The results in this paper could enhance the applicability of TV for solving very large scale imaging inverse problems.

In this paper, we propose an online radar imaging scheme that recovers a sparse scene and removes the multipath ringing induced by the front wall in a Through-the-Wall-Imaging (TWI) system without prior knowledge of the wall parameters. Our approach uses online measurements obtained from individual transmitter-receiver pairs to incrementally build the primary response of targets behind the front wall and find a corresponding delay convolution operator that generates the multi-path reflections available in the received signal. In order to perform online sparse imaging while removing wall clutter reflections, we developed a deconvolution extension of the Sparse Randomized Kaczmarz (SRK) algorithm that finds sparse solutions to under- and over-determined linear systems of equations. Our scheme allows for imaging with nonuniformly spaced antennas by building an explicit delay-and-sum imaging operator for each new measurement. Moreover, the active memory requirements remain small even for large scale MIMO systems since the imaging operators are only constructed for individual transmitter-receiver pairs. We test our approach on a simple FDTD simulated room with internal targets and demonstrate that our method successfully eliminates multipath reflections while correctly locating the targets.

The iterative Born approximation (IBA) is a well-known method for describing waves scattered by semitransparent objects. In this letter, we present a novel nonlinear inverse scattering method that combines IBA with an edge-preserving total variation regularizer. The proposed method is obtained by relating iterations of IBA to layers of an artificial multilayer neural network and developing a corresponding error backpropagation algorithm for efficiently estimating the permittivity of the object. Simulations illustrate that, by accounting for multiple scattering, the method successfully recovers the permittivity distribution where the traditional linear inverse scattering fails.

Iterative shrinkage/thresholding algorithm (ISTA) is a well-studied method for finding sparse solutions to ill-posed inverse problems. In this letter, we present a data-driven scheme for learning optimal thresholding functions for ISTA. The proposed scheme is obtained by relating iterations of ISTA to layers of a simple feedforward neural network and developing a corresponding error backpropagation algorithm for fine-tuning the thresholding functions. Simulations on sparse statistical signals illustrate potential gains in estimation quality due to the proposed data adaptive ISTA.

Generalized Linear Models (GLMs), where a random vector x is observed through a noisy, possibly nonlinear, function of a linear transform z = Ax arise in a range of applications in nonlinear filtering and regression. Approximate Message Passing (AMP) methods, based on loopy belief propagation, are a promising class of approaches for approximate inference in these models. AMP methods are computationally simple, general, and admit precise analyses with testable conditions for optimality for large i.i.d. transforms A. However, the algorithms can easily diverge for general transforms. This paper presents a convergent approach to the generalized AMP (GAMP) algorithm based on direct minimization of a large-system limit approximation of the Bethe Free Energy (LSL-BFE). The proposed method uses a double-loop procedure, where the outer loop successively linearizes the LSL-BFE and the inner loop minimizes the linearized LSL-BFE using the Alternating Direction Method of Multipliers (ADMM). The proposed method, called ADMM-GAMP, is similar in structure to the original GAMP method, but with an additional least-squares minimization. It is shown that for strictly convex, smooth penalties, ADMM-GAMP is guaranteed to converge to a local minima of the LSL-BFE, thus providing a convergent alternative to GAMP that is stable under arbitrary transforms. Simulations are also presented that demonstrate the robustness of the method for non-convex penalties as well.

Cycle spinning is a widely used approach for improving the performance of wavelet-based methods that solve linear inverse problems. Extensive numerical experiments have shown that it significantly improves the quality of the recovered signal without increasing the computational cost. In this letter, we provide the first theoretical convergence result for cycle spinning for solving general linear inverse problems. We prove that the sequence of reconstructed signals is guaranteed to converge to the minimizer of some global cost function that incorporates all wavelet shifts.

We consider the estimation of an independent and identically distributed (i.i.d.) (possibly non-Gaussian) vector x ∈ Rnfrom measurements y ∈ Rmobtained by a general cascade model consisting of a known linear transform followed by a probabilistic componentwise (possibly nonlinear) measurement channel. A novel method, called adaptive generalized approximate message passing (adaptive GAMP) is presented. It enables the joint learning of the statistics of the prior and measurement channel along with estimation of the unknown vector x. We prove that, for large i.i.d. Gaussian transform matrices, the asymptotic componentwise behavior of the adaptive GAMP is predicted by a simple set of scalar state evolution equations. In addition, we show that the adaptive GAMP yields asymptotically consistent parameter estimates, when a certain maximum-likelihood estimation can be performed in each step. This implies that the algorithm achieves a reconstruction quality equivalent to the oracle algorithm that knows the correct parameter values. Remarkably, this result applies to essentially arbitrary parametrizations of the unknown distributions, including nonlinear and non-Gaussian ones. The adaptive GAMP methodology thus provides a systematic, general and computationally efficient method applicable to a large range of linear-nonlinear models with provable guarantees.

In this paper, we reconstruct signals from underdetermined linear measurements where the componentwise gains of the measurement system are unknown a priori. The reconstruction is performed through an adaptation of the messagepassing algorithm called adaptive GAMP that enables joint gain calibration and signal estimation. To evaluate our approach, we apply it to the problem of sparse recovery and compare it against an ℓ1-based approach. We numerically show that adaptive GAMP yields excellent results even for a moderate amount of data. It approaches the performance of oracle GAMP where the gains are perfectly known asymptotically.

The steerable wavelet transform is a redundant image representation with the remarkable property that its basis functions can be adaptively rotated to a desired orientation. This makes the transform well-suited to the design of wavelet-based algorithms applicable to images with a high amount of directional features. However, arbitrary modification of the wavelet-domain coefficients may violate consistency constraints because a legitimate representation must be redundant. In this paper, by honoring the redundancy of the coefficients, we demonstrate that it is possible to improve the performance of regularized least-squares problems in the steerable wavelet domain. We illustrate that our consistent method significantly improves upon the performance of conventional denoising with steerable wavelets.

We introduce a new approach for the implementation of minimum mean-square error (MMSE) denoising for signals with decoupled derivatives. Our method casts the problem as a penalized least-squares regression in the redundant wavelet domain. It exploits the link between the discrete gradient and Haar-wavelet shrinkage with cycle spinning. The redundancy of the representation implies that some wavelet-domain estimates are inconsistent with the underlying signal model. However, by imposing additional constraints, our method finds wavelet-domain solutions that are mutually consistent. We confirm the MMSE performance of our method through statistical estimation of Lévy processes that have sparse derivatives.

We present a novel statistically-based discretization paradigm and derive a class of maximum a posteriori (MAP) estimators for solving ill-conditioned linear inverse problems. We are guided by the theory of sparse stochastic processes, which specifies continuous-domain signals as solutions of linear stochastic differential equations. Accordingly, we show that the class of admissible priors for the discretized version of the signal is confined to the family of infinitely divisible distributions. Our estimators not only cover the well-studied methods of Tikhonov and l1-type regularizations as particular cases, but also open the door to a broader class of sparsity-promoting regularization schemes that are typically nonconvex. We provide an algorithm that handles the corresponding nonconvex problems and illustrate the use of our formalism by applying it to deconvolution, magnetic resonance imaging, and X-ray tomographic reconstruction problems. Finally, we compare the performance of estimators associated with models of increasing sparsity.

We investigate the problem of the optimal reconstruction of a generalized Poisson process from its noisy samples. The process is known to have a finite rate of innovation since it is generated by a random stream of Diracs with a finite average number of impulses per unit interval. We formulate the recovery problem in a Bayesian framework and explicitly derive the joint probability density function (pdf) of the sampled signal. We compare the performance of the optimal Minimum Mean Square Error (MMSE) estimator with common regularization techniques such as ℓ1and Log penalty functions. The simulation results indicate that, under certain conditions, the regularization techniques can achieve a performance close to the MMSE method.

Many recent algorithms for sparse signal recovery can be interpreted as maximum-a-posteriori (MAP) estimators relying on some specific priors. From this Bayesian perspective, state-of-the-art methods based on discrete-gradient regularizers, such as total-variation (TV) minimization, implicitly assume the signals to be sampled instances of Lévy processes with independent Laplace-distributed increments. By extending the concept to more general Lévy processes, we propose an efficient minimum-mean-squared error (MMSE) estimation method based on message-passing algorithms on factor graphs. The resulting algorithm can be used to benchmark the performance of the existing or design new algorithms for the recovery of sparse signals.

We consider the denoising of signals and images using regularized least-squares method. In particular, we propose a simple minimization algorithm for regularizers that are functions of the discrete gradient. By exploiting the connection of the discrete gradient with the Haar-wavelet transform, the n-dimensional vector minimization can be decoupled into n scalar minimizations. The proposed method can efficiently solve total-variation (TV) denoising by iteratively shrinking shifted Haar-wavelet transforms. Furthermore, the decoupling naturally lends itself to extensions beyond ℓ1regularizers.

Motivated by the analog nature of real-world signals, we investigate continuous-time random processes. For this purpose, we consider the stochastic processes that can be whitened by linear transformations and we show that the distribution of their samples is necessarily infinitely divisible. As a consequence, such a modeling rules out the Bernoulli-Gaussian distribution since we are able to show in this paper that it is not infinitely divisible. In other words, while the Bernoulli-Gaussian distribution is among the most studied priors for modeling sparse signals, it cannot be associated with any continuous-time stochastic process. Instead, we propose to adapt the priors that correspond to the increments of compound Poisson processes, which are both sparse and infinitely divisible.

We consider the estimation of an i.i.d.\ vector $\xbf \in \R^n$ from measurements $\ybf \in \R^m$ obtained by a general cascade model consisting of a known linear transform followed by a probabilistic componentwise (possibly nonlinear) measurement channel. We present a method, called adaptive generalized approximate message passing (Adaptive GAMP), that enables joint learning of the statistics of the prior and measurement channel along with estimation of the unknown vector $\xbf$. The proposed algorithm is a generalization of a recently-developed method by Vila and Schniter that uses expectation-maximization (EM) iterations where the posteriors in the E-steps are computed via approximate message passing. The techniques can be applied to a large class of learning problems including the learning of sparse priors in compressed sensing or identification of linear-nonlinear cascade models in dynamical systems and neural spiking processes. We prove that for large i.i.d.\ Gaussian transform matrices the asymptotic componentwise behavior of the adaptive GAMP algorithm is predicted by a simple set of scalar state evolution equations. This analysis shows that the adaptive GAMP method can yield asymptotically consistent parameter estimates, which implies that the algorithm achieves a reconstruction quality equivalent to the oracle algorithm that knows the correct parameter values. The adaptive GAMP methodology thus provides a systematic, general and computationally efficient method applicable to a large range of complex linear-nonlinear models with provable guarantees.

We introduce a new method for adaptive one-bit quantization of linear measurements and propose an algorithm for the recovery of signals based on generalized approximate message passing (GAMP). Our method exploits the prior statistical information on the signal for estimating the minimum-mean-squared error solution from one-bit measurements. Our approach allows the one-bit quantizer to use thresholds on the real line. Given the previous measurements, each new threshold is selected so as to partition the consistent region along its centroid computed by GAMP. We demonstrate that the proposed adaptive-quantization scheme with GAMP reconstruction greatly improves the performance of signal and image recovery from one-bit measurements.

We introduce a new wavelet-based method for the implementation of Total-Variation-type denoising. The data term is least-squares, while the regularization term is gradient-based. The particularity of our method is to exploit a link between the discrete gradient and wavelet shrinkage with cycle spinning, which we express by using redundant wavelets. The redundancy of the representation gives us the freedom to enforce additional constraints (e.g., normalization) on the solution to the denoising problem. We perform optimization in an augmented-Lagrangian framework, which decouples the difficultn-dimensional constrained-optimization problem into a sequence ofneasier scalar unconstrained problems that we solve efficiently via traditional wavelet shrinkage. Our method can handle arbitrary gradient-based regularizers. In particular, it can be made to adhere to the popular principle of least total variation. It can also be used as a maximum a posteriori estimator for a variety of priors. We illustrate the performance of our method for image denoising and for the statistical estimation of sparse stochastic processes.

We consider the optimal quantization of compressive sensing measurements along with estimation from quantized samples using generalized approximate message passing (GAMP). GAMP is an iterative reconstruction scheme inspired by the belief propagation algorithm on bipartite graphs which generalizes approximate message passing (AMP) for arbitrary measurement channels. Its asymptotic error performance can be accurately predicted and tracked through the state evolution formalism. We utilize these results to design mean-square optimal scalar quantizers for GAMP signal reconstruction and empirically demonstrate the superior error performance of the resulting quantizers.