Image Restoration

The demand for robust face recognition in real-world surveillance cameras is increasing due to the needs of practical applications such as security and surveillance. Although face recognition has been studied extensively in the literature, achieving good performance in surveillance videos with unconstrained faces is inherently difficult. During the image acquisition process, the noncooperative subjects appear in arbitrary poses and resolutions in different lighting conditions, together with noise and blurriness of images. In addition, multiple cameras are usually distributed in a camera network and different cameras often capture a subject in different views. In this paper, we aim at tackling this unconstrained face recognition problem and utilizing multiple cameras to improve the recognition accuracy using a probabilistic approach. We propose a dynamic Bayesian network to incorporate the information from different cameras as well as the temporal clues from frames in a video sequence. The proposed method is tested on a public surveillance video dataset with a three-camera setup. We compare our method to different benchmark classifiers with various feature descriptors. The results demonstrate that by modeling the face in a dynamic manner the recognition performance in a multi-camera network is improved over the other classifiers with various feature descriptors and the recognition result is better than using any of the single camera.

Indexing ferms: Binary sequences, Currelutiun A method for thc synthcsik of an infinite family of binary sequences with perfcctly flat cross-correlation sidelobes is presented. The corrcaponding postprncessing arrays are similarly two-valued. and so the sequences may bc particularly suited to applications in high noise environments.

We present a framework based on modular integration and multiresolution for restoring images. We model the image as a Markov random field (MRF) and propose a restoration algorithm. In essence, the problem of image restoration requires learning of the MRF model and noise parameters which are used to restore degraded images. In the developed scheme, there exists interaction between the model \earning module and the image restoration module. A method based on homotopy continuation is used for unsupervised model learning and the restoration is achieved through the minimization of an energy function.

Generative Adversarial Networks are artificial neural networks that pit two different sets of neural networks against one another in order to generate data that isn't part of the training set. The Generative Adversarial Network (GAN) produces good outcomes when they are trained on image data that comes from the actual world. The generator and the discriminator make up the Generative Adversarial Network (GAN), which stands for "generative adversarial network." The parameters that were utilized to generate the data are completely arbitrary. The information is evaluated, and erroneous information is distinguished from true information by the discriminator. GAN has proven to be useful in various domains, such as object recognition, text synthesis, face ageing, image manipulation, image overpainting, image stitching, human pose synthesis, visual salience prediction, stenographic applications, and many more. Several researchers have investigated various types of GANs but comprehensive analysis and comparison of different types of recent GAN's like Deep Convolutional GAN, Wasserstein GAN,Auto Enocoder, Cycle GAN,Progressive GAN and Super Resoultion GAN have not been published so far in the literature. In this article, quantitative and qualitative analysis are carried on to evaluate their suitability of the GAN for a particular application.Examples of applications encompass texture production, facial reconstruction, facial recognition, high-resolution imaging, music composition, drawing creation, cosmetic enhancements, image transformation, voice synthesis, medical diagnostics, and video editing. No single GAN cannot fulfil the desired requirements for all the applications. The article concludes with a discussion of the possible uses of GANs as well as how these applications constitute a fascinating new area of research and prospective expansion.

Three-dimensional images of confocal laser scanning microscopy suffer from a depth-variant blur, due to refractive index mismatch between the different mediums composing the system as well as the specimen, leading to optical aberrations. Our goal is to develop an image restoration method for 3D confocal microscopy taking into account the blur variation with depth. The difficulty is that optical aberrations depend on the refractive index of the biological specimen. The depth-variant blur function or the Point Spread Function (PSF) is thus different for each observation. A blind or semi-blind restoration method needs to be developed for this system. For that purpose, we use a previously developed algorithm for the joint estimation of the specimen function (original image) and the 3D PSF, the continuously depth-variant PSF is approximated by a convex combination of a set of space-invariant PSFs taken at different depths. We propose to add to that algorithm a pupil-phase constraint for the PSF estimation, given by the the optical instrument geometry. We thus define a blind estimation algorithm by minimizing a regularized criterion in which we integrate the Gerchberg-Saxton algorithm allowing to include these physical constraints. We show the efficiency of this method relying on some numerical tests.

In this paper, we propose a maximum-entropy expectation-maximization (MEEM) algorithm. We use the proposed algorithm for density estimation. The maximum-entropy constraint is imposed for smoothness of the estimated density function. The derivation of the MEEM algorithm requires determination of the covariance matrix in the framework of the maximum-entropy likelihood function, which is difficult to solve analytically. We, therefore, derive the MEEM algorithm by optimizing a lower-bound of the maximum-entropy likelihood function. We note that the classical expectation-maximization (EM) algorithm has been employed previously for 2-D density estimation. We propose to extend the use of the classical EM algorithm for image recovery from randomly sampled data and sensor field estimation from randomly scattered sensor networks. We further propose to use our approach in density estimation, image recovery and sensor field estimation. Computer simulation experiments are used to demonstrate the superior performance of the proposed MEEM algorithm in comparison to existing methods.

Recent progress in the development of X-ray micro-computed tomography (μ-CT) makes it possible to directly visualize and quantify various features of the microstructure of complex porous media including sedimentary rocks, in particular [1-5]. In various porous media processes such as chemical dissolution and mechanical compaction, the conditions may be appropriate to allow local particle detachment and displacement within the porous medium. It is noted that for flow-throughporous-media experiments, a lack of observation of fines production does not preclude considerable local mobilization of particles within the porous medium at the pore-scale. Particle mobilization may be triggered by inertial forces of flow itself or releasing by chemical reactions or other forces and then displacing by flow. The local mobilization of fines is a largely unexplored phenomenon at the pore scale due to difficulty in direct access to local particle displacement.

Though there is a study concerning vision restoration of headlight-glare vision, enhancing visual quality is able to be improved. Current frame information would contain more video features instead of restoring it only from the previous frame in order to achieve higher quality of vision during dangerous-glare encounter. Therefore, this paper proposes a new method recovering information from the surrounding area of glare centroid which distorts the luminance. Three-dimensional recursive search (3DRS) is assigned to retrieve luminance information from Macroblock (MB) of the previous non-glare affected frames. Tone mapping is also applied to improve luminance of the distorted MB. The result shows that frames with the glare area under 15% implemented the proposed method acquires higher peak-signal-to-noise ratio (PSNR) than those applied glare simulation. Furthermore, the combination of restoration method and the proposed one results in the best performance of increasing PSNR among applying only a single method.

Photographs acquired under low-lighting conditions require long exposure times and therefore exhibit significant blurring due to the shaking of the camera. Using shorter exposure times results in sharper images but with a very high level of noise. By taking a pair of blurred/noisy images it is possible to reconstruct a sharp image without noise. This paper is devoted to the combination of observation models in the blurred/noisy image pair reconstruction problem. By examining the difference between the blurred image and the blurred version of the noisy image a third observation model is obtained. Based on the minimization of a linear convex combination of Kullback-Leibler divergences between posterior distributions, a procedure to combine the three observation models is proposed in the paper. The estimated images are compared with images provided by other reconstruction methods.

Photographs acquired under low-lighting conditions require long exposure times and therefore exhibit significant blurring due to the shaking of the camera. Using shorter exposure times results in sharper images but with a very high level of noise. By taking a pair of blurred/noisy images it is possible to reconstruct a sharp image without noise. This paper is devoted to the combination of observation models in the blurred/noisy image pair reconstruction problem. By examining the difference between the blurred image and the blurred version of the noisy image a third observation model is obtained. Based on the minimization of a linear convex combination of Kullback-Leibler divergences between posterior distributions, a procedure to combine the three observation models is proposed in the paper. The estimated images are compared with images provided by other reconstruction methods.

A prototype of the block-software system that is able to solve the main problems of image restoration and additional problems related to analysis/diagnostics of images and generation of databases of synthesized test images is proposed. The system can be used in the regime of emulation of the process of image transformation including blurring, analysis, and restoration (which allows tuning and training of the system for a specific device) and in the regime of analysis and restoration of blurred images. A method for estimation of the blurring operator using the observed blurred image is proposed. A learning algorithm for recognition of typical linear distortion operators is used to determine the type of blurring operator.

The causes of the appearance of distortions in the image formation and the approaches to correction of these distortions are considered. The motion blur, optical aberrations, and atmospheric scattering are used as distortion models. The applicability of metrical and topological characteristics of the distorted signal in distortion diagnostics and image restoration is examined.

Image restoration in astronomy has been considered a vital step in many ground-based observational programs that often suffer from sub-optimal seeing due to atmospheric turbulence, distortion of stellar shapes due to instrumental aberrations, trailing, and other issues. It holds importance for various tasks: improved astrometry, deblending of overlapping sources, faint source detection, and identification of point sources near bright extended objects, such as galaxies, to name a few. We conduct an empirical study by applying the Scaled Gradient Projection (SGP) iterative image deconvolution algorithm to restore distorted stellar shapes in our observed data. We investigate using a more flexible divergence measure, the β-divergence, which contains the commonly-used Kullback-Leibler (KL) divergence as a special case and allows automatic adaptation of the parameter β to the data. An extensive set of experiments comparing the performance of SGP and its β-divergence variant (β-SGP) is carried out on extracted star stamps and on images containing multiple stars (both crowded and relatively sparser fields). We show a consistent enhancement in the flux conservation across all considered scenarios using β-SGP compared to SGP. Using a few quantifiable metrics such as the Full-Width-at-Half-Maximum (FWHM) and ellipticity of stars, we observe that β-SGP improves restoration quality, compared to the SGP, in many cases and still preserves restoration quality in others. We conclude that generalized versions of image restoration algorithms are more robust due to their enhanced flexibility and could be a promising modification for astronomical image restoration.

Recently, sparse representation based methods have proven to be successful towards solving image restoration problems. The objective of these methods is to use sparsity prior of the underlying signal in terms of some dictionary and achieve optimal performance in terms of mean-squared error, a metric that has been widely criticized in the literature due to its poor performance as a visual quality predictor. In this work, we make one of the first attempts to employ structural similarity (SSIM) index, a more accurate perceptual image measure, by incorporating it into the framework of sparse signal representation and approximation. Specifically, the proposed optimization problem solves for coefficients with minimum L 0 norm and maximum SSIM index value. Furthermore, a gradient descent algorithm is developed to achieve SSIM-optimal compromise in combining the input and sparse dictionary reconstructed images. We demonstrate the performance of the proposed method by using image denoising and super-resolution methods as examples. Our experimental results show that the proposed SSIM-based sparse representation algorithm achieves better SSIM performance and better visual quality than the corresponding least square-based method.

This paper proposes accelerated subspace optimization methods in the context of image restoration. Subspace optimization methods belong to the class of iterative descent algorithms for unconstrained optimization. At each iteration of such methods, a stepsize vector allowing the best combination of several search directions is computed through a multi-dimensional search. It is usually obtained by an inner iterative second-order method ruled by a stopping criterion that guarantees the convergence of the outer algorithm. As an alternative, we propose an original multi-dimensional search strategy based on the majorize-minimize principle. It leads to a closed-form stepsize formula that ensures the convergence of the subspace algorithm whatever the number of inner iterations. The practical efficiency of the proposed scheme is illustrated in the context of edge-preserving image restoration.

Noise is an unavoidable aspect of modern camera technology, causing a decline in the overall visual quality of the images. Efforts are underway to diminish noise without compromising essential image features like edges, corners, and other intricate structures. Numerous techniques have already been suggested by many researchers for noise reduction, each with its unique set of benefits and drawbacks. Denoising images is a basic challenge in image processing. We describe a two-phase approach for removing impulse noise in this study. The adaptive median filter (AMF) for salt-and-pepper noise identifies noise candidates in the first phase. The second step minimizes an edge-preserving regularization function using a novel hybrid conjugate gradient approach. To generate the new improved search direction, the new algorithm takes advantage of two well-known successful conjugate gradient techniques. The descent property and global convergence are proven for the new methods. The obtained numerical results reveal that, when applied to image restoration, the new algorithms are superior to the classical fletcher reeves (FR) method in the same domain in terms of maintaining image quality and efficiency.

In this paper, a new extrapolation economy cascadic multigrid method is proposed to solve the image restoration model. The new method combines the new extrapolation formula and quadratic interpolation to design a nonlinear prolongation operator, which provides more accurate initial values for the fine grid level. An edge preserving denoising operator is constructed to remove noise and preserve image edges. The local smoothing operator reduces the influence of staircase effect. The experiment results show that the new method not only improves the computational efficiency but also ensures good recovery quality.

Early approaches to building mosaics by composing photographic images, assume the input images have similar exposures. Since this is unlikely to happen in practice, it became common to compensate for different exposures in the blending step, after the images have been registered, or aligned [1]. However, registration methods usually assume brightness constancy and fail to align images with different exposures. Recent approaches to this problem lead to computationally complex solutions that require either robust statistics or nonlinear optimization. In this paper we propose a computationally simple method to jointly estimate the registration parameters and the parameters describing the exposure correction, directly from the image intensity values. We obtain closedform solutions for the estimates of the exposure parameters. This enables the derivation of a simple two-step iterative algorithm to minimize the global cost. Our experiments show that this algorithm succeeds to register real images exhibiting simultaneously very distinct orientations and exposures.

Image rescaling is a commonly used bidirectional operation, which first downscales high-resolution images to fit various display screens or to be storage-and bandwidth-friendly, and afterward upscales the corresponding low-resolution images to recover the original resolution or the details in the zoom-in images. However, the noninjective downscaling mapping discards high-frequency contents, leading to the ill-posed problem for the inverse restoration task. This can be abstracted as a general image degradation-restoration problem with information loss. In this work, we propose a novel invertible framework to handle this general problem, which models the bidirectional degradation and restoration from a new perspective, i.e. invertible bijective transformation. The invertibility enables the framework to model the information loss of pre-degradation in the form of distribution, which could mitigate the ill-posed problem during post-restoration. To be specific, we develop invertible models to generate valid degraded images and meanwhile transform the distribution of lost contents to the fixed distribution of a latent variable during the forward degradation. Then restoration is made tractable by applying the inverse transformation on the generated degraded image together with a randomly-drawn latent variable. We start from image rescaling and instantiate the model as Invertible Rescaling Network (IRN), which can be easily extended to the similar decolorization-colorization task. We further propose to combine the invertible framework with existing degradation methods such as image compression for wider applications. Experimental results demonstrate the significant improvement of our model over existing methods in terms of both quantitative and qualitative evaluations of upscaling and colorizing reconstruction from downscaled and decolorized images, and ratedistortion of image compression. Code is available at .

This paper presents an image, then a video, compression scheme based on a restoration process applied to a transform domain. Most of the method relies on the ability to suppress, then automatically restore some transformed coefficients in a coding scheme. We use a total variation (TV) minimization model in order to voluntarily predict canceled coefficients. Thus, the method can be assimilated to a new prediction step by inpainting of transformed coefficients. The method has successfully been tested in a DCT-based JPEG coder, but with results that do not overpass the actual state-of-the-art. The main contribution of our work stands in the fact that this method has been introduced into actual video coding standards based on a predicting step. In these standards, a residual error is coded to recover the initial signal. In this case we noticed that the residual error has much less energy than the original coefficients and therefore that the compression rate could significantly be improved by using the same cancellation/restoration method, as a new prediction step. This has been (efficiently) tested on the H.264/AVC standard.

We consider in this paper the $l_0$-norm based dictionary learning approach combined with total variation regularization for the image restoration problem. It is formulated as a nonconvex nonsmooth optimization problem. Despite that this image restoration model has been proposed in many works, it remains important to ensure that the considered minimization method satisfies the global convergence property, which is the main objective of this work. Therefore, we employ the proximal alternating linearized minimization method whereby we demonstrate the global convergence of the generated sequence to a critical point. The results of several experiments demonstrate the performance of the proposed algorithm for image restoration.

Space Situational Awareness has become a key issue both for defense and civilian/industrial applications. Identification of potential or active threats and monitoring of key assets and operations are at stake. This activity also includes the follow up of dedicated satellites (such as telecommunication, observation), traffic handling, debris identification and tracking. We present current work performed at Onera on Adaptive Optics (AO) for Low Earth Orbit (LEO) satellite imaging. We describe briefly our AO system and its performance. We also address recent developments on image post-processing based on marginalized blind deconvolution combined to parsimonious PSF modelling. Finally, we present on sky results.

Image restoration is the process of obtaining a distorted/noise image and giving an approximate clear image of the original image. False focus, motion blur and noise are forms of distortion. Image restoration can be done by reversing the process called Point Extension Function (PSF). In this process, the blurred image is generated by point source imaging and can be used to restore the image lost due to the blur process. Like to form. Modern artificial intelligence (AI) applied to image processing includes facial recognition, object recognition and detection, video, image action, and visual search. It helps to develop smart applications in digital image processing.

This paper presents an image deblurring method that finds application in a broad scientific field such as image deblurring. A method for image deblurring, based on the pseudoinverse matrix is applied for removal of blur in an image caused by linear motion. This method assumes that linear motion corresponds to an integer number of pixels. Compared to other classical methods, this method attains higher values of the Improvement in Signal to Noise Ratio (ISNR) and the Peak Signal to Noise Ratio (PSNR) parameter. The values for the Mean Square Error (MSE) is lower and computational time has been decreased considerably with respect to the other methods. The presented experimental results are implemented in MATLAB.

An Image Inpainting is the art to reconstruct an image by removing scratch, noise, object or any other defect. Fill the region of missing information from a signal using surrounding information and reform signal is the basic work of Inpainting algorithms. There are various methods to do image Inpainting like exemplar based image Inpainting in which patch of particular size is get selected and that patch is used to fill missing part of an image by calculating highest priority. In this confidence value and data value that is collectively known as priority value of that patch is calculated. And after this most similar patch from the source region is detected and pasted on highest priority patch. Existing technique gives high computation time by searching similar patches in whole source region again and again. Also target region needs to be selected by user. So proposing method can provide better computation efficiency on traditional image Inpainting algorithm for automatic detection and removal of scratch in an image. In this approach whole image clustered using Iterative Threshold Selection algorithm and scratch will be detect automatically. After detecting scratch automatically highest priority patch will be calculated. Due to clustering highest priority patch need not to be search in whole source region again and again, it just need to take most similar patch from this cluster. So due to this source region can be minimized for optimal patch to search. So proposing method can provides better speedy approach for removal of scratch in an image than existing method.

In this report we present a brief outline of our technological approaches to developing a comprehensive imaging platform suitable for the investigation of the dynamics of the hemoglobin signal in large tissue structures using NIRS imaging techniques. Our approach includes a combined hardware and software development effort that provides for i) hardware integration, ii) system calibration, iii) data integrity checks, iv) image recovery, v) image enhancement and vi) signal processing. Presented are representative results obtained from human subjects that explore the sensitivity and other capabilities of the measuring system to detect focal hemodynamic responses in the head, breast and limb of volunteers. Results obtained support the contention that time-series NIRS imaging is a powerful and sensitive technique for exploring the hemodynamics of healthy and diseased tissues.

We develop a system for three-dimensional whole-body optoacoustic tomography of small animals for applications in preclinical research. The tomographic images are obtained while the objects of study ͑phantoms or mice͒ are rotated within a sphere outlined by a concave arc-shaped array of 64 piezocomposite transducers. Two pulsed lasers operating in the near-IR spectral range ͑755 and 1064 nm͒ with an average pulsed energy of about 100 mJ, a repetition rate of 10 Hz, and a pulse duration of 15 to 75 ns are used as optical illumination sources. During the scan, the mouse is illuminated orthogonally to the array with two wide beams of light from a bifurcated fiber bundle. The system is capable of generating images of individual organs and blood vessels through the entire body of a mouse with spatial resolution of ϳ0.5 mm.

Curve evolution schemes for image segmentation based on a region based contour model allowing for junctions, vector-valued images and topology changes are introduced. Together with an a posteriori denoising in the segmented homogeneous regions this leads to a fast and efficient method for image segmentation and restoration. An uneven spread of mesh points is avoided by using the tangential degrees of freedom. Several numerical simulations on artificial test problems and on real images illustrate the performance of the method.

Acquired images for medical and other purposes can be affected by noise from both the equipment used in the capturing or the environment. This can have adverse effect on the information therein. Thus, the need to restore the image to its original state by removing the noise. To effectively remove such noise, pre knowledge of the type of noise model present is necessary. This work explores different noise removal filters by first introducing noise to an image and then applying different spatial domain filtering techniques to the image to get rid of the noise. Different evaluation techniques such as Peak to Signal Noise Ratio(PSNR) and Root Mean Square Error(RMSE) were adopted to determine how effective each filter is on a given image noise. Result showed that some filters are more effective on some noise models than others.

Patch-based prior learning algorithm is capable of delivering state-of-the-art performance in image denoising. The major concern that affects the patch-based restoration algorithms is the accuracy of the patch priors. Our work is based on the extension of the Gaussian Mixture Model (GMM) estimation which is developed to iteratively capture and process data incrementally. In this paper, we introduce an image denoising framework that uses patch prior learned online refines Gaussian mixture model incrementally. This model decides if we can merge two given Gaussians without drifting from the real data distribution. An incrementally learning mixture model helps to reduce the complexity of the model while still giving a precise description of the observations. The results show that the proposed method generally achieves comparable performance to the conventional approach, while producing models at lower training time without any compromise with the model accuracy.

We introduce a new self‐adaptive algorithm for applications to image restoration problems. In order to study an image restoration, we consider the algorithm that contains inertial effects and step sizes, which is independent from the norm of the bounded linear operator. With some control conditions, the strong convergence to the minimum norm solution of the algorithm is obtained. Convergence analysis of the proposed algorithm is also discussed. Moreover, numerical results of image restoration problems illustrate that the proposed algorithm is efficient and outperforms other ones.

The Earth's atmosphere heavily affects the remote sensing images collected by spaceborne passive optical sensors due to radiation-matter interaction phenomena like radiation absorption, scattering, and thermal emission. A complex phenomenon is the adjacency effect, i.e., radiation reflected by the ground that, due to the atmospheric scattering, is being seen in a viewing direction different from that corresponding to the ground location that reflected it. Adjacency gives rise to crosstalk between neighboring picture elements up to a distance that depends on the width of the integral kernel function employed for the mathematical modeling of the problem. As long as the atmosphere is a linear spaceinvariant system, the adjacency can be modeled as a low-pass filter, with the atmospheric point spread function (APSF) applied to the initial image. In this paper, a direct method of estimating the discrete normalized APSF (NAPSF) using images gathered by high-resolution optical sensors is discussed. We discuss the use of the NAPSF estimate for deducing the Correction Spatial high-pass Filter (CSF)-a correction filter that removes the adjacency effect. The NAPSF estimation procedure has been investigated using statistical simulations, whose outcomes permitted us to identify the conditions under which the NAPSF could be measured with acceptable errors. The NAPSF estimation is examined for various natural images acquired by MOMS-2P, CHRIS, AVIRIS, and MIVIS.

Despite the breakthroughs in quality of image enhancement, an end-to-end solution for simultaneous recovery of the finer texture details and sharpness for degraded images with low resolution is still unsolved. Some existing approaches focus on minimizing the pixel-wise reconstruction error which results in a high peak signal-to-noise ratio. The enhanced images fail to provide high-frequency details and are perceptually unsatisfying, i.e., they fail to match the quality expected in a photo-realistic image. In this paper, we present Image Enhancement Generative Adversarial Network (IEGAN), a versatile framework capable of inferring photo-realistic natural images for both artifact removal and super-resolution simultaneously. Moreover, we propose a new loss function consisting of a combination of reconstruction loss, feature loss and an edge loss counterpart. The feature loss helps to push the output image to the natural image manifold and the edge loss preserves the sharpness of the output image. The reconstruction loss provides low-level semantic information to the generator regarding the quality of the generated images compared to the original. Our approach has been experimentally proven to recover photo-realistic textures from heavily compressed low-resolution images on public benchmarks and our proposed high-resolution World100 dataset.

Classification of moving objects in imaging through longdistance atmospheric path may be affected by distortions such as blur and spatiotemporal movements caused by air turbulence. This work aims to study and quantify the effects of these distortions on the ability to classify moving objects in atmospherically degraded video signals. For this purpose, we perform simulations and examine real long-range thermal video cases. In the simulation, we evaluate various geometrical (shape-based) object features for classification at different distortion levels. Furthermore, we examine the influence of image restoration on the classification performances in the real-degraded videos, using geometrical and textural features (combined and in separate) of the objects. Principal component analysis together with both k-nearest neighbor and support vector machines is used for the classification process. Results show how classification performances decrease as the level of blur increases, and how successful digital image restoration for real cases can significantly improve the classification performances.

Context. LINC-NIRVANA, the Fizeau interferometer of the Large Binocular Telescope (LBT), will require routine use of image reconstruction methods for data reduction. To this purpose our group has already developed the software package AIRY (Astronomical Image Restoration in interferometrY). Aims. Observations of a target, with different orientations of the baseline of LINC-NIRVANA, will provide images with different orientations with respect to the CCD camera. This rotation effect was not taken into account in our previous work. Therefore in this paper we propose a method able to compensate for the rotation of the field of view. Moreover we investigate acceleration techniques for reducing the computational burden of multiple image deconvolution. Methods. The basic method is a suitable modification of the Richardson-Lucy algorithm, also implementing an approach we proposed for reducing boundary effects. Acceleration techniques, proposed by Biggs & Andrews, are extended and applied to this new algorithm. Finally a method for estimating the unknown point spread function (PSF) by extracting and extrapolating the image of a reference star is developed and implemented. Results. The method introduced for compensating object rotation and reducing boundary effects, as well as its accelerated versions, are tested on simulated LINC-NIRVANA images, using the VLT image of the Crab Nebula as test object. The results are very promising. Moreover the method for PSFs extraction is tested on simulated images, derived from the LBT image of the galaxy NGC 6946 and obtained by convolving this image with PSFs computed by means of the numerical code LOST (Layer Oriented Simulation Tool).

Images taken in a low light condition with the presence of camera shake suffer from motion blur and photon shot noise. While state-of-the-art image restoration networks show promising results, they are largely limited to well-illuminated scenes and their performance drops significantly when photon shot noise is strong. In this paper, we propose a new blur estimation technique customized for photon-limited conditions. The proposed method employs a gradient-based backpropagation method to estimate the blur kernel. By modeling the blur kernel using a low-dimensional representation with the key points on the motion trajectory, we significantly reduce the search space and improve the regularity of the kernel estimation problem. When plugged into an iterative framework, our novel low-dimensional representation provides improved kernel estimates and hence significantly better deconvolution performance when compared to end-to-end trained neural networks. The source code and pretrained models are available at https://github. com / sanghviyashiitb / structured-kernel-cvpr23

Recent advances in solving imaging inverse problems have witnessed the combination of deep learning models with classical image models for better signal representation. One such approach, DeepRED, combines the deep image prior (DIP) with the regularization by denoising (RED) framework to boost the performance of image deblurring and super resolution tasks. In this paper, we formulate DeepRED as a consensus equilibrium problem and set up a fixedpoint algorithm for solving the equilibrium equations. We also derive sufficient conditions that the DIP generative prior should satisfy to ensure that the corresponding fixed-point operator is nonexpansive. We then demonstrate that the fixed-point algorithm that solves the CE equations results in improved image reconstruction quality in a deblurring setting compared to state-of-the-art methods