Real-world Blurred Images Research Articles

Blind multi-Poissonian image deconvolution with sparse log-step gradient prior.

Blind image deconvolution plays a very important role in the fields such as astronomical observation and fluorescence microscopy imaging, in which the noise follows Poisson distribution. However, due to the ill-posedness, it is a very challenging task to reach a satisfactory result from a single blurred image especially when the power of the Poisson noise is at a high level. Therefore, in this paper, we try to achieve high-quality restoration results with multi-blurred images which are contaminated by Poisson noise. Firstly, we design a novel sparse log-step gradient prior which adopts a mixture of logarithm and step functions to regularize the image gradients and combine it with the Poisson distribution to formulate the blind multi-image deconvolution problem. Secondly, we incorporate the methods of variable splitting and Lagrange multiplier to convert the original problem into sub-problems, then we alternately solve them to achieve the estimation of all the blur kernels of corresponding blurred images. Besides, we also design a non-blind multi-image deconvolution algorithm which is based on the log-step gradient prior to reach the final restored image. Experimental results on both synthetic and real-world blurred images show that the proposed prior is very capable of suppressing negative artifacts caused by ill-posedness. The algorithm can achieve restored image of very high quality which is competitive with some state-of-the-art methods.

Real Image Deblurring Based on Implicit Degradation Representations and Reblur Estimation

Most existing image deblurring methods are based on the estimation of blur kernels and end-to-end learning of the mapping relationship between blurred and sharp images. However, since different real-world blurred images typically have completely different blurring patterns, the performance of these methods in real image deblurring tasks is limited without explicitly modeling blurring as degradation representations. In this paper, we propose IDR2ENet, which is the Implicit Degradation Representations and Reblur Estimation Network, for real image deblurring. IDR2ENet consists of a degradation estimation process, a reblurring process, and a deblurring process. The degradation estimation process takes the real blurred image as input and outputs the implicit degradation representations estimated on it, which are used as the inputs of both reblurring and deblurring processes to better estimate the features of the blurred image. The experimental results show that whether compared with traditional or deep-learning-based deblurring algorithms, IDR2ENet achieves stable and efficient deblurring results on real blurred images.

An adaptive two phase blind image deconvolution algorithm for an iterative regularization model

This paper proposes a blind image deconvolution method which consists of two sequential phases, i.e., blur kernel estimation and image restoration. In the first phase, we adopt the L0-norm of image gradients and total variation (TV) to regularize the latent image and blur kernel, respectively. Then we design an alternating optimization algorithm which jointly incorporates the estimation of intermediately restored image, blur kernel and regularization parameters into account. In the second phase, we propose to take the mixture of L0-norm of image gradients and TV to regularize the latent image, and design an efficient non-blind deconvolution algorithm to achieve the restored image. Experimental results on both a benchmark image dataset and real-world blurred images show that the proposed method can effectively restore image details while suppress noise and ringing artifacts, the result is of high quality which is competitive with some state of the art methods.

Blind image deconvolution via salient edge selection and mean curvature regularization

Blind image deconvolution is an ill-posed problem since there exists infinite pairs of blur kernels and latent images. To obtain reasonable results of this problem, most previous methods have emphasized the importance of selecting salient edges for blur kernel estimation. In this paper, a blind deconvolution method based on explicit and implicit salient edges selection is proposed. Explicit edge selection is achieved by using mutually guided image filtering, while mean curvature regularization is adopted to remove disadvantageous structures implicitly. Besides, the proposed model can be efficiently optimized by using half-quadratic penalty method and mean curvature filter. Extensive experiments show that the proposed method performs favorably against the state-of-the-art blind deconvolution methods on both synthetic and real-world blurred images.

Gradient-based discriminative modeling for blind image deblurring

Blind image deconvolution is a fundamental task in image processing, computational imaging, and computer vision. It has earned intensive attention in the past decade since the seminal work of Fergus et al. [1] for camera shake removal. In spite of the recent great progress in this field, this paper aims to formulate the blind problem with a simpler modeling perspective. What is more important, the newly proposed approach is expected to achieve comparable or even better performance towards the real blurred images. Specifically, the core critical idea is the proposal of a pure gradient-based discriminative prior for accurate and robust blur kernel estimation. Numerous experimental results on both the benchmark datasets and real-world blurred images in various imaging scenarios, e.g., natural, manmade, low-illumination, text, or people, demonstrate well the effectiveness and robustness of the proposed approach.

Image deblurring using tri-segment intensity prior

Camera shake during exposure often introduces annoying blur of objects and deteriorates image quality. Existing image deblurring algorithms usually use intensity and gradient priors to alleviate the degree of blurring. However, these methods only consider the changes caused by the blur process in the low intensity range, omitting the changes caused by the blur process in the high and middle part of the intensity range. In this paper, we propose an effective blind image deblurring algorithm based on the three segments of intensity prior, i.e., low, middle and high parts. This work is motivated by the observation that the blur process destroys the sparsity of both ends of intensity, and meanwhile shrinks the distance between the two distinct gray levels. A fast numerical scheme is deployed for alternatingly computing the sharp image and the blur kernel using an image pyramid at the stage of kernel estimation. Extensive experiments on both synthetic and real-world blurred images demonstrate that our method performs favorably against the state-of-the-art image deblurring methods.

Scale-Iterative Upscaling Network for Image Deblurring

Machine learning based methods for blind deblurring are efficient to handle real-world blurred images, whose blur may be caused by various combined distortions. However, existing multi-level architectures fail to fit images of various scenarios. In this paper, we propose a scale-iterative upscaling network (SIUN) that restores sharp images in an iterative manner. It is not only able to preserve the advantages of weights sharing across scales but also more flexible when training and predicting with different iterations to fit different images. Specifically, we bring in the super-resolution structure instead of the upsampling layer between two consecutive scales to restore a detailed image. Besides, we explore different curriculum learning strategies for both training and prediction of the network and introduce a widely applicable strategy to make SIUN compatible with different scenarios, including text and face. Experimental results on both benchmark datasets and real blurred images show that our method can produce better results than state-of-the-art methods. Code is available at https://github.com/minyuanye/SIUN.

Adapting total generalized variation for blind image restoration

In this paper, a fast blind deconvolution approach is proposed for image deblurring by modifying a recent well-known natural image model, i.e., the total generalized variation (TGV). As a generalization of total variation, TGV aims at reconstructing a higher-quality image with high-order smoothness as well as sharp edge structures. However, when it turns to the blind issue, as demonstrated either empirically or theoretically by several previous blind deblurring works, natural image models including TGV actually prefer the blurred images rather than their counterpart sharp ones. Inspired by the discovery, a simple, yet effective modification strategy is applied to the second-order TGV, resulting in a novel L0–L1-norm-based image regularization adaptable to the blind deblurring problem. Then, a fast numerical scheme is deduced with O(NlogN) complexity for alternatingly estimating the intermediate sharp images and blur kernels via coupling operator splitting, augmented Lagrangian and also fast Fourier transform. Experiment results on a benchmark dataset and real-world blurred images demonstrate the superiority or comparable performance of the proposed approach to state-of-the-art ones, in terms of both deblurring quality and speed. Another contribution in this paper is the application of the newly proposed image prior to single image nonparametric blind super-resolution, which is a fairly more challenging inverse imaging task than blind deblurring. In spite of that, we have shown that both blind deblurring and blind super-resolution (SR) can be formulated into a common regularization framework. Experimental results demonstrate well the feasibility and effectiveness of the proposed blind SR approach, and also its advantage over the recent method by Michaeli and Irani in terms of estimation accuracy.

Boosting normalized sparsity regularization for blind image deconvolution

We focus on blind image deconvolution, which has attracted intensive attentions since Fergus et al.’s influential work in 2006. Among the current literature, the daring idea of imposing the normalized sparsity measure on blind image deblurring is a recent spotlight, which is, nevertheless, far from practical use in terms of estimating accuracy, efficiency, as well as robustness. To boost its performance, we propose a novel method via coupling the normalized sparsity measure with the total generalized variation, which, however, does not fit the blind deblurring problem. By use of operator splitting and alternating direction method of multipliers, a numerical scheme is derived, leading to a more accurate, efficient, and robust blind deblurring algorithm. Numerous blind deblurring results on both benchmark data and real-world blurred images demonstrate the competitive or even better performance compared against the state of the art.

Non-blind image deconvolution using natural image gradient prior

Since non-blind image deconvolution is inherently ill-posed, the results of unregularized methods are often contaminated by noise and ringing artifacts. To reach a stable solution, we adopt the natural image gradient prior to regularize the latent image and obtain an improved version of the Richardson–Lucy (RL) algorithm. We use both synthetic and real world blurred images to test the proposed method. Experimental results show that the negative artifacts are significantly suppressed and the restored images are of high quality.

Blind image deconvolution using the Fields of Experts prior

In this paper, we present a method for single image blind deconvolution. To improve its ill-posedness, we formulate the problem under Bayesian probabilistic framework and use a prior named Fields of Experts (FoE) which is learnt from natural images to regularize the latent image. Furthermore, due to the sparse distribution of the point spread function (PSF), we adopt a Student-t prior to regularize it. An improved alternating minimization (AM) approach is proposed to solve the resulted optimization problem. Experiments on both synthetic and real world blurred images show that the proposed method can achieve results of high quality.

Restoration of Images Degraded by Linear Motion Blurred Objects in Still Background

A blur restoration scheme for images with linear motion blurred objects in still background is proposed. The proposed scheme starts from a rough detection of locations of blurred objects. This rough segmentation of the blurred regions is based on an analysis of local orientation map. Then, parameters of blur are identified based on a linear constant-velocity motion blur model for every detected blurred region. After the blur parameters are estimated, the locations of blurred objects can be refined before going to a restoration process by using information from the identified blur parameters. Blur locations are refined by observing local power of the blurred image which is filtered by a high-pass filter. The high-pass filter has approximately a frequency characteristic that is complementary to the identified blur point spread function. As a final step, the image is restored by using the estimated blur parameters and locations based on an iterative deconvolution scheme applied with a regularization concept. Experimental examples of simulated and real world blurred images are demonstrated to confirm the performance of the proposed scheme.

Multiple-Image-Based Restoration for Motion Blur with Non-uniform Point Spread Function

A blind image restoration for non-linear motion blurs with non-uniform point spread functions based on multiple blurred versions of a same scene is proposed. The restoration is separately considered as identification and deconvolution problems. In the proposed identification process, an identification difficulty is introduced to rank an order of blur identification. A blurred image with the lowest identification difficulty is initially identified by using a single-image-based scheme. Then, other images are identified based on a cross convolution relation between each pair of blurred images. In addition, an iterative feedback scheme is applied to improve the identification results. For the deconvolution process, a spatial adaptive scheme using regional optimal terminating points is modified from a conventional iterative deconvolution scheme. The images are decomposed into sub-regions based on smoothness. The regional optimal terminating points are independently assigned to suppress a noise in smooth regions and sharpen the image in edgy regions. The optimal terminating point for each region is decided by considering a discrepancy error. Restoration examples of simulated and real world blurred images are experimented to demonstrate the performance of the proposed method.

Identification of Piecewise Linear Uniform Motion Blur

A motion blur identification scheme is proposed for non-linear uniform motion blurs approximated by piecewise linear models which consist of more than one linear motion component. The proposed scheme includes three modules that are a motion direction estimator, a motion length estimator and a motion combination selector. In order to identify the motion directions, the proposed scheme is based on a trial restoration by using directional forward ramp motion blurs along different directions and an analysis of directional information via frequency domain by using a Radon transform. Autocorrelation functions of image derivatives along several directions are employed for estimation of the motion lengths. A proper motion combination is identified by analyzing local autocorrelation functions of non-flat component of trial restored results. Experimental examples of simulated and real world blurred images are given to demonstrate a promising performance of the proposed scheme.