Image Deblurring and Super-Resolution by Adaptive Sparse Domain Selection and Adaptive Regularization

By Weisheng Dong, Lei Zhang, Guangming Shi, Xiaolin Wu. 2010

1,389 citations recorded by the source corpus.

As a powerful statistical image modeling technique, sparse representation has been successfully used in various image restoration applications. The success of sparse representation owes to the development of the l1-norm optimization techniques and the fact that natural images are intrinsically sparse in some domains. The image restoration quality largely depends on whether the employed sparse domain can represent well the underlying image. Considering that the contents can vary significantly across different images or different patches in a single image, we propose to learn various sets of bases from a precollected dataset of example image patches, and then, for a given patch to be processed, one set of bases are adaptively selected to characterize the local sparse domain. We further introduce two adaptive regularization terms into the sparse representation framework. First, a set of autoregressive (AR) models are learned from the dataset of example image patches. The best fitted AR models to a given patch are adaptively selected to regularize the image local structures. Second, the image nonlocal self-similarity is introduced as another regularization term. In addition, the sparsity regularization parameter is adaptively estimated for better image restoration performance. Extensive experiments on image deblurring and super-resolution validate that by using adaptive sparse domain selection and adaptive regularization, the proposed method achieves much better results than many state-of-the-art algorithms in terms of both PSNR and visual perception.

Subfields: Computational Mechanics, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Mechanics

In Computer Vision and Pattern Recognition

Citations in the corpus

Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network (2016, neutral): “Related approaches to the SR problem originate in compressed sensing and aim to estimate a sparse patch representation with respect to an over-complete dictionary [54, 9, 59].”
Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network (2016, methodology): “This approach is proposed in the methods of [47, 8].”
Learning a No-Reference Quality Metric for Single-Image Super-Resolution (2016, methodology): “Shan08 [2] Glasner09 [7] Yang13 [12] Dong11 [9] 2.”
The Little Engine That Could: Regularization by Denoising (RED) (2016, neutral): “566 Table 1 provides the restoration performance of the three RED schemes – the steepest567 descent (SD), the ADMM, and the fixed-point (FP) methods – along with the results of the 568 P 3, the state-of-the-art NCSR and IDD-BM3D [27], and two additional baseline deblurring 569 methods [67,68].”
ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting (2023, background): “Traditional image SR methods primarily focus on designing more rational image priors based on our subjective knowledge, such as non-local similarity [26], low-rankness [27], sparsity [28, 29], and so on.”
NTIRE 2017 Challenge on Single Image Super-Resolution: Dataset and Study (2017, background): “Single image super-resolution as well as image restoration research literature spans over decades [36, 20, 4, 13, 16, 3, 15, 14, 6, 32, 54, 30, 17, 23, 12, 47, 48, 10, 21].”
GAN-Based Projector for Faster Recovery With Convergence Guarantees in Linear Inverse Problems (2019, background): “Traditionally, for signal priors, natural images are considered sparse in some fixed or learnable basis [11, 8, 36, 22, 7, 38, 10, 21].”
Single-Image Super-Resolution: A Benchmark (2014, methodology): “We generate the SR images by the originally released code [27,17,46,5,43,36] or our implementation [16,30,11,9] if the code is not available.”
Solving Linear Inverse Problems Using Gan Priors: An Algorithm with Provable Guarantees (2018, background): “Several works (including [22, 23, 24]) employ sparsity priors to solve inverse problems such as denoising, super-resolution and inpainting.”
Sparse Modeling for Image and Vision Processing (2014, methodology): “…inpainting and demosaicking [Mairal et al., 2008c, 2009], super-resolution and deblurring [Yang et al., 2010a, 2012a, Couzinie-Devy et al., 2011, Dong et al., 2011b, Zeyde et al., 2012, Wang et al., 2012], face compression [Bryt and Elad, 2008], or for inverting non-linear local transformations…”
One Network to Solve Them All — Solving Linear Inverse Problems Using Deep Projection Models (2017, background): “The second approach learns a direct mapping from the linear measurement y to the solution x, with the help of large training datasets and deep neural nets.”
Deep Generalized Unfolding Networks for Image Restoration (2022, background): “, [7,12,17,25,32,56]) usually formulate IR as a Bayesian problem, solving Eq.”
Interval-wise testing for functional data (2017, background): “In the very recent literature some examples of domain selection can be found in cluster analysis (Chen et al. 2014), in functional regression with LASSO penalization (Zhao et al. 2012), and thresholded wavelet and sparse representations (Donoho and Johnstone 1994; Dong et al. 2011).”
Super-resolution: a comprehensive survey (2014, neutral): “[498], [503], [505], [517], [557], [571], [572], [576], [578], [579], [583], [595], [597], [601], [602], [604], [608], [614] usually assume that there are two overcomplete dictio-”
Semi-coupled dictionary learning with applications to image super-resolution and photo-sketch synthesis (2012, background): “Recently, sparse coding (or sparse representation) and dictionary learning have proven to be very effective in image reconstruction [20, 9, 6, 5], while the dictionary plays an important role to successfully accomplish such tasks.”