Text image super resolution using within-scale repetition of characters and strokes

In text images, there are some frequently used characters repeating more than others. Likewise, some characters have common strokes. This characteristic is used in this paper for machine-printed text-image super resolution. After segmenting the input low-resolution image into text lines and characters, 1) the characters are clustered and the clusters with large number of members, corresponding to the frequent characters, are detected. 2) A text-specific multiple-image super resolution is applied to the members of each large cluster and the result is verified by the recognition confidence of an OCR system. 3) A training example set is then constructed by extracting patches from the low-resolution frequent characters and their verified super resolution. Using this example set, infrequent characters are super resolved through the neighbor embedding SR algorithm. By placing all the super-resolved characters on their corresponding positions in the high-resolution grid, the final high-resolution image is generated. Our method achieves significant improvements in visual image quality and OCR character accuracy compared to related SR methods.     

A novel image Denoising approach using super resolution densely connected convolutional networks

Multimedia Tools and Applications - Image distortion effects, called noise, may occur due to various reasons such as image acquisition, transfer, and duplication. Image denoising is a preliminary...     

Single image super resolution based on sparse representation using discrete wavelet transform

Multimedia Tools and Applications - Single image super resolution (SR) based on sparse representation is a promising technique where the SR problem is solved by searching for the most robust...     

Texture image prior for SAR image super resolution based on total variation regularization using split Bregman iteration

In this article, we propose a total variation (TV) regularization approach for the reconstruction of super-resolution synthetic aperture radar (SAR) image based on gradient profile prior or other texture image prior in the maximum a posteriori framework. We also design a novel super-resolution reconstruction algorithm via split Bregman iteration with the known degradation matrix, thereby enhancing the resolution of the SAR image. The parameter adaptation of the TV regularization is performed based on the high-resolution (HR) SAR image at each step. Several evaluation indices are tested on SAR images for objective assessment of the performance of SAR image super-resolution reconstruction. This computationally efficient algorithm is robust to noise in SAR scenes in HR image estimation. Experimental results show that the proposed split Bregman super-resolution approach can effectively avoid the speckle noise generated due to some strange textures and has good effect of noise suppression, while effectively maintaining the SAR image content, the structure of the SAR image is more apparent. Additionally, the experimental results on real SAR scenes also demonstrate the effectiveness of the proposed algorithm and demonstrate its superiority to other super-resolution algorithms.     

An edge preserving IBP based super resolution image reconstruction using P-spline and MuCSO-QPSO algorithm

Super resolution (SR) reconstruction based on iterative back projection (IBP) is a widely used image reconstruction method. IBP approach is easy to implement and allows easy inclusion of the spatial domain with low computational complexity. However, local minima trapping; slow rate of convergence; sensitive to the initial guess; prone to ringing and jaggy artifacts are some major bottlenecks which restrict its performance. The present paper aims to enhance the performance of IBP based SR reconstruction (IBP-SRR) of image by exploring an effective method. The proposed method has fast convergence rate, a global optimal solution, capability to lessen the effect of artifacts and a noble generalization performance. In the present work, P-spline interpolation scheme imposes additional penalty in the inherently smooth B-spline interpolation process to provide a proper initial guess. An adaptive edge regularization technique is used in the constraint optimization of the reconstruction problem to minimize the effect of ringing artifacts. Finally, the overall reconstruction error of the reconstruction system is optimized using a hybrid meta-heuristic optimization technique. The optimization algorithm hybridizes the notion of Cuckoo search optimization (CSO) algorithm with a mutation operator (MuCSO) and the quantum behaved particle swarm optimization (QPSO). The MuCSO-QPSO algorithm is compared with other significant optimization algorithms such as GA, PSO, QPSO, CSO, MuCSO and found to be outperforming. Experimental results demonstrate the superiority of the proposed edge preserving IBP-SRR method in terms of enhanced spatial resolution, and more detail reconstruction.

     

Blind single image super resolution with low computational complexity

This paper proposes a single image super resolution algorithm with the aim of satisfying three desirable characteristics, namely, high quality of the produced images, adaptability to image contents and unknown blurring conditions used to generate given input images, and low computational complexity. After the given input image is up-scaled using a conventional reconstruction operator, the missing high frequency components estimated from lower resolution versions of the input image are added for improved quality and, moreover, the amount of the high frequency components to be added is adaptively determined. No computationally intensive operation is involved in the whole process, which makes the method computationally cheap. Experimental results show that the proposed method yields good subjective and objective image quality consistently across different blurring conditions and contents, and operates fast in comparison to existing state-of-the-art algorithms. In addition, it is also demonstrated that the proposed method can be used in combination with the existing algorithms in order to improve further their performance in terms of image quality.

     

一种自适应视频图像超分辨率重构算法

依据图像局部二阶统计量能够反映图像区域变化的特性,以参考帧的局部方差为参数建立凸集投影算子和区域修复阈值的条件,提出了一种自适应选取运动估计误差阈值和图像修复误差阈值的视频图像超分辨率重构方法.实验结果表明,与传统方法相比,该算法重构图像的主观质量得到了明显增强,提高了峰值信噪比.     

基于多种正则化的改进超分辨率重建算法

为了解决超分辨率图像重建过程中无法同时降低平滑区域噪声和保持图像细节的问题,结合改进的非局部变分（NLTV）和全变分（TV）正则项方法提出一种新的超分辨率重建算法。首先,根据图像重尾分布特性,结合高斯分布、拉普拉斯分布及柯西分布改进了传统NLTV正则项系数,提出了改进的ANLTV正则项。然后利用ANLTV正则项基于分裂Bregman算法重建了初始的高分辨率图像。最后结合TV正则项对重建的高分辨率图像进行去模糊操作,进而得到最终的超分辨率图像重建结果。为验证所提算法的性能,分别利用该算法与传统的TV和NLTV算法进行超分辨率图像重建并对比。实验结果表明,所提出的方法相比于传统的TV和NLTV重建算法,其峰值信噪比、信噪比和结构相似度均有所提高,能够同时满足超分辨率图像重建过程中抑制噪声和保持边缘细节的需求。     

Single image super resolution for texture images through neighbor embedding

This article proposes an improved learning based super resolution scheme using manifold learning for texture images. Pseudo Zernike moment (PZM) has been employed to extract features from the texture images. In order to efficiently retrieve similar patches from the training patches, feature similarity index matrix (FSIM) has been used. Subsequently, for reconstruction of the high resolution (HR) patch, a collaborative optimal weight is generated from the least square (LS) and non-negative matrix factorization (NMF) methods. The proposed method is tested on some color texture, gray texture, and some standard images. Results of the proposed method on texture images advocate its superior performance over established state-of-the-art methods.

     

基于深度学习的单图像超分辨率重建研究综述

为深入了解基于深度学习的单图像超分辨率重建（SISR）的发展,把握当前研究的热点和方向,针对现有基于深度学习的单图像超分辨率重建模型进行了梳理。介绍了相关深度学习算法和基于深度学习的模型以及评价指标,并通过实验对比分析现有模型的性能,其目的在于从本质上了解基于深度学习的单图像超分辨率重建模型的优势;对单图像超分辨率重建的关键问题进行了总结,并对未来的发展趋势进行了展望。     

Learning-based super resolution using kernel partial least squares

In this paper, we propose a learning-based super resolution approach consisting of two steps. The first step uses the kernel partial least squares (KPLS) method to implement the regression between the low-resolution (LR) and high-resolution (HR) images in the training set. With the built KPLS regression model, a primitive super-resolved image can be obtained. However, this primitive HR image loses some detailed information and does not guarantee the compatibility with the LR one. Therefore, the second step compensates the primitive HR image with a residual HR image, which is the subtraction of the original and primitive HR images. Similarly, the residual LR image is obtained from the down-sampled version of the primitive HR and original LR image. The relation of the residual LR and HR images is again modeled with KPLS. Integration of the primitive and the residual HR image will achieve the final super-resolved image. The experiments with face, vehicle plate, and natural scene images demonstrate the effectiveness of the proposed approach in terms of visual quality and selected image quality metrics.     

Bayesian combination of sparse and non-sparse priors in image super resolution

In this paper the application of image prior combinations to the Bayesian Super Resolution (SR) image registration and reconstruction problem is studied. Two sparse image priors, a Total Variation (TV) prior and a prior based on the ?1 norm of horizontal and vertical first-order differences (f.o.d.), are combined with a non-sparse Simultaneous Auto Regressive (SAR) prior. Since, for a given observation model, each prior produces a different posterior distribution of the underlying High Resolution (HR) image, the use of variational approximation will produce as many posterior approximations as priors we want to combine. A unique approximation is obtained here by finding the distribution on the HR image given the observations that minimizes a linear convex combination of Kullback–Leibler (KL) divergences. We find this distribution in closed form. The estimated HR images are compared with the ones obtained by other SR reconstruction methods.     

引入超分辨率下采样误差的图像边信息估计隐写

目的由于空域图像下采样过程中提供的量化误差边信息能够有效提升隐写安全性,为了得到下采样之前的高分辨率图像,提出一种基于超分辨率网络的空域图像边信息估计隐写方法。方法受原始下采样边信息隐写方法的启发,使用超分辨率网络生成被称为预载体的高分辨率图像。同时利用现有的空域图像对称失真算法得到每个像素点的修改失真,然后以浮点型精度对预载体下采样,得到和载体同分辨率的图像形式,利用对应像素点间的差值指导像素点的修改方向,实现基于初始失真的非对称失真调整。首先以峰值信噪比和极性估计准确率为指标对比了多种超分辨率网络以及基于传统插值方法的上采样性能,并通过调整初始失真分别进行隐写和隐写分析实验,选择使安全性提升最大的残差通道注意力机制网络及其对应调整系数作为本文的下采样边信息估计隐写方法。结果使用隐写领域中常用的3个数据库、两种传统初始失真函数以及两类隐写分析方法进行实验。在跨数据集的隐写安全性上,相比传统隐写方法,在对抗基于手工特征和基于深度学习的隐写分析时,本文方法的安全性均有显著提升,如在测试集载体图像上,嵌入率为0.5 bit/像素时,安全性分别提升6.67%和6.9%;在训练集载体图像上,本...     

基于残差神经网络的图像超分辨率改进算法

为更有效地提升图像的超分辨率（SR）效果,提出了一种多阶段级联残差卷积神经网络模型。首先,该模型采用了两阶段超分辨率图像重建方法先重建2倍超分辨率图像,再重建4倍超分辨率图像;其次,第一阶段与第二阶段皆使用残差层和跳层结构预测出高分辨率空间的纹理信息,由反卷积层分别重建出2倍与4倍大小的超分辨率图像;最后,以两阶段的结果分别构建多任务损失函数,利用第一阶段的损失指导第二阶段的损失,从而提高网络的训练速度,加强网络学习中的监督指导。实验结果表明,与bilinear算法、bicubic算法、基于卷积神经网络的图像超分辨率（SRCNN）算法和加速的超分辨率卷积神经网络（FSRCNN）算法相比,所提模型能更好地重建出图像的细节和纹理,避免了经过迭代之后造成的图像过度平滑,获得更高的峰值信噪比（PSNR）和平均结构相似度（MSSIM）。     

Frequency separation-based multi-scale cascading residual block network for image super resolution

Deep Convolutional Neural Network (CNN) has recently obtained remarkable achievements in single image super-resolution (SISR). Whereas, these existing methods are usually associated with abundant parameters or computational complexity, which highly limits the real-time application. To solve this problem, we propose a lightweight network named FSCRNet. In general, the proposed network consists of three parts: division schema, feature extraction block, and reconstruction block. Specifically, we decouple the image into two parts: content features and detail features, and then perform different operations separately. Concretely, for detailed features, by combining multi-scale strategy and cascading residual block (MSCRB), the model can explore features and propagate messages efficiently. Also, we introduce channel attention to enhance high-frequency feature representation ability. We use a content feature module (CFM) for content features, consisting of asymmetric convolutions to fetch the tensor elements from the horizontal and vertical directions. We demonstrate that the proposed method with few parameters performs favorably on the benchmarks in quantitative and qualitative results.

     

结合结构自相似性和卷积网络的单幅图像超分辨率

针对单幅图像超分辨率（SR）复原病态逆问题,在重建过程边缘细节丢失导致的模糊,提出一种结合结构自相似和卷积网络的单幅图像超分辨率算法。首先,通过将尺度分解获得待重构图片样本的自身结构相似性,结合外部数据库样本结合作为训练样本,可以解决样本过于分散的问题;其次,将样本输入卷积神经网络（CNN）进行训练学习,得到单幅图像超分辨率的先验知识;然后,利用非局部约束项自适应选择最优字典重建图像;最后,利用迭代反投影算法对图像超分辨率效果进一步提升。实验结果表明,与双三次插值（Bicubic）方法、K-SVD算法和基于卷积神经网络的图像超分辨率（SRCNN）方法等优秀算法相比,所提算法可以得到边缘更为清晰的超分辨率重建效果。     

Improving image resolution using subpixel motion

Given a low resolution camera, we would like to get an image with improved resolution using camera motion. From several low resolution images, with subpixel relative displacements, we get an improved resolution image. We find a high resolution image such that when simulating the imaging process we get low resolution images closest to the observed images. The method is tested on computer simulations, and simulated annealing is also used for the optimization process.     

基于结构自相似性和形变块特征的单幅图像超分辨率算法

针对单幅图像超分辨率（SR）复原样本资源不足和抗噪性差的问题，提出一种基于结构自相似和形变块特征的单幅图像超分辨率算法。首先，该方法通过构建尺度模型，尽可能地扩展搜索空间，克服单幅图像超分辨率训练样本不足的缺陷；接着，通过样例块的几何形变提升了局限性的内部字典大小；最后，为了提升重建图片的抗噪性，利用组稀疏学习字典来重建图像。实验结果表明：与Bicubic、稀疏字典学习（ScSR）算法和基于卷积神经网络的超分辨率（SRCNN）等优秀字典学习算法相比，所提算法可以得到主观视觉效果更为清晰和客观评价更高的超分辨率图像，峰值信噪比（PSNR）平均约提升了0.35 dB。另外所提算法通过几何形变的方式扩展了字典规模和搜索的准确性，在算法时间消耗上平均约减少了80 s。     

Criterion for optimal image resolution using SIFT

Image processing is used to check products in many factories. If we use down-sampled images, we can reduce the calculation time and the image noise. However, the accuracy of the detection also becomes low. The purpose of this article is to estimate the optimal image resolution for detection while keeping the accuracy of detection high. To achieve our purpose, we adopt the scale invariant feature transform (SIFT) as the criterion of the optimal image resolution. Finally, we confirm that our proposed method is useful with a simulation.     

Image super resolution using distributed locality sensitive hashing for manifold learning

In this paper we propose a distributed locality sensitive hashing based framework for image super resolution exploiting computational and storage efficiency of cloud. Now days huge multimedia data is available on the cloud which can be utilized using store anywhere and excess anywhere model. It may be noted that super resolution is required for consumer electronics display devices due to various reasons. The propose framework exploits the image correlation for image super resolution using locality sensitive hashing (LSH) for manifold learning. In our work we have exploited the benefits of manifold learning for image super resolution, which in-turn is a highly time complex operation. The time complexity is involved due to finding the approximate nearest neighbors from trillion of image patches for locally linear embedding (LLE) operation. In our approach it is mitigated by using a distributed framework which internally uses hash tables for mapping of patches in the target image from a database of internet picture collection. The proposed framework for super resolution provides promising results in comparison to existing approaches.