Example-based image super-resolution with class-specific predictors

Example-based super-resolution is a promising approach to solving the image super-resolution problem. However, the learning process can be slow and prediction can be inaccurate. In this paper, we present a novel learning-based algorithm for image super-resolution to improve the computational speed and prediction accuracy. Our new method classifies image patches into several classes, for each class, a class-specific predictor is designed. A class-specific predictor takes a low-resolution image patch as input and predicts a corresponding high-resolution patch as output. The performances of the class-specific predictors are evaluated using different datasets formed by face images and natural-scene images. We present experimental results which demonstrate that the new method provides improved performances over existing methods.     

Accurate image registration for MAP image super-resolution

The accuracy of image registration plays a dominant role in image super-resolution methods and in the related literature, landmark-based registration methods have gained increasing acceptance in this framework. In this work, we take advantage of a maximum a posteriori (MAP) scheme for image super-resolution in conjunction with the maximization of mutual information to improve image registration for super-resolution imaging. Local as well as global motion in the low-resolution images is considered. The overall scheme consists of two steps. At first, the low-resolution images are registered by establishing correspondences between image features. The second step is to fine-tune the registration parameters along with the high-resolution image estimation, using the maximization of mutual information criterion. Quantitative and qualitative results are reported indicating the effectiveness of the proposed scheme, which is evaluated with different image features and MAP image super-resolution computation methods.     

遥感图像的MAP超分辨重建

遥感根本目的就是获得清晰的高空间分辨率的图像,从而可以进一步地分析处理。为了在遥感测量中获得更高空间分辨率、更高信噪比、更清晰的图像,本文对图像处理领域超分辨算法进行了研究。建立了一套拟合模拟现实的成像系统模型,在这种模型的基础之上,利用最大后验概率系统理论,讨论了现实情况中的运动模糊,噪声等情况,改进了MAP超分辨算法。实验结果表明:使用本文改进的基于MAP理论的Markov随机场约束的多帧超分辨重建算法,可以较好提高超分辨效果,与三次立方插值方法相比,PSNR至少提高约5.1dB左右,与未改进的MAP方法相比,PSNR提高约0.2dB左右。本文提出了动态的先验约束方法,给约束函数添加与迭代次数相关的约束项,该改进创新可以加快收敛并且更加逼近真实图像,实验表明该方法收敛速度更快,约束效果良好,更适合实际应用。     

An improved regularization method for artifact rejection in image super-resolution

The main aim of this paper is to employ an improved regularization method to super-resolution problems. Super-resolution refers to a process that increases spatial resolution by fusing information from a sequence of images acquired in one or more of several possible ways. This process is an inverse problem, one that is known to be highly ill-conditioned. Total Variation regularization is one of the well-known techniques used to deal with such problems, which has some disadvantages like staircase effect artifacts and nonphysical dissipation. To enhance the robustness of processing against these artifacts, this paper proposes a new regularization method based on the coupling of fourth order PDE and a type of newly designed shock filtering based on complex diffusion in addition to previous Total Variation. In order to have sharp corner structures like edges, this work also considers a corner shock filter. The proposed scheme is not only able to remove the jittering effect artifacts along the edge directions but also able to restrain the rounding artifacts around the corner structures and most importantly, the stabilization of the overall process is assured.     

Learning knowledge representation with meta knowledge distillation for single image super-resolution

Although the deep CNN-based super-resolution methods have achieved outstanding performance, their memory cost and computational complexity severely limit their practical employment. Knowledge distillation (KD), which can efficiently transfer knowledge from a cumbersome network (teacher) to a compact network (student), has demonstrated its advantages in some computer vision applications. The representation of knowledge is vital for knowledge transferring and student learning, which is generally defined in hand-crafted manners or uses the intermediate features directly. In this paper, we propose a model-agnostic meta knowledge distillation method under the teacher–student architecture for the single image super-resolution task. It provides a more flexible and accurate way to help teachers transmit knowledge in accordance with the abilities of students via knowledge representation networks (KRNets) with learnable parameters. Specifically, the texture-aware dynamic kernels are generated from local information to decompose the distillation problem into texture-wise supervision for further promoting the recovery quality of high-frequency details. In addition, the KRNets are optimized in a meta-learning manner to ensure the knowledge transferring and the student learning are beneficial to improving the reconstructed quality of the student. Experiments conducted on various single image super-resolution datasets demonstrate that our proposed method outperforms existing defined knowledge representation-related distillation methods and can help super-resolution algorithms achieve better reconstruction quality without introducing any extra inference complexity.     

Solving a two-colour problem by applying probabilistic approach to a full-colour multi-frame image super-resolution

Reconstruction based algorithms play an important role in the multi-frame super-resolution problem. A group of images of the same scene are fused together to produce an image with higher spatial resolution, or with more visible details in the high spatial frequency features. Demosaicing algorithms interpolate missing pixels in a raw image taken from one Charged Coupled Device (CCD) array, upsampling the number of the pixels present in the image. Since super-resolution (SR) and demosaicing are the two faces of the same problem it is natural to address them together. In this paper it is: (i) shown that correct modelling of the Bayer pattern in the generative process improves the super-resolution performance for colour images, and (ii) an algorithm that incorporates the two colour prior into the probabilistic model is designed. The algorithm presented in this paper focuses on the classes of images that have two dominant colours, i.e. most of the areas in the image are uniformly coloured. A convex optimization procedure for joint super-resolution and demosaicing is developed which outperforms state-of-the-art algorithms.     

一种改进的退化文本图像二值化方法

文本图像二值化是文本图像识别的重要步骤,由于光照不均或文档水渍等原因导致文本图像退化,增加了文本图像识别的难度。本文对一种局部阈值算法进行了改进,首先对图像进行水平投影,根据直方图的极小点对版面进行简单划分,再利用全局阈值法估算出更为准确的各区域字符笔画宽度,从而自适应地得到适当的窗口尺寸,再利用对比图和局部阈值进行图像二值化,并结合OTSU图像消除原算法产生的伪轮廓。实验与分析表明,改进后的方法能够明显消除因笔画粗细不均、字符大小不同而产生的前景像素误识问题。     

Curvature preserving image super-resolution with gradient-consistency-anisotropic-regularization prior

Single image super-resolution (SR) often suffers from annoying interpolation artifacts such as blur, jagged edges, and ringing. In this paper, we aim to achieve artifact-free SR reconstruction from an input low resolution (LR) image using adaptive de-convolution and curvature refinement. To achieve this, we propose a curvature preserving image SR method based on a gradient-consistency-anisotropic-regularization (GCAR) prior. The gradient consistency term effectively suppresses visual artifacts such as ringing and preserves sharp edges in images while the anisotropic regularization term adaptively preserves the high frequency information according to the gradient magnitude. The complementary two terms are elaborately combined into the GCAR prior for the SR reconstruction. The GCAR prior is very effective in preserving image details and recovering high frequency information. Moreover, we use curvature refinement to remove jagged artifacts caused by aliasing due to decimation. The proposed method employs an effective feedback-control loop which contains adaptive de-convolution, re-convolution, pixel substitution, and curvature refinement. The GCAR prior is utilized in the adaptive de-convolution step. Extensive experiments on various test images demonstrate that the proposed method produces natural-looking and artifact-free SR results in terms of both visual quality and quantitative performance.     

Spatiotemporal approach for time-varying global image motionestimation

Image motion estimation using the spatiotemporal approach has largely relied on the constant velocity assumption, and thus becomes inappropriate when the velocity of the imaged scene or the camera changes during the data acquisition time. Using a polynomial or a trigonometric polynomial model for the time variation of the image motion, spatiotemporal algorithms are developed in this paper to handle time-varying (but space-invariant) motion. Under these models, it is shown that time-varying image motion estimation is equivalent to parameter estimation of one-dimensional (1-D) polynomial phase or phase-modulated signals, which allows one to exploit well-established results in radar signal processing. When compared with alternative approaches, the resulting motion estimation algorithms produce more accurate estimates. Simulation results are provided to demonstrate the proposed schemes.     

生成式对抗神经网络的多帧红外图像超分辨率重建

现有的红外图像超分辨率重建方法主要依赖实验数据进行设计,但在面对真实环境中的复杂退化情况时,它们往往无法稳定地表现。针对这一挑战,本文提出了一种基于深度学习的新颖方法,专门针对真实场景下的红外图像超分辨率重建,构建了一个模拟真实场景下红外图像退化的模型,并提出了一个融合通道注意力与密集连接的网络结构。该结构旨在增强特征提取和图像重建能力,从而有效地提升真实场景下低分辨率红外图像的空间分辨率。通过一系列消融实验和与现有超分辨率方法的对比实验,本文方法展现了其在真实场景下红外图像处理中的有效性和优越性。实验结果显示,本文方法能够生成更锐利的边缘,并有效地消除噪声和模糊,从而显著提高图像的视觉质量。     

从单幅图像恢复立体景象的新方法

在研究单幅平面图像内在特性的基础上,提出了一种恢复立体视觉景象建模的新方法。对图像进行智能识别处理,可以求得许多线段的特征参数,并由此计算出消隐点和消隐线,从而可自动获得场景的立体结构信息。本算法的特点在于用一个代数表达式统一了三种典型的度量方法,无需传统的相机内校正参数,直接可计算出建模用立体信息。建模结果用VRML格式保存、输出,以便于网上浏览。众多的图像验证了该方法的有效性、适用性。     

DPA: a deterministic approach to the MAP problem

Deterministic pseudo-annealing (DPA) is a new deterministic optimization method for finding the maximum a posteriori (MAP) labeling in a Markov random field, in which the probability of a tentative labeling is extended to a merit function on continuous labelings. This function is made convex by changing its definition domain. This unambiguous maximization problem is solved, and the solution is followed down to the original domain, yielding a good, if suboptimal, solution to the original labeling assignment problem. The performance of DPA is analyzed on randomly weighted graphs.     

Hyperspectral image super-resolution combining with deep learning and spectral unmixing

In recent years, hyperspectral image super-resolution has attracted the attention of many researchers and has become a hot topic in the field of computer vision. However, it is difficult to obtain high-resolution images due to imaging hardware devices. At present, many existing hyperspectral image super-resolution methods have not achieved good results. In this paper, we propose a hyperspectral image super-resolution method combining with deep residual convolutional neural network (DRCNN) and spectral unmixing. Firstly, the spatial resolution of the image is enhanced by learning a priori knowledge of natural images. The DRCNN reconstructs high spatial resolution hyperspectral images by concatenating multiple residual blocks, each containing two convolutional layers. Secondly, the spectral features of low-resolution and high-resolution hyperspectral images are linked by spectral unmixing. This approach aims to obtain the endmember matrix and the abundance matrix. The final reconstruction result is obtained by multiplying the endmember matrix and the abundance matrix. In addition, in order to improve the visual effect of the reconstructed image, the total variation regularity is used to impose constraints on the abundance matrix to enhance the relationship between the pixels. The experimental results of remote sensing data based on ground facts show that the proposed method has good performance and preserves spatial information and spectral information without the need for auxiliary images.     

A low-complexity image compression approach with single spatial prediction mode and transform

The well-known low-complexity JPEG and the newer JPEG-XR systems are based on block-based transform and simple transform-domain coefficient prediction algorithms. Higher complexity image compression algorithms, obtainable from intra-frame coding tools of video coders H.264 or HEVC, are based on multiple block-based spatial-domain prediction modes and transforms. This paper explores an alternative low-complexity image compression approach based on a single spatial-domain prediction mode and transform, which are designed based on a global image model. In our experiments, the proposed single-mode approach uses an average 20.5 % lower bit-rate than a standard low-complexity single-mode image coder that uses only conventional DC spatial prediction and 2-D DCT. It also does not suffer from blocking effects at low bit-rates.     

MRANet: Multi-atrous residual attention Network for stereo image super-resolution

In recent years, stereo cameras have been widely used in various fields. Due to the limited resolution of real equipments, stereo image super-resolution (SR) is a very important and hot topic. Recent studies have shown that deep network structures can directly affect feature expression and extraction and thus influence the final results. In this paper, we propose a multi-atrous residual attention stereo super-resolution network (MRANet) with parallax extraction and strong discriminative ability. Specifically, we propose a multi-scale atrous residual attention (MARA) block to obtain receptive fields of different scales through a multi-scale atrous convolution and then combine them with attention mechanisms to extract more diverse and meaningful information. Moreover, we propose a stereo feature fusion unit for stereo parallax extraction and single viewpoint feature refinement and integration. Experiments on benchmark datasets show that MRANet achieves state-of-the-art performance in terms of quantitative metrics and visual quality compared with several SR methods.     

Single sideband FSS with high image rejection

The design of a quasi-optical single sideband filter, which provides more than 30 dB of isolation between the frequency bands 294-305.5 and 329.5-341.5 GHz in the TM plane at 45deg incidence, is described. The structure, which consists of three free-standing arrays of dipole slot elements, generates a bandpass spectral response with an insertion loss below 0.5 dB at resonance. Simulated and measured transmission coefficients in the range 250-400 GHz are shown to be in good agreement     

Anisotropic nonlinear diffusion approach to single image resolution enhancement

In many image resolution enhancement applications, the blurring process of the imaging system is unknown. This paper discusses the problem of single image blind resolution enhancement without estimating the point spread function （PSF）. A regularization model is constructed for image enhancement based on the prior informaton of image error and image gray value,which does not need any prior in- formation of PSF. Moreover, through the solution of Euler equations, an an isotropic nonlinear diffusion equation are obtained,which can avoid the high computational cost of regularization model. Furthermore,in order to get sub-pixel superresolved image, the regularization model for image enhancement is extended to the enlargement of image, which can enlarge and enhance image at the same time. Last,to get clearer edges, a high frequency enhancement filter is used on the superresolved image. Numerical results show that the new model can get much clearer super-resolution images than traditional methods, and the peak signal to noise ratios （PSNRs） are also higher than traditional methods.     

A modified aperture-synthesis method for super-resolution by a single digital hologram

A modified aperture-synthesis method is reported to improve the resolution of a reconstructed image in digital holography by a single hologram. A series of sub-holograms incoherently overlapped on a CCD are recorded as a single aperture-synthesized digital hologram. The angular division multiplexing is introduced to both the object and the reference path for holographic recording. The final super-resolved image is obtained by synthesizing the several reconstructed images. In the experiment, a two-dimensional transparent USAF resolution test target is used. The result demonstrates that the resolution of the reconstructed image is improved from 16.00 lp/mm to 22.64 lp/mm in the horizontal direction.