Image super resolution based on residual dense CNN and guided filters

Convolutional neural networks (CNNs) have recently made impressive results for image super-resolution (SR). Our goal is to introduce a new image SR framework rely on a CNN. In this paper, the input image is decomposed into luminance channel and chromatic channels. A designed network based on a residual dense network is introduced to extract the hierarchical features from luminance part. The bicubic interpolation is simply used to upscale low resolution (LR) chromatic channels. However, this step degrades the chromatic channels. To tackle this issue, the SR reconstructed luminance channel is applied as the reference image in guided filters to promote the interpolated chromatic channels. Guided filters technique has ability to retain sharp edges and fine details from the reference image and carry them to the target images. Extensive experiments on several commonly used image SR testing datasets demonstrate that our framework has the ability to extract features and outperforms existing well-known techniques for image SR by LR image into the high resolution (HR) image efficiently.

     

超分辨率图像配准方法研究

介绍了空间域和频率域图像配准原理,在总结已有成果的基础上,对几种典型的算法进行了分析和比较,最后给出了超分辨率图像配准方法的发展方向.     

Image super resolution model enabled by wavelet lifting with optimized deep convolutional neural network

This paper plans to develop an intelligent super resolution model with the linkage of Wavelet lifting scheme and Deep learning algorithm. Before initiating the resolution procedure, the entire HR images are converted into Low Resolution (LR) images using bicubic interpolation-based downsampling and upsampling. Further, the Wavelet lifting scheme helps to generate the four subbands of each image like LR wavelet Sub-Bands for LR images, and High Resolution (HR) wavelet Sub-Bands for HR images. The residual image is generated by taking the difference between the LR wavelet Sub-Bands and HR wavelet Sub-Bands images. The proposed model involves two main phases: Training phase and Testing. The training phase trains the residual image of all images by Deep Convolutional Neural Network with LR wavelet Sub-Bands as input and residual image as target. On the other hand, in testing phase, the LR wavelet Sub-Bands query image is subjected to Deep Convolutional Neural Network, which outputs the concerned residual image. This generated residual image is summed with LR wavelet Sub-Bands image, followed by inverse wavelet lifting scheme to obtain the final super resolution image. The main contribution of this paper is to improve the conventional Deep Convolutional Neural Network by optimizing the number of hidden layer, and hidden neurons using modified Whale Optimization Algorithm called Average Fitness Enabled Whale Optimization Algorithm by considering the objective of maximizing the Peak Signal-to-Noise Ratio. Finally, the proposed method achieves an improved quality of the results which is comparable the existing models.     

利用残差密集网络的运动模糊复原方法

针对图像生成过程中由于物体运动或相机抖动产生的运动模糊问题,提出了利用残差密集网络的运动模糊图像复原方法。设计对抗网络结构,以残差密集网络为生成器,通过长短连接实现不同层次特征的融合,生成复原图像,以深度卷积网络为判别器,判断图像真伪,在生成器和判别器的对抗中提高网络性能;采用对抗损失和内容损失结合的损失函数,提高网络的复原效果;以端到端的方式,省略模糊核的估计过程,输入模糊图像直接获取复原图像。实验结果表明,该方法能够取得较好的复原效果。     

基于对偶字典学习的图像超分辨技术研究

提出一种基于对偶字典学习的图像超分辨方法,通过稀疏重建的方法得到重建的图像,对偶字典通过稀疏表示将低分辨图像和高分辨图像联系起来.在稀疏表示过程中,低分辨图像在低分辨字典上的稀疏表示能够很好地提高对应的高分辨图像在高分辨字典上的稀疏表示效果.将字典的学习建模为包含l1范数优化问题的双层最优化问题,采用隐微分法计算随机梯度下降的期望梯度.仿真实验结果表明,该方法能够达到和联合字典学习方法相同的速度和质量,同时,在实际应用中可以通过神经网络模型学习方法提高算法的速度.与现有的算法比较,表明了该算法的有效性.     

基于向量量化的图像超分辨率方法

受分形编码思想启发,提出了一种新的基于向量量化的图像超分辨率方法。该方法使用学习算法来获取单幅输入图像中的高频信息和低频信息之间的对应关系,并利用此关系对输入图像的一个倍频程的空间频率内添加图像细节以获得高分辨率图像。该方法克服了传统插值方法中因过度平滑导致图像模糊和纹理保持较差的缺点,能够重现出传统插值方法不能复原出的一些高频图像细节。实验结果显示该算法在客观和主观上都比传统插值方法有更好的评价。     

基于自适应规整化方法的图像超分辨率重建

以Gauss-Gibbs随机场模型为图像的先验概率模型,运用自适应规整化的最大后验概率(MAP)方法进行图像超分辨率重建.通过对先验概率分布参数的估计,对图像超分辨率重建求解进行自适应规整化,从而提高重建图像的质量.实验结果表明,该算法能较好地再现图像的各种边缘信息,重建的高分辨率图像在峰值信噪比和视觉效果方面都得到明显提高.     

基于稠密残差网络的图像超分辨率重建算法

图像超分辨率重建作为一种廉价方便的图像增强手段,在视频监控、医学成像、卫星遥感等领域有着重要的研究意义.为此结合深度学习在图像重建的性能优势,提出了一种基于增强稠密残差网络(ERDN)的图像超分辨率重建模型.首先使用多卷积核的稠密残差神经网络模块,提取图像的细节信息;然后通过跳跃连接和特征复用模块对多层图像信息进行筛选...     

基于加权随机森林的图像超分辨率算法研究

针对目前图像超分辨率重建效果欠佳的问题,提出一种基于加权随机森林的图像超分辨率重建算法.利用随机森林对图像块的特征进行聚类,并引入岭回归模型建立每类叶子结点中高、低分辨率图像块的映射关系,重建时根据测试低分辨率图像块所属的类别以及在每类叶子结点中的K近邻近似拟合误差,进行加权预测获得高分辨率图像块.将图像的非局部自相似性与迭代反投影算法相结合对预测的高分辨率图像进行后处理以提高重建质量.实验结果表明,所提出算法可以有效提高峰值信噪比,具有较好的可视效果.     

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.

     

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.

     

基于残差网络的蛋白质超二级结构图像分类

为解决SVM、Bayes、RNN(recurrent neural network)等传统算法在蛋白质结构分类任务中精度低的问题,提出一种基于残差网络的蛋白质超二级结构图像分类方法.将PDB(protein data bank)和SCOP(structural classification of pro-teins)数据库中的4类蛋白质超二级结构3D模型转化为14角度拍摄的2D图像,针对每类图像,通过残差网络单元进行深度特征提取和优化,利用神经网络模型训练,将验证精度最高的模型保存下来并进行测试.实验结果表明,分类精度达到了90.2％,验证了模型的可行性和算法的有效性.     

基于多尺度密集网络的肺结节图像检索算法

现有基于内容的医学图像检索（CBMIR）算法存在特征提取的不足，导致图像的语义信息表达不完善、图像检索性能较差，为此提出一种多尺度密集网络算法以提高检索精度。首先，将512×512的肺结节图像降维到64×64，同时加入密集模块以解决提取的低层特征和高层语义特征之间的差距；其次，由于网络的不同层提取的肺结节图像信息不同，为了提高检索精度和效率，采用多尺度方法结合图像的全局特征和结节局部特征生成检索哈希码。实验结果分析表明，与自适应比特位的检索（ABR）算法相比，提出的算法在64位哈希码编码长度下的肺结节图像检索查准率可以达到91.17%，提高了3.5个百分点；检索一张肺切片需要平均时间为48 μs。所提算法的检索结果在表达图像丰富的语义特征和检索效率方面，优于其他对比的网络结构，适用于为医生临床辅助诊断提供依据、帮助患者有效治疗。     

基于深度渐进式反投影注意力网络的图像超分辨率重建

针对多数单帧图像超分辨率（SISR）方法在重建预测图像时存在高频信息丢失和上采样过程中会引入噪声以及特征图各通道之间的相互依赖关系难以确定等问题,提出了深度渐进式反投影注意力网络。首先使用渐进式上采样方法将低分辨率（LR）图像逐步缩放至给定的倍率,缓解上采样过程中造成的高频信息丢失等问题;然后在渐进式上采样的每个阶段融合迭代反投影思想,学习高分辨率（HR）和LR特征图之间的映射关系并减少上采样过程中引入的噪声;最后使用注意力机制为渐进式反投影网络不同阶段产生的特征图动态分配注意力资源,使网络模型学习到各特征图之间的相互依赖关系。实验结果表明,所提出的方法相比主流的超分辨率方法,峰值信噪比（PSNR）最高可增加3.16 dB,结构相似性最高可提升0.218 4。     

Single-frame super resolution of remote-sensing images by convolutional neural networks

Super resolution (SR) refers to generation of a high-resolution (HR) image from a decimated, blurred, low-resolution (LR) image set, which can be either a single-frame or multi-frame that contains a collection of images acquired from slightly different views of the same observation area. In this study, two convolutional neural network (CNN)-based deep learning techniques are adapted in single-frame SR to increase the resolution of remote sensing (RS) images by a factor of 2, 3, and 4. In order to both preserve the colour information and speed up the algorithm, first an intensity hue saturation (IHS) transform is utilized and the SR techniques are only applied to the intensity channel of the images. Colour information is then restored with an inverse IHS transformation. We demonstrate the results of the proposed method on RS images acquired from Satellites Pour l’Observation de la Terre (SPOT) or Earth-observing satellites and Pleiades satellites with different spatial resolution. First synthetic LR images are created by downsampling, then structural similarity (SSIM) Index, peak signal-to-noise ratio (PSNR), Spectral Angle Mapper (SAM) and Erreur Relative Globale Adimensionnelle de Synthese (ERGAS) values are calculated for a quantitative evaluation of the methods. Finally, the method, with better performance results, is tested within a real scenario, that is, with original LR images as the input. The obtained HR images demonstrated visible qualitative enhancements.     

Robust super resolution of compressed video

This paper presents a robust algorithm to recover high-frequency information from compressed low-resolution (LR) video sequences. Previous super-resolution (SR) approaches have succeeded in resolution enhancement when the motion in the LR sequence is simple. However, when the motion is complex, new artifacts will be introduced in the SR processing. To solve this problem, we develop a robust Bayesian SR algorithm with two steps. We first isolate the frames individually to get their corresponding initial SR solutions within the Bayesian framework. Secondly, with a robust cost function to reject outliers and noise, final SR images are achieved with multiple LR frames. In the mean time, we impose the constraint that the distribution of high-resolution (HR) image gradient should be equal to one of the corresponding decompressed LR images to sharpen the edges of the results. As a result of these steps, we are able to produce high-quality deblurred results, which show a suppressing of high-frequency artifacts and less ringing artifacts, with a higher peak signal-to-noise ratio (PSNR).     

Motion block based video super resolution

This paper presents a new video super resolution technique, based on the motion and static areas of the low resolution video frames. In order to separate the motion and static blocks, a block motion estimation method is performed between a reference and its neighboring frames. Among the motion blocks, the occluded blocks are identified using an adaptive threshold applied on each block individually. Structure-adaptive normalized convolution (SANC) reconstruction method is used to generate the high resolution static and motion blocks where discrete wavelet transform (DWT) based interpolation is used to produce the high resolution occluded blocks. The static and motion blocks are combined into a high resolution frame. Finally, a sharpening process is performed on the high resolution frame in order to generate the super resolved high resolution output frame. The experimental results show that the proposed technique provides significantly better qualitative visual results as well as quantitative higher PSNR than the state of the art video super resolution algorithms.     

A progressive framework for dense stereo matching

A progressive framework is proposed for dense stereo matching to solve problems caused by weaktexture and occlusion in this paper. The main idea is that disparity is extracted progressively, from coarse to fine, from sparse to dense. First, a coarse disparity map is obtained by the segment-based pre-matching method, in which horizontal and vertical segment matching are performed in parallel and pre-matching results are merged to preserve more details. Second, disparity diffusion is performed to roughly estimate disparity values for miss-matched points. Third, a probabilistic approach is used for disparity refinement, taking into account stereo prior, image likehood and disparity smoothness. Experiments are made on the Middlebury benchmark to demostrate the effectiveness of the proposed algorithm.