A progressive framework for rotary motion deblurring

The rotary motion deblurring is an inevitable procedure when the imaging seeker is mounted in the rotating missiles.Traditional rotary motion deblurring methods suffer from ringing artifacts and noise,especially for large blur extents.To solve the above problems,we propose a progressive rotary motion deblurring framework consisting of a coarse deblurring stage and a refinement stage.In the first stage,we design an adaptive blur extents factor(BE factor)to balance noise suppression and details reconstruction.And a novel deconvolution model is proposed based on BE factor.In the second stage,a triplescale deformable module CNN(TDM-CNN)is designed to reduce the ringing artifacts,which can exploit the 2D information of an image and adaptively adjust spatial sampling locations.To establish a standard evaluation benchmark,a real-world rotary motion blur dataset is proposed and released,which includes rotary blurred images and corresponding ground truth images with different blur angles.Experimental results demonstrate that the proposed method outperforms the state-of-the-art models on synthetic and real-world rotary motion blur datasets.The code and dataset are available at https://github.com/JinhuiQin/RotaryDeblurring.

MIDNet:Deblurring Network for Material Microstructure Images

Scanning electron microscopy(SEM)is a crucial tool in the field of materials science,providing valuable insightsinto the microstructural characteristics of materials.Unfortunately,SEM images often suffer from blurrinesscaused by improper hardware calibration or imaging automation errors,which present challenges in analyzingand interpretingmaterial characteristics.Consequently,rectifying the blurring of these images assumes paramountsignificance to enable subsequent analysis.To address this issue,we introduce a Material Images DeblurringNetwork(MIDNet)built upon the foundation of the Nonlinear Activation Free Network(NAFNet).MIDNetis meticulously tailored to address the blurring in images capturing the microstructure of materials.The keycontributions include enhancing the NAFNet architecture for better feature extraction and representation,integratinga novel soft attention mechanism to uncover important correlations between encoder and decoder,andintroducing newmulti-loss functions to improve training effectiveness and overallmodel performance.We conducta comprehensive set of experiments utilizing the material blurry dataset and compare them to several state-of-theartdeblurring methods.The experimental results demonstrate the applicability and effectiveness of MIDNet in thedomain of deblurring material microstructure images,with a PSNR(Peak Signal-to-Noise Ratio)reaching 35.26 dBand an SSIM(Structural Similarity)of 0.946.Our dataset is available at:https://github.com/woshigui/MIDNet.

Deblurring,artifact-free optical coherence tomography with deconvolution-random phase modulation

Deconvolution is a commonly employed technique for enhancing image quality in optical imaging methods.Unfortu-nately,its application in optical coherence tomography(OCT)is often hindered by sensitivity to noise,which leads to ad-ditive ringing artifacts.These artifacts considerably degrade the quality of deconvolved images,thereby limiting its effect-iveness in OCT imaging.In this study,we propose a framework that integrates numerical random phase masks into the deconvolution process,effectively eliminating these artifacts and enhancing image clarity.The optimized joint operation of an iterative Richardson-Lucy deconvolution and numerical synthesis of random phase masks(RPM),termed as De-conv-RPM,enables a 2.5-fold reduction in full width at half-maximum(FWHM).We demonstrate that the Deconv-RPM method significantly enhances image clarity,allowing for the discernment of previously unresolved cellular-level details in nonkeratinized epithelial cells ex vivo and moving blood cells in vivo.

Image defocus deblurring method based on gradient difference of boundary neighborhood

Background For static scenes with multiple depth layers,existing defocused image deblurring methods have the problems of edge-ringing artifacts or insufficient deblurring owing to inaccurate estimation of the blur amount,and prior knowledge in nonblind deconvolution is not strong,which leads to image detail recovery challenges.Methods To this end,this study proposes a blur map estimation method for defocused images based on the gradient difference of the boundary neighborhood,which uses the gradient difference of the boundary neighborhood to accurately obtain the amount of blurring,thereby preventing boundary ringing artifacts.The obtained blur map is then used for blur detection to determine whether the image needs to be deblurred,thereby improving the efficiency of deblurring without manual intervention and judgment.Finally,a nonblind deconvolution algorithm was designed to achieve image deblurring based on the blur amount selection strategy and sparse prior.Results Experimental results showed that our method improves PSNR(Peak Signal-to-Noise Ratio)and SSIM(Structural Similarity Index)by an average of 4.6%and 7.3%,respectively,compared to existing methods.Conclusions Experimental results showed that the proposed method outperforms existing methods.Compared to existing methods,our method can better solve the problems of boundary ringing artifacts and detail information preservation in defocused image deblurring.

Image Motion Deblurring Based on Salient Structure Selection and L0-2 Norm Kernel Estimation

Single image motion deblurring has been a very challenging problem in the field of image processing. Although there are many researches had been proposed to solve this problem, it still has problems on kernel accuracy. In order to improve the kernel accuracy, an effective structure selection method was used to select the salient structure of the blur image. Then a novel kernel estimation method based on L0-2 norm was proposed. To guarantee the sparse kernel and eliminate the negative influence of details L0-norm was used. And L2-norm was used to ensure the continuity of kernel. Many experiments were done to compare proposed method and state-of-the-art methods. The results show that our method can estimate a better kernel and use less time than previous work, especially when the size of blur kernel is large.

BaMBNet:A Blur-Aware Multi-Branch Network for Dual-Pixel Defocus Deblurring

Reducing the defocus blur that arises from the finite aperture size and short exposure time is an essential problem in computational photography.It is very challenging because the blur kernel is spatially varying and difficult to estimate by traditional methods.Due to its great breakthrough in low-level tasks,convolutional neural networks(CNNs)have been introdu-ced to the defocus deblurring problem and achieved significant progress.However,previous methods apply the same learned kernel for different regions of the defocus blurred images,thus it is difficult to handle nonuniform blurred images.To this end,this study designs a novel blur-aware multi-branch network(Ba-MBNet),in which different regions are treated differentially.In particular,we estimate the blur amounts of different regions by the internal geometric constraint of the dual-pixel(DP)data,which measures the defocus disparity between the left and right views.Based on the assumption that different image regions with different blur amounts have different deblurring difficulties,we leverage different networks with different capacities to treat different image regions.Moreover,we introduce a meta-learning defocus mask generation algorithm to assign each pixel to a proper branch.In this way,we can expect to maintain the information of the clear regions well while recovering the missing details of the blurred regions.Both quantitative and qualitative experiments demonstrate that our BaMBNet outperforms the state-of-the-art(SOTA)methods.For the dual-pixel defocus deblurring(DPD)-blur dataset,the proposed BaMBNet achieves 1.20 dB gain over the previous SOTA method in term of peak signal-to-noise ratio(PSNR)and reduces learnable parameters by 85%.The details of the code and dataset are available at https://github.com/junjun-jiang/BaMBNet.

On weak solutions for an image denoising-deblurring model

A new denoising-deblurring model in image restoration is proposed,in which the regularization term carries out anisotropic diffusion on the edges and isotropic diffusion on the regular regions.The existence and uniqueness of weak solutions for this model are proved,and the numerical model is also testified.Compared with the TV diffusion,this model preferably reduces the staircase appearing in the restored images.

Remote Sensing Image Deblurring Based on Grid Computation

In general, there is a demand for real-time processing of mass quantity remote sensing images. However, the task is not only data-intensive but also computating-intensive. Distributed processing is a hot topic in remote sensing processing and image deblurring is also one of the most important needs. In order to satisfy the demand for quick proc- essing and deblurring of mass quantity satellite images, we developed a distributed, grid computation-based platform as well as a corresponding middleware for grid computation. Both a constrained power spectrum equalization algorithm and effective block processing measures, which can avoid boundary effect, were applied during the processing. The re- sult is satisfactory since computation efficiency and visual effect were greatly improved. It can be concluded that the technology of spatial information grids is effective for mass quantity remote sensing image processing.

Deblurring Texture Extraction from Digital Aerial Image by Reforming “Steep Edge” Curve

Texture extract from digital aerial image is widely used in three-dimensional city modeling to generate “photo-realistic” views. In this paper, a method based on reforming “Steep edge” curve, which clearly explains how the diffraction of the sunlight makes digital aerial image blurring, is proposed to deblur the texture extraction from digital aerial image, and the experiment shows a good result in visualization and automation.

Blind Deblurring Based on L_0 Norm from Salient Edges

Motion deblurring is a basic problem in the field of image processing and analysis. This paper proposes a new method of single image blind deblurring which can be significant to kernel estimation and non-blind deconvolution. Experiments show that the details of the image destroy the structure of the kernel, especially when the blur kernel is large. So we extract the image structure with salient edges by the method based on RTV. In addition, the traditional method for motion blur kernel estimation based on sparse priors is conducive to gain a sparse blur kernel. But these priors do not ensure the continuity of blur kernel and sometimes induce noisy estimated results. Therefore we propose the kernel refinement method based on L0 to overcome the above shortcomings. In terms of non-blind deconvolution we adopt the L1/L2 regularization term. Compared with the traditional method, the method based on L1/L2 norm has better adaptability to image structure, and the constructed energy functional can better describe the sharp image. For this model, an effective algorithm is presented based on alternating minimization algorithm.

Blind Motion Deblurring for Online Defect Visual Inspection

Online defect visual inspection (ODVI) works while the object has to be static, otherwise the relative motion between camera and object will create motion blur in images. In order to implement ODVI in dynamic scene, it developes one blind motion deblurring method whose objective is to estimate blur kernel parameters precisely. In the proposed method, Radon transform on superpixels determinated the blur angle, and the autocorrelation function based on magnitude (AFM) of the preprocessed blurred image was utilized to identify the blur length. With the projection relationship discussed in this study, it will be unnecessary to rotate the blurred image or the axis. The proposed method is of high accuracy and robustness to noise, and it can somehow handle saturated pixels. To validate the proposed method, experiments have been carried out on synthetic images both in noise free and noisy situations. The results show that the method outperforms existing approaches. With the modified Richardson– Lucy deconvolution, it demonstrates that the proposed method is effective for ODVI in terms of subjective visual quality.

基于DeblurGAN的运动图像去模糊方法分析

阐述DeblurGAN盲运动模糊移动方法,对运动图像进行去模糊化处理。试验结果表明,与bur影像相比,通过运用DeblurGAN,可以确保运动图像清晰度得以显著提升,同时,还能实现对图像细节纹理的清晰显示。

基于映射空间编码的高速运动轨道图像去模糊研究

针对轨道缺陷检测系统因镜头抖动或相机快速移动而导致所采集图像较为模糊的问题,提出一种基于最大后验概率估计思想的映射空间编码的高速运动轨道图像去模糊算法。首先,该算法使用深度编解码器和残差网络分别对数据集中清晰图像到模糊图像的映射关系和模糊核进行编码,为了保证编码时频率信息的完整性,算法在传统的残差模块上引入快速傅里叶变换通道构成双通道残差网络,以补偿多次特征提取带来的频率损失;其次,算法采用深度图像先验(Deep Image Prior,DIP)将潜在的清晰图像和模糊核进行参数化,再利用先验得到的模糊核和清晰图像来调用编码空间中的映射关系;最后,通过交替优化潜在的清晰图像和模糊核,从而去逼近一个真实未知的映射,进而实现真实场景下高速运动轨道图像的去模糊。实验结果表明,双通道残差模块提取的特征图频率信息分量强度普遍高于传统的残差模块,相较于使用传统残差模块实现该算法,采用双通道残差模块可使峰值信噪比(Peak Signal-to-Noise Ratio,PSNR)提升0.84 dB,结构相似性(Structural Similarity,SSIM)提高0.025 1。与现有的深度学习去模糊算法相比,提出的去模糊算法对高速轨道检测系统所采集图像的去模糊效果更佳,在性能方面相较于最好的去模糊算法,PSNR提高了1.84 dB,SSIM提升了0.017 3,显著提升了采集图像的质量。研究结果可为下一步识别轨道部件是否存在缺陷提供清晰图像。

基于分数阶全变分和低秩正则化的彩色图像去模糊方法

针对现有的彩色图像去模糊过程中存在色彩失衡、阶梯效应和伪影等现象,提出了一种基于分数阶全变分和低秩正则的图像去模糊优化方法。首先,将传统的RGB彩色图像转换到YCbCr颜色空间,利用其亮度通道特征解决色彩失衡问题;其次,利用分数阶全变分的特征消除图像恢复任务中出现的阶梯效应,并且引入加权核范数低秩正则进一步抑制伪影及噪声;最后,利用交替方向乘子法设计出高效的求解方法,通过迭代优化得到纯净图像的最优估计。对彩色图像测试的实验结果表明,所提出的方法对图像去模糊任务取得较好的视觉恢复效果,客观评价指标良好。

基于NSST与稀疏先验的遥感图像去模糊方法

针对遥感图像的模糊问题,设计一种基于非下采样剪切波变换与稀疏先验的图像复原算法.首先,利用遥感图像在非下采样剪切波分解下的高频图像的稀疏特性设置先验条件构造图像复原模型;其次,采用交替方向乘子法求解模型;再次,采用软阈值方法对高频图像进行约束处理,在低频图像进行导向滤波处理,以最大可能保留图像的细节信息;最后,将高频图像与低频图像进行重构,对重构后的图像采用卷积神经网络进行深度去噪,最终复原出清晰的图像.将该去模糊算法与H-PNP,GSR,L2TV算法进行实验对比.实验结果表明,该算法能有效去除遥感图像中的模糊和噪声,保留图像的边缘细节,客观评价指标均高于其他3种对比实验算法.

基于Elastic-net正则化的神经网络方法求解反问题

神经网络已经成为求解反问题的热点方法之一,引入elastic-net正则项作为神经网络中损失函数的惩罚项防止求解过程的过度拟合,并通过交叉训练实现基于elastic-net正则项的神经网络的算法.通过压缩感知和图像去模糊2个数值实验,验证elastic-net正则项防止过度拟合的可行性和有效性.此外,当变换矩阵条件数较大时,在较低的训练轮次下可以达到较好的训练效果.

基于半监督算法的高光谱影像特征提取仿真

高光谱影像包括待测物的空间、光谱和辐射三重信息,且图像信息具有维度高、空间相关性弱、特征非线性强的特点,导致其空间特征序列混乱,特征提取难度大。于是提出基于半监督算法的高光谱影像特征提取方法。应用半监督算法对高光谱图像中的高维数据降维处理,并基于降维结果完成高光谱图像的去模糊。高光谱图像完成降维去模糊后,根据特征学习模型学习高光谱影像数据,获取图像深层特征。在像元空间内对深度特征以及空间信息完成空、谱的联合,实现高光谱影像特征的提取。实验结果表明,所提方法应用下影像特征点在特征空间内聚类效果好,查全率和查准率均能达到95%以上,说明上述方法的应用性能更优。

生成对抗网络在图像修复中的应用综述

随着生成对抗网络的迅猛发展,许多基于传统方法难以较好解决的图像修复问题获得了新的研究途径。生成对抗网络凭借强大的生成能力,能从受损图像中恢复出完好的图像,故而在图像修复中得到较为广泛的应用。总结了近年来利用生成对抗网络修复受损图像问题的相关理论与研究,以受损图像的类别及其所适配的修复方法为主要划分依据,将图像修复的应用划分为图像补全、图像去模糊、图像去噪三个主要方面。针对每一方面,通过技术原理、应用对象等维度对图像修复的应用进一步细分。对于图像补全领域,从使用条件引导与潜在编码等角度探讨了基于生成对抗网络的不同图像补全方法;对于图像去模糊领域,阐释了运动模糊图像与静态模糊图像的本质不同及其修复方法;对于图像去噪领域,归纳了不同类别图像的个性化去噪方法。同时,对于每一类应用,分析了所采用的具体生成对抗网络模型的特点及其贡献。最后,总结了生成对抗网络应用于图像修复的优势与不足,并对未来应用场景进行了展望。

结合梯度指导和局部增强Transformer的图像去模糊网络

模糊图像不仅影响人类感知还会影响后续计算机视觉任务的性能,例如自动驾驶系统和户外监控系统中的视觉算法.针对以往基于深度学习的去模糊方法感受野较小,不能动态适应输入内容和重建图像细节信息困难等问题,提出了一种基于Transformer的图像去模糊网络.网络包含两个分支:图像内容分支和梯度分支,每条分支均以具有窗口机制的Transformer作为主干,通过梯度分支的信息指导图像去模糊重建,能够更好地恢复图像的边缘和纹理.同时,为了充分利用图像的内容信息和梯度信息,本文还设计了一个交互式融合模块来有效融合特征信息.此外,本文通过在Transformer块的自注意力机制和前馈网络中引入卷积来解决Transformer对局部信息建模不足的问题.在合成数据集和真实数据集上的大量实验结果表明,提出的算法能有效去除复杂模糊并且恢复清晰的细节,在定量指标和视觉效果上均优于目前的主流去模糊算法.

去模糊网络复层运动图像恢复算法仿真

复层运动图像由于模糊而导致细节纹理差,无法有效传递信息。为解决上述问题,基于AED健美操数据集,在数字图像处理的基础上,构建出一种去模糊网络与运动目标检测相结合的图像恢复算法,即DGV2-YOL3算法。算法首先对AED数据样本进行截取并预处理,利用下采样方法提升图像特征提取能力;然后基于四组通道采集图像特征,减少25%系统参数量,并利用细粒度模块对特征进行残差学习;接着对学习特征进行上采用融合处理,通过全局跳跃方法重建图像;最后基于DBSACAN算法,自适应规划图像锚点与IOU阈值,识别检测出运动类型,提升检测准确率。仿真主观分析表明,较其它算法相比,DGV2-YOL3算法处理后的图像,其纹理清晰,边缘显著,且“袜子”细节被恢复明显;客观分析结果显示,DGV2-YOL3算法具有最高的峰值信噪比与分类指标,较其它8类叠加算法相比,分别平均提升了13.48与5.01%,且数据处理效率排名第三,具有较高的实时性。综上所述,DGV2-YOL3算法有效的提升了模糊图像的边缘细节纹理恢复能力,具有较高的仿真研究价值。