基于SENet双路径多尺度特征融合的ECT图像重建

针对单一神经网络在电容层析成像图像重建过程中难以捕捉复杂、深层电容向量特征的问题，提出一种基于压缩激励网络(squeeze-and-excitation networks, SENet)双路径多尺度特征融合的电容层析成像图像重建算法。构建多尺度密集深度空洞卷积模块，使模型获得更大的局部感受野的同时可以保持较低计算复杂度，并实现多尺度特征融合，以捕获电容向量的多尺度细节特征，增强模型的表征能力；采用残差神经网络解决深层网络提取特征时出现的退化现象，并添加SENet模块重新标定电容特征张量所属通道对应权重，校准特征响应。形成具有双向特征提取能力的双通道多特征融合的混合模型，以更好的拟合电容张量与介电常数之间的非线性映射关系。试验结果表明，BSFF算法相对于Landweber迭代算法、CNN算法等具有更高的图像重建质量，更好的鲁棒性。     

基于动态加权的多尺度残差网络旋转机械故障诊断算法EI北大核心CSCD

针对传统的机械故障诊断方法特征提取困难问题,提出一种新颖的基于特征通道重标定的动态加权多尺度残差网络旋转机械故障诊断方法。将原始数据作为网络的输入,设计宽卷积层进行信息初步融合扩大模型的感受野;再分别构建三个独立的以残差块为基础的并行分支网络,通过设计多尺度卷积核分别从并行分支网络提取深度特征;接着设计动态加权层建立全局信息建模特征通道之间的动态非线性关系,对每个尺度的特征通道进行重标定,提高网络对故障信息的敏感性;将三个尺度的特征进行特征融合,通过分类器实现故障诊断。在多个数据集上进行试验,验证了该算法的有效性。     

基于动态加权的多尺度残差网络旋转机械故障诊断算法

针对传统的机械故障诊断方法特征提取困难问题,提出一种新颖的基于特征通道重标定的动态加权多尺度残差网络旋转机械故障诊断方法。将原始数据作为网络的输入,设计宽卷积层进行信息初步融合扩大模型的感受野;再分别构建三个独立的以残差块为基础的并行分支网络,通过设计多尺度卷积核分别从并行分支网络提取深度特征;接着设计动态加权层建立全局信息建模特征通道之间的动态非线性关系,对每个尺度的特征通道进行重标定,提高网络对故障信息的敏感性;将三个尺度的特征进行特征融合,通过分类器实现故障诊断。在多个数据集上进行试验,验证了该算法的有效性。     

基于多尺度稠密卷积网络的单图像超分辨率重建

目的针对已有网络对于卷积特征图利用率低下,从而导致高倍数图像重建质量不高的情况,提出一种多尺度稠密卷积网络(SRMD)。方法对SRDenseNet的稠密连接模块进行改进,去除批规范化层,参考已有网络,设计多尺度特征提取层和1×1的信息整合层,从而构成多尺度稠密卷积模块。SRMD通过一个多尺度特征提取层堆叠64个底层特征图,再由8个多尺度稠密卷积模块经过稠密连接堆叠1024个特征图,最后通过信息整合和子像素卷积模块输出超分辨率重建图像。结果在Set5,Set14,B100和U100数据集上进行测试,SRMD重建图像的峰值信噪比分别为30.1570,26.9952,25.7860, 23.4821 dB,结构相似性分别为0.8813,0.7758,0.7243,0.7452。结论与已有网络相比,SRMD与DRCN,VDSR表现相当,优于SRDenseNet和BiCubic方法。     

基于Transformer的多尺度物体检测

目前,Transformer基本模型对同一场景内不同尺寸物体的检测能力不足,其主要原因为各层等尺度的输入嵌入无法提取跨尺度特征,导致网络不具备在不同尺度的特征之间建立交互的能力。基于此,提出一种基于Transformer的多尺度物体检测网络,该网络采用跨尺度嵌入层初步对图像特征进行嵌入处理;利用多分支空洞卷积对输入进行下采样,通过调整并行分支的膨胀率使该结构具有多样的感受野;然后,由残差自注意力模块对输出嵌入结果进行处理,为特征图的局部和全局信息构建联系,使注意力计算融入有效的多尺度语义信息,最终实现多尺度物体检测。模型在COCO等数据集上进行训练,实验结果表明该方法与其他物体检测方法相比具有显著优势。     

结合多尺度及密集特征图融合的阴影检测方法

为了提高图像中阴影检测的准确性,提出一种利用深度神经网络实现阴影检测的方法.首先,构造了一种密集特征图融合结构,将不同卷积层产生的特征图进行融合;其次,针对图像中阴影的多种尺度特征,设计了一种串并联结合的扩张卷积结构提取图像中阴影多尺度特征;最后,将串并联结合的扩张卷积结构和密集特征图融合结构进行结合,设计出一种端到端...     

融合HSV与方向梯度特征的多尺度图像检索

针对现有彩色图像检索算法存在旋转变化鲁棒性差、特征维度高和检索时间长的问题,通过融合主曲率的改进方向梯度特征与HSV颜色特征,提出了一种创新的多尺度图像检索方法.该方法从多个尺度将图像表面的几何曲率信息融合到FHOG描述符中,得到基于主曲率的改进方向梯度算法(P-FHOG),在此基础上进一步融合图像的颜色信息,得到基于...     

基于多尺度Transformer的多视图三维形状分析方法

基于多视图的三维形状分析方法是三维计算机视觉领域中的重要研究分支,通过整合三维形状在多个视角下的二维图像的特征来完成三维形状的识别、检索等任务。然而,如何有效地探索不同视角之间的关联性,并运用这些关联性来聚合多视图图像的特征仍然是三维形状分析中一个亟待解决的核心问题。受到最近兴起的Transformer网络在关系建模问题上成功应用的启发,研究工作引入了一种创新的多尺度Transformer架构,提出了基于多尺度Transformer的多视图三维形状分析方法(Multi-View Multi-Scale Transformer, MVMST)。此方法能够有效地学习不同视角之间的关联性,将多视图图像的特征聚合为一个具有强大表达能力的整体描述符。与以往方法使用感受野为全局的Transformer建模多视图特征的关系不同,该方法受到多尺度学习方法的启发,使用多尺度的Transformer来建模不同尺度下的多视图图像特征之间的关系,并设计了一个多尺度融合模块将多个尺度下经过Transformer处理的特征进行融合,得到一个相比单一尺度更加有效的多尺度表示。多个视图的多尺度表示最终经过视角池化模块...     

空频域结合的多尺度扩张卷积注意力数字水印

目的 将深度学习应用于数字水印,在隐藏信息的同时,不断提高图像的不可见性和鲁棒性,提出一种结合空间域和频率域的多尺度扩张卷积注意力数字水印算法(SF-ACA)。方法 SF-ACA算法的网络框架包含由ACA和SFE构成的生成器、解码器2个部分组成。其中,ACA网络中的MCA模块将3个不同扩张率的扩张卷积对载体图像以多尺度融合的方式进行特征提取,使载体图像能更有效地隐藏水印信息;SFE结合快速傅里叶卷积块,在空域和频域中通过不同大小的感受野捕获互补信息,更精准地获取水印的特征信息,增强了秘密信息的不可见性和鲁棒性。结果 本文提出的水印方法在隐藏与载体图像尺寸相等的三通道彩色图像时,PSNR值为38.81 dB,较UDH方法的PSNR值提高了7.78%。水印图像的隐藏容量是4 096比特,该算法与UDH方法在Dropout、Gaussian噪声、JPEG攻击下,提取精度分别提升了5.38%、10.5%、1.65%,满足不可见性要求的同时实现了强鲁棒性。结论 本文方法在隐藏容量较大时,不可见性和鲁棒性都达到了较好的性能。     

多特征聚合的红外-可见光行人重识别

红外-可见光行人重识别在视频监控、智能交通、安防等领域具有广泛应用。但是不同图像模态间的差异，给该领域带来了巨大的挑战。现有方法主要集中于缓解模态间差异以获得更具鉴别性的特征，但却忽略了邻级特征之间的关系以及多尺度信息对全局特征的影响。因此，本文提出一种基于多特征聚合的红外-可见光行人重识别方法(MFANet)解决现有方法的缺陷。首先在特征提取阶段融合邻级特征，引导低级特征信息的融入，以强化高级特征，使得特征更具健壮性；然后聚合不同感受野的多尺度特征以获得丰富的上下文信息；最后，以多尺度特征作为引导，强化特征以获得更具鉴别性的特征。在SYSU-MM01和RegDB数据集上的实验结果证明了所提方法的有效性，其中SYSU-MM01数据集在最困难的全搜索单镜头模式下平均精度达到了71.77%。     

Investigation of multi-frame image super-resolution based on adaptive self-learning methods

The goal of this paper is to introduce and demonstrate a new high-performance super-resolution (SR) method for multi-frame images. By combining learning-based and reconstruction-based SR methods, this paper proposes a multi-frame image super-resolution method based on adaptive self-learning. Using the adaptive self-learning method and recovery of high-frequency edge information, an initial high-resolution (HR) image containing effective texture information is obtained. The edge smoothness prior is then used to satisfy the global reconstruction constraint and enhance the quality of the HR image. Our results indicate that this method achieves better performance than several other methods for both simulated data and real-scene images.     

Residual Feature Attentional Fusion Network for Lightweight Chest CT Image Super-Resolution

The diagnosis of COVID-19 requires chest computed tomography (CT). High-resolution CT images can provide more diagnostic information to help doctors better diagnose the disease, so it is of clinical importance to study super-resolution (SR) algorithms applied to CT images to improve the resolution of CT images. However, most of the existing SR algorithms are studied based on natural images, which are not suitable for medical images; and most of these algorithms improve the reconstruction quality by increasing the network depth, which is not suitable for machines with limited resources. To alleviate these issues, we propose a residual feature attentional fusion network for lightweight chest CT image super-resolution (RFAFN). Specifically, we design a contextual feature extraction block (CFEB) that can extract CT image features more efficiently and accurately than ordinary residual blocks. In addition, we propose a feature-weighted cascading strategy (FWCS) based on attentional feature fusion blocks (AFFB) to utilize the high-frequency detail information extracted by CFEB as much as possible via selectively fusing adjacent level feature information. Finally, we suggest a global hierarchical feature fusion strategy (GHFFS), which can utilize the hierarchical features more effectively than dense concatenation by progressively aggregating the feature information at various levels. Numerous experiments show that our method performs better than most of the state-of-the-art (SOTA) methods on the COVID-19 chest CT dataset. In detail, the peak signal-to-noise ratio (PSNR) is 0.11 dB and 0.47 dB higher on CTtest1 and CTtest2 at SR compared to the suboptimal method, but the number of parameters and multi-adds are reduced by 22K and 0.43G, respectively. Our method can better recover chest CT image quality with fewer computational resources and effectively assist in COVID-19.     

Harnessing defocus blur to recover high-resolution information in shape-from-focus technique

Traditional shape-from-focus (SFF) uses focus as the singular cue to derive the shape profile of a 3D object from a sequence of images. However, the stack of low-resolution (LR) observations is space-variantly blurred because of the finite depth of field of the camera. The authors propose to exploit the defocus information in the stack of LR images to obtain a super-resolved image as well as a high-resolution (HR) depth map of the underlying 3D object. Appropriate observation models are used to describe the image formation process in SFF. Local spatial dependencies of the intensities of pixels and their depth values are accounted for by modelling the HR image and the HR structure as independent Markov random fields. Taking as input the LR images from the stack and the LR depth map, the authors first obtain the super-resolved image of the 3D specimen and use it subsequently to reconstruct a HR depth profile of the object.     

Efficient document-image super-resolution using convolutional neural network

Experiments performed by us using optical character recognizers (OCRs) show that the character level accuracy of the OCR reduces significantly with decrease in the spatial resolution of document images. There are real life scenarios, where high-resolution (HR) images are not available, where it is desirable to enhance the resolution of the low-resolution (LR) document image. In this paper, our objective is to construct a HR image, given a single LR binary image. The works reported in the literature mostly deal with super-resolution of natural images, whereas we try to overcome the spatial resolution problem in document images. We have trained and obtained a novel convolutional model based on neural networks, which achieves significant improvement in terms of the peak-signal-to-noise ratio (PSNR) of the reconstructed HR images. Using parametric rectified linear units, mean PSNR improvements of 2.32, 4.38, 6.43 and 8.92 dB have been achieved over those of LR input images of 50, 75, 100 and 150 dots per inch (dpi) resolution and average word level accuracy of almost 43%, 45% and 57% on 75 dpi Tamil, English and Kannada images, respectively.     

Super-Resolution Reconstruction of Images Based on Microarray Camera

In the field of images and imaging, super-resolution (SR) reconstruction of images is a technique that converts one or more low-resolution (LR) images into a highresolution (HR) image. The classical two types of SR methods are mainly based on applying a single image or multiple images captured by a single camera. Microarray camera has the characteristics of small size, multi views, and the possibility of applying to portable devices. It has become a research hotspot in image processing. In this paper, we propose a SR reconstruction of images based on a microarray camera for sharpening and registration processing of array images. The array images are interpolated to obtain a HR image initially followed by a convolution neural network (CNN) procedure for enhancement. The convolution layers of our convolution neural network are 3×3 or 1×1 layers, of which the 1×1 layers are used to improve the network performance particularly. A bottleneck structure is applied to reduce the parameter numbers of the nonlinear mapping and to improve the nonlinear capability of the whole network. Finally, we use a 3×3 deconvolution layer to significantly reduce the number of parameters compared to the deconvolution layer of FSRCNN-s. The experiments show that the proposed method can not only ameliorate effectively the texture quality of the target image based on the array images information, but also further enhance the quality of the initial high resolution image by the improved CNN.     

增强稀疏编码的超分辨率重建

本文提出一种基于稀疏字典编码的超分辫率方法.该方法有效地建立高、低分辫率图像高频块间的稀疏关联,并将这种关联作为先验知识来指导基于稀疏字典的超分辫率重建.较超完备字典,稀疏字典对先验知识的表达更紧凑、更高效.字典训练过程中,本文选用高频信息作为高分辫率图像的特征,更有效地建立高、低分辫率图像决间的稀疏关联,所需的训练样...     

Robust MR image super-resolution reconstruction with cross-modal edge-preserving regularization

In this article, we propose a novel image super-resolution (SR) reconstruction method in the field of magnetic resonance imaging, which is based on a cross-modal edge-preserving regularization integrating the internal gradient prior from the target-modal image itself and the external gradient prior from the reference-modal image obtained by pre-scan in many medical imaging scenes. The reference-modal image is a high-resolution guidance image that has much shareable information such as gradient orientation on edge regions, which can be used to improve the image resolution of the target modal. In addition, to be robust against the misalignment between the target-modal image and reference-modal image, a multimodal registration is incorporated in the SR reconstruction process. In this work, the proposed SR method can be formulated as an alternating optimization problem, that is, the target-modal and reference-modal images are alternately updated through iterations. Experimental results on simulated and realistic images show the superior performance of the proposed approach over several state-of-the-art SR techniques.     

A robust variational approach to super-resolution with nonlocal TV regularisation term

Abstract

Multi frame super-resolution (SR) reconstruction algorithms make use of complimentary information among low-resolution (LR) images to yield a high-resolution (HR) image. Inspired by recent development on the video denoising problem, we propose a robust variational approach for SR-based on a constrained variational model that uses the nonlocal total variation (TV) as a regularisation term. In our method, a weighted fidelity term is proposed to take into account inaccurate estimates of the registration parameters and the point spread function. Moreover, we introduce the nonlocal TV as a regularisation term in order to take into account complex spatial interactions within images. In this way, important features and fine details are enhanced simultaneously with noise reduction. Furthermore, an alternative nonlocal TV regularisation is proposed based on a better weight function which integrates gradient similarity and radiometric similarity. Experiments show the effectiveness and practicability of the proposed method.     

基于变分贝叶斯多图像超分辨的平面复眼空间分辨率增强

平面复眼成像系统利用多个子孔径对场景进行成像,由于子孔径大小和图像传感器空间采样率的限制,各子孔径图像质量较差。如何融合多个子孔径图像来获得高分辨率图像是亟需解决的问题。多图像超分辨理论利用多幅具有互补信息的图像来重构高空间分辨率图像,然而现有理论通常采用过于简化的运动模型,这种简化的运动模型对平面复眼成像并不完全适用。若直接把现有多图像超分辨理论用于平面复眼分辨率增强,不准确的相对运动估计将降低图像分辨率增强性能。针对这些问题,本文在变分贝叶斯框架下改进了现有多图像超分辨理论中的运动模型,并把导出的联合估计算法用于平面复眼分辨率增强。仿真数据实验和真实复眼数据实验验证了推荐方法的正确性和有效性。     

基于自相似性与稀疏表示的超分辨率算法

目的为了解决当前稀疏表示的超分辨率算法效果依赖参与训练的数据的问题,结合图像的自相似性,提出一种基于自相似性与稀疏表示相结合的超分辨率算法。方法算法利用图像的多维自相似性,构建多维图像金字塔,采用改进的相似块搜索策略,得到对应的高低分辨率图像块作为训练样本,然后对样本进行字典训练,最后根据稀疏表示得到超分辨率图像。结果实验结果显示,文中算法在峰值信噪比(PSNR)和结构相似度(SSIM)上优于其他算法,对于实验图像而言,PSNR平均提升了0.5 dB。结论提出的超分辨率算法未引入外部数据库,具有较好的效果,能够用于超分辨率重建。