基于多尺度残差网络的单应估计方法

单应估计是许多计算机视觉任务中一个基础且重要的步骤。传统单应估计方法基于特征点匹配,难以在弱纹理图像中工作。深度学习已经应用于单应估计以提高其鲁棒性,但现有方法均未考虑到由于物体尺度差异导致的多尺度问题,所以精度受限。针对上述问题,提出了一种用于单应估计的多尺度残差网络。该网络能够提取图像的多尺度特征信息,并使用多尺度特征融合模块对特征进行有效融合,此外还通过估计四角点归一化偏移进一步降低了网络优化难度。实验表明,在MS-COCO数据集上,该方法平均角点误差仅为0.788个像素,达到了亚像素级的精度,并且在99%情况下能够保持较高的精度。由于综合利用了多尺度特征信息且更容易优化,该方法精度显著提高,并具有更强的鲁棒性。     

基于不同尺度增强融合的视网膜血管分割算法

针对现有视网膜血管图像提取细小血管准确率较低的问题,提出了一种基于多尺度线性检测器与局部和全局增强相结合的视网膜血管分割方法.对多尺度线检测器进行研究,将其分为小尺度和大尺度两部分;利用小尺度对局部增强后的图像与大尺度对全局增强后的图像分别进行检测,得到不同尺度下的响应函数;将不同尺度下的响应函数进行融合,得到最终的视网膜血管结构.在STARE和DRIVE两个数据库上进行实验,结果表明:该算法得到的平均血管准确率分别达到96.62％和96.45％,平均真阳性率分别达到75.52％和83.07％,分割准确率高,能够得到较好的血管分割结果.     

基于Treelet变换的模拟电路故障诊断

针对模拟电路故障诊断中特征向量冗余的问题,提出一种基于Treelet变换的模拟电路故障诊断方法.Treelet变换是一种自适应的多尺度的数据分析方法,适用于对高维数据降维和特征选择。文中首先对被测电路的输出信号采样,将采集到的信号进行Treelet变换,提取故障特征向量,最后将得到的特征向量输入BP神经网络进行故障模式识别。仿真实验结果表明,该方法能够有效地提取电路故障特征。与其他故障特征提取方法相比较,基于Treelet变换的模拟电路故障诊断方法具有较高的故障诊断率和收敛速度。     

Retinal vessel segmentation using enhanced fuzzy min-max neural network

Automated segmentation of retinal vessels plays a pivotal role in early diagnosis of ophthalmic disorders. In this paper, a blood vessel segmentation algorithm using an enhanced fuzzy min-max neural network supervised classifier is proposed. The input to the network is an optimal 11-D feature vector which consists of spatial as well as frequency domain features extracted from each pixel of a fundus image. The essence of the method is its hyperbox classifier which performs online learning and gives binary output without any need of post-processing. The method is tested on publicly available databases DRIVE and STARE. The results are compared with the existing methods in the literature. The proposed method exhibits efficient performance and can be implemented in computer aided screening and diagnosis of retinal diseases. The method attains an average accuracy, sensitivity and specificity of 95.73%, 74.75% and 97.81% on DRIVE database and 95.51%, 74.65% and 97.11% on STARE database, respectively.

     

联合同步挤压小波变换和多尺度排列熵的局部放电类型识别

为稳定提取变压器局部放电信号的特征,提出一种基于同步挤压小波变换和多尺度排列熵的局部放电特征提取方法,再通过ＧＫ模糊聚类方法对局部放电信号的特征进行识别分类。首先,通过同步挤压小波变换对4种典型变压器故障产生的局部放电信号进行分解,将其分解为一组含有局部放电特征信息的模态分量;然后,通过多尺度排列熵量化各模态分量的局部放电特征信息,使用各模态分量多尺度排列熵的平均值作为识别特征向量;最后,利用模糊聚类得到的局部放电样本标准聚类中心,采用欧式贴近度进行局部放电识别分类。将提出的方法应用于变压器局部放电的实验数据上,并与基于小波分解方法和经验模态分解的识别方法进行对比分析,实验结果表明,所提出的方法具有更好的分类性,对变压器局部放电分类具有更高的识别精度,平均识别精度达到93.60%。     

基于多尺度多粒度融合的行人重识别方法

行人重识别是指利用计算机视觉技术在给定监控的图像中识别目标行人,受拍摄场景视角和姿势变化、遮挡等因素的影响,现有基于局部特征的行人重识别方法所提取的特征辨别力差,从而导致重识别精度较低。为有效地利用特征信息,提出一种多尺度多粒度融合的行人重识别方法MMF-Net。通过多个分支结构学习不同尺度和不同粒度的特征,并利用局部特征学习优化全局特征,以加强全局特征和局部特征的关联性。同时,在网络的低层引入语义监督模块以提取低层特征,并将其作为行人图像相似性度量的补充,实现低层特征和高层特征的优势互补。基于改进的池化层,通过结合最大池化和平均池化的特点获取具有强辨别力的特征。实验结果表明,MMF-Net方法在Market-1501数据集上的首位命中率和mAP分别为95.7%和89.1%,相比FPR、MGN、BDB等方法,其具有较优的鲁棒性。     

Construction and verification of retinal vessel segmentation algorithm for color fundus image under BP neural network model

To improve the accuracy of retinal vessel segmentation, a retinal vessel segmentation algorithm for color fundus images based on back-propagation (BP) neural network is proposed according to the characteristics of retinal blood vessels. Four kinds of green channel image enhancement results of adaptive histogram equalization, morphological processing, Gaussian matched filtering, and Hessian matrix filtering are used to form feature vectors. The BP neural network is input to segment blood vessels. Experiments on the color fundus image libraries DRIVE and STARE show that this algorithm can obtain complete retinal blood vessel segmentation as well as connected vessel stems and terminals. When segmenting most small blood vessels, the average accuracy on the DRIVE library reaches 0.9477, and the average accuracy on the STARE library reaches 0.9498, which has a good segmentation effect. Through verification, the algorithm is feasible and effective for blood vessel segmentation of color fundus images and can detect more capillaries.

     

基于编解码卷积神经网络的单张图像深度估计

摘 要：针对传统方法在单目视觉图像深度估计时存在鲁棒性差、精度低等问题,提出一 种基于卷积神经网络(CNN)的单张图像深度估计方法。首先,提出层级融合编码器-解码器网络, 该网络是对端到端的编码器-解码器网络结构的一种改进。编码器端引入层级融合模块,并通过 对多层级特征进行融合,提升网络对多尺度信息的利用率。其次,提出多感受野残差模块,其 作为解码器的主要组成部分,负责从高级语义信息中估计深度信息。同时,多感受野残差模块 可灵活地调整网络感受野大小,提高网络对多尺度特征的提取能力。在 NYUD v2 数据集上完 成网络模型有效性验证。实验结果表明,与多尺度卷积神经网络相比,该方法在精度 δ<1.25 上 提高约 4.4%,在平均相对误差指标上降低约 8.2%。证明其在单张图像深度估计的可行性。     

High speed detection of retinal blood vessels in fundus image using phase congruency

Detection of blood vessels in retinal fundus image is the preliminary step to diagnose several retinal diseases. There exist several methods to automatically detect blood vessels from retinal image with the aid of different computational methods. However, all these methods require lengthy processing time. The method proposed here acquires binary vessels from a RGB retinal fundus image in almost real time. Initially, the phase congruency of a retinal image is generated, which is a soft-classification of blood vessels. Phase congruency is a dimensionless quantity that is invariant to changes in image brightness or contrast; hence, it provides an absolute measure of the significance of feature points. This experiment acquires phase congruency of an image using Log-Gabor wavelets. To acquire a binary segmentation, thresholds are applied on the phase congruency image. The process of determining the best threshold value is based on area under the relative operating characteristic (ROC) curve. The proposed method is able to detect blood vessels in a retinal fundus image within 10 s on a PC with (accuracy, area under ROC curve) = (0.91, 0.92), and (0.92, 0.94) for the STARE and the DRIVE databases, respectively.     

基于多尺度自注意卷积的足迹压力图像检索算法

为了提高足迹压力图像检索的精度,提出基于多尺度自注意卷积的足迹压力图像检索算法.首先,对足迹压力图像进行角度校正、对齐、擦除等预处理操作,减小图像角度等因素对特征提取的影响.再由多个并行分支的空洞卷积和自适应注意模块构成的多尺度自注意卷积模块自适应地提取可判别特征.最后,由全局特征分支、残缺性评分掩模分支构成残缺性评分模块,得到共同残缺性评分矩阵,利用该评分矩阵对可判别特征进行加权组合,提高网络对残缺足迹共同可见区域的关注程度.实验表明,在构建的FootPrintImage数据集上,文中算法具有较高的首中准确率和平均检索精度.     

Computer-aided retinal vessel segmentation in retinal images: convolutional neural networks

In medicine, diagnosis is as important as treatment. Retinal blood vessels are the most easily visible vessels in the whole body, and therefore, play a key role in the diagnosis of numerous diseases and eye disorders. Systematic and eye diseases cause morphologic variations, such as the growing, narrowing or branching of retinal blood vessels. Imaging-based screening of retinal blood vessels plays an important role in the identification and follow-up of eye diseases. Therefore, automatic retinal vessel segmentation can be used to diagnose and monitor those diseases. Computer-aided algorithms are required for the analysis of progression of eye diseases. This study proposes a hybrid method that provides a combination of pre-processing and data augmentation methods with a deep learning model. Pre-processing was used to solve the irregular clarification problems and to form a contrast between the background and retinal blood vessels. After pre-processing step, a convolutional neural network (CNN) was designed and then trained for the extraction of retinal blood vessels. In the training phase, data augmentation was performed to improve training performance. The CNN was trained and tested in the DRIVE database, which is commonly used in retinal blood vessel segmentation and publicly available for studies in this area. Results showed that the proposed system extracted vessels with a sensitivity of 77.78%, specificity of 97,84%, precision of 84.17% and accuracy of 95.27%.

This study also compared the results to those of previous studies. The comparison showed that the proposed method is an efficient and successful method for extracting retinal blood vessels. Moreover, the pre-processing phases improved the system performance. We believe that the proposed method and results will make contribution to the literature.

     

基于多尺度变形卷积的特征金字塔光流计算方法

针对现有深度学习光流计算方法的运动边缘模糊问题,提出了一种基于多尺度变形卷积的特征金字塔光流计算方法.首先,构造基于多尺度变形卷积的特征提取模型,显著提高图像边缘区域特征提取的准确性;然后,将多尺度变形卷积特征提取模型与特征金字塔光流计算网络耦合,提出一种基于多尺度变形卷积的特征金字塔光流计算模型;最后,设计一种结合图像与运动边缘约束的混合损失函数,通过指导模型学习更加精准的边缘信息,克服了光流计算运动边缘模糊问题.分别采用MPI-Sintel和KITTI2015测试图像集对该方法与代表性的深度学习光流计算方法进行综合对比分析.实验结果表明,该方法具有更高的光流计算精度,有效解决了光流计算的边缘模糊问题.     

基于边缘特征点对对齐度的图像配准方法

针对基于特征的图像配准方法存在特征提取的多样性和相似度计算的复杂性等问题,在定义边缘特征点对的角度直方图和对齐度的基础上,提出了一种基于边缘特征点对对齐度的图像配准方法。该方法首先利用小波多尺度积准确地提取边缘图像和特征点,然后根据特征点的角度直方图得到的旋转角度,并通过计算所有特征点对在边缘图像中的对齐度来精确地确定匹配点对。大量的实验结果表明,该方法具有较强的适用性、精确性和有效性。     

基于自适应四元数奇异值分解的图像拼接检测

图像拼接将来源不同的图像合并成一幅图,由此引起图像中光照方向、噪声等特性出现不一致的情况。目前多数方法根据拼接图像中噪声的不一致性来检测伪造区域,但是普遍对不同大小图像块的噪声估计准确性不高,导致真阳性率较低,且当噪声差异较小时会检测失败。针对该问题,提出一种基于自适应四元数奇异值分解（QSVD）的噪声估计方法。对图像进行超像素分割,利用自适应QSVD估计超像素的噪声,结合图像亮度并利用多项式拟合建立图像噪声-亮度函数,得到各超像素到该函数曲线的最小距离测度。为提高检测精确率,利用色温估计算法提取超像素的色温特征,将距离测度与色温特征相融合作为最终的特征向量,利用FCM模糊聚类定位拼接区域。在Columbia IPDED拼接图像数据集上进行实验,结果表明,该方法在未经后处理图像集上的检测TPR值较对比方法至少提升8.21个百分点,且对高斯模糊、JPEG压缩和伽马校正表现出较好的鲁棒性。     

Enhanced SSD with interactive multi-scale attention features for object detection

Single Shot MultiBox Detector (SSD) method using multi-scale feature maps for object detection, showing outstanding performance in object detection task. However, as a one-stage detection method, it’s difficult for SSD methods to quickly notice significant areas of objects in the image. In the SSD network structure, feature maps of different scales are used to independently predict object, and there is a lack of interaction between low-level feature maps and high-level feature maps. In this paper we propose an enhanced SSD method using interactive multi-scale attention features (MA-SSD). Our method uses the attention mechanism to generate attention features of multiple scales and adds it to the original detection branch of the SSD method, which effectively enhances the feature representation ability and improves the detection accuracy. At the same time, the feature of different detection scales interacts with each other, and all the detection branches in our method have a parallel structure, which ensures the detection efficiency. Our proposed method achieves competitive performance on the public dataset PascalVOC.

     

多尺度卷积特征融合的SSD手势识别算法

为了提高对中小占比手势识别的准确性与稳定性,提出了一种多尺度卷积特征融合的SSD(single shot multibox detector)手势识别方法。该方法突出表现在两大方面,其一,在原始的SSD算法的多尺度卷积检测方法基础上,引入了不同卷积层的特征融合思想,经过空洞卷积下采样操作与反卷积上采样操作,实现网络结构中的浅层视觉卷积层与深层语义卷积层的融合,代替原有的卷积层用于手势识别,以提高模型对中小目标手势的识别精度;其二,为了解决正负样本不均衡导致分类性能差的问题,提出一种改进的损失函数,以提升模型对目标手势的分类能力。在手势识别公开的数据集上的实验结果表明,与SSD和Faster R-CNN等识别方法相比,能够在保持较高的手势检测精度的同时,又具有较好的鲁棒性与检测速度。     

基于YOLO算法的眼底图像视盘定位方法

视盘的各个参数是衡量眼底健康状况和病灶的重要指标,视盘的检测和定位对于观察视盘的形态尤为重要。在以往的视盘定位研究中,主要根据视盘的形状、亮度、眼底血管的走向等特征使用图像处理的方法对眼底图像中视盘进行定位。由于人为因素影响较大,特征提取时间较长,且视盘定位效率低,因此提出一种基于YOLO算法的眼底图像视盘定位方法。利用YOLO算法将眼底图像划分为N×N的格子,每个格子负责检测视盘中心点是否落入该格子中,通过多尺度的方式和残差层融合低级特征对视盘进行定位,得到不同大小的边界框,最后通过非极大抑制的方式筛选出得分最高的边界框。通过在3个公开的眼底图像数据集(DRIVE、DRISHTI-GS1和MESSIDOR)上,对所提出的视盘定位方法进行测试,定位准确率均为100%,实验同时定位出视盘的中心点坐标,与标准中心点的平均欧氏距离分别为22.36 px、2.52 px、21.42 px,验证了该方法的准确性和通用性。     

Logo detection with extendibility and discrimination

Logos are specially designed marks that identify goods, services, and organizations using distinguished characters, graphs, signals, and colors. Identifying logos can facilitate scene understanding, intelligent navigation, and object recognition. Although numerous logo recognition methods have been proposed for printed logos, a few methods have been specifically designed for logos in photos. Furthermore, most recognition methods use codebook-based approaches for the logos in photos. A codebook-based method is concerned with the generation of visual words for all the logo models. When new logos are added, the codebook reconstruction is required if effectiveness is a crucial factor. Moreover, logo detection in natural scenes is difficult because of perspective tilt and non-rigid deformation. Therefore, this study develops an extendable, but discriminating, model-based logo detection method. The proposed logo detection method is based on a support vector machine (SVM) using edge-based histograms of oriented gradient (HOGE) as features through multi-scale sliding window scanning. Thereafter, anti-distortion affine scale invariant feature transform (ASIFT) is used for logo verification with constraints on the ASIFT matching pairs and neighbors. The experimental results using the public Flickr-Logo database confirm that the proposed method has a higher retrieval and precision accuracy compared to existing model-based methods.

     

基于特征点类别可分性判断准则的图像分类

针对图像分类特征点特性界定模糊,导致相似性度量误差较大的问题,提出采用特征点类别可分性判断准则的图像分类方法。结合信息熵理论提取图像特征点的可分性特性,根据图像特征向量标识决策属性的不同性质,计算特征向量间的可分性距离值,得到最近邻特征向量集,从待分图像各特征向量与最近邻特征向量集标识类别的平均距离,及平均可分性度量值两方面定义新的图像类别判断准则。理论分析与Caltech256图像库仿真实验表明,基于特征点类别可分性判断准则有效地提高了图像的分类准确率。     

快速优化筛选多尺度矩形域的二进制描述

目的 为更好地兼顾基于手动设置的二进制特征描述子优越的实时性能和基于优化学习的二进制特征描述子鲁棒的区分性能,提出一种快速优化筛选多尺度矩形域的二进制描述算法(MRFO),运用于识别卫星装配时所需的典型工件目标。方法 按像素的灰度值和梯度方向划分图像并利用不同的高斯核函数进行平滑,建立多尺度的子图像集合;从多尺度的子图像中,快速通过约束条件提取候选矩形域;在训练阶段,通过优化学习计算候选矩形域的相关得分及最优阈值,筛选出其中具有强区分性和低相关性的集合;在测试阶段,计算筛选出的矩形域响应值并利用最优阈值进行二值化,将结果依次串联构成二进制描述向量。结果 实验通过ROC曲线图和80%精确率条件下的召回率统计结果证明MRFO描述算法具有优越的区分性能,平均的精确度能够高出对比算法8%~12%;并在真实的视频图像中利用MRFO描述算法识别出典型工件目标;根据训练阶段的执行时间只有传统优化学习算法的4.35%,只是在测试阶段略高于手动设置的二进制描述算法,证明MRFO描述算法具有优良的实时性能。结论 MRFO描述算法能够更好地克服各种视角、尺度和旋转变换的干扰以及周围相似背景信息的影响,准确识别出典型工件目标,有助于提高卫星的地面装配精度和效率,改善国内相关行业的自动化水平。普遍适用性较强,具有良好的应用前景。