LSVSH描述符:高鉴别强鲁棒的点云局部特征统计直方图

三维局部特征描述是三维计算机视觉中的重要任务.现实场景中包含噪声、遮挡和杂波等干扰,使得准确和鲁棒的三维局部特征描述具有很大的挑战性.为提高特征描述的性能,提出一种局部曲面变化统计直方图(local sur-face variation based statistics histogram,LSVSH)描述符.首先设计一种不依赖于局部参考轴(local reference axis,LRA)的新属性(称为曲率属性),增强描述符对LRA误差的稳健性;然后沿径向剖分局部空间,在每个子空间中统计3个角度属性和1个曲率属性生成LSVSH描述符,实现对局部曲面信息的全面稳健描述.在B3R,U3M,U3OR和QuLD这4个数据集上进行大量的实验,结果表明,LSVSH在4个数据集上的RPC下面积(the area under the recall-precision curve,AUCpr)值分别为0.95,0.70,0.54和0.10,优于现有的局部特征描述符的性能;在U3M数据集上的正确配准率和在U3OR数据集上的正确识别率分别达到70%和100%,验证了LSVSH应用于物体配准和识别任务上的有效性.

基于关键特征增强机制的3D人脸识别

3D人脸识别是计算机视觉领域的重要组成部分,Pointnet依靠深度学习解决了点云的无序性,实现了3D点云的全局特征提取,但由于点云数据缺乏细节纹理,仅靠全局特征很难实现复杂情况下的人脸识别。针对以上问题,基于Pointnet提出了一种局部特征描述子,用于描述点云局部空间的几何特征,并引入关键特征增强机制,通过特征概率分布增强人脸关键信息,该机制能减少不必要特征对任务的干扰,有效提升模型的准确率。在公共数据集CASIA-3D、Lock3DFace、Bosphorus上进行实验测试,结果表明该方法能很好地应对表情变化、部分遮挡以及头部姿态的干扰,在弱光环境下其准确率高于RP-Net 1.1%,并具有良好的实时性。

An Angle Structure Descriptor for Image Retrieval

This paper presents an efficient image feature representation method, namely angle structure descriptor(ASD), which is built based on the angle structures of images. According to the diversity in directions, angle structures are defined in local blocks. Combining color information in HSV color space, we use angle structures to detect images. The internal correlations between neighboring pixels in angle structures are explored to form a feature vector. With angle structures as bridges, ASD extracts image features by integrating multiple information as a whole, such as color, texture, shape and spatial layout information. In addition, the proposed algorithm is efficient for image retrieval without any clustering implementation or model training. Experimental results demonstrate that ASD outperforms the other related algorithms.

基于局部特征和视觉词袋模型的大豆叶片病害识别

病害检测对提高大豆作物产量至关重要。针对传统视觉法诊断大豆作物病害而导致病害识别效率和分类准确率不高的问题,提出一种基于局部描述符和视觉词袋技术以数据表征大豆叶片图像的分类算法,同时保留有关潜在疾病的视觉信息。采用SIFT、DSIFT、PHOW和SURF 4种算法对大豆叶片的霜霉病、锈病TAN和锈病RB进行分类识别。结果表明,局部描述符PHOW表现出最佳的分类识别结果,其正确分类率为96.25%。进一步研究PHOW在不同颜色空间下的大豆病害识别效果。结果表明,与灰度图像相比,使用HSV、Opponent颜色空间可有效提升对大豆叶片病害检测的正确分类率,其正确分类率分别可达99.83%和99.58%,验证采用局部描述符和视觉词袋技术识别大豆叶片病害方法的可行性和高效性,并为其他作物的病害识别提供一种通用的分类识别方法。

基于YOLOv5和少样本学习的高空作业安全带检测

针对电厂高空作业人员安全带佩戴检测问题,现有研究大多利用深度检测模型直接检测,不仅需要大量样本训练模型,而且由于高空作业背景杂乱、人员目标小,导致不易检测等。为此,提出一种基于目标检测和少样本细粒度分类的两阶段检测方法:首先利用YOLOv5检测视频图像中的高空作业人员,再用少样本细粒度分类方法识别其是否佩戴安全带。针对佩戴和不佩戴安全带人员的细微差别,设计了一种基于局部描述符的少样本度量学习模型,在公用数据集预训练模型基础上,利用少量训练样本对模型微调,用于安全带佩戴识别。实验结果表明,在支持集图像数为60时,识别精度达到了97.86%。所提方法可实现少样本情况下对高空作业人员安全带佩戴情况的精确检测。

一种高效的二进制点云局部特征描述算法

针对现有点云局部特征描述算法时效性和紧凑型不能满足实际应用需求的问题,提出一种二进制多层切片距离特征描述算法(B-MSD)。首先在特征点处建立了一种稳健的局部参考坐标系;然后设计了两种多层切片和子区域划分方式并计算平均距离,以提取点云局部三维形状特征;最后通过量化的方法将各子区域计算的平均距离转化为二进制码串,串联所有码串生成最终的二进制描述子。在多个公开点云数据集上与几个经典描述子进行仿真对比,结果表明,在鉴别力更强的基础上,上述算法的紧凑性、计算效率和匹配速度方面综合表现更好。3D目标检测和点云配准应用实验也验证了上述算法的有效性。

Matching DSIFT Descriptors Extracted from CSLM Images

The matching of local descriptors represents at this moment a key tool in computer vision, with a wide variety of methods designed for tasks such as image classification, object recognition and tracking, image stitching, or data mining relying on it. Local feature description techniques are usually developed so as to provide invariance to photometric variations specific to the acquisition of natural images, but are nonetheless used in association with biomedical imaging as well. It has been previously shown that the matching of gradient based descriptors is affected by image modifications specific to Confocal Scanning Laser Microscopy (CSLM). In this paper we extend our previous work in this direction and show how specific acquisition or post-processing methods alleviate or accentuate this problem.

Novel 3D local feature descriptor of point clouds based on spatial voxel homogenization for feature matching

Obtaining a 3D feature description with high descriptiveness and robustness under complicated nuisances is a significant and challenging task in 3D feature matching.This paper proposes a novel feature description consisting of a stable local reference frame(LRF)and a feature descriptor based on local spatial voxels.First,an improved LRF was designed by incorporating distance weights into Z-and X-axis calculations.Subsequently,based on the LRF and voxel segmentation,a feature descriptor based on voxel homogenization was proposed.Moreover,uniform segmentation of cube voxels was performed,considering the eigenvalues of each voxel and its neighboring voxels,thereby enhancing the stability of the description.The performance of the descriptor was strictly tested and evaluated on three public datasets,which exhibited high descriptiveness,robustness,and superior performance compared with other current methods.Furthermore,the descriptor was applied to a 3D registration trial,and the results demonstrated the reliability of our approach.

面向小样本学习的双重度量孪生神经网络

为了解决孪生神经网络因使用图像级特征度量,存在对位置、复杂背景及类内差异比较敏感的问题,提出了一种双重度量孪生神经网络(DM-SiameseNet)。具体来说,DM-SiameseNet使用图像级的特征和局部特征(局部描述符)共同表示每幅图像,然后根据两种不同级别的相似度度量学习特征映射,最后使用自适应融合策略,自适应整合两种不同级别特征表示的度量结果。实验结果表明,改进后模型的准确率分别提高了5.04%和9.66%,并且在miniImageNet、TieredImageNet、Stanford Dogs、Stanford Cars和CUB-200数据集上高于只使用图像级的特征表示或者局部描述符表示的度量方法。实验结果证明所提出的模型不仅考虑了图像的全局特征,还可以捕获图像中更多有效的局部特征信息,提高了模型的泛化能力。

RANDOM FOREST FOR INTERMEDIATE DESCRIPTOR FUSION IN SHOT BOUNDARY DETECTION

Shot boundary detection is the fundamental part in many real applications as video retrieval and so on. This paper tackles the problem of video segment obtaining in complex movie videos. Firstly, intermediate descriptor is proposed to depict the variation of both abrupt and gradual change in shot boundaries, which is formed by distance vector on Local Binary Pattern(LBP), GIST(GIST) or their fusion. Instead of just using the adjacent frames distance, intermediate descriptor keeps the distances between current frame and consecutive frames. It comprehensively characterizes local temporal structure, which is especially important for gradual change. For the excellent ability for feature fusion in random forests, it is adopted here to verify the fusion effect of intermediate descriptor on LBP and GIST. The whole experiments are designed on the subset of TRECVid 2013 INS(INstance Search) task to verify the effectiveness of proposed intermediate descriptor and the fusion ability for random forest. Compared with static and adaptive thresholds approaches, the best performance can be achieved by post-fusion of intermediate descriptor on LBP and GIST.

GNSS失锁下融合描述子与粒子滤波的LiDAR定位方法研究

在GNSS失锁情况下,如何基于三维激光雷达进行定位,是值得关注的问题。通过点云匹配或蒙特卡洛定位(MCL)的方法进行定位,存在初值敏感、计算时间长和搜索空间大的问题。本文提出了一种定位方法,首先利用描述子初始化,确定粗略的位姿,缩小搜索空间,然后使用MCL进行精定位。通过给定描述子初值,预先计算三维网格概率,达到了缩小搜索空间、减少计算时间的效果。初步的试验结果表明,该方法在GNSS失锁情况下定位精度保持较好。

多层次特征和粒子群优化的场景分类

针对遥感图像的场景分类精度问题,提出多层次特征和粒子群算法优化分类器的场景分类算法。利用聚集局部描述符编码算法对尺度不变特征变换算法提取的局部特征编码,获得中层特征,通过卷积神经网络提取高层特征,将提取的特征作为支持向量机的输入数据,引入粒子群算法优化该分类器的参数,进行场景分类。在RSC11和WHU-RS19两个公开的遥感图像数据集上进行实验,分类精度分别达到95.28和97.20。将WHU-RS19数据集的结果与其它方法比较,精度有明显提高。实验结果表明,在分类时对分类器参数进行优化,分类效果更佳。

改进AKAZE算法的泥石流区无人机影像特征匹配

针对在泥石流区灾害应急中使用无人机高分辨率影像特征匹配时时效性较低的问题,本文提出了一种改进AKAZE无人机影像特征匹配的算法。该算法首先使用AKAZE特征点检测算法提取局部稳定不变特征,用二进制描述符BEBLID描述检测到的特征点,采用最近邻次近邻距离比(NNDR)完成初步匹配;然后采用核线几何约束计算变换矩阵,达到内点提纯、提高匹配质量的目的;最后选取5组同一无人机序列影像进行特征匹配试验,分别与经典SIFT算法、AKAZE算法、ORB算法进行比较。试验结果表明,该方法的匹配准确率与SIFT算法接近,略高于AKAZE算法,明显优于ORB算法,计算速度明显优于SIFT算法和AKAZE算法,基本达到ORB算法的计算效率。本文方法能较好地应用于对匹配精度和匹配时效均要求较高的泥石流场景无人机影像数据处理中。

一种快速变电站设备三维点云识别方法

针对变电站设备点云识别的问题,提出一种新的识别方法。首先,建立设备点云的局部坐标系,利用设备点云在空间中的对称性和分布密度,确定局部坐标系的x、y轴,该坐标系具有平移和旋转不变性,并且对噪声鲁棒。其次,结合变电站设备点云的形状和设备视图的差异,定义一种新的特征描述子,用于设备点云的描述和识别,并建立一个包含避雷器、断路器、隔离开关等54种电气设备的模板库,库中包含了每个模板设备的型号、编号以及模板设备的特征描述子信息。通过计算待识别的设备点云的特征描述子,将其和模板库中的特征描述子进行匹配,找到匹配误差最小的模板,完成对设备点云的识别。最后,在90个待识别设备点云上对本文算法和另外两种变电站点云识别算法进行验证,结果表明:本文算法可以达到90%的识别准确率,识别一个设备的平均时间为3.2 s,可以较好地平衡识别准确率和识别效率,并且当待识别点云中存在噪声和遮挡时,本文算法的识别准确率略高于另外两种算法,当待识别点云密度不均匀时,本文算法仍然能保持70%以上的识别准确率。

基于四叉树的ORB-LBP改进算法

针对ORB算法存在图像分布不均匀、匹配程度不高、匹配精度差的问题,通过划分网格计算图像灰度值的方法计算角点提取阈值。通过在金字塔层上构建四叉树的方法,在不同金字塔层分别构建不同深度的四叉树以提高计算效率,最后融合BRIEF-LBP特征描述子以提升ORB算法匹配精度。实验结果表明:对比传统ORB算法在速度上降低了5%,但均匀度提升了66左右,召回率也提升了10%;对比其他改进算法,速度提升了2%和5%,特征点分布均匀度提升了48和49,召回率也提升了36.63%和4.925%,实现了在少量增加计算量的同时,特征点均匀度和匹配精度效果有较大提升。

结合区域全局特征和Ann-SIFT的二阶段快速景象匹配算法

在GPS拒止环境下,提出了一种基于高清航拍图像的视觉定位方法。为了保证高分辨率图像匹配的实时性,结合粗粒度高级语义全局特征和细粒度点特征实现两阶段快速景象匹配。第一阶段,设计三元组网络针对图像区域提取具有尺度和旋转不变性的全局描述符进行区域预匹配。第二阶段,快速最近邻点搜索的Ann-SIFT对实时图和参考区域进行精确点匹配以实现定位。对于3 000×3 000的参考图像和500×500到1 000×1 000尺寸范围的实时图像,平均耗时可以从66.24 s降低到0.97 s。

基于RFNA和改进LBD的镜像线特征匹配方法

针对物体和镜像之间的匹配问题,引入RNFA(Relative Number of False Alarms)边缘链检测方法获取更丰富的线段。文中提出一种改进的LBD(Line Band Descriptor)算法用于构建局部不变特征描述符,通过比较局部不变特征描述符获得初始匹配对。采用全局投影角度的筛选方式,并通过拟合投影中线的方式剔除初始匹配对中误匹配项。在完成全局投影角度的选取和投影中线的拟合后,放宽对局部不变特征描述符阈值的筛选以获得更多的匹配对,提升召回率。图像集仿真实验结果表明,文中所提算法在纹理较弱区域能够更好地识别线段,且能够在保证原算法性能的基础上获得更多的匹配对,提高5%左右的正确匹配率,并达到90%以上的召回率。

Supervised Feature Learning for Offline Writer Identification Using VLAD and Double Power Normalization

As an indispensable part of identity authentication,offline writer identification plays a notable role in biology,forensics,and historical document analysis.However,identifying handwriting efficiently,stably,and quickly is still challenging due to the method of extracting and processing handwriting features.In this paper,we propose an efficient system to identify writers through handwritten images,which integrates local and global features from similar handwritten images.The local features are modeled by effective aggregate processing,and global features are extracted through transfer learning.Specifically,the proposed system employs a pre-trained Residual Network to mine the relationship between large image sets and specific handwritten images,while the vector of locally aggregated descriptors with double power normalization is employed in aggregating local and global features.Moreover,handwritten image segmentation,preprocessing,enhancement,optimization of neural network architecture,and normalization for local and global features are exploited,significantly improving system performance.The proposed system is evaluated on Computer Vision Lab(CVL)datasets and the International Conference on Document Analysis and Recognition(ICDAR)2013 datasets.The results show that it represents good generalizability and achieves state-of-the-art performance.Furthermore,the system performs better when training complete handwriting patches with the normalization method.The experimental result indicates that it’s significant to segment handwriting reasonably while dealing with handwriting overlap,which reduces visual burstiness.

Detailed Quantum Mechanical QSAR Analysis of Certain Aminopyrimidoisoquinolinequinones with Anticancer Activity

A detailed quantum mechanical analysis of electronic disposition of five aminopyrimidoisoquinolinequinones (APIQs) was performed after extraction of this subset of compounds from a larger data set of APIQs via a reported clustering methodology (Elfaki, et al. 2020). Both semi empirical PM3 method and DFT quantum mechanical methods were used to calculate global and local quantum mechanical descriptors (QMDs) to define the electronic environment of these molecules in attempt to rationalize their observed anti-cancer response variability. The biological response is the anticancer activity against human gastric adenocarcenoma (AGS) cell line. The correlation matrix between the calculated global electronic descriptors and biological activity demonstrated that the global dipole moment gives the highest correlation. The local electronic environment was analysed by The Mullikan charges (MC) and Fukui functions for N-5, C-6, C-8 in addition to the N atom of phenylamino side group at C-8. MCs furnished no useful information as each of these atoms had almost identical MC values for all the five compounds with exception of C-6 which gave varied values. Regressing MCs of C-6 against the response traces 60% of the latter variability. As C-6 is an extra annular methyl carbon adjacent to N-5 in isoquinoline residue of APIQ, we reasoned that the chemical reactivities of 4 out of the 5 APIQs might be due to a Chichibabin-type tautomerism implying a possible alkylation aspect in their mechanism of action. The corresponding Fukui functions (f-, f+ and f0) showed a considerable consistency with the patterns of chemical reactivity exhibited by this small set of APIQs.

基于POS数据的无人机高精度匹配定位方法

针对无人机高精度多目标快速定位的应用需求,提出了一种基于POS数据的无人机高精度匹配定位方法。基于共线定位方法初步定位解算,获取航拍校正图像,裁剪基准匹配底图;提取SIFT特征关键点,采用改进SIFT描述符描述关键点;根据暴力匹配方法求解匹配点对,基于特征向量欧式距离、地理位置距离的二次误匹配剔除方法,充分抑制匹配点对的误匹配;最后,结合RANSAC算法解算变换矩阵,实现基准图像和航拍图像的匹配融合,并将融合图像引导至地理信息系统,获知航拍图像上任意点的位置坐标和态势信息。设计了无人机匹配定位软件,进行了测试验证,与共线定位方法相比,基于POS数据的匹配定位方法的定位精度明显提高。