深度特征融合的非刚性三维形状局部匹配

针对海量、异构三维形状匹配与智能检索技术的需求,提出了一种基于级联卷积神经网络(F-PointCNN)深度特征融合的三维形状局部匹配方法.首先,采用特征袋模型,提出几何图像表示方法,该几何图像不仅能够有效区分同类异构的非刚性三维模型,而且能够揭示大尺度不完整三维模型的结构相似性.其次,构建级联卷积神经网络学习框架F-P...     

Hierarchical aggregation for efficient shape extraction

This paper presents an efficient framework which supports both automatic and interactive shape extraction from surfaces. Unlike most of the existing hierarchical shape extraction methods, which are based on computationally expensive top-down algorithms, our framework employs a fast bottom-up hierarchical method with multiscale aggregation. We introduce a geometric similarity measure, which operates at multiple scales and guarantees that a hierarchy of high-level features are automatically found through local adaptive aggregation. We also show that the aggregation process allows easy incorporation of user-specified constraints, enabling users to interactively extract features of interest. Both our automatic and the interactive shape extraction methods do not require explicit connectivity information, and thus are applicable to unorganized point sets. Additionally, with the hierarchical feature representation, we design a simple and effective method to perform partial shape matching, allowing efficient search of self-similar features across the entire surface. Experiments show that our methods robustly extract visually meaningful features and are significantly faster than related methods.     

结合可变形卷积与双边网格的立体匹配网络

双目立体匹配被广泛应用于无人驾驶、机器人导航、增强现实等三维重建领域。在基于深度学习的立体匹配网络中采用多尺度2D卷积进行代价聚合,存在对目标边缘处的视差预测鲁棒性较差以及特征提取性能较低的问题。提出将可变形卷积与双边网格相结合的立体匹配网络。通过改进的特征金字塔网络进行特征提取,并将注意力特征增强、注意力机制、Meta-ACON激活函数引入到改进的特征金字塔网络中,以充分提取图像特征并减少语义信息丢失,从而提升特征提取性能。利用互相关层进行匹配计算,获得多尺度3D代价卷,采用2D可变形卷积代价聚合结构对多尺度3D代价卷进行聚合,以解决边缘膨胀问题,使用双边网格对聚合后的低分辨率代价卷进行上采样,经过视差回归得到视差图。实验结果表明,该网络在Scene Flow数据集中的端点误差为0.75,相比AANet降低13.8%,在KITTI2012数据集中3px的非遮挡区域误差率为1.81%,能准确预测目标边缘及小区域处的视差。     

Object-aware deep feature extraction for feature matching

Feature extraction is a fundamental step in the feature matching task. A lot of studies are devoted to feature extraction. Recent researches propose to extract features by pre-trained neural networks, and the output is used for feature matching. However, the quality and the quantity of the features extracted by these methods are difficult to meet the requirements for the practical applications. In this article, we propose a two-stage object-aware-based feature matching method. Specifically, the proposed object-aware block predicts a weighted feature map through a mask predictor and a prefeature extractor, so that the subsequent feature extractor pays more attention to the key regions by using the weighted feature map. In addition, we introduce a state-of-the-art model estimation algorithm to align image pair as the input of the object-aware block. Furthermore, our method also employs an advanced outlier removal algorithm to further improve matching quality. Experimental results show that our object-aware-based feature matching method improves the performance of feature matching compared with several state-of-the-art methods.     

基于SIFT算法的目标特征检测与提取技术研究

验证了一种能够在不同图像之间进行同一个物体相匹配的方法,具有很强的可靠性,称之为SIFT算法(尺度不变特征变换).SIFT算法能够处理图像间发生的尺度变换、旋转、很大范围内的仿射形变、视角变换、噪声以及光照变换.它的功能十分强大,甚至可以仅仅根据一个简单的物体特征,在一个大型数据库中的许多高品质图像中进行相应目标的寻找...     

图像匹配方法研究综述

目的 图像匹配作为计算机视觉的核心任务,是后续高级图像处理的关键,如目标识别、图像拼接、3维重建、视觉定位、场景深度计算等。本文从局部不变特征点、直线、区域匹配3个方面对图像匹配方法予以综述。方法 局部不变特征点匹配在图像匹配领域发展中最早出现,对这类方法中经典的算法本文仅予以简述,对于近年来新出现的方法予以重点介绍,尤其是基于深度学习的匹配方法,包括时间不变特征检测器（TILDE）、Quad-networks、深度卷积特征点描述符（DeepDesc）、基于学习的不变特征变换（LIFT）等。由于外点剔除类方法常用于提高局部不变点特征匹配的准确率,因此也对这类方法予以介绍,包括用于全局运动建模的双边函数（BF）、基于网格的运动统计（GMS）、向量场一致性估计（VFC）等。与局部不变特征点相比,线包含更多场景和对象的结构信息,更适用于具有重复纹理信息的像对匹配中,线匹配的研究需要克服包括端点位置不准确、线段外观不明显、线段碎片等问题,解决这类问题的方法有线带描述符（LBD）、基于上下文和表面的线匹配（CA）、基于点对应的线匹配（LP）、共面线点投影不变量法等,本文从问题解决过程的角度对这类方法予以介绍。区域匹配从区域特征提取与匹配、模板匹配两个角度对这类算法予以介绍,典型的区域特征提取与匹配方法包括最大稳定极值区域（MSER）、基于树的莫尔斯区域（TBMR）,模板匹配包括快速仿射模板匹配（FAsT-Match）、彩色图像的快速仿射模板匹配（CFAST-Match）、具有变形和多样性的相似性度量（DDIS）、遮挡感知模板匹配（OATM）,以及深度学习类的方法MatchNet、L2-Net、PN-Net、DeepCD等。结果 本文从局部不变特征点、直线、区域3个方面对图像匹配方法进行总结对比,包括特征匹配方法中影响因素的比较、基于深度学习类匹配方法的比较等,给出这类方法对应的论文及代码下载地址,并对未来的研究方向予以展望。结论 图像匹配是计算机视觉领域后续高级处理的基础,目前在宽基线匹配、实时匹配方面仍需进一步深入研究。     

普通话发音训练多级音素模板综合评价方法

提出了一种应用于普通话声韵母发音评价的多级音素模板综合评分法,该方法在单模板匹配的基础上,通过寻找汉语音素发音过程中的浊化、摩擦、爆破等特征,形成多个子模式,进行多模式匹配,最后给出加权评分。实验结果表明多级音素模板综合评分法有效地改善了汉语发音中几组相似音之间的区分度。也介绍了MFCC等语音特征提取、DTW模板匹配算法、基于聚类的模板训练以及综合加权评价机制的实现方法。     

Structure-oriented contour representation and matching for engineering shapes

Conventional shape matching for engineering models primarily considers rigid shape similarity. They do not seek global shape similarity while considering large local deformations. However, engineering models created by some parametric-based design can involve large parametric changes. As a result, they do not share similarity in their global shape. Hence our goal is to develop shape representations for global matching of part models that can have large dissimilarity through stretching and/or bending.This paper presents a strategy of an integrated shape matching for contours of engineering drawings inspired by the divide-and-conquer paradigm. The original shape is decoupled into two levels of shape representations namely, higher-level structure and lower-level geometry. The higher-level structure matching is then achieved driven by optimal integrated solutions from matching of lower-level local geometry. Feature points are first extracted using curve evolution to attain the two levels of representations. In order to suit engineering semantics, a new significance function for a point is defined to suppress small features using discrete curve evolution. To conduct the integrated shape matching, a mechanism of using lookup tables is employed to associate these two levels of representations. Dynamic Time Warping and Elastic Matching are employed at different levels of shape representations in order to achieve the optimal integration.To demonstrate the advantages of the proposed work for engineering shapes, experiments for contour evolution, feature point registration, and shape-based similarity for retrieval are conducted. They are also compared with the existing methods. The experimental results show that the structure-oriented contour representation and matching are more meaningful and consistent from an engineering perspective.     

结合空间结构卷积和注意力机制的三维点云分类网络

目的 三维点云分类作为一项关键任务,在计算机视觉、机器人和自动驾驶等领域有着广泛的应用场景。现有的三维点云分类网络在使用边卷积进行局部特征提取时通常存在输入特征差异性小,空间结构信息提取、融合不充分等问题。针对上述问题,设计了一种结合空间结构卷积和注意力机制的点云分类网络。方法 首先,提出一种空间结构卷积,在边卷积的基础上引入邻接点之间的相对位置信息来降低输入特征相似性,而后从结构和位置两个角度分别进行特征编码,实现更具多样性的局部几何结构捕获。其次,设计了全局特征编码模块,从坐标信息中提炼全局特征信息,同时在网络中融合了注意力机制,用于关联局部和全局特征表示,有效保留了全局特征信息,实现全局特征的适应性调整。最后,将局部几何结构信息和全局位置信息进行有效的融合,获得更具代表性和差异性的特征表征。结果 设计实验在公开数据集ModelNet40上对提出的网络模型的性能进行评估,点云分类总体准确率和平均准确率分别达到93.0%和89.7%,具备良好的分类性能和预测效率。实验结果表明,空间结构卷积的使用有效增加了输入特征的多样性,位置和结构的单独编码有效提高了局部特征的表达能力。同时,提出的注意力加权方式在保留全局特征前提下实现了局部特征和全局特征的关联。结论 提出的网络有较强的细粒度特征提取能力,具有良好的分类性能。     

基于稀疏图像序列的雕塑点自动云三维重构方法

为了提高对雕塑点稀疏图像的点云三维重建的分析能力,提出一种基于稀疏图像序列的雕塑点自动云三维重构方法,基于稀疏散乱点三维重建和锐化模板特征匹配方法进行图像三维重建。采用三维角点检测和边缘轮廓特征提取方法,进行雕塑点稀疏图像三维点云特征检测,对检测的雕塑点稀疏图像点云数据进行信息融合处理,采用梯度运算方法进行特征分解,实现对雕塑点稀疏图像的信息增强和融合滤波。结合局部均值降噪方法进行图像的提纯处理,提高雕塑点稀疏图像轮廓重建能力,采用锐化模板特征匹配和块分割技术,实现雕塑点自动云三维重构。仿真结果表明,采用该方法进行雕塑点自动云三维重构的准确性较高,图像匹配能力较好,且重构输出信噪比较高。     

基于深度学习SuperGlue的枪弹痕迹自动识别方法

为实现枪弹痕迹自动匹配的准确性,提出了将基于深度学习的SuperPoint特征提取和SuperGlue匹配算法引入枪弹痕迹自动识别研究。通过SuperPoint网络提取弹痕图像特征点位置与描述子向量;研究了SuperGlue的匹配机制,包括注意力机制的图神经网络(GNN)及优化匹配层,将提取的弹底窝痕迹的特征点和描述子使用SuperGlue算法进行匹配。实验表明SuperPoint特征通过SuperGlue匹配,相较于机器学习算法实现了更高的匹配准确度,正确匹配数量提高,为枪支鉴定增加科学性。     

Real-Time 3D Motion Tracking with Known Geometric Models

In this paper a new model-based tracking algorithm is proposed for real-time performance. The matching process includes two aspects of: (1) feature extraction using local minimum energy and (2) global matching of known 3D models against the projected features. The algorithm is robust to change in lighting and background. The small motion hypothesis was used for fitting of feature energy which is defined as the negative absolute value of the edge strength. An autoregressiveAR(1) model is employed for detecting incorrect matches in terms of the feature energy. We have found a new invariance-based method to eliminate false matches caused by strong shadow or occlusion. The invariance is the ratio of trigonometric functions of the angles formed by a polygon. In order to calculate the vertices of the object surface in an image, regression technique in terms of matched features is efficient in our approach. A linear least squares method and the orthonormal rotation matrix are used for motion estimation and pose update of the six degrees of freedom. Also, an Extended Kalman Filter is introduced to guarantee a smooth motion estimation and prediction.     

基于双目被动立体视觉的三维人脸重构与识别

提出一种基于双目被动视觉的三维人脸识别方法, 该方法采用非接触式的人脸信息采集技术, 利用图像中弱特征检测方法实现双目视觉中的人脸检测与初步视差估计, 运用基于复小波的相位相关技术对人脸表面进行亚像素级小区域匹配, 重建人脸三维点云信息. 通过可调训练次数的神经网络技术实现多层次人脸曲面重建, 并结合人脸2D图像对重构曲面进行仿射归一, 继而迭代地进行特征提取与识别过程. 实验结果表明, 双目视觉方法使人脸信息采集过程友好隐蔽; 在对应点匹配中, 运用复小波的相位相关算法可获得密集的亚像素精度配准点对, 用神经网络方法可正确重建人脸曲面. 识别过程对环境以及人脸位姿表情等鲁棒性强. 该系统成本十分低廉, 适合在许多领域推广应用.     

Robust palmprint verification using 2D and 3D features

This paper presents a new personal authentication system that simultaneously exploits 2D and 3D palmprint features. The objective of our work is to improve accuracy and robustness of existing palmprint authentication systems using 3D palmprint features. The proposed multilevel framework for personal authentication efficiently utilizes the robustness (against spoof attacks) of the 3D features and the high discriminating power of the 2D features. The developed system uses an active stereo technique, structured light, to simultaneously capture 3D image or range data and a registered intensity image of the palm. The surface curvature feature based method is investigated for 3D palmprint feature extraction while Gabor feature based competitive coding scheme is used for 2D representation. We comparatively analyze these representations for their individual performance and attempt to achieve performance improvement using the proposed multilevel matcher that utilizes fixed score level combination scheme to integrate information. Our experiments on a database of 108 subjects achieved significant improvement in performance with the integration of 3D features as compared to the case when 2D palmprint features alone are employed. We also present experimental results to demonstrate that the proposed biometric system is extremely difficult to circumvent, as compared to the currently proposed palmprint authentication approaches in the literature.     

基于多尺度分析的三维曲线匹配技术研究

从提取三维物体碎片轮廓曲线出发,提出一种基于多尺度分析的三维曲线匹配技术。轮廓曲线经多尺度平滑后,计算曲线的特征矢量,通过比较特征矢量以判断轮廓曲线的相似性,实现三维曲线匹配。实验表明提出的算法具有准确性、鲁棒性和容错性。     

多种匹配模式在下视景象匹配中的应用

本采用多种匹配模式进行下视景象的匹配，利用小波分解生成多分辨率影像、提取边缘特征作为匹配的基础，同对中改进了一种小波兴趣算子来生成兴趣特征图像，并用于初匹配。通过试验及统计，分析了不司匹配模式的效果，以及多种模式相结合对下视景象匹配的改善。     

耳廓点云形状特征匹配的路径跟随算法

路径跟随算法结合凸松弛方法与凹松弛方法,通过跟随凸凹问题的解路径,近似地求解图匹配问题,具有较高的匹配精度.将路径跟随算法用于耳廓特征图的匹配问题:首先,基于PCA方法构造耳廓点云的显著性关键点集合;然后,采用乘积型参数域上的单值二次曲面方法拟合关键点邻域内的点集,并将曲面的局部形状特征定义为耳廓的局部形状相似测度;第三,对关键点集合进行Delaunay三角剖分,得到关键点集合在三维空间内的拓扑结构图,并定义关键点图的整体结构差异测度;最后,记耳廓关键点图的组合差异测度为关键点图的整体结构差异测度与关键点上的局部形状相似测度的线性组合,并基于路径跟随算法快速求解关键点图之间的精确匹配.相关实验结果表明:与其他相关算法相比,该算法具有较高的匹配效率和匹配精度.