WO2020248096A1 - 基于局部特征的三维人脸识别方法和系统 - Google Patents
基于局部特征的三维人脸识别方法和系统 Download PDFInfo
- Publication number
- WO2020248096A1 WO2020248096A1 PCT/CN2019/090551 CN2019090551W WO2020248096A1 WO 2020248096 A1 WO2020248096 A1 WO 2020248096A1 CN 2019090551 W CN2019090551 W CN 2019090551W WO 2020248096 A1 WO2020248096 A1 WO 2020248096A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- dimensional face
- dimensional
- extracted
- face
- face image
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
Definitions
- the present invention relates to the technical field of face recognition, in particular to a three-dimensional face recognition method and system based on local features.
- Two-dimensional face recognition technology has achieved satisfactory results since it was proposed and developed in the last century, and two-dimensional face recognition technology has been applied to various fields, but it is still affected by changes in facial posture, changes in environmental lighting, etc. influences.
- 3D face recognition is not affected by face pose changes and environmental lighting changes, it is easily affected by changes in facial expressions, and each 3D face point cloud contains tens of thousands of points, and the calculation speed is greatly affected.
- the present invention provides a three-dimensional face recognition method and system based on local features, aiming to reduce the number of points involved in calculation, improve algorithm speed, and improve robustness to changes in three-dimensional facial expressions.
- the present invention provides a three-dimensional face recognition method based on local features, characterized in that the method includes the following steps:
- the similarity between the face area to be extracted and the corresponding curve of the pre-stored three-dimensional face image is obtained according to the geodesic distance.
- the step of acquiring the to-be-extracted face region of the to-be-recognized three-dimensional face image according to the face nose tip detection algorithm based on the support vector machine includes:
- the Euclidean distance algorithm is used to smooth the recognized three-dimensional face image.
- the step of acquiring the to-be-extracted face region of the processed three-dimensional face image according to the face nose tip detection algorithm based on the support vector machine includes:
- the step of acquiring the nose tip point in the processed three-dimensional face image according to the shape index includes:
- the nose tip point is obtained according to the density of neighbor nodes of each point in the candidate points.
- the shape index is calculated according to the maximum principal curvature and the minimum principal curvature of each point in the three-dimensional point cloud in the processed three-dimensional face image, and its calculation formula is:
- SI represents the shape index
- K_(i_max) represents the maximum curvature of the i-th point
- K_(i_min) represents the minimum curvature of the i-th point.
- the step of obtaining the isogeometric contour line of the face region to be extracted includes:
- the contour line of the middle geodesic distance of the face area to be extracted is extracted.
- the step of obtaining the radial curve of the face region to be extracted includes:
- each intersection constitutes at least two radial curves.
- the present invention also provides a three-dimensional face recognition system, the system includes a memory, a processor, and a computer program stored on the memory, the computer program is executed by the processor to achieve the above The steps of the method described.
- the local feature-based three-dimensional face recognition method and system of the present invention obtain the to-be-extracted face region of the three-dimensional face image to be recognized according to the above-mentioned technical solution according to the face nose tip detection algorithm based on the support vector machine; obtain the to-be-extracted face The iso-geodetic contour line and radial curve of the region; according to the iso-geometric contour line and the radial curve, the shape analysis algorithm is used to obtain the geodesic of the corresponding curve of the face area to be extracted and the pre-stored three-dimensional face image Line distance; according to the geodesic distance, the similarity between the face area to be extracted and the corresponding curve of the pre-stored three-dimensional face image is obtained.
- the number of points involved in the calculation of the three-dimensional face recognition process is reduced Increased the speed of the algorithm and the robustness to changes in 3D facial expressions.
- FIG. 1 is a schematic flowchart of a method for three-dimensional face recognition based on local features in an embodiment of the present invention
- Fig. 2 is a flowchart of the entire face recognition of the three-dimensional face recognition method based on local features of the present invention.
- FIG. 3 is a schematic diagram of the preprocessing steps in the data preprocessing stage of the three-dimensional face recognition method based on local features of the present invention
- Figure 4 is a schematic diagram of point cloud distribution in sharp areas
- Figure 5 is a schematic diagram of geodesic curvature
- Fig. 6 is a schematic diagram of the result of isogeometric contour line detection
- Figure 7 is a schematic diagram of a radial curve
- Figure 8 is the original point cloud image
- Figure 9 is a radial curve rotated 360 degrees
- Figure 10 is a radial curve rotated 270 degrees
- Figure 11 is a radial curve rotated 180 degrees
- Figure 12 is a radial curve rotated 90 degrees
- Figure 13 is a radial curve from 270 degrees to 360 degrees
- Figure 14 is a schematic diagram showing that the point clouds of two faces (different people) are simplified into 10 equal geodesic contour lines;
- FIG. 15 is a schematic diagram of the shape analysis result of the fifth contour line of two human faces.
- the calculation method of matching three-dimensional faces with overall features can only be operated on a complete face, and it is more sensitive to lack of face data and occlusion.
- the invention proposes a solution.
- the present invention proposes a three-dimensional face recognition method and system based on local features.
- the technical solution adopted by the present invention is mainly in the data preprocessing stage.
- a support vector-based The nose point detection algorithm of the machine in view of the excessive amount of 3D face point cloud data, a method of simplifying the 3D face point cloud using isogeometric contour lines and radial curves based on the nose point point is proposed to improve the 3D point cloud
- the expression of the face reduces the number of points involved in the calculation, improves the speed of the algorithm, and improves the robustness to the changes of three-dimensional facial expressions.
- FIG. 1 is a schematic flowchart of a preferred embodiment of a method for 3D face recognition based on local features of the present invention.
- the method for 3D face recognition based on local features includes the following steps:
- Step S10 Obtain the to-be-extracted face region of the three-dimensional face image to be recognized according to the face nose tip detection algorithm based on the support vector machine.
- the execution subject of this embodiment may be, for example, a smart terminal with a face recognition function such as a mobile phone, a computer, or an IPAD.
- step S10 may specifically include:
- Step S101 Smoothing the three-dimensional face image to be recognized to obtain a processed three-dimensional face image.
- the Euclidean distance algorithm may be used to smooth the recognized three-dimensional face image to obtain a processed three-dimensional face image.
- Step S102 obtains the to-be-extracted face region of the processed three-dimensional face image according to the face nose tip detection algorithm based on the support vector machine.
- the tip of the nose in the processed three-dimensional face image may be acquired first according to the shape index, and then the tip of the nose is taken as the center of the sphere, and the pending radius of the processed three-dimensional face image is acquired according to the preset radius. Extract the face area.
- the step of acquiring the nose tip point in the processed three-dimensional face image according to the shape index may include:
- the nose tip point is obtained according to the density of neighbor nodes of each point in the candidate points.
- the shape index is calculated according to the maximum principal curvature and minimum principal curvature of each point in the three-dimensional point cloud in the processed three-dimensional face image, and its calculation formula is:
- S1 represents the shape index
- K_(i_max) represents the maximum curvature of the i-th point
- K_(i_min) represents the minimum curvature of the i-th point.
- This embodiment considers that three-dimensional face preprocessing is a very important part of the face recognition system. This is because the quality of the face data is not very good due to the influence of the machine and the surrounding environment when the three-dimensional face is collected. This will affect the effect of the next feature extraction stage. Therefore, the present invention adopts two measures, one is to smooth the three-dimensional face point cloud data, and the other is to propose a new method for detecting the tip of the nose to facilitate cutting human face.
- Step S20 Obtain isogeometric contour lines and radial curves of the face region to be extracted.
- the step of obtaining the isogeometric contour line of the face area to be extracted may specifically include: taking the nose tip point as the center and the geodesic line as the measuring distance, extracting the isogeometric contour line of the face area to be extracted The contour line of the ground distance.
- the step of obtaining the radial curve of the face region to be extracted may include:
- step S30 a shape analysis algorithm is used to obtain the geodesic distance between the face area to be extracted and the curve corresponding to the pre-stored three-dimensional face image according to the isogeometric contour line and the radial curve.
- a shape analysis algorithm is used to obtain the curve corresponding to the face area to be extracted and the pre-stored three-dimensional face image. Geodesic distance.
- Step S40 Obtain the similarity between the face area to be extracted and the corresponding curve of the pre-stored three-dimensional face image according to the geodesic distance.
- the three-dimensional face to be recognized can be compared according to the similarity.
- Image for face recognition After the similarity of the corresponding curve between the face area to be extracted and the pre-stored three-dimensional face image is obtained, the three-dimensional face to be recognized can be compared according to the similarity. Image for face recognition.
- the present embodiment obtains the to-be-extracted face area of the three-dimensional face image to be recognized according to the above-mentioned technical solution according to the face-nose-tip detection algorithm based on the support vector machine; obtains the iso-geometric contour line of the to-be-extracted face area And a radial curve; according to the iso-geometric contour line and the radial curve, a shape analysis algorithm is used to obtain the geodesic distance between the face area to be extracted and the curve corresponding to the pre-stored three-dimensional face image; according to the survey The ground distance obtains the similarity between the face area to be extracted and the corresponding curve of the pre-stored 3D face image.
- the number of points involved in the calculation in the 3D face recognition process is reduced and the algorithm speed is improved. , Improved robustness to changes in 3D facial expressions.
- the main experience of 3D face recognition algorithm development so far can be divided into two major stages: One is the matching of 3D faces based on overall features in the early days. This type of method is computationally intensive and can only be operated on a complete face. , It is more sensitive to the lack of face data and occlusion; the other stage is the 3D face recognition algorithm based on local features, which is also the mainstream direction of current research. This type of algorithm is more flexible and has stronger robustness against occlusion and loss. .
- the present invention is a three-dimensional face recognition algorithm aimed at local features.
- the present invention consists of three parts in total.
- Figure 2 shows the entire face recognition process.
- the present invention is a three-dimensional face recognition algorithm based on shape analysis. By extracting the contour lines and radial curves of the face, the distance of the corresponding curve is calculated by the method of shape analysis, and the KNN algorithm is used for classification.
- the key module technology of this method is data preprocessing and feature extraction.
- 3D face preprocessing is a very important part of the face recognition system. This is because the quality of the face data is not very good due to the influence of the machine and the surrounding environment when the 3D face is collected, which will affect the next The effect of the feature extraction stage.
- This method adopts two measures, one is to smooth the three-dimensional face point cloud data, and the other is to propose a new method for detecting the tip of the nose to facilitate the cutting of the face. Please refer to FIG. 3, which is a schematic diagram of the preprocessing steps in the data preprocessing stage of the present invention.
- the depth value of some points in the 3D point cloud that is, the value of the z-axis, will have a large error due to calculation errors. These points have a great influence on the subsequent nose point detection and feature extraction, and will reduce the recognition of the entire system. effect.
- Figure 4 intercepts the point cloud distribution in the sharp area. These point cloud distributions indicate that the point distribution in the sharp area is more scattered, and the distance between the points will be much larger than the normal area.
- Shape Index can be used to describe the concave and convex information of the local shape on the three-dimensional surface, and the local information will not change with the change of the face pose.
- shape index is used to first filter out a part of the point cloud data containing the tip of the nose, and then the trained SVM classifier is used for further screening, and finally the tip of the nose is determined according to the density of the neighbor nodes of each point in the candidate point, the density is the largest
- the point of the nose is set as
- Shape Index is calculated based on the maximum principal curvature and minimum principal curvature of each point in the three-dimensional point cloud.
- the calculation formula is shown in formula (3-2):
- SI represents the shape index
- Ki_max represents the maximum curvature of the i-th point
- Ki_min represents the minimum curvature of the i-th point.
- the position of the center of the sphere can be regarded as the tip of the nose.
- the radius r is selected, and all the spheres in this radius are r
- the three-dimensional point inside is the face area to be extracted (radius r is usually an empirical value).
- the 3D face point cloud data is expressed in the form of a curve, which will greatly reduce the amount of calculation.
- the geodesic distance refers to the shortest distance between two points on a three-dimensional surface, not the straight line distance in Euclidean space.
- the meaning of equidistance transformation refers to the three-dimensional surface The distance between two points will not change basically.
- the three-dimensional surface When the three-dimensional surface is deformed, it corresponds to the surface of the three-dimensional face, that is, when the expression of the three-dimensional face changes, the geodesic distance property on the surface of the three-dimensional face can still remain unchanged .
- the expression changes after equidistant transformation, the geodesic distance hardly changes due to its geometric properties, so the geodesic distance can be regarded as a fixed geometric information of the three-dimensional face .
- the geodesic curvature is also calculated on a three-dimensional surface, so suppose it represents a three-dimensional surface. Since the geodesic curvature is calculated on a three-dimensional curve, any three-dimensional curve on the three-dimensional surface is represented by C .
- the unit tangent vector, the main normal vector and the subnormal vector at a certain point p of the three-dimensional curve C are represented by a, ⁇ and ⁇ , respectively.
- the unit normal vector of the point p on the three-dimensional surface S is represented by n.
- the angle between the principal normal vector ⁇ and the unit normal vector n is represented by ⁇ .
- the normal curvature of S in the direction ⁇ of the tangent vector of point P is defined as:
- k represents the curvature
- ⁇ represents the angle between the principal normal vector ⁇ and the unit normal vector n.
- ⁇ represents the projection coefficient
- k g represents the geodesic curvature
- the geodesic line indicates that if the geodesic curvature of any point on a curve is 0, then this curve is a geodesic line.
- t is the arc length parameter of C, and the tangent vector of C at point P can be obtained. See formula (3-5).
- the simplified formula (3-7) is the first basic form of the surface.
- the coefficients E, F, G are functions of u and v, which are called the first basic quantity of the surface.
- T p represents the tangent surface at point p on the three-dimensional surface S, according to the unit normal vector n and the tangent surface T p
- the curvature vector KN can be decomposed into two directions, one is the geodesic curvature vector, and the other is the normal area rate vector.
- v and ⁇ respectively represent the geodesic curvature vector and the normal area rate vector.
- k n and k ⁇ represent curvature
- T represents a tangent plane
- n represents a unit vector
- the formula for calculating the curvature of the geodesic line adopted by the present invention is the Liuweier formula, and the specific equation expression is as follows:
- the calculation of the geodesic involves the differential form of the geodesic curvature vector.
- the geodesic curvature vector is 0, it means that the point p belongs to a geodesic.
- the differential of the geodesic can be obtained from the formula (3-7) The equation is:
- the above formula is a set of second-order ordinary differential equations.
- r(u 1 (s), u 2 (s)) represents a line passing the initial point r(u 1 (s) 0 , u 2 (s) 0 ) and is in line with the initial direction Tangent geodesic.
- the algorithm for calculating the geodesic distance of the present invention is the Dijkstra algorithm shown in 2.
- the Dijkstra algorithm is originally used to find the shortest path in the network model, and can calculate the shortest path from one node to other nodes.
- the characteristic of Dijkstra's algorithm is that the requested node is the center point, and then expands outwards in a ring until the target is calculated.
- Geodesic distance which is the shortest distance between any two points on a three-dimensional surface along the surface.
- the geodesic distance is hardly affected by changes in expressions.
- contour lines on the three-dimensional face surface including horizontal contour lines.
- Side contour line and circular contour line here is the circular contour line, with the nose tip point detected in the preprocessing stage as the center, and the geodesic line as the measured distance, extract the contour line equal to the geodesic distance.
- any two three-dimensional points p and p 0 is the shortest distance between the two points, that is, the geodesic distance on the three-dimensional surface
- the formula (3 -18) Represents the function of the geodesic distance between two points:
- k represents the geodesic distance between two points
- p and p 0 represent any two three-dimensional points
- S represents the three-dimensional face surface
- the curve of the three-dimensional curved face that is, the geodesic, can be defined as the formula (3-19):
- k can have three values, as shown in Table 3.
- Fig. 6 is a schematic diagram of the result of isogeometric contour line detection.
- the radial curve is shown in Figure 7.
- the radial curve is a curve segment that starts from the tip of the nose towards the periphery of the face.
- the present invention is obtained by finding the symmetry plane of the three-dimensional human face, and then finding the intersection line of this symmetry plane and the three-dimensional human face curved surface.
- the face symmetry plane refers to a three-dimensional plane that separates the entire three-dimensional face from the left and right by the tip of the nose.
- the face symmetry plane is extracted according to the symmetry properties of the three-dimensional face.
- the symmetry plane needs to be detected in this area.
- the central area is selected around the tip of the nose.
- the tip of the nose is regarded as the center of a sphere.
- the radius of the sphere is 40mm.
- the three-dimensional face is the central area to be determined. The specific calculation steps are shown in Table 4.
- the first step is to extract the symmetry surface of the three-dimensional face, and then calculate the intersection of the symmetrical surface and the three-dimensional face.
- a curve formed by the points in this intersection is a radial curve of the three-dimensional face passing the tip of the nose.
- A/n symmetry planes have been obtained, find the intersection of each symmetry plane and the three-dimensional face surface, so that each intersection composes at least two radial curves, so the number of radial curves obtained is at least 2a /n.
- the radial curve extracted by the present invention can freely set the maximum rotation angle, can accurately extract the radial curve of a certain area, and each radial curve is centered on the nose point and passes through the half of the three-dimensional face surface. Rigid area and non-rigid area, this shows that each radial curve can extract very detailed three-dimensional face information.
- Figures 8 to 13 show the radial curves extracted by the present invention.
- Figure 8 is the original point cloud image
- Figure 9 is the radial curve rotated 360 degrees
- Figure 10 is the radial curve rotated 270 degrees
- Figure 11 is the radial curve rotated 180 degrees
- Figure 12 is the radial curve rotated 90 degrees
- the curve, Figure 13 is a radial curve from 270 degrees to 360 degrees.
- the geodesic distance d of the corresponding curves of the two faces is calculated through the method of shape analysis.
- Figure 14 shows that the point clouds of two faces (different people) are simplified into 10 iso-geometric contour lines, and the shape analysis is performed on the fifth contour line of the two faces. The result is shown in Figure 15.
- the three-dimensional face recognition method based on local features of the present invention obtains the to-be-extracted face region of the three-dimensional face image to be recognized according to the above-mentioned technical solution according to the face nose detection algorithm based on support vector machine; Extract the iso-geodetic contour line and radial curve of the face area; according to the iso-geodetic contour line and the radial curve, use a shape analysis algorithm to obtain the corresponding curve between the face area to be extracted and the pre-stored three-dimensional face image According to the geodesic distance, the similarity between the face area to be extracted and the corresponding curve of the pre-stored three-dimensional face image is obtained.
- the participation in the three-dimensional face recognition process is reduced The number of points calculated improves the speed of the algorithm and the robustness to changes in three-dimensional facial expressions.
- the present invention also provides a three-dimensional face recognition system, the system includes a memory, a processor, and a computer program stored on the memory, and the computer program is executed by the processor to implement the above implementation
- the computer program is executed by the processor to implement the above implementation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Image Analysis (AREA)
Abstract
本发明公开了一种基于局部特征的三维人脸识别方法和系统，该方法包括：根据基于支持向量机的人脸鼻尖检测算法获取待识别三维人脸图像的待提取人脸区域；获取所述待提取人脸区域的等测地轮廓线和径向曲线；根据所述等测地轮廓线和径向曲线，采用形状分析算法获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的测地线距离；根据所述测地线距离获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的相似度。相对于现有技术，本发明减少了三维人脸识别过程中参与计算的点的数量，提高了算法速度，提高了对三维人脸表情变化的鲁棒性。
Description
本发明涉及人脸识别技术领域，尤其涉及一种基于局部特征的三维人脸识别方法和系统。
二维人脸识别技术自上个世纪提出发展至今已经取得了令人满意的结果，并且二维人脸识别技术已经应用到了各个领域，但是它仍旧受着人脸姿态变化、环境光照变化等的影响。
三维人脸识别技术的快速发展就是为了解决二维人脸识别的上述瓶颈问题。三维人脸识别虽然不受人脸姿态变化、环境光照变化的影响，但是容易受到表情变化的影响，并且每张三维人脸点云都包含上万个点，计算速度受到了很大的影响。
发明内容
本发明提供一种基于局部特征的三维人脸识别方法和系统，旨在减少参与计算的点的数量，提高算法速度，提高对三维人脸表情变化的鲁棒性。
为实现上述目的，本发明提供一种基于局部特征的三维人脸识别方法，其特征在于，所述方法包括以下步骤：
根据基于支持向量机的人脸鼻尖检测算法获取待识别三维人脸图像的待提取人脸区域；
获取所述待提取人脸区域的等测地轮廓线和径向曲线；
根据所述等测地轮廓线和径向曲线，采用形状分析算法获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的测地线距离；
根据所述测地线距离获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的相似度。
其中，所述根据基于支持向量机的人脸鼻尖检测算法获取待识别三维人脸图像的待提取人脸区域的步骤包括：
对所述待识别三维人脸图像进行平滑处理，得到处理后的三维人脸图像；
根据基于支持向量机的人脸鼻尖检测算法获取所述处理后的三维人脸图像的待提取人脸区域。
其中，所述对所述待识别三维人脸图像进行平滑处理的步骤中，采用欧式距离算法对所述识别三维人脸图像进行平滑处理。
其中，所述根据基于支持向量机的人脸鼻尖检测算法获取所述处理后的三维人脸图像的待提取人脸区域的步骤包括：
根据形状指数获取所述处理后的三维人脸图像中的鼻尖点；
以所述鼻尖点为球心，并根据预设半径获取所述处理后的三维人脸图像的待提取人脸区域。
其中，所述根据形状指数获取所述处理后的三维人脸图像中的鼻尖点的步骤包括：
采用形状指数筛选出所述处理后的三维人脸图像中包括鼻尖点的一部分点云数据；
采用预先训练好的SVM分类器对筛选出来的点云数据进一步筛选；
根据候选点里面每个点的邻居节点的密度获取所述鼻尖点。
其中，所述形状指数是根据所述处理后的三维人脸图像中的三维点云中每个点的最大主曲率和最小主曲率计算得来的，其计算公式为：
其中，SI表示的是形状指数，K_(i_max)表示的是第i个点的最大曲率，K_(i_min)表示的是第i个点的最小曲率。
其中，所述获取所述待提取人脸区域的等测地轮廓线的步骤包括：
以所述鼻尖点为中心，测地线为衡量距离，提取出所述待提取人脸区域中等测地线距离的轮廓线。
其中，所述获取所述待提取人脸区域的径向曲线的步骤包括：
获取所述待提取区域中三维人脸的对称面；
选择每条径向曲线的偏转角为n度，设置最大的旋转角度为a，将初始的对称面M通过旋转n度，获得到个对称面；
获取每个对称面与三维人脸表面的交集，每个交集组成至少两条径向曲线。
为实现上述目的，本发明还提出一种三维人脸识别系统，所述系统包括存储器、处理器以及存储在所述存储器上的计算机程序，所述计算机程序被所述处理器运行时实现如上所述的方法的步骤。
本发明基于局部特征的三维人脸识别方法和系统通过上述技术方案，根据基于支持向量机的人脸鼻尖检测算法获取待识别三维人脸图像的待提取人脸区域；获取所述待提取人脸区域的等测地轮廓线和径向曲线；根据所述等测地轮廓线和径向曲线，采用形状分析算法获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的测地线距离；根据所述测地线距离获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的相似度，相对于现有技术，减少了三维人脸识别过程中参与计算的点的数量，提高了算法速度，提高了对三维人脸表情变化的鲁棒性。
图1是本发明实施例中基于局部特征的三维人脸识别方法的流程示意图；
图2是本发明基于局部特征的三维人脸识别方法整个人脸识别的流程图。
图3是本发明基于局部特征的三维人脸识别方法在数据预处理阶段的预处理步骤示意图；
图4是尖锐区域点云分布示意图；
图5是测地曲率示意图；
图6是等测地轮廓线检测结果示意图；
图7是径向曲线示意图；
图8为原始点云图像；
图9为旋转360度的径向曲线；
图10为旋转270度的径向曲线；
图11为旋转180度的径向曲线；
图12为旋转90度的径向曲线；
图13为从270度到360度的径向曲线；
图14是表示两张人脸(不同人)点云被简化为10条等测地轮廓线的示意图；
图15是对两张人脸的第5条轮廓线进行形状分析的结果示意图。
本发明目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
考虑到目前的人脸识别算法中，以整体特征进行三维人脸的匹配的计算方法，只有在一个完整的人脸上才能够进行操作，对人脸数据缺失、遮挡比较敏感，由此，本发明提出一种解决方案。
具体的，本发明提出一种基于局部特征的三维人脸识别方法和系统，本发明所采用的技术方案主要是在数据预处理阶段，根据鼻尖点的特殊地位和几何性质，提出了基于支持向量机的鼻尖点检测算法；针对三维人脸点云数据量过大，提出了使用基于鼻尖点的等测地轮廓线和径向曲线来简化三维人脸点云的方法，由此改进三维点云人脸的表达方式，减少参与计算的点的数量，提高算法的速度，并提高对三维人脸表情变化的鲁棒性。
请参照图1，图1是本发明基于局部特征的三维人脸识别方法较佳实施例的流程示意图。
如图1所示，本实施例中，该基于局部特征的三维人脸识别方法包括以下步骤：
步骤S10，根据基于支持向量机的人脸鼻尖检测算法获取待识别三维人脸图像的待提取人脸区域。
可以理解的是，本实施例的执行主体例如可以为手机、电脑、IPAD等具有人脸识别功能的智能终端。
其中，所述步骤S10具体可以包括：
步骤S101，对所述待识别三维人脸图像进行平滑处理，得到处理后的三维人脸图像。
具体实施时，可以采用采用欧式距离算法对所述识别三维人脸图像进行平滑处理，以得到处理后的三维人脸图像。
步骤S102根据基于支持向量机的人脸鼻尖检测算法获取所述处理后的三维人脸图像的待提取人脸区域。
具体的，可以先根据形状指数获取所述处理后的三维人脸图像中的鼻尖点，再以所述鼻尖点为球心，并根据预设半径获取所述处理后的三维人脸图像的待提取人脸区域。
其中，所述根据形状指数获取所述处理后的三维人脸图像中的鼻尖点的步骤可以包括：
采用形状指数筛选出所述处理后的三维人脸图像中包括鼻尖点的一部分点云数据；
采用预先训练好的SVM分类器对筛选出来的点云数据进一步筛选；
根据候选点里面每个点的邻居节点的密度获取所述鼻尖点。
所述形状指数是根据所述处理后的三维人脸图像中的三维点云中每个点的最大主曲率和最小主曲 率计算得来的，其计算公式为：
其中，S1表示的是形状指数，K_(i_max)表示的是第i个点的最大曲率，K_(i_min)表示的是第i个点的最小曲率。
本实施例考虑到三维人脸预处理是人脸识别系统中很重要的一个环节，这是由于三维人脸在采集的时候因为机器和周围环境的影响造成了人脸数据的质量不是很好，这将影响接下来的特征提取阶段的效果，因此，本发明采取两方面的措施，一是对三维人脸点云数据进行平滑处理，二是提出一个新的检测鼻尖点的方法，方便裁切人脸。
步骤S20，获取所述待提取人脸区域的等测地轮廓线和径向曲线。
其中，所述获取所述待提取人脸区域的等测地轮廓线的步骤具体可以包括：以所述鼻尖点为中心，测地线为衡量距离，提取出所述待提取人脸区域中等测地线距离的轮廓线。
所述获取所述待提取人脸区域的径向曲线的步骤可以包括：
先获取所述待提取区域中三维人脸的对称面；再选择每条径向曲线的偏转角为n度，设置最大的旋转角度为a，将初始的对称面M通过旋转n度，获得到a/n个对称面；然后获取每个对称面与三维人脸表面的交集，每个交集组成至少两条径向曲线。
步骤S30，根据所述等测地轮廓线和径向曲线，采用形状分析算法获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的测地线距离。
本实施例中，在获取到所述待提取人脸区域的等测地轮廓线和径向曲线后，采用形状分析算法获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的测地线距离。
步骤S40，根据所述测地线距离获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的相似度。
其中，若所述测地线距离越小，表示对应曲线的相似度越高。
可以理解的是，本实施例中，在获取到所述待提取人脸区域与预先存储的三维人脸图像对应曲线的相似度后，即可以根据所述相似度对所述待识别三维人脸图像进行人脸识别。
由此，本实施例通过上述技术方案，根据基于支持向量机的人脸鼻尖检测算法获取待识别三维人脸图像的待提取人脸区域；获取所述待提取人脸区域的等测地轮廓线和径向曲线；根据所述等测地轮廓线和径向曲线，采用形状分析算法获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的测地线距离；根据所述测地线距离获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的相似度，相对于现有技术，减少了三维人脸识别过程中参与计算的点的数量，提高了算法速度，提高了对三维人脸表情变化的鲁棒性。
以下对本发明基于局部特征的三维人脸识别方法做进一步的详细阐述。
一、总体思想：
三维人脸识别算法发展至今主要经历可以划分为两大阶段：一是早期的时候以整体特征进行三维人 脸的匹配，这类方法计算量大并且只有在一个完整的人脸上才能够进行操作，对人脸数据缺失、遮挡比较敏感；另一个阶段是基于局部特征的三维人脸识别算法，也是现在研究的主流方向，这类算法更加的灵活，对于遮挡、缺失有更强的鲁棒性。本发明是针对于局部特征的三维人脸识别算法。
本发明的实现思路是：
(1)在数据预处理阶段，根据鼻尖点的特殊地位和几何性质，提出了基于支持向量机的鼻尖点检测算法。
(2)针对三维人脸点云数据量过大，提出了使用基于鼻尖点的等测地轮廓线和径向曲线来简化三维人脸点云的方法，然后引入形状分析的算法对简化后的人脸计算对应的三维曲线的测地线距离作为分类的依据。
(3)采用KNN分类。
二、本发明提供的操作及实现流程：
本发明总共由三大部分组成，图2展示了整个人脸识别流程。本发明是基于形状分析的三维人脸识别算法。通过提取人脸上的轮廓线和径向曲线，通过形状分析的方法计算对应曲线的距离，使用KNN的算法进行分类。本方法的关键模块技术是数据预处理和特征提取部分。
三、关键模块及技术：
1、数据预处理阶段：
三维人脸预处理是人脸识别系统中很重要的一个环节，这是由于三维人脸在采集的时候因为机器和周围环境的影响造成了人脸数据的质量不是很好，这将影响接下来的特征提取阶段的效果。本方法采取两方面的措施，一是对三维人脸点云数据进行平滑处理，二是提出一个新的检测鼻尖点的方法，方便裁切人脸。请参照图3，图3是本发明在数据预处理阶段的预处理步骤示意图。
1.1、人脸平滑处理
三维点云中的某些点的深度值即z轴的值会由于计算的错误会出现较大的误差这些点对接下来的鼻尖点检测和特征提取有很大的影响，会降低整个系统的识别效果。
大部分深度值的误差表现形式为突起，图4截取了尖锐区域的点云分布，这些点云分布表明了尖锐区域的点分布比较分散，和正常区域相比点之间的距离会大很多。
尖锐区域的点之间的距离当对于正常点大，所以本节提出了根据点之间的欧式距离的算法，具体算法步骤如表1所示：
表1 人脸数据平滑步骤
公式(3-1)表示如下：
公式(3-1)中的函数
表示在欧式空间中两个三维点之间的直线距离，K表示的是k个邻居点。
1.2、基于支持向量机的人脸鼻尖点检测
形状指数(Shape Index)可以用来描述三维曲面上局部形状的凹凸信息，这些局部信息不会随着人脸姿态的变化而改变。此处利用形状指数先筛选出包含鼻尖点的一部分点云数据，然后使用训练好的SVM分类器进行进一步的筛选，最后根据候选点里面每个点的邻居节点的密度来确定鼻尖点，密度最大的的定为鼻尖点。
Shape Index是根据三维点云中每个点的最大主曲率和最小主曲率计算得来的，计算公式见公式(3-2)：
公式(3-2)中SI表示的是形状指数，K
i_max表示的是第i个点的最大曲率，K
i_min表示的是第i个点的最小曲率。
1.3、人脸区域提取
因为三维人脸可以看作是位于一个球体内，球心的位置近似可以看作是鼻尖点，根据前一小节求得的鼻尖点作为球心，选择半径r，所有在这个半径为r的球体里面的三维点就是要提取的人脸区域(半径r通常是一个经验值)。
2、基于局部曲线的三维人脸识别算法
因为大部分三维人脸数据库中的三维人脸数据都至少有几万个点，常见的一些基于关键点的局部提取特征的方法因为会计算到每个点并且会有很大的重复计算，这导致了特征提取阶段十分耗时。此处把三维人脸点云数据用曲线的形式表达出来，这会极大的减少计算量。
测地线距离蕴含的一个数学思想是等距变换，测地线距离指的是在三维曲面上两点间最短的距离，并非欧式空间种的直线距离，等距变换的含义是指三维曲面上两点的距离基本不会发生变换当三维曲面发生形变的时候，对应到三维人脸这个曲面，即三维人脸发生表情变化时，三维人脸表面上的测地线距离性质仍能保持不变。对于三维人脸来说经过等距变换即发生表情变化后，测地线距离由于其几何性质几乎不会发生变化，所以测地线距离可以看作时三维人脸上一个固定不变的几何信息。
因为人脸旋转不同角度以及受到遮挡的人脸会影响人脸识别的效果，此处通过提取人脸上的轮廓线 和径向曲线，通过形状分析的方法计算对应曲线的距离，使用KNN的算法进行分类。
2.1、测地线距离计算
介绍测地线之前，先简单介绍下测地曲率的概念。
如图5所示，测地曲率也是在三维曲面上进行计算的，所以假设表示一个三维曲面，由于测地曲率是在三维曲线上求得的，所以三维曲面上的任意一条三维曲线用C表示。假设有三维曲线C上某一点p，还有在三维曲线C的某一点p处的单位切向量、主法向量和副法向量分别用a，β和γ表示。而点p在三维曲面S上的单位法向量用n来表示，如图5所示，主法向量β和单位法向量n的夹角用θ表示。则S在点P的切向量方向α上的法方向曲率定义为：
k
n＝k c o s θ＝k β·n (3-3)
公式(3-3)中k表示的是曲率，θ表示的是主法向量β和单位法向量n的夹角。
那么测地曲率的真正意义是什么？首先求出一条三维曲线C，对于三维曲线C上的某一点P来讲，曲线C在该点的曲率法向量kβ，然后把kβ投影到三维曲线C在点P处的切平面上得到的结果即为点P的测地曲率k
g：
k
g＝k β ε (3-4)
式(3-4)中ε表示投影系数，k
g表示的是测地曲率。
得出测地曲率后，才定义了什么是测地线，测地线表示的是如果一条曲线上的任意一个点的测地曲率为0，那么这条曲线就是测地线。
其次要了解一下测地线的数学原理。
(1)测地线计算的数学原理。
首先定义一个三维曲面S，并且方程表达式如下：
S：r＝r(u，v) (u，v)∈D
然后定义三维曲线C，方程表达式如下：
由以上表达式中可以推导出一个向量方程用来表示三维曲线C：
C：r＝r(u(t)，v(t)) (α＜t＜β)
上式中的t是C的弧长参数，可得出C在点P处的切向量，见公式(3-5)。
如果将其写成微分形式的话，则可表示成如下形式：
dr＝r
udu+r
vdv
设s是曲线C的弧长，由于ds＝|dr|，所以
ds
2＝dr
2＝(r
udu+r
vdv)
2＝r
u
2du
2+2r
ur
vdudv+r
v
2dv
2 (3-6)
记
E＝r
u
2，F＝r
ur
v，G＝r
v
2
则公式(3-5)简化为公式(3-7)：
ds
2＝Edu
2+2Fdudv+Gdv
2 (3-7)
简化后的公式(3-7)就是曲面的第一基本形式，系数E、F、G是u、v的函数，称为曲面的第一基本量。
使用KN表示某一点p在三维曲线C上的曲率向量，如公式(3-8)所示，T
p表示在三维曲面S上点p处的切曲面，根据单位法向量n和切曲面T
p就可以把曲率向量KN分解到两个方向上，一个是测地线曲率向量，另一个是法区率向量。
kN＝v×τ (3-8)
其中v和τ分别表示的是测地线曲率向量和法区率向量。
v＝k
nn (3-9)
τ＝k
η(n×T) (3-10)
式中k
n和k
η表示的是曲率，T表示的是切平面，n表示的是单位向量。
由曲面论的基本公式
可以得到：
及
本发明采用的计算测地线曲率的公式是刘维尔公式，具体方程表达式如下所示：
测地线的计算涉及到测地曲率向量的微分形式，当测地曲率向量为0时，表示该点p属于一条测地线，从公式(3-7)中可得出测地线的微分方程为：
上式是一组二阶的常微分方程。
若给出初始条件：
当S＝S
0时，
方程组(3-16)有唯一解的前提是必须给定三维曲面上某一点和这个点位置的切向量。
u
i＝u
i(s) (i＝1，2) (3-17)
因此，r(u
1(s)，u
2(s))表示一条过初始点r(u
1(s)
0，u
2(s)
0)且和初始方向
相切的测地线。
(2)测地线的计算。
测地线距离这一个三维曲面上的几何性质在人脸识别中越来越受到重视，与此同时许多计算测地线距离的算法被提出来了。此处引入了迪杰斯特拉算法计算测地线距离。
表2 迪杰斯特拉算法计算测地线距离
本发明计算测地线距离的算法如2所示的迪杰斯特拉算法，迪杰斯特拉算法本来是用来求网络模型 中最短路径的，可以计算一个节点到其它节点的最短路径。迪杰斯特拉算法的特点是以所求节点为中心点，然后呈环形向外扩展，直到计算出目标为止。
2.2、等测地轮廓线提取方法
测地线距离，它是表示三维曲面上的任意两点沿曲面前进的最短距离，测地线距离几乎不会受到表情变化的影响，三维人脸曲面上有多种轮廓线，包括水平轮廓线、侧轮廓线和环形轮廓线，这里使用的是环形轮廓线，以在预处理阶段检测出的鼻尖点为中心，测地线作为衡量的距离，提取出等测地线距离的轮廓线。
具体分析如下：
首先获得一张三维人脸曲面S，任意两个三维点p和p
0，点p和p
0之间的距离即为两点间的最短距离即三维曲面上的测地线距离，公式(3-18)表示两点间测地线距离的函数表示：
F(p，p
0)＝k；k∈[0，+∞]and(p，p
0)∈S (3-18)
其中k表示两点间的测地线距离，p和p
0表示的是任意的两个三维点，S表示的三维人脸曲面。
因此，可以定义三维曲面人脸上的曲线即测地线为公式(3-19)：
其中k可以有三种取值方式，如表3所示。
表3 测地线距离k的三种取值范围
以预处理阶段检测出的鼻尖点为中心，计算到鼻尖点测地线半径为k的点，这些点组成的线称为等测地轮廓线。如图6所示，图6为等测地轮廓线检测结果示意图。
2.3、径向曲线提取方法
径向曲线如图7所示，径向曲线就是以鼻尖点为起点向人脸外围方向的一条曲线段。本发明是通过求出三维人脸的对称面，然后求出这个对称面和三维人脸曲面的交线得出的。
2.3.1、人脸对称面计算
人脸对称面指的是过鼻尖点把整个三维人脸左右分开的一个三维平面，根据三维人脸的对称性质来提取人脸对称面。首先需要确定三维人脸的一个中心区域，需要在这个区域检测对称面，中心区域是围绕鼻尖点进行选区的，把鼻尖点看作是一个球体的中心，这个球体的半径是40mm，球体内的三维人脸就是要确定的中心区域。具体的计算步骤如表4所示。
表4 人脸对称面提取
2.3.2、径向曲线提取
首先是提取出三维人脸的对称面，然后计算这个对称曲面和三维人脸的交集，这个交集中的点组成的一条曲线即为三维人脸上过鼻尖点的一条径向曲线。具体的提取径向曲线的步骤如下：
(1)使用小节2.3.1中介绍的方法提取人脸中心对称面；
(2)选择每条径向曲线的偏转角为n度，设置最大的旋转角度为a，把初始的对称面M通过旋转n度，就可以得到a/n个对称面；
(3)已经得到了a/n个对称面，求每个对称面和三维人脸表面的交集，这样每个交集的组成至少两条径向曲线，这样得到的径向曲线条数至少有2a/n条。
本发明提取的径向曲线可以自由设定最大旋转角度，可以准确地取出某块区域的径向曲线，并且每条径向曲线都以鼻尖点为中心，并且穿过了三维人脸曲面的半刚性区域和非刚性区域，这样表明了每条径向曲线都可以提取到十分详细的三维人脸的金和信息。如图8至13展示了本发明提取的径向曲线。图8为原始点云图像，图9为旋转360度的径向曲线，图10为旋转270度的径向曲线，图11为旋转180度的径向曲线，图12为旋转90度的径向曲线，图13为从270度到360度的径向曲线。
2.4、径向曲线提取
根据前面方法求得的等测地轮廓线和径向曲线，通过形状分析的方法，计算两张人脸对应曲线的测地线距离d，d越小表示两条曲线相似度越高。图14表示两张人脸(不同人)点云被简化为10条等测地轮廓线，对两张人脸的第5条轮廓线进行形状分析，结果如图15所示。
综上所述，本发明基于局部特征的三维人脸识别方法通过上述技术方案，根据基于支持向量机的人脸鼻尖检测算法获取待识别三维人脸图像的待提取人脸区域；获取所述待提取人脸区域的等测地轮廓线和径向曲线；根据所述等测地轮廓线和径向曲线，采用形状分析算法获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的测地线距离；根据所述测地线距离获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的相似度，相对于现有技术，减少了三维人脸识别过程中参与计算的点的数量，提高了算法速度，提高了对三维人脸表情变化的鲁棒性。
为实现上述目的，本发明还提出一种三维人脸识别系统，所述系统包括存储器、处理器以及存储在 所述存储器上的计算机程序，所述计算机程序被所述处理器运行时实现如上实施例所述的方法的步骤，这里不再赘述。
以上所述仅为本发明的优选实施例，并非因此限制本发明的专利范围，凡是利用本发明说明书及附图内容所作的等效结构或流程变换，或直接或间接运用在其它相关的技术领域，均同理包括在本发明的专利保护范围内。
Claims (16)
- 一种基于局部特征的三维人脸识别方法，其特征在于，所述方法包括以下步骤：根据基于支持向量机的人脸鼻尖检测算法获取待识别三维人脸图像的待提取人脸区域；获取所述待提取人脸区域的等测地轮廓线和径向曲线；根据所述等测地轮廓线和径向曲线，采用形状分析算法获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的测地线距离；根据所述测地线距离获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的相似度。
- 根据权利要求1所述的三维人脸识别方法，其特征在于，所述根据基于支持向量机的人脸鼻尖检测算法获取待识别三维人脸图像的待提取人脸区域的步骤包括：对所述待识别三维人脸图像进行平滑处理，得到处理后的三维人脸图像；根据基于支持向量机的人脸鼻尖检测算法获取所述处理后的三维人脸图像的待提取人脸区域。
- 根据权利要求2所述的三维人脸识别方法，其特征在于，所述对所述待识别三维人脸图像进行平滑处理的步骤中，采用欧式距离算法对所述识别三维人脸图像进行平滑处理。
- 根据权利要求3所述的三维人脸识别方法，其特征在于，所述根据基于支持向量机的人脸鼻尖检测算法获取所述处理后的三维人脸图像的待提取人脸区域的步骤包括：根据形状指数获取所述处理后的三维人脸图像中的鼻尖点；以所述鼻尖点为球心，并根据预设半径获取所述处理后的三维人脸图像的待提取人脸区域。
- 根据权利要求4所述的三维人脸识别方法，其特征在于，所述根据形状指数获取所述处理后的三维人脸图像中的鼻尖点的步骤包括：采用形状指数筛选出所述处理后的三维人脸图像中包括鼻尖点的一部分点云数据；采用预先训练好的SVM分类器对筛选出来的点云数据进一步筛选；根据候选点里面每个点的邻居节点的密度获取所述鼻尖点。
- 根据权利要求5所述的三维人脸识别方法，其特征在于，所述形状指数是根据所述处理后的三维人脸图像中的三维点云中每个点的最大主曲率和最小主曲率计算得来的，其计算公式为：
其中，SI表示的是形状指数，K_(i_max)表示的是第i个点的最大曲率，K_(i_min)表示的是第i个点的最小曲率。 - 根据权利要求5所述的三维人脸识别方法，其特征在于，所述获取所述待提取人脸区域的等测地轮廓线的步骤包括：以所述鼻尖点为中心，测地线为衡量距离，提取出所述待提取人脸区域中等测地线距离的轮廓线。
- 根据权利要求5所述的三维人脸识别方法，其特征在于，所述获取所述待提取人脸区域的径向曲线的步骤包括：获取所述待提取区域中三维人脸的对称面；选择每条径向曲线的偏转角为n度，设置最大的旋转角度为a，将初始的对称面M通过旋转n度，获得到a/n个对称面；获取每个对称面与三维人脸表面的交集，每个交集组成至少两条径向曲线。
- 一种三维人脸识别系统，其特征在于，所述系统包括存储器、处理器以及存储在所述存储器上的计算机程序，所述计算机程序被所述处理器运行时实现三维人脸识别算法，所述算法包括以下步骤：根据基于支持向量机的人脸鼻尖检测算法获取待识别三维人脸图像的待提取人脸区域；获取所述待提取人脸区域的等测地轮廓线和径向曲线；根据所述等测地轮廓线和径向曲线，采用形状分析算法获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的测地线距离；根据所述测地线距离获取所述待提取人脸区域与预先存储的三维人脸图像对应曲线的相似度。
- 根据权利要求9所述的三维人脸识别系统，其特征在于，所述根据基于支持向量机的人脸鼻尖检测算法获取待识别三维人脸图像的待提取人脸区域的步骤包括：对所述待识别三维人脸图像进行平滑处理，得到处理后的三维人脸图像；根据基于支持向量机的人脸鼻尖检测算法获取所述处理后的三维人脸图像的待提取人脸区域。
- 根据权利要求10所述的三维人脸识别系统，其特征在于，所述对所述待识别三维人脸图像进行平滑处理的步骤中，采用欧式距离算法对所述识别三维人脸图像进行平滑处理。
- 根据权利要求11所述的三维人脸识别系统，其特征在于，所述根据基于支持向量机的人脸鼻尖检测算法获取所述处理后的三维人脸图像的待提取人脸区域的步骤包括：根据形状指数获取所述处理后的三维人脸图像中的鼻尖点；以所述鼻尖点为球心，并根据预设半径获取所述处理后的三维人脸图像的待提取人脸区域。
- 根据权利要求12所述的三维人脸识别系统，其特征在于，所述根据形状指数获取所述处理后的三维人脸图像中的鼻尖点的步骤包括：采用形状指数筛选出所述处理后的三维人脸图像中包括鼻尖点的一部分点云数据；采用预先训练好的SVM分类器对筛选出来的点云数据进一步筛选；根据候选点里面每个点的邻居节点的密度获取所述鼻尖点。
- 根据权利要求13所述的三维人脸识别系统，其特征在于，所述形状指数是根据所述处理后的三维人脸图像中的三维点云中每个点的最大主曲率和最小主曲率计算得来的，其计算公式为：
其中，SI表示的是形状指数，K_(i_max)表示的是第i个点的最大曲率，K_(i_min)表示的是第i个点的最小曲率。 - 根据权利要求13所述的三维人脸识别系统，其特征在于，所述获取所述待提取人脸区域的等测地轮廓线的步骤包括：以所述鼻尖点为中心，测地线为衡量距离，提取出所述待提取人脸区域中等测地线距离的轮廓线。
- 根据权利要求13所述的三维人脸识别系统，其特征在于，所述获取所述待提取人脸区域的径向曲线的步骤包括：获取所述待提取区域中三维人脸的对称面；选择每条径向曲线的偏转角为n度，设置最大的旋转角度为a，将初始的对称面M通过旋转n度，获得到a/n个对称面；获取每个对称面与三维人脸表面的交集，每个交集组成至少两条径向曲线。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2019/090551 WO2020248096A1 (zh) | 2019-06-10 | 2019-06-10 | 基于局部特征的三维人脸识别方法和系统 |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2019/090551 WO2020248096A1 (zh) | 2019-06-10 | 2019-06-10 | 基于局部特征的三维人脸识别方法和系统 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2020248096A1 true WO2020248096A1 (zh) | 2020-12-17 |
Family
ID=73781133
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2019/090551 WO2020248096A1 (zh) | 2019-06-10 | 2019-06-10 | 基于局部特征的三维人脸识别方法和系统 |
Country Status (1)
| Country | Link |
|---|---|
| WO (1) | WO2020248096A1 (zh) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113158892A (zh) * | 2021-04-20 | 2021-07-23 | 南京大学 | 一种与纹理和表情无关的人脸识别方法 |
| CN113837067A (zh) * | 2021-09-18 | 2021-12-24 | 成都数字天空科技有限公司 | 器官轮廓检测方法、装置、电子设备及可读存储介质 |
| CN114359998A (zh) * | 2021-12-06 | 2022-04-15 | 江苏理工学院 | 一种人脸口罩佩戴状态下的识别方法 |
| CN117422710A (zh) * | 2023-12-14 | 2024-01-19 | 泉州师范学院 | 一种点云测地线的测量方法 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101650777A (zh) * | 2009-09-07 | 2010-02-17 | 东南大学 | 一种基于密集点对应的快速三维人脸识别方法 |
| CN101986328A (zh) * | 2010-12-06 | 2011-03-16 | 东南大学 | 一种基于局部描述符的三维人脸识别方法 |
| CN106096503A (zh) * | 2016-05-30 | 2016-11-09 | 东南大学 | 一种基于关键点和局部特征的三维人脸识别方法 |
| CN107748859A (zh) * | 2017-08-10 | 2018-03-02 | 东南大学 | 一种基于径向线的局部遮挡下的三维人脸识别方法 |
-
2019
- 2019-06-10 WO PCT/CN2019/090551 patent/WO2020248096A1/zh active Application Filing
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101650777A (zh) * | 2009-09-07 | 2010-02-17 | 东南大学 | 一种基于密集点对应的快速三维人脸识别方法 |
| CN101986328A (zh) * | 2010-12-06 | 2011-03-16 | 东南大学 | 一种基于局部描述符的三维人脸识别方法 |
| CN106096503A (zh) * | 2016-05-30 | 2016-11-09 | 东南大学 | 一种基于关键点和局部特征的三维人脸识别方法 |
| CN107748859A (zh) * | 2017-08-10 | 2018-03-02 | 东南大学 | 一种基于径向线的局部遮挡下的三维人脸识别方法 |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113158892A (zh) * | 2021-04-20 | 2021-07-23 | 南京大学 | 一种与纹理和表情无关的人脸识别方法 |
| CN113158892B (zh) * | 2021-04-20 | 2024-01-26 | 南京大学 | 一种与纹理和表情无关的人脸识别方法 |
| CN113837067A (zh) * | 2021-09-18 | 2021-12-24 | 成都数字天空科技有限公司 | 器官轮廓检测方法、装置、电子设备及可读存储介质 |
| CN113837067B (zh) * | 2021-09-18 | 2023-06-02 | 成都数字天空科技有限公司 | 器官轮廓检测方法、装置、电子设备及可读存储介质 |
| CN114359998A (zh) * | 2021-12-06 | 2022-04-15 | 江苏理工学院 | 一种人脸口罩佩戴状态下的识别方法 |
| CN114359998B (zh) * | 2021-12-06 | 2024-03-15 | 江苏理工学院 | 一种人脸口罩佩戴状态下的识别方法 |
| CN117422710A (zh) * | 2023-12-14 | 2024-01-19 | 泉州师范学院 | 一种点云测地线的测量方法 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2020248096A1 (zh) | 基于局部特征的三维人脸识别方法和系统 | |
| CN107742102B (zh) | 一种基于深度传感器的手势识别方法 | |
| WO2018107979A1 (zh) | 一种基于级联回归的多姿态的人脸特征点检测方法 | |
| WO2017219391A1 (zh) | 一种基于三维数据的人脸识别系统 | |
| WO2020063527A1 (zh) | 基于多特征检索和形变的人体发型生成方法 | |
| WO2016023264A1 (zh) | 指纹识别方法及其指纹识别装置 | |
| JP5845365B2 (ja) | ３次元の近接インタラクションにおける、又は３次元の近接インタラクションに関連する改善 | |
| CN106780557B (zh) | 一种基于光流法和关键点特征的运动目标跟踪方法 | |
| CN100559398C (zh) | 自动的深度图像配准方法 | |
| CN108052942B (zh) | 一种飞机飞行姿态的视觉图像识别方法 | |
| CN101650777B (zh) | 一种基于密集点对应的快速三维人脸识别方法 | |
| WO2015067084A1 (zh) | 人眼定位方法和装置 | |
| CN104091162A (zh) | 基于特征点的三维人脸识别方法 | |
| CN106446773B (zh) | 一种全自动、鲁棒的三维人脸检测方法 | |
| CN108830170B (zh) | 一种基于分层特征表示的端到端目标跟踪方法 | |
| CN105046694A (zh) | 一种基于曲面拟合系数特征的点云快速配准方法 | |
| CN113628263A (zh) | 一种基于局部曲率及其近邻特征的点云配准方法 | |
| WO2015165227A1 (zh) | 人脸识别方法 | |
| JP7354468B1 (ja) | 幾何学的特性に基づく軽量化された単結晶シリコン楕円パラメータの認識方法及びシステム | |
| CN110288640A (zh) | 基于凸密度极值的点云配准方法 | |
| CN114648445B (zh) | 一种基于特征点提取及精配准优化的多视角高分辨率点云拼接方法 | |
| CN106682575A (zh) | Elm算法的人眼点云特征定位 | |
| CN106980845B (zh) | 基于结构化建模的人脸关键点定位方法 | |
| Xu et al. | 3D joints estimation of the human body in single-frame point cloud | |
| Xin et al. | Accurate and complete line segment extraction for large-scale point clouds |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19932826 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 19932826 Country of ref document: EP Kind code of ref document: A1 |