WO2017000816A1 - 一种面部识别系统及面部识别方法 - Google Patents

一种面部识别系统及面部识别方法 Download PDF

Info

Publication number
WO2017000816A1
WO2017000816A1 PCT/CN2016/086631 CN2016086631W WO2017000816A1 WO 2017000816 A1 WO2017000816 A1 WO 2017000816A1 CN 2016086631 W CN2016086631 W CN 2016086631W WO 2017000816 A1 WO2017000816 A1 WO 2017000816A1
Authority
WO
WIPO (PCT)
Prior art keywords
feature
face
module
recognized
image
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2016/086631
Other languages
English (en)
French (fr)
Inventor
梁宁清
张宏鑫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Yutou Technology Hangzhou Co Ltd
Original Assignee
Yutou Technology Hangzhou Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Yutou Technology Hangzhou Co Ltd filed Critical Yutou Technology Hangzhou Co Ltd
Priority to EP16817180.9A priority Critical patent/EP3319010A4/en
Priority to JP2017567755A priority patent/JP2018525718A/ja
Priority to US15/741,063 priority patent/US10438056B2/en
Publication of WO2017000816A1 publication Critical patent/WO2017000816A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172—Classification, e.g. identification
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/161—Detection; Localisation; Normalisation
    • G06V40/166—Detection; Localisation; Normalisation using acquisition arrangements
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172—Classification, e.g. identification
    • G06V40/173—Classification, e.g. identification face re-identification, e.g. recognising unknown faces across different face tracks
    • G—PHYSICS
    • G07—CHECKING-DEVICES
    • G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
    • G07C9/00—Individual registration on entry or exit
    • G07C9/30—Individual registration on entry or exit not involving the use of a pass
    • G07C9/32—Individual registration on entry or exit not involving the use of a pass in combination with an identity check
    • G07C9/37—Individual registration on entry or exit not involving the use of a pass in combination with an identity check using biometric data, e.g. fingerprints, iris scans or voice recognition

Definitions

  • the present invention relates to the field of identity recognition, and in particular, to a face recognition system and a face recognition method.
  • the face recognition technology solution applied in the access security system solves the problem of identification of a face image of a specific angle acquired at a specific location and under specific lighting conditions.
  • This solution can be deployed on high-performance computers or on embedded computers with low computing resources. Since the access security system allows result responses with delays of up to 1 second or longer, and is generally a one-time authentication requirement, there are few applications that are continuously processed, so algorithms with higher computational complexity must be used.
  • the face recognition technology scheme used in the access security system has the following defects: 1. It is necessary to fix the light condition by compensating the light source, which is very sensitive to light; 2. It is required to acquire a fixed face pose, such as a positive face; 3. Usually, the computational complexity Higher, it takes more time to give a result, and the real-time response requirement cannot be achieved in continuous detection.
  • the present invention provides a face recognition system and a face recognition method applied to a robot vision system, which realize face recognition work after detecting a face region.
  • a facial recognition system comprising:
  • An illumination pre-processing module receives the input face image to be recognized, and optimizes the illumination difference degree of the to-be-recognized face image, and outputs the image to be recognized after the illumination pre-processing to the next module;
  • a feature generating module configured to receive the light-preprocessed face image to be recognized, and perform feature generation, and output a feature vector to be identified for describing a face detail feature
  • a face feature library module pre-stored feature vectors of known faces
  • a feature matching and identification module coupled to the feature generation module and the face feature library module, to match the feature vector to be identified output by the feature generation module with all feature vectors in the feature library, and output The identity result corresponding to the identified feature vector is described.
  • the facial recognition system described above wherein the illumination pre-processing module processes the to-be-recognized face image using a Gaussian difference method.
  • the feature generation module uses a local binary pattern feature to describe a feature vector of the face detail feature.
  • the face recognition system described above wherein the face feature library module further provides a new face feature adding interface, by which a new face feature of a person with a known identity can be added to the face feature library module. Or the facial features of an unknown newcomer.
  • the present invention also provides a face recognition method, characterized by the face recognition system according to any one of claims 1 to 4, the method comprising the steps of:
  • Step 1 initializing the face feature library module, so that the face feature library module pre-stores a known face feature vector
  • Step 2 the illumination pre-processing module receives the input face image to be recognized, and optimizes the illumination difference degree of the to-be-identified face image, and outputs the image to be recognized after the illumination pre-processing;
  • Step 3 The feature generation module receives the image of the face to be recognized processed by the illumination preprocessing module, and performs a feature generation operation on the face image to be recognized, and outputs a feature vector for describing the feature of the face detail. ;
  • Step 4 The feature matching and identifying module performs matching calculation on the feature vector to be identified output by the feature generating module and all feature vectors in the feature library, and outputs an identity result corresponding to the feature vector to be identified.
  • the illumination pre-processing module processes the face image using a Gaussian difference method.
  • the feature vector of the face detail feature is described by using a local binary pattern feature.
  • the step 4 if the feature matching and recognition module determines that the face to be recognized is an unknown face, the unknown identification knot is output. And storing the feature vector of the unknown face into the face feature library module.
  • the technical solution of the present invention can realize the face recognition function after the detection of the face region, and can be applied to various illumination conditions, including the case of polarized light or uncompensated light source, and can recognize the face of the multi-pose, for example, the left side, Faces on the right side, up, down, and even sideways, while reducing computational complexity, reducing computational resource consumption, and high response speed, so that faces appearing in the robot's field of view can be recognized in real time, and in the face Real-time feedback is achieved through continuous recognition when the identity changes.
  • FIG. 1 is a functional block diagram of a face recognition system of the present invention.
  • the facial recognition system of the present invention is mainly applied to a vision system of a robot, and is mainly composed of an illumination pre-processing module 1, a feature generation module 2, a face feature library module 3, and a feature matching and recognition module 4.
  • the illumination pre-processing module 1 optimizes the illumination difference degree of the input face image to be recognized, and reduces the influence of the illumination change on the system recognition rate.
  • the module processes the image using Gaussian difference.
  • the effect of illumination changes on the same face image can be seen as superimposing a low frequency signal on the image, while Gaussian convolution on the face image can be considered as low pass filtering, using two different size volumes
  • the Gaussian convolution operation of the kernel can construct a Gaussian difference operation, and the effect can be regarded as a band-pass filter. Since the frequency of the effective features for face recognition is usually higher than the frequency of the illumination signal, two convolutions are appropriately selected.
  • the size of the kernel constitutes a differential filter that removes the effects of light changes in the face image.
  • the feature generation module 2 is connected to the illumination pre-processing module 1 for performing feature generation on the illumination pre-processed face image output by the illumination pre-processing module 1 and outputting a feature vector describing the face detail feature.
  • the types of features currently used to describe faces include principal component analysis (PCA), independent component analysis (ICA), local binary pattern features (LBP), and deep neural network training features, for our application needs.
  • PCA principal component analysis
  • ICA independent component analysis
  • LBP local binary pattern features
  • deep neural network training features for our application needs.
  • LBP is the best choice.
  • LBP has two advantages. The first is pure fixed-point calculation.
  • the calculation amount is much smaller than other features, especially in ARM (a microprocessor with a reduced instruction set, here also refers to the computing system built with this processor). On the processor with no floating-point computing capability, the computational advantage is more obvious.
  • LBP is essentially a differential-like feature that resists certain light changes. This can continue to weaken the adverse effect of the light change information that cannot be eliminated in the light pre-processing module on the recognition rate of the system.
  • the system identification logic compares the feature vector of the face image to be recognized with the face feature vector of the known identity in the face feature library module 3 to find the face with the most similar features. .
  • the identification step needs to be completed by querying the face feature library module 3. Since the LBP feature of a single face picture can only be used to identify a face picture with a small range of angles, the technical solution of the present invention manages the feature vector set of different angles of the same person by collecting more face image features. .
  • the feature matching and identifying module 4 is connected to the face feature library module 3 and the feature generating module 2. When the identification request comes, the feature matching and identifying module 4 outputs the face feature vector to be recognized by the feature generating module 2 and the face.
  • All the feature vectors in the feature library module 3 perform matching calculation, find the closest one feature vector and give the similarity value, and determine whether the face to be recognized is the person with the closest feature according to the preset threshold value of the similarity value. face. If yes, the corresponding identity result is given; if not, determining that the face to be recognized is an unknown face, if there is a demand at this time, the feature vector of the unknown face may be added to the face feature library module 3, and need to be added Identity information for next identification.
  • the above functional modules constitute a facial recognition system.
  • a face posture correction technique can be used to reduce the influence of different face poses on the recognition accuracy of face image acquisition, but to achieve better results, the function module requires more computing resources. Therefore, the calculation amount exceeds the real-time processing capability of the low resource system, so we use the face to collect more angles, and at the same time, through the matching method in the feature matching and recognition module 4, the matching result can still solve the face recognition problem of different postures. Achieve similar results.
  • a face recognition system and a face recognition method of the present invention will be described in detail below with reference to a specific embodiment.
  • the low-requirement facial recognition system of the present invention can be applied to the facial recognition of the onboard vision system of the robot.
  • the hardware resource used in this embodiment is the Samsung Exynos 5410 processing platform, and the software solution used is based on the C++ language implementation scheme. All functional modules are cured onto the processing platform.
  • the face feature library module 3 is initialized, that is, the five positions of the left, left 1, middle, right, and right 2 of each known identity face are collected, and the upper, middle, and lower positions are 8 in total.
  • the feature vector of the face image of the location is stored in the face feature library module 3, and then a face image is arbitrarily input to determine the identity of the face image. Its recognition speed is about 200 milliseconds, which satisfies the application scenario of robot interaction.
  • the present invention discloses a face recognition system and a face recognition method, which solves the face recognition problem when the illumination and the shooting angle are significantly changed, and the computing resources are relatively limited.
  • the technical solution of the present invention can be calculated.
  • the system with relatively limited resources can determine the most likely identity of the face to be recognized in real time and give a confidence rate; the scheme supports the identification of 20 to 50 people.
  • the illumination brightness on the face image changes uniformly, that is, the brightness may be different in different face images, but the face light in each face image is the same as the light, not the light side condition, the angle condition
  • the left and right deflections are within 40 degrees and the upper and lower deflections are within 30 degrees
  • the recognition accuracy of 20 people is over 90%
  • the accuracy rate of 50 people is over 80%.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Collating Specific Patterns (AREA)
  • Image Analysis (AREA)

Abstract

本发明涉及身份识别领域，尤其涉及一种面部识别系统及面部识别方法，主要包括光照预处理模块、特征生成模块、人脸特征库模块以及特征匹配与识别模块。该面部识别方法首先通过光照预处理模块对输入的人脸图像进行光线差异度优化，然后通过特征生成模块生成该人脸图像的特征向量，再将待识别的特征向量与特征库中的所有特征向量进行匹配计算，给出待识别的特征向量对应的身份结果。本发明的技术方案可以在计算资源较为有限的系统上实时判断出待识别的人脸图像的身份，耗时短准确率高。

Description

一种面部识别系统及面部识别方法 技术领域
本发明涉及身份识别领域，尤其涉及一种面部识别系统及面部识别方法。
背景技术
现有技术中，应用在门禁安防系统中的人脸识别技术方案解决的是特定位置、特定光线条件下采集的特定角度的人脸图像的身份识别问题。这一方案可以部署在高性能的计算机上，也可以部署在低计算资源的嵌入式计算机上。由于门禁安防系统允许结果响应有长至1秒甚至更长的延时，而且一般是一次性的身份验证需求，很少有连续处理的应用场景，所以必须使用计算复杂度较高的算法。
门禁安防系统使用的人脸识别技术方案具有以下缺陷：1、需要通过补偿光源来固定光线条件，对光线非常敏感；2、要求采集固定的人脸姿势，例如正脸；3、通常计算复杂度较高，给出一次结果耗时较多，连续检测时达不到实时响应需求。
发明内容
鉴于上述问题，本发明提供一种应用在机器人视觉系统的面部识别系统及面部识别方法，实现检测到人脸区域以后的人脸身份识别功 能，可以运用于各种光照条件，包括偏光或者无补偿光源的情况，且能识别多姿势的人脸，例如左侧、右侧、上仰、下俯甚至歪向一边的人脸，同时能降低计算复杂度使计算资源消耗降低到可以实时识别出现在机器人视野中的人脸，对响应速度要求较高，而且在人脸身份变化时通过连续识别实现实时反馈。
本发明解决上述技术问题所采用的技术方案为：
提供一种面部识别系统，其特征在于，包括：
光照预处理模块，该模块接收输入的待识别人脸图像，对所述待识别人脸图像进行光照差异度的优化后，将经过光照预处理的待识别人脸图像输出至下一模块；
特征生成模块，与所述光照预处理模块连接，以接收所述经过光照预处理的待识别人脸图像，并进行特征生成后，输出用于描述人脸细节特点的待识别的特征向量；
人脸特征库模块，预存储有已知人脸的特征向量；
特征匹配与识别模块，与所述特征生成模块以及人脸特征库模块连接，以将所述特征生成模块输出的待识别的特征向量与所述特征库中的所有特征向量进行匹配计算，输出所述待识别的特征向量对应的身份结果。
优选的，上述的面部识别系统，其中，所述光照预处理模块使用高斯差分的方法对所述待识别人脸图像进行处理。
优选的，上述的面部识别系统，其中，所述特征生成模块采用局部二值模式特征来描述所述人脸细节特点的特征向量。
优选的，上述的面部识别系统，其中，所述人脸特征库模块还提供新的人脸特征添加接口，通过该接口可以向所述人脸特征库模块加入已知身份的人的新人脸特征或未知身份的新人的人脸特征。
本发明还提供一种面部识别方法，其特征在于，基于权利要求1-4中任意一项所述的面部识别系统，所述方法包括以下步骤：
步骤1，初始化所述人脸特征库模块，以使所述人脸特征库模块中预存储已知的人脸特征向量；
步骤2，所述光照预处理模块接收输入的待识别人脸图像，对所述待识别人脸图像进行光照差异度的优化后，将经过光照预处理的待识别人脸图像输出；
步骤3，所述特征生成模块接收经过所述光照预处理模块处理过的待识别人脸图像，并对该待识别人脸图像进行特征生成操作后，输出用于描述人脸细节特点的特征向量；
步骤4，所述特征匹配与识别模块将所述特征生成模块输出的待识别的特征向量与所述特征库中的所有特征向量进行匹配计算，输出所述待识别的特征向量对应的身份结果。
优选的，上述的面部识别方法，其中，在所述步骤2中，所述光照预处理模块使用高斯差分的方法对所述人脸图像进行处理。
优选的，上述的面部识别方法，其中，在所述步骤3中，采用局部二值模式特征描述所述人脸细节特点的特征向量。
优选的，上述的面部识别方法，其中，所述步骤4中，若所述特征匹配与识别模块判定待识别人脸为未知人脸，则输出未知识别结 果，并将该未知人脸的特征向量存储至人脸特征库模块中。
上述技术方案具有如下优点或有益效果：
本发明的技术方案可以实现检测到人脸区域以后的人脸身份识别功能，可以运用于各种光照条件，包括偏光或者无补偿光源的情况，且能识别多姿势的人脸，例如左侧、右侧、上仰、下俯甚至歪向一边的人脸，同时能降低计算复杂度使计算资源消耗降低，响应速度较高，从而可以实时识别出现在机器人视野中的人脸，而且在人脸身份变化时通过连续识别实现实时反馈。
附图说明
通过阅读参照以下附图对非限制性实施例所作的详细描述，本发明及其特征、外形和优点将会变得更加明显。在全部附图中相同的标记指示相同的部分。并未可以按照比例绘制附图，重点在于示出本发明的主旨。
图1是本发明的面部识别系统的功能模块图。
具体实施方式
下面结合附图和具体的实施例对本发明作进一步的说明，但是不作为本发明的限定。
如图1所示，本发明的面部识别系统，主要应用于机器人的视觉系统，主要由光照预处理模块1、特征生成模块2、人脸特征库模块3以及特征匹配与识别模块4组成。
其中，光照预处理模块1对输入的待识别人脸图像进行光照差异度的优化，减少光照变化对系统识别率的影响。该模块使用高斯差分的方法对图像进行处理。光照变化在同一人脸图像上的影响可被视为在该图像上叠加了一个低频的信号，而对人脸图像进行高斯卷积运算可被视为进行低通滤波，利用两个不同大小卷积核的高斯卷积运算可以构造一个高斯差分运算，效果可视为一个带通滤波器，由于用于人脸识别的有效特征的频率通常比光照信号的频率高，所以适当选择两个卷积核的大小组成差分滤波器可以除去人脸图像中光线变化的影响。图像处理或图像增强技术中用于对光线变化进行处理的方法超过10种，在综合效果和计算资源后，在本实施例中选用了高斯差分的方法。特征生成模块2与光照预处理模块1连接，用以对光照预处理模块1输出的经过光照预处理的人脸图像进行特征生成，并输出描述人脸细节特点的特征向量。目前用来描述人脸的特征类型有主要成分分析(PCA)、独立成分分析(ICA)、局部二值模式特征(LBP)以及深度神经网络训练出的特征等几种，针对我们的应用需求，LBP是最佳的选择。LBP的优势有两点，第一是纯定点计算，计算量较其他特征大为减少，特别是在ARM(一种精简指令集的微处理器，这里也指利用该处理器进行搭建的计算系统)框架无浮点计算能力的处理器上，计算量优势更为明显；第二点是LBP本质上是一个类似差分的特征，能抵抗一定的光线变化。这样可以继续弱化光线预处理模块中未能消除的光线变化信息对系统识别率的不利影响。在特征生成模块2对待识别的人脸图像进行特征生成后，就将待识别的特征向量存储至 人脸特征库模块3中，人脸特征库模块3中预存储有已知人脸的特征向量。当需要识别人脸图像时，系统识别逻辑是把需要识别的人脸图像的特征向量与人脸特征库模块3中已知身份的人脸特征向量进行比对，找出特征最相似的人脸。这时需要通过查询人脸特征库模块3来完成识别步骤。由于单人脸图片的LBP特征只能被用于识别角度范围很小的人脸图片，因此本发明的技术方案通过收集更多角度的人脸图像特征，来管理同一个人不同角度的特征向量集。特征匹配与识别模块4与人脸特征库模块3和特征生成模块2连接，当有识别请求到来时，特征匹配与识别模块4将特征生成模块2输出的待识别的人脸特征向量与人脸特征库模块3中的所有特征向量进行匹配计算，找出最接近的一个特征向量并给出相似度值，根据相似度值的预设阈值判定待识别的人脸是否就是特征最接近的那个人脸。若是，给出相应的身份结果；若否，判定待识别人脸为未知人脸，如果此时有需求，可以将该未知人脸的特征向量添加到人脸特征库模块3中，同时需要添加身份信息，以便下次识别。
上述的功能模块组成了面部识别系统。现有的技术中还有一个人脸姿势矫正的技术可以用来减少人脸图像采集时不同的人脸姿势对识别精度的影响，但是要达到较好的效果，该功能模块需要较多的计算资源，使得计算量超出低资源系统的实时处理能力，所以我们用采集更多角度的人脸，同时通过特征匹配与识别模块4中的匹配方法使得匹配结果仍然能够解决不同姿势的人脸识别问题，达到相近的效果。
下面结合一具体实施例对本发明的一种面部识别系统及面部识别方法作详细说明。
本发明的低资源需求的面部识别系统，可运用于机器人板载视觉系统的面部识别，在本实施例中所用的硬件资源为三星Exynos 5410处理平台，所用的软件方案为基于c++语言实现方案的所有功能模块，固化到处理平台上。然后对人脸特征库模块3做初始化，即采集每个已知身份人脸左2、左1、中、右1和右2这5个位置，上、中下、这3个位置共8个位置的人脸图像的特征向量，存储到人脸特征库模块3中，然后任意输入一个人脸图像，判断出该人脸图像的身份。其识别速度大约在200毫秒，满足机器人交互的应用场景。
综上所述，本发明公开了一种面部识别系统及面部识别方法，解决了在光照和拍摄角度明显变化，而计算资源又较为有限时的人脸识别问题，本发明的技术方案可以在计算资源较为有限的系统上实时判断出待识别人脸最可能的身份，并给出置信率；该方案支持20～50个人的识别。当限定以下光照条件为在人脸图像上的光照亮度均匀变化，即不同人脸图像中亮度可以不同，但一张人脸图像中各处要同光照而非有明暗的侧光情况，角度条件为左右偏转各40度以内，上下偏转各30度以内时，20人识别准确率达90％以上，50人准确率达80％以上。
本领域技术人员应该理解，本领域技术人员在结合现有技术以及上述实施例可以实现所述变化例，在此不做赘述。这样的变化例并不影响本发明的实质内容，在此不予赘述。
以上对本发明的较佳实施例进行了描述。需要理解的是，本发明并不局限于上述特定实施方式，其中未尽详细描述的设备和结构应该理解为用本领域中的普通方式予以实施；任何熟悉本领域的技术人员，在不脱离本发明技术方案范围情况下，都可利用上述揭示的方法和技术内容对本发明技术方案做出许多可能的变动和修饰，或修改为等同变化的等效实施例，这并不影响本发明的实质内容。因此，凡是未脱离本发明技术方案的内容，依据本发明的技术实质对以上实施例所做的任何简单修改、等同变化及修饰，均仍属于本发明技术方案保护的范围内。

Claims (8)

  1. 一种面部识别系统，其特征在于，包括：
    光照预处理模块，该模块接收输入的待识别人脸图像，对所述待识别人脸图像进行光照差异度的优化后，将经过光照预处理的待识别人脸图像输出至下一模块；
    特征生成模块，与所述光照预处理模块连接，以接收所述经过光照预处理的待识别人脸图像，并进行特征生成后，输出用于描述人脸细节特点的待识别的特征向量；
    人脸特征库模块，预存储有已知人脸的特征向量；
    特征匹配与识别模块，与所述特征生成模块以及人脸特征库模块连接，以将所述特征生成模块输出的待识别的特征向量与所述特征库中的所有特征向量进行匹配计算，输出所述待识别的特征向量对应的身份结果。
  2. 根据权利要求1所述的面部识别系统，其特征在于，所述光照预处理模块使用高斯差分的方法对所述待识别人脸图像进行处理。
  3. 根据权利要求1所述的面部识别系统，其特征在于，所述特征生成模块采用局部二值模式特征来描述所述人脸细节特点的特征向量。
  4. 根据权利要求1所述的面部识别系统，其特征在于，所述人脸特征库模块还提供新的人脸特征添加接口，通过该接口可以向所述人脸特征库模块加入已知身份的人的新人脸特征或未知身份的新人的人脸特征。
  5. 一种面部识别方法，其特征在于，基于权利要求1-4中任意 一项所述的面部识别系统，所述方法包括以下步骤：
    步骤1，初始化所述人脸特征库模块，以使所述人脸特征库模块中预存储已知的人脸特征向量；
    步骤2，所述光照预处理模块接收输入的待识别人脸图像，对所述待识别人脸图像进行光照差异度的优化后，将经过光照预处理的待识别人脸图像输出；
    步骤3，所述特征生成模块接收经过所述光照预处理模块处理过的待识别人脸图像，并对该待识别人脸图像进行特征生成操作后，输出用于描述人脸细节特点的特征向量；
    步骤4，所述特征匹配与识别模块将所述特征生成模块输出的待识别的特征向量与所述特征库中的所有特征向量进行匹配计算，输出所述待识别的特征向量对应的身份结果。
  6. 根据权利要求5所述的面部识别方法，其特征在于，在所述步骤2中，所述光照预处理模块使用高斯差分的方法对所述人脸图像进行处理。
  7. 根据权利要求5所述的面部识别方法，其特征在于，在所述步骤3中，采用局部二值模式特征描述所述人脸细节特点的特征向量。
  8. 根据权利要求5所述的面部识别方法，其特征在于，所述步骤4中，若所述特征匹配与识别模块判定待识别人脸为未知人脸，则输出未知识别结果，并将该未知人脸的特征向量存储至人脸特征库模块中。
PCT/CN2016/086631 2015-06-30 2016-06-21 一种面部识别系统及面部识别方法 Ceased WO2017000816A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
EP16817180.9A EP3319010A4 (en) 2015-06-30 2016-06-21 FACE DETECTION SYSTEM AND FACIAL RECOGNITION METHOD
JP2017567755A JP2018525718A (ja) 2015-06-30 2016-06-21 顔認識システム及び顔認識方法
US15/741,063 US10438056B2 (en) 2015-06-30 2016-06-21 Face recognition system and face recognition method

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510383515.9 2015-06-30
CN201510383515.9A CN106326816A (zh) 2015-06-30 2015-06-30 一种面部识别系统及面部识别方法

Publications (1)

Publication Number Publication Date
WO2017000816A1 true WO2017000816A1 (zh) 2017-01-05

Family

ID=57607768

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/086631 Ceased WO2017000816A1 (zh) 2015-06-30 2016-06-21 一种面部识别系统及面部识别方法

Country Status (7)

Country Link
US (1) US10438056B2 (zh)
EP (1) EP3319010A4 (zh)
JP (1) JP2018525718A (zh)
CN (1) CN106326816A (zh)
HK (1) HK1231601A1 (zh)
TW (1) TW201701192A (zh)
WO (1) WO2017000816A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109063691A (zh) * 2018-09-03 2018-12-21 武汉普利商用机器有限公司 一种人脸识别底库优化方法及系统

Families Citing this family (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9971847B2 (en) * 2016-01-07 2018-05-15 International Business Machines Corporation Automating browser tab groupings based on the similarity of facial features in images
CN108228871A (zh) 2017-07-21 2018-06-29 北京市商汤科技开发有限公司 人脸图像动态入库方法和装置、电子设备、介质、程序
CN107679570A (zh) * 2017-09-27 2018-02-09 广东欧珀移动通信有限公司 解锁控制方法及相关产品
US10621416B2 (en) * 2017-10-02 2020-04-14 Microsoft Technology Licensing, Llc Image processing for person recognition
US10747989B2 (en) * 2018-08-21 2020-08-18 Software Ag Systems and/or methods for accelerating facial feature vector matching with supervised machine learning
TWI688902B (zh) * 2018-08-31 2020-03-21 國立中正大學 應用於表情辨識之拓展式局部二值模式方法及其系統
CN108932783A (zh) * 2018-09-19 2018-12-04 南京邮电大学 一种基于二维人脸识别的面向大流量场景的门禁系统
CN109377616B (zh) * 2018-10-30 2021-02-12 南京邮电大学 一种基于二维人脸识别的门禁控制系统
CN111259684B (zh) * 2018-11-30 2023-07-28 Tcl科技集团股份有限公司 一种确定摄像头到人脸距离的方法及装置
CN109614962B (zh) * 2019-01-24 2022-11-18 深圳市梦网视讯有限公司 一种偏光光源人脸图像检测方法和系统
CN109919041A (zh) * 2019-02-16 2019-06-21 天津大学 一种基于智能机器人的人脸识别方法
TWI701607B (zh) * 2019-02-26 2020-08-11 英屬開曼群島商麥迪創科技股份有限公司 動態臉部識別系統與動態臉部識別方法
CN110097668A (zh) * 2019-04-19 2019-08-06 合肥宇科电子科技有限公司 一种人脸识别监测控制系统
US10650824B1 (en) * 2019-05-10 2020-05-12 Fmr Llc Computer systems and methods for securing access to content provided by virtual assistants
CN112395436B (zh) * 2019-08-14 2024-07-02 天津极豪科技有限公司 一种底库录入方法及装置
CN110807403B (zh) * 2019-10-29 2022-12-02 中新智擎科技有限公司 一种用户身份识别方法、装置及电子设备
CN112668514A (zh) * 2020-12-31 2021-04-16 云从科技集团股份有限公司 人脸识别的采集控制方法及系统、控制装置、存储介质
CN113743308B (zh) * 2021-09-06 2023-12-12 汇纳科技股份有限公司 基于特征质量的人脸识别方法、装置、存储介质及系统
CN118172860B (zh) * 2024-05-13 2024-07-12 深圳市西伦土木结构有限公司 一种基于身份识别的智能校园门禁系统

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102663400A (zh) * 2012-04-16 2012-09-12 北京博研新创数码科技有限公司 一种结合预处理的lbp特征提取方法
CN103761514A (zh) * 2014-01-26 2014-04-30 公安部第三研究所 基于广角枪机和多球机实现人脸识别的系统及方法
CN104268539A (zh) * 2014-10-17 2015-01-07 中国科学技术大学 一种高性能的人脸识别方法及系统

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP3469031B2 (ja) * 1997-02-18 2003-11-25 株式会社東芝 顔画像登録装置及びその方法
JP3886660B2 (ja) * 1999-03-11 2007-02-28 株式会社東芝 人物認識装置における登録装置及びその方法
CN101561874B (zh) * 2008-07-17 2011-10-26 清华大学 一种人脸虚拟图像生成的方法
WO2011055164A1 (en) * 2009-11-06 2011-05-12 Vesalis Method for illumination normalization on a digital image for performing face recognition
US9082235B2 (en) * 2011-07-12 2015-07-14 Microsoft Technology Licensing, Llc Using facial data for device authentication or subject identification
CN102779273B (zh) * 2012-06-29 2016-08-03 重庆邮电大学 一种基于局部对比模式的人脸识别方法
JP6074182B2 (ja) * 2012-07-09 2017-02-01 キヤノン株式会社 画像処理装置及び画像処理方法、プログラム
JP6268960B2 (ja) * 2013-11-15 2018-01-31 オムロン株式会社 画像認識装置及び画像認識装置に対するデータ登録方法
CN103714347B (zh) * 2013-12-30 2017-08-25 汉王科技股份有限公司 人脸识别方法及人脸识别装置

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102663400A (zh) * 2012-04-16 2012-09-12 北京博研新创数码科技有限公司 一种结合预处理的lbp特征提取方法
CN103761514A (zh) * 2014-01-26 2014-04-30 公安部第三研究所 基于广角枪机和多球机实现人脸识别的系统及方法
CN104268539A (zh) * 2014-10-17 2015-01-07 中国科学技术大学 一种高性能的人脸识别方法及系统

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3319010A4 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109063691A (zh) * 2018-09-03 2018-12-21 武汉普利商用机器有限公司 一种人脸识别底库优化方法及系统

Also Published As

Publication number Publication date
HK1231601A1 (zh) 2017-12-22
EP3319010A1 (en) 2018-05-09
US20180121714A1 (en) 2018-05-03
US10438056B2 (en) 2019-10-08
JP2018525718A (ja) 2018-09-06
CN106326816A (zh) 2017-01-11
TW201701192A (zh) 2017-01-01
EP3319010A4 (en) 2019-02-27

Similar Documents

Publication Publication Date Title
WO2017000816A1 (zh) 一种面部识别系统及面部识别方法
CN113302620B (zh) 使用机器学习模型确定对象与人之间的关联
US10776470B2 (en) Verifying identity based on facial dynamics
KR102477190B1 (ko) 얼굴 인식 방법 및 장치
TWI677825B (zh) 視頻目標跟蹤方法和裝置以及非易失性電腦可讀儲存介質
Fard et al. Asmnet: A lightweight deep neural network for face alignment and pose estimation
WO2019232862A1 (zh) 嘴巴模型训练方法、嘴巴识别方法、装置、设备及介质
WO2019232866A1 (zh) 人眼模型训练方法、人眼识别方法、装置、设备及介质
Kalas Real time face detection and tracking using OpenCV
WO2017000764A1 (zh) 一种手势检测识别方法及系统
US10311287B2 (en) Face recognition system and method
Dahmane et al. Head pose estimation based on face symmetry analysis
Tathe et al. Human face detection and recognition in videos
Uke et al. Optimal video processing and soft computing algorithms for human hand gesture recognition from real-time video
CN110348272B (zh) 动态人脸识别的方法、装置、系统和介质
CN110807391A (zh) 基于视觉的人-无人机交互用人体姿态指令识别方法
Rana et al. Real time deep learning based face recognition system using Raspberry PI
Monisha et al. A real-time embedded system for human action recognition using template matching
Geetha et al. 3D face recognition using Hadoop
Mariappan et al. A labVIEW design for frontal and non-frontal human face detection system in complex background
Tang et al. Research on real-time face recognition and tracking model based on embedded platform
Liu et al. Robust face detection with eyes occluded by the shadow from dazzling avoidance system
Alex et al. Gradient feature matching for in-plane rotation invariant face sketch recognition
Ni et al. Fast iris segmentation under partly occlusion based on MTCNN and weighted FCN
Chang et al. Design and optimization of multiple heads detection for embedded system

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: 16817180

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2017567755

Country of ref document: JP

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: 15741063

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE

WWE Wipo information: entry into national phase

Ref document number: 2016817180

Country of ref document: EP