WO2017000807A1 - 一种人脸图像识别方法 - Google Patents

一种人脸图像识别方法 Download PDF

Info

Publication number
WO2017000807A1
WO2017000807A1 PCT/CN2016/086481 CN2016086481W WO2017000807A1 WO 2017000807 A1 WO2017000807 A1 WO 2017000807A1 CN 2016086481 W CN2016086481 W CN 2016086481W WO 2017000807 A1 WO2017000807 A1 WO 2017000807A1
Authority
WO
WIPO (PCT)
Prior art keywords
face image
training sample
user
album
smart terminal
Prior art date
Application number
PCT/CN2016/086481
Other languages
English (en)
French (fr)
Inventor
黄超
蔡明峻
Original Assignee
芋头科技(杭州)有限公司
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 芋头科技(杭州)有限公司 filed Critical 芋头科技(杭州)有限公司
Publication of WO2017000807A1 publication Critical patent/WO2017000807A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16Human faces, e.g. facial parts, sketches or expressions
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting

Definitions

  • the present invention relates to the field of image recognition technologies, and in particular, to a face image recognition method.
  • face recognition is increasingly used in smart terminals.
  • the user can call up some other related data including the name of the person, the contact method, the user account of the social network, etc., and the user can also implement the encryption and decryption operation by recognizing the face of the other person.
  • the existing face recognition technology usually requires the user to manually input a large number of image samples required for recognition in advance, and the input manner may be batch import, or pre-production acquisition.
  • the user's entire operation is very cumbersome, greatly reducing the user experience.
  • a technical solution for a face image recognition method which aims to solve the problem that the user in the prior art needs to manually input a large amount of image sample data for face recognition, and set one by one.
  • a face image recognition method is applicable to a smart terminal; wherein a pre-training step is included to obtain a corresponding training sample;
  • the pre-training step specifically includes:
  • Step S1 The smart terminal remotely acquires a communication directory and a user album stored in the user's mobile terminal, and uses the face image in the user album to match the personal avatar in the communication directory to establish the a first training sample associated with a personal image corresponding to the personal avatar in the communication directory; and/or
  • the smart terminal remotely searches for and obtains a preset friend list corresponding to a user account associated with the social network and a friend album, and according to the face image and the friend list in the friend album Matching the personal avatar in the middle to establish a second training sample associated with the personal name of the face image corresponding to the personal avatar in the buddy list;
  • Step S2 combining the first training sample and/or the second training sample to form a preliminary training sample
  • Step S3 adopting the face image included in the preliminary training sample, and performing identification confirmation according to the preliminary training sample to complete and form a final identification data;
  • the smart terminal collects a face image that needs to be recognized, and performs recognition according to the identification data, and outputs a recognition result.
  • the method for recognizing a face image wherein the step of forming the first training sample in the step S1 specifically includes:
  • Step S11a remotely acquiring the face image in the user album
  • Step S12a remotely obtaining the communication directory
  • Step S13a Perform matching in the communication directory by using the face image to find a matching personal avatar in the communication directory;
  • Step S14a associating the face image with a personal name corresponding to the matched personal avatar
  • Step S15a the steps S11a-S14a are cyclically executed to form the first training sample according to all the face images in the user album.
  • the method for recognizing a face image wherein the step of forming the second training sample in the step S1 specifically includes:
  • Step S11b remotely obtaining the face image in the friend album associated with the user account by using the user account associated with the at least one social networking site;
  • Step S12b remotely obtaining a buddy list associated with the user account by using the user account associated with the at least one social networking site;
  • Step S13b Perform matching in the buddy list by using the face image to find a matching personal avatar in the buddy list;
  • Step S14b associating the face image with a personal name corresponding to the matching personal avatar in the buddy list
  • Step S15b the steps S11b-14b are cyclically executed to form the second training sample according to all the face images in the friend album.
  • the method for identifying a face image wherein, in the step S1, the smart terminal accesses the mobile terminal by using a wireless connection, and obtains the corresponding communication directory and the User photo album.
  • the method for recognizing a face image wherein the step S3 specifically includes:
  • Step S31 collecting, in the scene, a face image included in the preliminary training sample
  • Step S32 identifying and identifying the face image according to the preliminary training sample:
  • step S33 If not recognized, updating the preliminary training sample according to the face image; then moving to step S33;
  • step S33 If it can be identified, go directly to step S33;
  • Step S33 the steps S31-32 are performed cyclically to complete the preliminary training sample according to the plurality of the face images;
  • Step S34 integrating the improved preliminary training samples to form the identification data, and storing the identification data in the smart terminal.
  • the smart terminal confirms by the user corresponding to the face image collected by the scene by issuing a prompt sound.
  • the face image recognition method wherein the smart terminal is an intelligent terminal having a robot appearance.
  • the above technical solution has the beneficial effects of providing a face image recognition method, so that the smart terminal can automatically recognize and train the face recognition data according to the user's relationship network, thereby avoiding the need for the user to manually input a large amount of face recognition.
  • the cumbersome operation of the image sample enhances the user experience.
  • 1-4 are schematic flowcharts of a face image recognition method in a preferred embodiment of the present invention.
  • a technical solution of a face image recognition method is provided, which is applicable to a smart terminal.
  • the above-described face image recognition method includes a pre-training step.
  • the so-called pre-training step refers to a process of establishing identification data required for recognition before actually putting the smart terminal into an actual face recognition operation.
  • a pre-training step can be used to establish a library of information for identification in the smart terminal.
  • the pre-training step specifically includes:
  • Step S1 The smart terminal remotely acquires a communication directory and a user album stored in the user's mobile terminal, and uses the image sample in the user album to match the personal avatar in the communication directory to establish an image avatar and a personal avatar in the communication directory. a first training sample associated with the corresponding individual name; and/or
  • the smart terminal remotely searches for and obtains a preset friend list corresponding to the user account associated with the social network and the friend album, and matches the personal image in the friend list according to the image sample in the friend album to establish an image sample and a friend list.
  • a second training sample associated with the personal name of the corresponding personal avatar;
  • only the first training sample or the second training sample may be selected.
  • the first training sample and the second training sample described above may be simultaneously formed.
  • the step of forming the first training sample specifically includes:
  • Step S11a remotely obtaining a face image in the user album
  • the preset face image may be a face image collected in the field, or may be a face image that has been collected before, such as a photo of a friend who has been posted on a social network.
  • Step S12a remotely obtaining a mailing list
  • the so-called communication directory refers to a preset list of personal names and personal avatars associated therewith, and correspondingly, other information related to individuals, such as contact information, may be included in the directory list. And/or home address, and/or user accounts for social networks, etc.
  • the communication directory is stored in a user's mobile terminal.
  • the smart terminal can remotely obtain the above-mentioned communication directory. Specifically, the smart terminal establishes a wireless connection with the user's mobile terminal, and obtains a communication directory stored in the mobile terminal through the wireless connection, that is, the user authorizes the smart terminal to remotely access the user's communication directory.
  • the smart terminal and the user's mobile terminal are usually connected by a short distance, data can be transmitted by using a Bluetooth connection. In other embodiments of the present invention, other wireless connection methods may also be used to transmit data.
  • Step S13a matching the face image in the communication directory to find a matching personal avatar in the communication directory;
  • Step S14a associating a face image with a personal name corresponding to the matched personal avatar
  • step S15a steps S11a-S14a are cyclically executed to form a first training sample according to all face images in the user album.
  • the face image included in the user album refers to a photo of the user album, and the photo having the face shape can be clearly recognized.
  • each face image in the user album is matched with the personal avatar included in the communication directory to find a personal avatar with a high degree of matching, and the corresponding face is obtained.
  • the image is associated with the person's name corresponding to the person's avatar.
  • the face image is associated with corresponding personal information (including personal name, contact information, user account on social network, email, etc.).
  • all face images in the user's album are traversed and a first training sample is finally formed.
  • the smart terminal automatically forms a sample library for face recognition by accessing the communication directory and the user album in the mobile terminal.
  • the method for forming the second training sample specifically includes:
  • Step S11b remotely obtaining the user account by using a user account associated with at least one social networking website The face image in the associated friend's album;
  • Step S12b remotely obtaining a buddy list associated with the user account by using a user account associated with the at least one social networking site;
  • Step S13b matching the face image in the buddy list to find the matching personal avatar in the buddy list;
  • Step S14b associating the face image with the personal name corresponding to the matching personal avatar in the buddy list;
  • steps S11b-14b are cyclically executed to form a second training sample according to all face images in the friend album.
  • the foregoing steps S11b-15b are similar to the steps S11a-15a, except that the smart terminal forms the second training by referring to the buddy list and the buddy album in the user account on the user's social network. sample.
  • the smart terminal connects to the social networking website by accessing the Internet, and obtains the buddy list and the buddy album of the corresponding user account after being authorized.
  • the smart terminal extracts a face image in the friend album, and matches the personal face image in the buddy list according to the extracted face image to match the face image with the matched personal avatar.
  • Personal name is associated.
  • the face image is associated with personal information corresponding to the matching personal avatar (including personal name, contact information, user account on the social network, email, etc.).
  • the face image in the friend album is also an image in the photo of the friend album that can clearly recognize the shape of the face.
  • the smart terminal traverses all face images in the friend's album and finally forms a second training sample.
  • Step S2 combining the first training sample and/or the second training sample to form a preliminary training sample
  • the corresponding training sample is set as the preliminary training sample.
  • the two training samples are combined to form a preliminary training sample.
  • Step S3 adopting a face image included in the preliminary training sample, and performing identification confirmation according to the preliminary training sample to complete and form a final identification data;
  • step S3 specifically includes:
  • Step S31 collecting a face image included in the preliminary training sample on the spot
  • the so-called on-site acquisition refers to the on-site collection of a face image associated with a specific person associated with the preliminary training sample, such as shooting acquisition, etc., and then taking the face image as an input, Face image recognition based on preliminary training samples.
  • the object collected in the field may be limited to The person included in the user's relationship network.
  • the face image collected in the above manner can also be associated with a stranger, that is, a face image that does not exist in the preliminary training sample, so that self-learning of the preliminary training sample can be realized.
  • a stranger that is, a face image that does not exist in the preliminary training sample, so that self-learning of the preliminary training sample can be realized.
  • Step S32 identifying and confirming the face image according to the preliminary training sample:
  • step S33 If not recognized, the preliminary training sample is updated according to the face image; then the process proceeds to step S33;
  • step S33 If it can be identified, go directly to step S33;
  • the so-called identification confirmation refers to confirming the recognition accuracy of the preliminary training sample.
  • a face image is collected in the field and input into the smart terminal, and the smart terminal obtains a corresponding result according to the preliminary training sample matching (for example, the corresponding personal name).
  • the smart terminal will emit a prompt tone, for example, prompting the user to "Is this identification correct?". If the user confirms, the identification is correct; otherwise, the identification error needs to be corrected.
  • Step S33 performing steps S31-32 cyclically to complete the preliminary training samples according to the plurality of face images
  • the work of performing identification confirmation is performed according to a plurality of face image loops. If an identification error occurs, the user may manually input the matched personal information or modify the sample.
  • the data of the library and other methods improve the preliminary training samples. The above steps S31-32 are cycled, and the preliminary training samples are identified and the sample library is perfected with as many face images as possible.
  • step S34 the improved preliminary training samples are integrated to form identification data, and are stored in the smart terminal.
  • the second training sample and the perfect training sample are obtained by acquiring the training samples, performing pattern recognition on the sample of the image samples corresponding to each individual name, screening the repeated results after cross matching, and finally merging the preliminary training samples after the evening.
  • identification data ie, the identification model formed after the combination and its associated information, such as personal information, etc.
  • the identification data is finally archived to the relationship of the user saved in the smart terminal. In the network.
  • the smart terminal can collect the face to be recognized, and identify according to the identification data, and output the recognition result.
  • the smart terminal can identify the face using the finally formed identification data.
  • the smart terminal may be a smart terminal having a robot appearance.
  • the above-described face image recognition method can be applied to a robot device that can exchange information with a user.

Abstract

本发明公开了一种人脸图像识别方法,属于图像识别技术领域;方法包块:步骤S1,采用通讯名录和用户相册训练生成第一训练样本;和/或采用社交网站的用户账号的好友列表和好友相册训练生成第二训练样本;步骤S2,结合第一训练样本和/或第二训练样本以形成一初步训练样本;步骤S3,采用包括于初步训练样本中的人脸图像,并根据初步训练样本进行识别确认,以完善并形成一最终的识别数据;智能终端根据识别数据执行人脸图像的识别操作。上述技术方案的有益效果是:使得智能终端能够根据使用者的关系网络自动识别并训练人脸识别数据,从而避免使用者需要手动输入大量人脸识别所需的图片样本的繁琐操作,提升使用者的使用体验。

Description

一种人脸图像识别方法 技术领域
本发明涉及图像识别技术领域,尤其涉及一种人脸图像识别方法。
背景技术
人脸识别作为一种新兴的识别技术,越来越多地应用于智能终端中。例如使用者可以通过辨别人脸调出一些关联数据包括人名、联系方式、社交网络的用户账号等,使用者同样可以通过辨别人脸的方式实现加解密操作等等。
但是,现有的人脸识别技术,通常需要使用者预先手动输入识别所需的大量图像样本,输入的方式可以为批量导入,或者预先摄制采集等。但是无论哪种手动输入方式,都使得使用者的整个操作非常繁琐,大大降低了使用者的使用体验。
发明内容
根据现有技术中存在的问题,现提供一种人脸图像识别方法的技术方案,旨在解决现有技术中存在的使用者需要手动输入大量供人脸识别的图像样本数据,并一一设置图像与人名的关系,从而导致操作非常繁琐的缺陷;
上述技术方案具体包括:
一种人脸图像识别方法,适用于智能终端;其中,包括一预训练步骤,以得到相应的训练样本;
所述预训练步骤具体包括:
步骤S1,所述智能终端远程获取保存于使用者的移动终端内的通讯名录以及用户相册,采用所述用户相册中的人脸图像与所述通讯名录中的个人头像进行匹配,以建立所述人脸图像与所述通讯名录中的所述个人头像对应的个人姓名相关联的第一训练样本;和/或
所述智能终端远程查找并获取预设的关联于社交网络的用户账号对应的好友列表以及好友相册,并根据所述好友相册中的人脸图像与所述好友列表 中的个人头像进行匹配,以建立所述人脸图像与所述好友列表中的所述个人头像相对应的个人姓名相关联的第二训练样本;
步骤S2,结合所述第一训练样本和/或所述第二训练样本以形成一初步训练样本;
步骤S3,采用包括于所述初步训练样本中的所述人脸图像,并根据所述初步训练样本进行识别确认,以完善并形成一最终的识别数据;
形成所述识别数据后,所述智能终端采集需要识别的人脸图像,并根据所述识别数据进行识别,输出识别结果。
优选的,该人脸图像识别方法,其中,所述步骤S1中,形成所述第一训练样本的步骤具体包括:
步骤S11a,远程获取所述用户相册中的所述人脸图像;
步骤S12a,远程获取所述通讯名录;
步骤S13a,采用所述人脸图像在所述通讯名录中进行匹配,以找到所述通讯名录中相匹配的个人头像;
步骤S14a,将所述人脸图像与相匹配的个人头像对应的个人姓名相关联;
步骤S15a,循环执行所述步骤S11a-S14a,以根据所述用户相册中的所有所述人脸图像训练形成所述第一训练样本。
优选的,该人脸图像识别方法,其中,所述步骤S1中,形成所述第二训练样本的步骤具体包括:
步骤S11b,通过关联于至少一个社交网站的所述用户账号,远程获取所述用户账号相关联的所述好友相册中的所述人脸图像;
步骤S12b,通过关联于至少一个社交网站的所述用户账号,远程获取所述用户账号相关联的好友列表;
步骤S13b,采用所述人脸图像在所述好友列表中进行匹配,以找到所述好友列表中相匹配的个人头像;
步骤S14b,将所述人脸图像与所述好友列表中相匹配的个人头像对应的个人姓名相关联;
步骤S15b,循环执行所述步骤S11b-14b,以根据所述好友相册中的所有所述人脸图像训练形成所述第二训练样本。
优选的,该人脸图像识别方法,其中,所述步骤S1中,所述智能终端通过无线连接方式接入所述移动终端,并从所述移动终端内获取相应的所述通讯名录和所述用户相册。
优选的,该人脸图像识别方法,其中,所述步骤S3具体包括:
步骤S31,现场采集包括于所述初步训练样本中的一人脸图像;
步骤S32,根据所述初步训练样本对所述人脸图像进行识别确认:
若无法识别,则根据所述人脸图像更新所述初步训练样本;随后转向步骤S33;
若能够识别,则直接转向步骤S33;
步骤S33,循环执行所述步骤S31-32,以根据多张所述人脸图像完善所述初步训练样本;
步骤S34,将经过完善的所述初步训练样本整合形成所述识别数据,并保存于所述智能终端内。
优选的,该人脸图像识别方法,其中,所述步骤S32中,所述智能终端通过发出提示音以供现场采集的所述人脸图像对应的使用者确认。
优选的,该人脸图像识别方法,其中,所述智能终端为具有机器人外观的智能终端。
上述技术方案的有益效果是:提供一种人脸图像识别方法,使得智能终端能够根据使用者的关系网络自动识别并训练人脸识别数据,从而避免使用者需要手动输入大量人脸识别所需的图片样本的繁琐操作,提升使用者的使用体验。
附图说明
图1-4是本发明的较佳的实施例中,一种人脸图像识别方法的流程示意图。
具体实施方式
下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本发明一部分实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有作 出创造性劳动的前提下所获得的所有其他实施例,都属于本发明保护的范围。
需要说明的是,在不冲突的情况下,本发明中的实施例及实施例中的特征可以相互组合。
下面结合附图和具体实施例对本发明作进一步说明,但不作为本发明的限定。
本发明的较佳的实施例中,基于现有技术中存在的上述问题,现提供一种人脸图像识别方法的技术方案,适用于智能终端。
本发明的较佳的实施例中,上述人脸图像识别方法中包括一预训练步骤。所谓预训练步骤,是指在实际将智能终端投入实际的人脸识别操作之前,建立识别所需的识别数据的过程。换言之,本发明的较佳的实施例中,通过预训练步骤,可以预先在智能终端中建立一个用于识别的信息库。
则本发明的较佳的实施例中,如图1所示,上述预训练步骤具体包括:
步骤S1,智能终端远程获取保存于使用者的移动终端内的通讯名录以及用户相册,采用用户相册中的图像样本与通讯名录中的个人头像进行匹配,以建立图像样本与通讯名录中的个人头像对应的个人姓名相关联的第一训练样本;和/或
智能终端远程查找并获取预设的关联于社交网络的用户账号对应的好友列表以及好友相册,并根据好友相册中的图像样本与好友列表中的个人头像进行匹配,以建立图像样本与好友列表中的个人头像相对应的个人姓名相关联的第二训练样本;
本发明的一个较佳的实施例中,可以选择仅形成上述第一训练样本或者上述第二训练样本。
本发明的另一个较佳的实施例中,可以同时形成上述第一训练样本和第二训练样本。
则本发明的较佳的实施例中,如图2所示,上述步骤S1中,形成第一训练样本的步骤具体包括:
步骤S11a,远程获取用户相册中的人脸图像;
本发明的较佳的实施例中,预设的人脸图像可以是现场采集的人脸图像,也可以是之前已经采集完成的人脸图像,例如社交网络上已经发布的朋友的大头照等。
步骤S12a,远程获取通讯名录;
本发明的较佳的实施例中,所谓通讯名录,是指预设的包括个人姓名以及与其关联的个人头像的列表,相应地在名录列表中可以包括其他关联于个人的信息,例如联系方式,和/或家庭住址,和/或社交网络的用户账号等。本发明的较佳的实施例中,上述通讯名录保存于使用者的移动终端内。
本发明的较佳的实施例中,智能终端可以远程获取上述通讯名录。具体地,智能终端与使用者的移动终端之间建立无线连接,并通过无线连接获取移动终端内保存的通讯名录,即使用者授权智能终端远程访问使用者的通讯名录。本发明的较佳的实施例中,由于智能终端与使用者的移动终端之间通常是短距离连接,因此可以采用蓝牙连接方式传输数据。本发明的其他实施例中,同样可以采用其他无线连接方式传输数据。
步骤S13a,采用人脸图像在通讯名录中进行匹配,以找到通讯名录中相匹配的个人头像;
步骤S14a,将人脸图像与相匹配的个人头像对应的个人姓名相关联;
步骤S15a,循环执行步骤S11a-S14a,以根据用户相册中的所有人脸图像训练形成第一训练样本。
本发明的较佳的实施例中,上述用户相册中包括的人脸图像,是指用户相册的照片中,能够清晰识别出其中具有人脸形状的照片。
本发明的较佳的实施例中,利用上述用户相册中的每个人脸图像,与通讯名录中包括的个人头像进行一一匹配,以找到匹配程度较高的个人头像,并将相应的人脸图像与该个人头像对应的个人姓名相关联。进一步地,本发明的较佳的实施例中,经过匹配后,将人脸图像与相应的个人信息(包括个人姓名、联系方式、社交网络上的用户账号、电子邮件等)进行关联。
本发明的较佳的实施例中,遍历用户相册中的所有人脸图像,并最终形成一第一训练样本。
换言之,本发明的较佳的实施例中,智能终端通过访问移动终端内的通讯名录和用户相册自动形成一人脸识别的样本库。
本发明的较佳的实施例中,如图3所示,上述步骤1中,形成第二训练样本的方法具体包括:
步骤S11b,通过关联于至少一个社交网站的用户账号,远程获取用户账 号相关联的好友相册中的人脸图像;
步骤S12b,通过关联于至少一个社交网站的用户账号,远程获取用户账号相关联的好友列表;
步骤S13b,采用人脸图像在好友列表中进行匹配,以找到好友列表中相匹配的个人头像;
步骤S14b,将人脸图像与好友列表中相匹配的个人头像对应的个人姓名相关联;
步骤S15b,循环执行步骤S11b-14b,以根据好友相册中的所有人脸图像训练形成第二训练样本。
本发明的较佳的实施例中,上述步骤S11b-15b与步骤S11a-15a类似,区别在于:智能终端通过借鉴使用者的社交网络上的用户账号中的好友列表以及好友相册来形成第二训练样本。本发明的较佳的实施例中,智能终端通过接入互联网的方式连接至社交网站,经过授权后获取相应的用户账号的好友列表和好友相册。
本发明的较佳的实施例中,智能终端提取好友相册中的人脸图像,并根据被提取的人脸图像匹配好友列表中的个人头像,以将人脸图像与相匹配的个人头像对应的个人姓名相关联。进一步地,将人脸图像与相匹配的个人头像对应的个人信息(包括个人姓名、联系方式、社交网络上的用户账号、电子邮件等)相关联。
本发明的较佳的实施例中,好友相册中的人脸图像,同样为好友相册的照片中能够清晰识别人脸形状的图像。
本发明的较佳的实施例中,智能终端遍历好友相册中所有的人脸图像并最终形成一第二训练样本。
步骤S2,结合第一训练样本和/或第二训练样本以形成一初步训练样本;
本发明的一个较佳的实施例中,若只有第一训练样本或者第二训练样本,则将相应的训练样本设置为初步训练样本。
本发明的另一个较佳的实施例中,若同时形成第一训练样本和第二训练样本,则将这两个训练样本进行合并,以形成初步训练样本。
步骤S3,采用包括于初步训练样本中的人脸图像,并根据初步训练样本进行识别确认,以完善并形成一最终的识别数据;
本发明的较佳的实施例中,如图4所示,上述步骤S3具体包括:
步骤S31,现场采集包括于初步训练样本中的一人脸图像;
本发明的较佳的实施例中,所谓现场采集,是指现场对关联于初步训练样本的一个特定的人进行人脸图像的采集,例如拍摄采集等,随后将该人脸图像作为输入,以根据初步训练样本进行人脸图像识别。
本发明的一个较佳的实施例中,由于上述训练样本的形成依赖于使用者的关系网络(例如通讯名录和社交网络的用户账号对应的好友列表),因此上述现场采集的对象可以被限定为使用者的关系网络中包括的人。
本发明的另一个较佳的实施例中,上述现场采集的人脸图像同样可以被关联于一个陌生人,即不存在于初步训练样本中的人脸图像,这样可以实现初步训练样本的自学习功能,以扩充样本库。
步骤S32,根据初步训练样本对人脸图像进行识别确认:
若无法识别,则根据人脸图像更新初步训练样本;随后转向步骤S33;
若能够识别,则直接转向步骤S33;
本发明的较佳的实施例中,所谓识别确认,是指确认初步训练样本的识别准确度。下面给出一个示例:
现场采集一个人脸图像并输入到智能终端内,智能终端根据初步训练样本匹配得到一个相应的结果(例如相应的个人姓名)。此时智能终端会发出一个提示音,例如提示使用者“此次识别是否正确?”。若使用者进行确认,则此次识别正确;反之,此次识别错误,需要修正。
步骤S33,循环执行步骤S31-32,以根据多张人脸图像完善初步训练样本;
本发明的较佳的实施例中,如上文中所述,根据多张人脸图像循环执行进行识别确认的工作,若出现识别错误,则可以采用使用者手动输入相匹配的个人信息,或者修正样本库的数据等方式对初步训练样本进行完善。循环上述步骤S31-32,尽可能以较多张人脸图像来对初步训练样本进行识别确认和样本库的完善。
步骤S34,将经过完善的初步训练样本整合形成识别数据,并保存于智能终端内。
本发明的较佳的实施例中,经过以上三种方法(即形成第一训练样本、 第二训练样本以及完善训练样本)获取得到的训练样本,通过对对应每个个人姓名的图像样本抽样进行模式匹识别,交叉匹配后筛选重复的结果,并最终将经过晚上的初步训练样本进行合并,以得到相应的识别数据(即合并后形成的识别模型及其关联信息,例如个人信息等),并将这些识别数据归档到智能终端中。具体地,本发明的较佳的实施例中,由于最初依赖于使用者的关系网络建立相应的人脸识别的训练样本,因此,最终将这些识别数据归档到智能终端中保存的使用者的关系网络中。
本发明的较佳的实施例中,形成识别数据后,智能终端即可以采集需要识别的人脸,并根据识别数据进行识别,输出识别结果。换言之,完成上述一系列训练样本的预先形成操作后,智能终端即可以使用最终形成的识别数据对人脸进行识别。
本发明的较佳的实施例中,上述智能终端可以为具有机器人外观的智能终端。换言之,本发明的较佳的实施例中,可以将上述人脸图像识别方法应用于可与使用者进行信息交互的机器人设备中。
以上所述仅为本发明较佳的实施例,并非因此限制本发明的实施方式及保护范围,对于本领域技术人员而言,应当能够意识到凡运用本发明说明书及图示内容所作出的等同替换和显而易见的变化所得到的方案,均应当包含在本发明的保护范围内。

Claims (7)

  1. 一种人脸图像识别方法,适用于智能终端;其特征在于,包括一预训练步骤,以得到相应的训练样本;
    所述预训练步骤具体包括:
    步骤S1,所述智能终端远程获取保存于使用者的移动终端内的通讯名录以及用户相册,采用所述用户相册中的人脸图像与所述通讯名录中的个人头像进行匹配,以建立所述人脸图像与所述通讯名录中的所述个人头像对应的个人姓名相关联的第一训练样本;和/或
    所述智能终端远程查找并获取预设的关联于社交网络的用户账号对应的好友列表以及好友相册,并根据所述好友相册中的人脸图像与所述好友列表中的个人头像进行匹配,以建立所述人脸图像与所述好友列表中的所述个人头像相对应的个人姓名相关联的第二训练样本;
    步骤S2,结合所述第一训练样本和/或所述第二训练样本以形成一初步训练样本;
    步骤S3,采用包括于所述初步训练样本中的所述人脸图像,并根据所述初步训练样本进行识别确认,以完善并形成一最终的识别数据;
    形成所述识别数据后,所述智能终端采集需要识别的人脸图像,并根据所述识别数据进行识别,输出识别结果。
  2. 如权利要求1所述的人脸图像识别方法,其特征在于,所述步骤S1中,形成所述第一训练样本的步骤具体包括:
    步骤S11a,远程获取所述用户相册中的所述人脸图像;
    步骤S12a,远程获取所述通讯名录;
    步骤S13a,采用所述人脸图像在所述通讯名录中进行匹配,以找到所述通讯名录中相匹配的个人头像;
    步骤S14a,将所述人脸图像与相匹配的个人头像对应的个人姓名相关联;
    步骤S15a,循环执行所述步骤S11a-S14a,以根据所述用户相册中的所有所述人脸图像训练形成所述第一训练样本。
  3. 如权利要求1所述的人脸图像识别方法,其特征在于,所述步骤S1 中,形成所述第二训练样本的步骤具体包括:
    步骤S11b,通过关联于至少一个社交网站的所述用户账号,远程获取所述用户账号相关联的所述好友相册中的所述人脸图像;
    步骤S12b,通过关联于至少一个社交网站的所述用户账号,远程获取所述用户账号相关联的好友列表;
    步骤S13b,采用所述人脸图像在所述好友列表中进行匹配,以找到所述好友列表中相匹配的个人头像;
    步骤S14b,将所述人脸图像与所述好友列表中相匹配的个人头像对应的个人姓名相关联;
    步骤S15b,循环执行所述步骤S11b-14b,以根据所述好友相册中的所有所述人脸图像训练形成所述第二训练样本。
  4. 如权利要求1所述的人脸图像识别方法,其特征在于,所述步骤S1中,所述智能终端通过无线连接方式接入所述移动终端,并从所述移动终端内获取相应的所述通讯名录和所述用户相册。
  5. 如权利要求1所述的人脸图像识别方法,其特征在于,所述步骤S3具体包括:
    步骤S31,现场采集包括于所述初步训练样本中的一人脸图像;
    步骤S32,根据所述初步训练样本对所述人脸图像进行识别确认:
    若无法识别,则根据所述人脸图像更新所述初步训练样本;随后转向步骤S33;
    若能够识别,则直接转向步骤S33;
    步骤S33,循环执行所述步骤S31-32,以根据多张所述人脸图像完善所述初步训练样本;
    步骤S34,将经过完善的所述初步训练样本整合形成所述识别数据,并保存于所述智能终端内。
  6. 如权利要求5所述的人脸图像识别方法,其特征在于,所述步骤S32中,所述智能终端通过发出提示音以供现场采集的所述人脸图像对应的使用者确认。
  7. 如权利要求1所述的人脸图像识别方法,其特征在于,所述智能终端为具有机器人外观的智能终端。
PCT/CN2016/086481 2015-06-30 2016-06-20 一种人脸图像识别方法 WO2017000807A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510381767.8A CN106326815B (zh) 2015-06-30 2015-06-30 一种人脸图像识别方法
CN201510381767.8 2015-06-30

Publications (1)

Publication Number Publication Date
WO2017000807A1 true WO2017000807A1 (zh) 2017-01-05

Family

ID=57607855

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/086481 WO2017000807A1 (zh) 2015-06-30 2016-06-20 一种人脸图像识别方法

Country Status (4)

Country Link
CN (1) CN106326815B (zh)
HK (1) HK1231600A1 (zh)
TW (1) TWI579773B (zh)
WO (1) WO2017000807A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110309691A (zh) * 2018-03-27 2019-10-08 腾讯科技(深圳)有限公司 一种人脸识别方法、装置、服务器及存储介质

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110490177A (zh) 2017-06-02 2019-11-22 腾讯科技(深圳)有限公司 一种人脸检测器训练方法及装置
CN108241734A (zh) * 2017-12-01 2018-07-03 国政通科技股份有限公司 一种基于照片库的信息核查一体化方法及装置
CN108305146A (zh) * 2018-01-30 2018-07-20 杨太立 一种基于图像识别的发型推荐方法及系统
CN108764149B (zh) * 2018-05-29 2022-02-18 北京中庆现代技术股份有限公司 一种针对班级学生人脸模型的训练方法
CN109493073B (zh) * 2018-10-25 2021-07-16 创新先进技术有限公司 一种基于人脸的身份识别方法、装置及电子设备
TWI759731B (zh) * 2020-04-27 2022-04-01 淡江大學 機器學習方法
CN112364733B (zh) * 2020-10-30 2022-07-26 重庆电子工程职业学院 智能安防人脸识别系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7274822B2 (en) * 2003-06-30 2007-09-25 Microsoft Corporation Face annotation for photo management
US20080175447A1 (en) * 2007-01-24 2008-07-24 Samsung Electronics Co., Ltd. Face view determining apparatus and method, and face detection apparatus and method employing the same
CN102637255A (zh) * 2011-02-12 2012-08-15 北京千橡网景科技发展有限公司 用于处理图像中包含的面部的方法和设备
CN102811286A (zh) * 2012-07-27 2012-12-05 广东欧珀移动通信有限公司 一种通信录的群组创建方法
CN102819726A (zh) * 2012-06-27 2012-12-12 宇龙计算机通信科技(深圳)有限公司 用于移动终端的照片处理系统及方法

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN201774591U (zh) * 2010-08-12 2011-03-23 天津三星光电子有限公司 一种带通信录的具有自动人脸识别功能的数码相机
TW201423454A (zh) * 2011-12-09 2014-06-16 Primax Electronics Ltd 相片管理系統
TW201348984A (zh) * 2012-05-18 2013-12-01 Primax Electronics Ltd 相片影像管理方法及相片影像管理系統
CN102779179B (zh) * 2012-06-29 2018-05-11 华为终端(东莞)有限公司 一种信息关联的方法及终端
CN102867173B (zh) * 2012-08-28 2015-01-28 华南理工大学 一种人脸识别方法及其系统
CN103399896B (zh) * 2013-07-19 2019-08-23 广州华多网络科技有限公司 识别用户间关联关系的方法及系统
CN103793697B (zh) * 2014-02-17 2018-05-01 北京旷视科技有限公司 一种人脸图像的身份标注方法及人脸身份识别方法
CN103970830B (zh) * 2014-03-31 2017-06-16 小米科技有限责任公司 信息推荐方法和装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7274822B2 (en) * 2003-06-30 2007-09-25 Microsoft Corporation Face annotation for photo management
US20080175447A1 (en) * 2007-01-24 2008-07-24 Samsung Electronics Co., Ltd. Face view determining apparatus and method, and face detection apparatus and method employing the same
CN102637255A (zh) * 2011-02-12 2012-08-15 北京千橡网景科技发展有限公司 用于处理图像中包含的面部的方法和设备
CN102819726A (zh) * 2012-06-27 2012-12-12 宇龙计算机通信科技(深圳)有限公司 用于移动终端的照片处理系统及方法
CN102811286A (zh) * 2012-07-27 2012-12-05 广东欧珀移动通信有限公司 一种通信录的群组创建方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110309691A (zh) * 2018-03-27 2019-10-08 腾讯科技(深圳)有限公司 一种人脸识别方法、装置、服务器及存储介质
CN110309691B (zh) * 2018-03-27 2022-12-27 腾讯科技(深圳)有限公司 一种人脸识别方法、装置、服务器及存储介质

Also Published As

Publication number Publication date
HK1231600A1 (zh) 2017-12-22
TW201701188A (zh) 2017-01-01
TWI579773B (zh) 2017-04-21
CN106326815A (zh) 2017-01-11
CN106326815B (zh) 2019-09-13

Similar Documents

Publication Publication Date Title
WO2017000807A1 (zh) 一种人脸图像识别方法
CN104636501B (zh) 一种多媒体网络人工即时翻译系统及方法
CN103310142B (zh) 基于可穿戴设备的人机融合安全认证方法
CN205314704U (zh) 一种智能门锁及系统
WO2017215240A1 (zh) 基于神经网络的人脸特征提取建模、人脸识别方法及装置
US9730046B2 (en) Automatically inferring user signature through contextual learning
WO2019000777A1 (zh) 基于互联网的人脸美化系统
CN104935440B (zh) 远程设立金融账户的认证方法及认证设备
US20170374198A1 (en) Automated Use of Interactive Voice Response Systems
JP6128500B2 (ja) 情報管理方法
US9721079B2 (en) Image authenticity verification using speech
CN103973441A (zh) 基于音视频的用户认证方法和装置
WO2016004768A1 (zh) 一种社交关系管理的方法、设备及系统
CN110912893A (zh) 一种账号合并方法
CN105338183A (zh) 基于移动终端的人物识别提醒系统及方法
CN106161155A (zh) 一种信息处理方法及主终端
CN110149618A (zh) 基于声纹授权的智能设备接入方法、装置、设备及介质
CN103870735A (zh) 解锁处理方法及装置
CN104717127A (zh) 基于图像识别实现联系人触发的方法、终端及系统
TW201816646A (zh) 利用行動裝置應用程式的電子交易認證方法及系統
WO2015101317A1 (zh) 终端机器人安全系统及操作方法
CN112769872B (zh) 一种基于音频及视频特征融合的会议系统接入方法及系统
WO2017016035A1 (zh) 一种通信控制方法及装置
WO2017016027A1 (zh) 连接建立方法、连接建立装置和通信系统
US20180288373A1 (en) Treatment method for doorbell communication

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: 16817171

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: 16817171

Country of ref document: EP

Kind code of ref document: A1