WO2017162076A1 - Procédé et système d'identification de visage - Google Patents

Procédé et système d'identification de visage Download PDF

Info

Publication number
WO2017162076A1
WO2017162076A1 PCT/CN2017/076723 CN2017076723W WO2017162076A1 WO 2017162076 A1 WO2017162076 A1 WO 2017162076A1 CN 2017076723 W CN2017076723 W CN 2017076723W WO 2017162076 A1 WO2017162076 A1 WO 2017162076A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
similarity
face feature
feature template
face
Prior art date
Application number
PCT/CN2017/076723
Other languages
English (en)
Chinese (zh)
Inventor
陈�胜
Original Assignee
北京握奇数据股份有限公司
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 北京握奇数据股份有限公司 filed Critical 北京握奇数据股份有限公司
Publication of WO2017162076A1 publication Critical patent/WO2017162076A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06V40/179—Human faces, e.g. facial parts, sketches or expressions metadata assisted face recognition

Definitions

  • the invention belongs to the field of biometrics, and particularly relates to a face recognition method and system.
  • Face recognition is a very challenging research topic, especially for face recognition under unconstrained conditions. Under different lighting, posture, expression, age, makeup, etc., shooting the same person's face image is very different. How to effectively overcome the effects of these unfavorable factors is crucial for face recognition. In general, the impact of adverse factors on face recognition can be overcome in two ways.
  • the feature extraction process of face recognition is optimized: appropriate features are extracted from the face image, which should be intra-class changes of the same person while effectively depicting the inter-class differences between different people (intra-class Difference) has strong stability; on the other hand, it optimizes the feature template matching process of face recognition: the feature templates of two or more sets of face images that need to be compared are input into an appropriate classifier for comparison, A good classifier can further expand the distance between different people's face feature templates, reduce the distance between different face image feature templates of the same person, and better discriminate whether the two sets of face images are from the same person. In the case that the feature extraction process has been perfected, the optimization feature template comparison process can still greatly improve the recognition accuracy of the entire face recognition system.
  • the face recognition system can be divided into a single template comparison system and a multiple template comparison system.
  • the use of multiple template alignment techniques in a face recognition system can significantly improve recognition accuracy compared to a single template alignment system.
  • one solution is to use the average similarity as the decision basis for the feature template comparison.
  • user A registers k feature templates X A1 , X A2 , . . . , X Ak in the face recognition system, and the similarity between the feature template Y of the face image to be recognized and the k feature templates is S A1 .
  • the scheme averages the similarity of the multi-template alignment, and the distribution of the similarity after the average is more concentrated. It is easy to cause the FRY of the face recognition to increase compared with the single template. Another solution is to use maximum similarity as the basis for decision making of feature template comparisons.
  • the maximum value of the similarity S A1 , S A2 , . . . , S Ak of the feature template Y of the face image to be recognized and the k feature templates is set to S A , if S A is greater than or equal to a preset
  • the feature template compares the similarity threshold S 0 , and determines that the user in the face image corresponding to Y is the same person as A; if S A is smaller than the threshold S 0 , it determines the user in the face image corresponding to Y A is not the same person.
  • the similarity of the multi-template alignment is maximized.
  • the maximum similarity of multiple template comparisons between different people will be more similar than the single template comparison. It is easier to misjudge different people. For the same person, the false positive rate of face recognition, that is, the false acceptance rate FAR rises when compared with the single template.
  • an object of the present invention is to provide a face recognition method and system, by which the accuracy of face recognition can be effectively improved.
  • a face recognition method includes the following steps:
  • the face feature template library stores N user face feature templates and user basic information, each user corresponding to at least one face feature template; wherein, N ⁇ 1;
  • the maximum similarity between the face feature template to be identified and a certain user is the maximum similarity between the similarity between the face feature template to be identified and all the face feature templates of the user; the face feature template to be recognized and The average similarity of a user refers to the face feature template to be identified and the user's location. The average of the similarities of the face feature templates;
  • the maximum similarities between the N-users in the face feature template and the face feature template library are S Max1 , S Max2 ,..., S MaxN ; the face feature template to be recognized and the face feature template library
  • the average similarity of users is recorded as S Avg1 , S Avg2 ,...,S AvgN ;
  • step 3 Obtain the maximum value MAX Max in S Max1 , S Max2 ,..., S MaxN , and judge according to the user's face feature template and the face feature template to be recognized in the face feature template library corresponding to the maximum value MAX Max Whether the user to be identified is the user in the face feature template library corresponding to the maximum value MAX Max , if yes, the recognition ends, and if not, proceeds to step 3;
  • step 2 and step 3 according to the face feature template of a user in the face feature template library and the face feature template to be recognized, it is determined whether the user to be identified is a
  • the way to describe a user includes:
  • step 3 2) determining whether the maximum similarity S MaxA is greater than or equal to the preset maximum similarity threshold S max , and if so, determining that the user to be identified is the user, and if not, proceeding to step 3);
  • step 4 judging whether k ⁇ K S is satisfied, if yes, proceeding to step 4), if not, determining that the user to be identified is not the user; wherein K S is a positive integer, K S >1;
  • step 4 if the average similarity S AvgA ⁇ S avg , before the prompt recognition fails, the method further includes:
  • the method further includes the steps of determining the maximum similarity threshold S max , the average similarity threshold S avg , and the minimum similarity threshold S min , and the determining manner is:
  • a. Construct a face image test library, where the face image of the H person is stored in the test library, and the number of face images of each person is one;
  • the false positive rate threshold, the second false positive rate threshold, and the third false positive rate threshold are greater than the first when N>1, respectively.
  • the embodiment of the invention further provides a face recognition system, comprising:
  • a template library building module configured to construct a face feature template library, where the face feature template library stores N user face feature templates and user basic information, and each user corresponds to at least one face feature template; N ⁇ 1;
  • a feature extraction module to be used for extracting a face feature in a face image of the user to be identified, and obtaining a face feature template to be recognized
  • a face recognition module configured to perform identification of the user to be identified according to the face feature template to be identified and the face feature template library;
  • the face recognition module includes:
  • a first similarity calculation unit configured to respectively calculate a similarity between the face feature template to be recognized and each face feature template of each user in the face feature template library, to obtain a face feature template and a face feature to be recognized The maximum similarity and average similarity of each user in the template library;
  • the maximum similarity between the face feature template to be identified and a certain user is the maximum similarity between the similarity between the face feature template to be identified and all the face feature templates of the user; the face feature template to be recognized and The average similarity of a certain user refers to an average value of the similarity between the face feature template to be identified and all the face feature templates of the user;
  • the maximum similarities between the N-users in the face feature template and the face feature template library are S Max1 , S Max2 ,..., S MaxN ; the face feature template to be recognized and the face feature template library
  • the average similarity of users is recorded as S Avg1 , S Avg2 ,...,S AvgN ;
  • the first face recognition unit is configured to obtain a maximum value MAX Max in S Max1 , S Max2 , . . . , S MaxN , and a facial feature template of the user in the face feature template library corresponding to the maximum value MAX Max
  • the recognized face feature template determines whether the user to be identified is the user in the face feature template library corresponding to the maximum value MAX Max , and if so, the recognition ends; if not, the second face recognition unit is entered;
  • a second face recognition unit configured to acquire a maximum value AVG Max in S Avg1 , S Avg2 , . . . , S AvgN , according to a face feature template of the user in the face feature template library corresponding to the maximum value AVG Max
  • the recognized face feature template determines whether the user to be identified is the user in the face feature template library corresponding to the maximum value AVG Max , and if so, the recognition ends, and if not, the recognition fails.
  • a face recognition system as described above, there are k face feature templates of the user in the face feature template library, and the face feature template to be recognized is similar to the k face feature template.
  • the degrees are S A1 , S A2 , ..., S Ak ;
  • the first face recognition unit and the second face recognition unit each include:
  • the average similarity calculating sub-unit for taking the k similarities S A1, S A2, ..., S Ak maximum similarity S MaxA, k is calculated a similarity S A1, S A2, ..., the average value of S Ak , get the average similarity S AvgA ;
  • a first identifying subunit configured to determine whether the maximum similarity S MaxA is greater than or equal to a preset maximum similarity threshold S max , and if yes, determining that the user to be identified is the user, and if not, entering the second identifier unit;
  • a second identification subunit configured to determine whether k ⁇ K S is satisfied, and if yes, proceeds to step third identification subunit, and if not, determines that the user to be identified is not the user; wherein K S is a positive integer, K S >1;
  • a third identifying subunit configured to determine whether the average similarity S AvgA is greater than or equal to a preset average similarity threshold S avg , and if yes, determining that the user to be identified is the user, and if not, determining that the user is to be identified The user is not the user.
  • the third identification subunit further includes:
  • a face recognition system as described above, the system further comprising:
  • the similarity confirmation module includes:
  • the test library building unit is configured to construct a face image test library, wherein the face image of the H person is stored in the test library, and the number of face images of each person is one;
  • sub-library building unit for randomly selecting J images from each person's I face images to form a sub-library P, and all remaining images constitute a sub-library G;
  • a sub-library grouping unit which divides a face image of each person in the sub-library G into K groups, and obtains an H ⁇ K group image
  • a second similarity calculation unit configured to calculate, for each image P′ in the H ⁇ J images in the sub-library P, each image of the image P′ and the sub-library G other than the user to which the image P′ belongs
  • the maximum similarity, the minimum similarity and the average similarity are obtained as H ⁇ J ⁇ (H-1) ⁇ K maximum similarities, H ⁇ J ⁇ (H-1) ⁇ K minimum similarities and H ⁇ J ⁇ (H-1) ⁇ K average similarities;
  • a similarity threshold determining unit configured to arrange the H ⁇ J ⁇ (H ⁇ 1 ⁇ K maximum similarities, the minimum similarity, and the average similarity in descending order, respectively, in an order of H ⁇ J ⁇ ( H-1) ⁇ K maximum similarities, the ratio of the position number to H ⁇ J ⁇ (H-1) ⁇ K is equal to the preset first misrecognition rate threshold, the maximum similarity is the maximum similarity threshold S max Taking the ratio of the position number of H ⁇ J ⁇ (H-1) ⁇ K average similarities to H ⁇ J ⁇ (H-1) ⁇ K equal to the average similarity of the preset second misrecognition rate threshold is average Similarity threshold S avg ; taking H ⁇ J ⁇ (H-1) ⁇ K minimum similarities, the ratio of the position number to H ⁇ J ⁇ (H-1) ⁇ K is equal to the preset third misrecognition rate threshold The minimum similarity is the minimum similarity threshold S min .
  • the invention has the beneficial effects that the face recognition method provided by the present invention comprehensively utilizes the maximum similarity and average similarity between the face feature template to be recognized and the face feature template of the user in the template library, or the maximum similarity. Degree, average similarity and minimum similarity multiple reference factors can effectively improve the accuracy of face recognition compared with existing face recognition methods.
  • FIG. 1 is a flowchart of a face recognition method according to Embodiment 1 of the present invention.
  • FIG. 2 is a flowchart of determining whether a user to be identified is a user in a feature template library according to Embodiment 2 of the present invention
  • FIG. 3 is a flowchart of a face recognition method according to Embodiment 3 of the present invention.
  • FIG. 4 is a structural block diagram of a face recognition system according to Embodiment 6 of the present invention.
  • FIG. 1 is a flowchart of a face recognition method provided in this embodiment. As can be seen from the figure, the method may include the following steps:
  • Step S101 construct a face feature template library
  • a face feature template library for face recognition is constructed, and the feature library is used for storing basic information of the user and face feature information.
  • the existing facial feature extraction algorithm is used to extract the facial features of the user's face image
  • the deep neural network algorithm is used to extract the facial features
  • the extracted user facial features are used as follow-up purposes.
  • the face feature template identifying the user is stored in the template library.
  • the face feature template library stores N user face feature templates and user basic information (user name and other information), and each user corresponds to at least one face feature template; wherein, N ⁇ 1.
  • the number of face feature templates for different users can be different.
  • Step S102 Extract a face feature in a face image of the user to be identified, and obtain a face feature template to be identified;
  • the face image of the user to be recognized is first acquired, and the image is generally collected in real time, or may be an existing face image (for characters in the existing face image)
  • the recognition is performed), and then the facial features in the image of the user's face to be recognized are extracted, and the facial feature template to be recognized is obtained.
  • the manner of extracting the face features of the user to be recognized is the same as the manner of extracting the face features when constructing the template library in step S101.
  • Step S103 Perform identification of the user to be identified according to the face feature template to be identified and the face feature template library.
  • the face feature template library constructed in step S101 performs identification of the user to be identified.
  • the identification manner of the user to be identified is as shown in FIG. 1 , and includes the following steps:
  • the maximum similarity between the face feature template to be identified and a certain user refers to the maximum similarity between the similarity between the face feature template to be identified and all the face feature templates of the user; the face feature to be recognized
  • the average similarity between the template and a certain user refers to the average of the similarity between the face feature template to be identified and all the face feature templates of the user.
  • the face feature template library includes two users, user A and user B.
  • the template library user A corresponds to two face feature templates
  • user B corresponds to three face feature templates
  • the face feature templates to be recognized are respectively calculated.
  • the larger of the two similarities is the maximum similarity between the face feature template to be identified and the user A.
  • the average of the two similarities is the person to be identified.
  • the maximum value of the similarity between the face feature template to be identified and the three face feature templates of the user B is the maximum similarity between the face feature template to be recognized and the user B
  • the average of the three similarities is The average similarity between the face feature template to be identified and the user B.
  • the similarity of the two face template features is calculated by using the existing calculation method.
  • the similarity between the two face feature templates can be calculated by using the cosine similarity.
  • the maximum similarities of the N face users in the face feature template and the face feature template library are S Max1 , S Max2 , . . . , S MaxN respectively ; the face feature template and the face to be recognized are respectively
  • the average similarity of N users in the feature template library is recorded as S Avg1 , S Avg2 , ..., S AvgN , respectively .
  • step 3 Obtain the maximum value MAX Max in S Max1 , S Max2 ,..., S MaxN , and judge according to the user's face feature template and the face feature template to be recognized in the face feature template library corresponding to the maximum value MAX Max Whether the user to be identified is the user in the face feature template library corresponding to the maximum value MAX Max , if yes, the recognition ends, and if not, proceeds to step 3;
  • the face recognition method in this embodiment first calculates the similarity between the face feature template of the user to be identified and each face feature template of each user in the template library, and obtains the face feature template to be identified and each
  • the maximum similarity of the user is determined by selecting a maximum of the plurality of maximum similarities, and identifying the user to be identified based on the facial feature template in the template library of the user corresponding to the maximum value of the maximum similarity. If the user cannot be identified by the facial feature template in the template library of the user corresponding to the maximum value, and according to the average similarity of each facial feature template to each user, the maximum value among the multiple average similarities is obtained.
  • the face feature template in the corresponding user's template library is recognized again.
  • step 2 and step 3 after the maximum value MAX Max and the maximum value AVG Max are obtained , the face feature template of the user and the face feature to be recognized in the face feature template library corresponding to the maximum value MAX Max are obtained.
  • the manner in which the template is to be recognized by the user to be identified, or the method of identifying the user to be recognized according to the face feature template of the user in the face feature template library corresponding to the maximum AVG Max and the face feature template to be recognized may be used. Some face recognition methods are implemented.
  • the difference between this embodiment and the first embodiment is that after the maximum value MAX Max is obtained in step 2 or after the AVG Max is acquired in step 3, the embodiment further provides a maximum value MAX Max .
  • the face feature template of the user and the face feature template to be recognized in the corresponding face feature template library are used to identify the user to be recognized or the face feature template of the user in the face feature template library corresponding to the maximum AVG Max And the manner of identifying the user to be recognized by the face feature template to be identified.
  • the manner of determining whether the user to be identified is the user is as shown in FIG. 2, and includes the following Steps:
  • step 3 2) determining whether the maximum similarity S MaxA is greater than or equal to the preset maximum similarity threshold S max , and if so, determining that the user to be identified is the user, and if not, proceeding to step 3);
  • step 2 the user is the user in the face feature template library corresponding to the maximum value MAX Max .
  • step 3 the user is the user in the face feature template library corresponding to the maximum value AVG Max .
  • FIG. 3 is a flowchart of a method for recognizing a face provided in this embodiment.
  • the method in this embodiment of China may include the following steps:
  • Step S301 construct a face feature template library
  • Step S302 Extract a face feature in a face image of the user to be identified, and obtain a face feature template to be identified;
  • step S301 The specific manner of constructing the face feature template library in step S301 in the embodiment and the face feature in the face image of the user to be recognized in step S302, and obtaining the face feature template to be recognized, and the first embodiment
  • step S101 The manner of constructing the face feature template library in step S101 described in the same manner as the step S102 of obtaining the face feature template to be recognized is the same.
  • Step S303 Perform identification of the user to be identified according to the face feature template to be identified and the face feature template library.
  • the specific manner of performing the identification of the user to be identified according to the face feature template to be identified and the constructed face feature template library includes the following steps:
  • the maximum similarity between the face feature template to be identified and a certain user refers to the maximum similarity between the similarity between the face feature template to be identified and all the face feature templates of the user; the face feature to be recognized
  • the average similarity between the template and a certain user refers to the average of the similarity between the face feature template to be identified and all the face feature templates of the user.
  • the maximum similarity between the face feature template and the face feature template library in the face feature template library are respectively S Max1 , S Max2 , . . . , S MaxN ; the face feature template to be recognized and the person
  • the average similarity of N users in the face feature template library is recorded as S Avg1 , S Avg2 , ..., S AvgN , respectively .
  • step 3 acquisition S Max1, S Max2, ..., MAX Max S MaxN maximum value of the maximum value MAX Max and corresponding facial feature template library users, and the person to be identified with the face feature pattern of the maximum value MAX Max Corresponding the similarity of the user's face feature template in the face feature template library, determining whether the user to be identified is the user in the face feature template library corresponding to the maximum value MAX Max , and if so, the recognition ends, if not, Then proceed to step 3;
  • the face feature template of the user in the face feature template library corresponding to the maximum value MAX Max has k 1 , and the face feature template to be recognized and k 1 person
  • the similarity of the face feature templates are S A1 , S A2 , . . . , S Ak1 ; respectively, and the specific manner of determining whether the user to be identified is the user in the face feature template library corresponding to the maximum value MAX Max is as shown in FIG. 3 . Includes the following steps:
  • K S Is a positive integer, K S >1;
  • v, k 1 th take similarities S A1, S A2, ..., S Ak1 whether the minimum degree of similarity S MinA, S MinA determining the minimum degree of similarity equal to or greater than a preset minimum similarity threshold S min, and if yes, Then, the user to be identified is the user in the face feature template library corresponding to the MAX Max . If not, the user to be identified is not the user in the face feature template library corresponding to the MAX Max , and the process proceeds to step 3.
  • the manner of determining whether the user to be identified is the user in the face feature template library corresponding to the maximum value AVG Max and the face feature template corresponding to determining whether the user to be identified is the maximum value MAX Max in the above step 2
  • the users in the library are in the same way, as follows:
  • step III Determine whether k 2 ⁇ K S is satisfied. If yes, proceed to step IV. If not, determine that the user to be identified is not a user in the face feature template library corresponding to AVG Max , and prompt recognition failure; wherein, K S Is a positive integer, K S >1;
  • V, k 2 th take similarities S A1, S A2, ..., S Ak2 S MinB the minimum degree of similarity, determining the minimum degree of similarity is greater than or equal to S MinB preset minimum similarity threshold S min, and if yes, Then, the user to be identified is the user in the face feature template library corresponding to the AVG Max . If not, the user to be identified is not the user in the face feature template library corresponding to the AVG Max , and the recognition fails.
  • the face recognition method provided in this embodiment is a flowchart for using the method described in Embodiment 3 for a 1:1 face verification system, wherein 1:1 refers to only storing in the face feature template library.
  • a user's face feature template, ie N 1.
  • the R facial feature template of the user A is stored in the face feature template library, and the R personal face feature templates are respectively recorded as feature templates X A1 , X A2 , . . . , X AR , to be identified by the user to be identified.
  • the face feature template is recorded as the feature template Y.
  • K S is 3.
  • Step 1 Calculate the similarity between the feature template Y and the R feature templates X A1 , X A2 , . . . , X AR respectively.
  • the R similarities are respectively recorded as S A1 , S A2 , . . . , S AR , and the maximum similarity is taken.
  • A (S A1 + S A2 +... + S AR ) / R;
  • Step 2 determining whether SMax A is greater than or equal to the preset maximum similarity threshold S max , and if so, determining that the user to be identified corresponding to the feature template Y is the same person as the user A, and if not, proceeding to step 3;
  • Step 3 Determine whether R ⁇ 3 is satisfied. If yes, proceed to step 4. If the user to be identified corresponding to the judgment feature template Y is not the same person as the user A, the prompt recognition fails.
  • Step 4 Determine whether SAvg A is greater than or equal to the preset average similarity threshold S avg , and if so, determine that the user to be identified corresponding to the feature template Y is the same person as the user A, and if not, proceed to step 5;
  • Step 5 Determine whether the SMin A is greater than or equal to the preset minimum similarity threshold. If yes, determine that the user to be identified corresponding to the feature template Y is the same person as the user A. If not, determine the user to be identified corresponding to the feature template Y. The user is not the same person as User A, and the prompt recognition fails.
  • the maximum similarity threshold S max , the average similarity threshold S avg , and the minimum similarity threshold S min in the first embodiment, the second embodiment, the third embodiment, and the fourth embodiment may be directly applied according to the application scenario. Set directly to the experience value by the user.
  • a method for determining the maximum similarity threshold S max , the average similarity threshold S avg , and the minimum similarity threshold S min is provided, and the manner includes the following steps:
  • a. Construct a face image test library, where the face image of the H person is stored in the test library, and the number of face images of each person is one;
  • the first erroneous rate threshold, the second erroneous rate threshold, and the third erroneous rate threshold may be set to the same value or may be set to different values.
  • the size of the test library is also as large as possible.
  • the first false alarm rate threshold can be set to 0.01%
  • the second false positive rate threshold can be set to 0.01%
  • the third false positive rate threshold can be set to 0.01%
  • N>1 the first false positive rate threshold can be set to 0.001%
  • the second false positive rate The threshold can be set to 0.001%
  • the first The three false alarm rate threshold can be set to 0.001%.
  • a test library of 100 face images of 100 faces per person for a total of 100,000 face images can be constructed. 10 sub-databases P constituting a total of 10,000 face images are randomly selected from each of the 100 face images, and the remaining 90,000 face images constitute a sub-library G, and 90 of each of the sub-libraries G The face images are evenly divided into 9 groups, which are divided into 9 ⁇ 1000 groups.
  • FIG. 4 is a structural block diagram of a face recognition system provided in this embodiment.
  • the system includes a template library establishment module 100, a similarity threshold determination module 200, and a feature extraction module 300 to be verified. And a face recognition module 400. among them,
  • the template library building module 100 is configured to construct a face feature template library, where the face feature template library stores N user face feature templates and user basic information, and each user corresponds to at least one face feature template; , N ⁇ 1;
  • the to-be-verified feature extraction module 300 is configured to extract a facial feature in a face image of the user to be identified, and obtain a facial feature template to be identified;
  • the face recognition module 400 is configured to perform identification of the user to be identified according to the face feature template to be recognized and the face feature template library; the face recognition module 400 includes a first similarity calculation unit 401 The first face recognition unit 402 and the second face recognition unit 403. among them:
  • the first similarity calculation unit 401 is configured to separately calculate the similarity between the face feature template to be recognized and each face feature template of each user in the face feature template library, and obtain the face feature template and the face to be recognized.
  • the maximum similarity between the face feature template to be identified and a certain user is the maximum similarity between the similarity between the face feature template to be identified and all the face feature templates of the user; the face feature template to be recognized and The average similarity of a certain user refers to an average value of the similarity between the face feature template to be identified and all the face feature templates of the user;
  • the maximum similarities between the N-users in the face feature template and the face feature template library are S Max1 , S Max2 ,..., S MaxN ; the face feature template to be recognized and the face feature template library
  • the average similarity of users is recorded as S Avg1 , S Avg2 ,...,S AvgN ;
  • the first face recognition unit 402 is configured to obtain a maximum value MAX Max in S Max1 , S Max2 , . . . , S MaxN , according to a face feature template of the user in the face feature template library corresponding to the maximum value MAX Max a face feature template to be identified, determining whether the user to be identified is a user in the face feature template library corresponding to the maximum value MAX Max , and if so, the recognition ends; if not, entering the second face recognition unit;
  • a second face recognition unit 403 configured to acquire a maximum value AVG Max in S Avg1 , S Avg2 , . . . , S AvgN , according to a face feature template of the user in the face feature template library corresponding to the maximum value AVG Max
  • the face feature template to be identified determines whether the user to be identified is the user in the face feature template library corresponding to the maximum value AVG Max , and if so, the recognition ends, and if not, the recognition fails.
  • the first face recognition unit 402 and the second face recognition unit 403 each include an average similarity calculation subunit, a first identification subunit, a second identification subunit, and a third identification subunit.
  • the average similarity calculating sub-unit for taking the k similarities S A1, S A2, ..., S Ak maximum similarity S MaxA, k is calculated a similarity S A1, S A2, ..., the average value of S Ak , get the average similarity S AvgA ;
  • a first identifying subunit configured to determine whether the maximum similarity S MaxA is greater than or equal to a preset maximum similarity threshold S max , and if yes, determining that the user to be identified is the user, and if not, entering the second identifier unit;
  • a second identification subunit configured to determine whether k ⁇ K S is satisfied; if yes, proceed to step third identification subunit, if not, determine that the user to be identified is not the user, and prompt identification failure; wherein, K S Is a positive integer, K S >1;
  • a third identifying subunit configured to determine whether the average similarity S AvgA is greater than or equal to a preset average similarity threshold S avg , and if yes, determining that the user to be identified is the user, and if not, prompting the identification to fail .
  • the third identifier subunit may further include:
  • the face recognition system may further include:
  • the similarity threshold determining module 200 is configured to determine the maximum similarity threshold S max , the average similarity threshold S avg , and the minimum similarity threshold S min .
  • the similarity confirmation module 200 may include a test library construction unit 201, a sub-library construction unit 202, a sub-library grouping unit 203, a second similarity calculation unit 204, and a similarity threshold determination unit 205. among them:
  • the test library construction unit 201 is configured to construct a face image test library, where the face image of the H person is stored in the test library, and the number of face images of each person is one;
  • the sub-library building unit 202 is configured to randomly select J images from each of the I face images to form a sub-library P, and all the remaining images constitute a sub-library G;
  • the sub-library grouping unit 203 divides the face image of each person in the sub-library G into K groups, and obtains H. ⁇ K group image;
  • the second similarity calculation unit 204 is configured to calculate, for each image P′ in the H ⁇ J images in the sub-library P, each set of images other than the user of the image P′ in the sub-library G
  • the maximum similarity, minimum similarity and average similarity of the image are obtained as H ⁇ J ⁇ (H-1) ⁇ K maximum similarities, H ⁇ J ⁇ (H-1) ⁇ K minimum similarities and H ⁇ J ⁇ (H-1) ⁇ K average similarities;
  • the similarity threshold determining unit 205 is configured to arrange the H ⁇ J ⁇ (H ⁇ 1 ⁇ K maximum similarities, the minimum similarity, and the average similarity in descending order, respectively, in an order of H ⁇ J ⁇ (H-1) ⁇ K maximum similarities, the ratio of the position number to H ⁇ J ⁇ (H-1) ⁇ K is equal to the maximum similarity of the preset first misrecognition rate threshold is the maximum similarity threshold S Max ; taking H ⁇ J ⁇ (H-1) ⁇ K average similarities, the ratio of the position number to H ⁇ J ⁇ (H-1) ⁇ K is equal to the preset second misunderstanding rate threshold, the average similarity is The average similarity threshold S avg ; taking H ⁇ J ⁇ (H-1) ⁇ K minimum similarities, the ratio of the position number to H ⁇ J ⁇ (H-1) ⁇ K is equal to the preset third misrecognition rate threshold The minimum similarity is the minimum similarity threshold S min .

Landscapes

  • Engineering & Computer Science (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Human Computer Interaction (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)
  • Collating Specific Patterns (AREA)

Abstract

L'invention concerne un procédé et un système d'identification de visage. Le procédé d'identification comprend : la création d'une bibliothèque de modèles de caractéristiques de visages (S301), cette bibliothèque de modèles de caractéristiques de visages étant mémorisée avec des modèles de caractéristiques de visage et des informations d'utilisateurs de base de N utilisateurs, et chacun des utilisateurs correspondant à au moins un modèle de caractéristiques de visage ; l'extraction de caractéristiques de visage à partir d'une image de visage d'un utilisateur à identifier, de façon à obtenir un modèle de caractéristiques de visage à identifier (S302) ; le calcul de niveaux de similarité entre le modèle de caractéristiques de visage à identifier et les modèles de caractéristiques de visages des utilisateurs respectifs dans la bibliothèque de modèles de caractéristiques de visages, et la découverte d'une valeur maximale dans les niveaux de similarité les plus élevés entre le modèle de caractéristiques de visage à identifier et les utilisateurs respectifs ; une détermination indiquant si l'utilisateur est l'utilisateur correspondant à ladite valeur maximale ; et, si tel n'est pas le cas, la découverte d'une valeur maximale dans les niveaux de similarité moyens entre le modèle de caractéristiques de visage à identifier et les utilisateurs respectifs, et une détermination indiquant si l'utilisateur est l'utilisateur correspondant à ladite valeur maximale. Par rapport aux procédés d'identification de visage existants, le procédé et le système de la présente invention peuvent améliorer nettement la précision de l'identification de visage.
PCT/CN2017/076723 2016-03-24 2017-03-15 Procédé et système d'identification de visage WO2017162076A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610173912.8 2016-03-24
CN201610173912.8A CN105868695B (zh) 2016-03-24 2016-03-24 一种人脸识别方法及系统

Publications (1)

Publication Number Publication Date
WO2017162076A1 true WO2017162076A1 (fr) 2017-09-28

Family

ID=56625593

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/076723 WO2017162076A1 (fr) 2016-03-24 2017-03-15 Procédé et système d'identification de visage

Country Status (2)

Country Link
CN (1) CN105868695B (fr)
WO (1) WO2017162076A1 (fr)

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110298335A (zh) * 2019-07-09 2019-10-01 中控华运（厦门）集成电路有限公司 人脸识别系统及人脸识别方法
CN110843726A (zh) * 2019-12-19 2020-02-28 北京明朝万达科技股份有限公司 汽车门锁控制系统、方法及车辆
CN110858283A (zh) * 2018-08-23 2020-03-03 营利度富信息系统（上海）有限公司 人脸识别在智慧银行系统中的应用
CN111291596A (zh) * 2018-12-07 2020-06-16 杭州海康威视数字技术股份有限公司 一种基于人脸识别的预警方法及装置
CN111444886A (zh) * 2020-04-27 2020-07-24 无锡职业技术学院 一种基于可穿戴设备的智能人脸识别系统
CN111582305A (zh) * 2020-03-26 2020-08-25 平安科技（深圳）有限公司 生物特征识别方法、装置、计算机设备及存储介质
CN111597532A (zh) * 2020-04-10 2020-08-28 云知声智能科技股份有限公司 基于人脸识别实现儿童机器人童锁系统的方法和系统
CN111639689A (zh) * 2020-05-20 2020-09-08 杭州海康威视系统技术有限公司 人脸数据处理方法、装置及计算机可读存储介质
CN113327332A (zh) * 2018-09-29 2021-08-31 爱笔（北京）智能科技有限公司 基于生物信息的票务处理方法及系统、服务端、客户端
CN113420666A (zh) * 2021-06-23 2021-09-21 上海应用技术大学 一种基于人脸识别技术的记忆辅助方法及装置
CN113469012A (zh) * 2021-06-28 2021-10-01 广州云从鼎望科技有限公司 用户刷脸验证的方法、系统、介质及装置
CN114116849A (zh) * 2021-12-01 2022-03-01 南威软件股份有限公司 一种针对人像预警信息的二次研判方法及装置
CN115019365A (zh) * 2022-05-23 2022-09-06 北京计算机技术及应用研究所 一种基于模型适用性度量的分级人脸识别方法
CN115880745A (zh) * 2022-09-07 2023-03-31 以萨技术股份有限公司 一种获取人脸图像特征的数据处理系统
CN116597185A (zh) * 2023-07-17 2023-08-15 荣耀终端有限公司 模板更新方法及电子设备
CN117284131A (zh) * 2023-09-26 2023-12-26 广东安能电业有限公司 一种电动单车充电柜及其控制方法

Families Citing this family (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105868695B (zh) * 2016-03-24 2019-04-02 北京握奇数据系统有限公司 一种人脸识别方法及系统
CN107832657A (zh) * 2017-09-19 2018-03-23 浙江广锐信息科技有限公司 图像处理方法和装置
CN107563360B (zh) * 2017-09-30 2021-05-07 百度在线网络技术（北京）有限公司 信息获取方法和装置
CN107679493A (zh) * 2017-09-30 2018-02-09 百度在线网络技术（北京）有限公司 人脸识别方法和装置
CN108875493B (zh) * 2017-10-12 2021-04-27 北京旷视科技有限公司 人脸识别中相似度阈值的确定方法以及确定装置
CN107862301A (zh) * 2017-11-29 2018-03-30 厦门瑞为信息技术有限公司 降低人脸识别误识率的方法
CN108052879B (zh) * 2017-11-29 2020-08-07 厦门瑞为信息技术有限公司 一种降低人脸识别误识率的方法
CN108229322B (zh) * 2017-11-30 2021-02-12 北京市商汤科技开发有限公司 基于视频的人脸识别方法、装置、电子设备及存储介质
CN108090420B (zh) * 2017-11-30 2021-07-23 睿视智觉（深圳）算法技术有限公司 一种人脸识别方法
CN108564067B (zh) * 2018-04-28 2021-07-20 深圳卡安兴科技有限公司 人脸比对的阈值确定方法及系统
CN110619253B (zh) * 2018-06-19 2022-06-07 北京京东尚科信息技术有限公司 身份识别的方法和装置
CN109033988A (zh) * 2018-06-29 2018-12-18 江苏食品药品职业技术学院 一种基于人脸识别的图书馆出入管理系统
CN109657605A (zh) * 2018-12-17 2019-04-19 东方网力科技股份有限公司 一种人脸轨迹与mac轨迹碰撞的方法及装置
CN109815845B (zh) * 2018-12-29 2022-03-29 达闼机器人有限公司 人脸识别的方法和装置、存储介质
CN109766857A (zh) * 2019-01-16 2019-05-17 嘉兴学院 一种基于语义对齐多区域模板融合的三维人脸识别方法
CN109902644A (zh) * 2019-03-07 2019-06-18 北京海益同展信息科技有限公司 人脸识别方法、装置、设备和计算机可读介质
CN110866443B (zh) * 2019-10-11 2023-06-16 厦门身份宝网络科技有限公司 人像存储方法、人脸识别方法、设备及存储介质
CN110717458B (zh) * 2019-10-11 2022-07-22 北京地平线机器人技术研发有限公司 人脸识别方法及识别装置
CN112860931B (zh) * 2021-01-18 2023-11-03 广东便捷神科技股份有限公司 人脸识别库的构建方法
CN112818885B (zh) * 2021-02-07 2024-05-28 中国银联股份有限公司 人脸识别方法、装置、设备及存储介质
CN113158786A (zh) * 2021-03-11 2021-07-23 光控特斯联(上海)信息科技有限公司 人脸识别数据处理方法、装置、计算机设备和存储介质
CN113269125B (zh) * 2021-06-10 2024-05-14 北京中科闻歌科技股份有限公司 一种人脸识别方法、装置、设备及存储介质
CN115880761B (zh) * 2023-02-09 2023-05-05 数据空间研究院 基于策略优化的人脸识别方法、系统、存储介质及应用

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060204058A1 (en) * 2004-12-07 2006-09-14 Kim Do-Hyung User recognition system and method thereof
CN101964064A (zh) * 2010-07-27 2011-02-02 上海摩比源软件技术有限公司 一种人脸比对方法
CN102184391A (zh) * 2011-05-19 2011-09-14 汉王科技股份有限公司 一种分布式人脸识别方法及系统和人脸识别终端
CN103902961A (zh) * 2012-12-28 2014-07-02 汉王科技股份有限公司 一种人脸识别方法及装置
CN105184238A (zh) * 2015-08-26 2015-12-23 广西小草信息产业有限责任公司 一种人脸识别方法及系统
CN105868695A (zh) * 2016-03-24 2016-08-17 北京握奇数据系统有限公司 一种人脸识别方法及系统

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005196678A (ja) * 2004-01-09 2005-07-21 Neucore Technol Inc テンプレートマッチング方法および対象画像領域抽出装置
CN105320923B (zh) * 2014-08-05 2018-06-12 北京大学 车型识别方法及装置
CN105335726B (zh) * 2015-11-06 2018-11-27 广州视源电子科技股份有限公司 人脸识别置信度获取方法和系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060204058A1 (en) * 2004-12-07 2006-09-14 Kim Do-Hyung User recognition system and method thereof
CN101964064A (zh) * 2010-07-27 2011-02-02 上海摩比源软件技术有限公司 一种人脸比对方法
CN102184391A (zh) * 2011-05-19 2011-09-14 汉王科技股份有限公司 一种分布式人脸识别方法及系统和人脸识别终端
CN103902961A (zh) * 2012-12-28 2014-07-02 汉王科技股份有限公司 一种人脸识别方法及装置
CN105184238A (zh) * 2015-08-26 2015-12-23 广西小草信息产业有限责任公司 一种人脸识别方法及系统
CN105868695A (zh) * 2016-03-24 2016-08-17 北京握奇数据系统有限公司 一种人脸识别方法及系统

Cited By (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110858283A (zh) * 2018-08-23 2020-03-03 营利度富信息系统（上海）有限公司 人脸识别在智慧银行系统中的应用
CN113327332A (zh) * 2018-09-29 2021-08-31 爱笔（北京）智能科技有限公司 基于生物信息的票务处理方法及系统、服务端、客户端
CN111291596A (zh) * 2018-12-07 2020-06-16 杭州海康威视数字技术股份有限公司 一种基于人脸识别的预警方法及装置
CN110298335A (zh) * 2019-07-09 2019-10-01 中控华运（厦门）集成电路有限公司 人脸识别系统及人脸识别方法
CN110843726A (zh) * 2019-12-19 2020-02-28 北京明朝万达科技股份有限公司 汽车门锁控制系统、方法及车辆
CN111582305B (zh) * 2020-03-26 2023-08-18 平安科技（深圳）有限公司 生物特征识别方法、装置、计算机设备及存储介质
CN111582305A (zh) * 2020-03-26 2020-08-25 平安科技（深圳）有限公司 生物特征识别方法、装置、计算机设备及存储介质
CN111597532A (zh) * 2020-04-10 2020-08-28 云知声智能科技股份有限公司 基于人脸识别实现儿童机器人童锁系统的方法和系统
CN111597532B (zh) * 2020-04-10 2023-11-17 云知声智能科技股份有限公司 基于人脸识别实现儿童机器人童锁系统的方法和系统
CN111444886A (zh) * 2020-04-27 2020-07-24 无锡职业技术学院 一种基于可穿戴设备的智能人脸识别系统
CN111639689A (zh) * 2020-05-20 2020-09-08 杭州海康威视系统技术有限公司 人脸数据处理方法、装置及计算机可读存储介质
CN111639689B (zh) * 2020-05-20 2023-07-25 杭州海康威视系统技术有限公司 人脸数据处理方法、装置及计算机可读存储介质
CN113420666A (zh) * 2021-06-23 2021-09-21 上海应用技术大学 一种基于人脸识别技术的记忆辅助方法及装置
CN113469012A (zh) * 2021-06-28 2021-10-01 广州云从鼎望科技有限公司 用户刷脸验证的方法、系统、介质及装置
CN113469012B (zh) * 2021-06-28 2024-05-03 广州云从鼎望科技有限公司 用户刷脸验证的方法、系统、介质及装置
CN114116849A (zh) * 2021-12-01 2022-03-01 南威软件股份有限公司 一种针对人像预警信息的二次研判方法及装置
CN115019365A (zh) * 2022-05-23 2022-09-06 北京计算机技术及应用研究所 一种基于模型适用性度量的分级人脸识别方法
CN115019365B (zh) * 2022-05-23 2024-04-09 北京计算机技术及应用研究所 一种基于模型适用性度量的分级人脸识别方法
CN115880745A (zh) * 2022-09-07 2023-03-31 以萨技术股份有限公司 一种获取人脸图像特征的数据处理系统
CN116597185A (zh) * 2023-07-17 2023-08-15 荣耀终端有限公司 模板更新方法及电子设备
CN116597185B (zh) * 2023-07-17 2023-12-15 荣耀终端有限公司 模板更新方法及电子设备
CN117284131A (zh) * 2023-09-26 2023-12-26 广东安能电业有限公司 一种电动单车充电柜及其控制方法
CN117284131B (zh) * 2023-09-26 2024-05-03 广东安能电业有限公司 一种电动单车充电柜及其控制方法

Also Published As

Publication number Publication date
CN105868695B (zh) 2019-04-02
CN105868695A (zh) 2016-08-17

Similar Documents

Publication Publication Date Title
WO2017162076A1 (fr) Procédé et système d'identification de visage
JP7109634B2 (ja) アイデンティティ認証方法及び装置
WO2019119505A1 (fr) Procédé et dispositif de reconnaissance faciale, dispositif informatique et support d'enregistrement
WO2021036436A1 (fr) Procédé et appareil de reconnaissance faciale
WO2018001092A1 (fr) Procédé et appareil de reconnaissance d'un visage
CN101587485B (zh) 一种基于人脸识别技术的人脸信息自动登录方法
CN105825176A (zh) 基于多模态非接触身份特征的识别方法
CN107358699B (zh) 一种安全验证方法及系统
WO2019090834A1 (fr) Procédé et appareil de capture d'armoire d'expression basés sur une empreinte vocale
CN103841108A (zh) 用户生物特征的认证方法和系统
CN108875907B (zh) 一种基于深度学习的指纹识别方法和装置
WO2013178002A1 (fr) Procédé et dispositif de reconnaissance de la parole et de mise en correspondance, programme d'ordinateur et support d'informations
Baig et al. Fingerprint-Iris fusion based identification system using a single hamming distance matcher
WO2016201731A1 (fr) Appareil et procédé de reconnaissance d'empreinte digitale et dispositif électronique
CN101964056A (zh) 一种具有活体检测功能的双模态人脸认证方法和系统
Liu et al. Learning multi-boosted HMMs for lip-password based speaker verification
WO2020050413A1 (fr) Dispositif de détermination d'image candidate du visage pour une authentification, procédé de détermination d'image candidate du visage pour une authentification, programme, et support d'enregistrement
US9135562B2 (en) Method for gender verification of individuals based on multimodal data analysis utilizing an individual's expression prompted by a greeting
CN113505717A (zh) 一种基于人脸面部特征识别技术的在线通行系统
Arora et al. Palmhashnet: Palmprint hashing network for indexing large databases to boost identification
KR101070442B1 (ko) 다중 레벨 얼굴 인식을 이용한 얼굴 인증 시스템 및 인증 방법
KR20100041562A (ko) 인증 대상자의 얼굴 인식과 지문 인식을 통한 사용자 인증 수행 방법 및 시스템
CN112861742A (zh) 人脸识别方法、装置、电子设备及存储介质
KR20110100008A (ko) 연령 및 성별을 이용한 사용자 인식 장치 및 방법
CN107657289A (zh) 多模态生物识别装置及其识别方法

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17769353

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 17769353

Country of ref document: EP

Kind code of ref document: A1