CN108549883A - A kind of face recognition methods again - Google Patents
A kind of face recognition methods again Download PDFInfo
- Publication number
- CN108549883A CN108549883A CN201810486584.6A CN201810486584A CN108549883A CN 108549883 A CN108549883 A CN 108549883A CN 201810486584 A CN201810486584 A CN 201810486584A CN 108549883 A CN108549883 A CN 108549883A
- Authority
- CN
- China
- Prior art keywords
- characteristic vector
- depth characteristic
- cosine similarity
- distance metric
- similarity measurement
- 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.)
- Pending
Links
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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/30—Scenes; Scene-specific elements in albums, collections or shared content, e.g. social network photos or video
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to field of face identification more particularly to a kind of recognition methods again of face, include the following steps:Training set A, B are obtained, using training set A, B training convolutional neural networks, depth characteristic vector set M, N is extracted respectively, establishes corresponding depth characteristic vector space S, D of depth characteristic vector set M, N;Learn distance metric and cosine similarity measurement;Calculate the distance metric for needing two facial images to be tested and cosine similarity measurement;Judge the similarity between two facial images in conjunction with distance metric and cosine similarity measurement.By using the present invention, following effect may be implemented:Distance metric is combined into the similarity between can more fully judging two width different images with cosine similarity measurement, it is more accurate to judge.
Description
Technical field
The present invention relates to field of face identification more particularly to a kind of recognition methods again of face.
Background technology
As people are paid more and more attention the concern of social public security and the development of mass data storage technology, peace
Personnel in full monitoring system, which re-recognize, has become a hot issue.Angle and the light variation at different cameral visual angle are very
Greatly, the appearance of personage may be changed so that it is still a challenging problem that personage, which re-recognizes,.
Precision of the identification technology in identification process be not high again by existing personnel.Such as application number:
CN201710839181.0, denomination of invention:A kind of patent based on the face authentication method for quickly handling more learning distance metrics
Apply, the technical solution recorded in the patent is:Enter training set, wherein training set is made of two subsets, the sample in subset S
This is to (xi, xj) the same face is come from, the sample in subset D is to (xi, xj) from two different faces;To training subset
Each of face image extract K kind features, indicate sample xiK-th of feature, k=1,2 ... ..., K;Learn K kind features
Corresponding distance matrix metric and its weight;Two facial images are inputted as test sample;To two facial images of input
Extract same K kinds feature, and using the distance matrix metric that has learnt and weight calculation this two facial images away from
From d;Judge, is same person, if more than the second given threshold if the distance of two images is less than given first threshold
Value, then be not same person.In the technical scheme, only similar and different face is trained, so identifying not
Same angle shot, the different same person of face's light different photos when, be unable to get one and accurately judge structure.It is another
Aspect by a single distance metric for find two width different images between difference and contact often it is unilateral.
Invention content
To solve the above problems, the present invention proposes a kind of face recognition methods again, the phase for judging two facial images
Like degree.
A kind of face recognition methods again, includes the following steps:Obtain training set A, B, wherein the sample in training set A is
The facial image that same person is shot by different angle under different light environments, the sample in training set B are logical for different people
Cross the facial image that different angle is shot under different light environments;Using training set A, B training convolutional neural networks, carry respectively
Depth characteristic vector set M, N are taken, corresponding depth characteristic vector space S, D of depth characteristic vector set M, N is established;Pass through depth spy
Sign vector set M, N learn to obtain subspace W, pass through depth characteristic vector set M, N, subspace W, depth characteristic vector space S, D
Calculate distance metric and cosine similarity measurement;Calculating needs the distance metric and cosine similarity of two facial images to be tested
Measurement;Judge the similarity between two facial images in conjunction with the distance metric and cosine similarity measurement.
Preferably, the convolutional neural networks include pond layer, the first full articulamentum and the second full articulamentum, and described first
Full articulamentum includes 1536 neurons, and the neuronal quantity of the second full articulamentum is equal to the number of pedestrian image in training set
Amount.
Preferably, the calculating distance metric includes:The calculation formula of distance metric is:
Wherein,xi,yjIt is the depth characteristic vector of image i and j,
I ∈ A, j ∈ B, xi∈ N, yj∈M。
Preferably, the calculating cosine similarity measurement includes:With a depth characteristic in depth characteristic vector space S
The cosine value of the angle of a depth characteristic vector y in vector x and depth characteristic vector space D is measured as cosine similarity
dcos(x,y)。
Preferably, distance metric described in the combination and cosine similarity are measured to judge the phase between two facial images
Include like degree:The distance matrix d (x, y) after distance metric and cosine similarity measurement fusion is calculated,
D (x, y)=dW(x,y)+dcos(x,y)
Given threshold T, if the result of calculation of distance matrix d (x, y) is more than threshold value T, two facial image dissmilarities are
Different people;Otherwise, it means that two facial images are similar, it is same person.
By using the present invention, following effect may be implemented:
1. the facial image of different angle shooting and different light shootings is trained by convolutional neural networks,
When identifying the facial image that different angle shooting and different light are shot, the accuracy of discrimination is improved;
2. distance metric and cosine similarity measurement are combined between can more fully judging two width different images
Similarity, judge it is more accurate.
Description of the drawings
The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
Fig. 1 is the flow diagram of the present invention.
Specific implementation mode
Below in conjunction with attached drawing, technical scheme of the present invention will be further described, but the present invention is not limited to these realities
Apply example.
The basic thought of the present invention is trained to the facial image shot in different angle shooting and different light,
To improve in the human face photo under identifying varying environment, robustness is good;In addition, not being to judge two width not according to distance metric
With the similarity between image, but between being combined by distance metric and cosine similarity measurement and judging different images
Similarity.
A kind of face recognition methods again, includes the following steps:
Step 1 obtains training set A, B, wherein the sample in training set A be same person by different angle in difference
The facial image shot under light environment, the sample in training set B be different people by different angle under different light environments
The facial image of shooting.In the discrimination of actual face, handled facial image is often different angle shooting, due to being
Different angle shooting, cause the light of face face different, i.e. the difference of lightness and darkness.Consider these discrimination factors, it will
Training set A, B are trained by convolutional neural networks, are more corresponded to actual needs, while improving the accuracy of discrimination.
Step 2: using training set A, B training convolutional neural networks, depth characteristic vector set M, N are extracted respectively, are established deep
Spend corresponding depth characteristic vector space S, D of set of eigenvectors M, N.Trained convolutional neural networks include pond layer, first complete
Articulamentum and the second full articulamentum, the first full articulamentum include 1536 neurons, the nerve of the second full articulamentum
First quantity is equal to the quantity of pedestrian image in training set.Wherein, the number parameter of the first full articulamentum oneself setting as needed;
Second full articulamentum plays the role of classification in entire convolutional neural networks;The quantity of second full articulamentum is training set A, B
In image number;Then fc7 layers in the second full articulamentum of output is extracted as depth characteristic vector.
Step 3 learns to obtain subspace W by depth characteristic vector set M, N, passes through depth characteristic vector set M, N, son
Space W, depth characteristic vector space S, D calculate distance metric and cosine similarity measurement.Specifically, the calculating of distance metric is public
Formula is:
Wherein,xi,yjIt is the depth characteristic vector of image i and j,
I ∈ A, j ∈ B, xi∈ N, yj∈M.The calculation formula of subspace W is:W=(w1,w2,…,wr)∈Rd×r, d expression primitive characters
The dimension in space, r indicate the dimension in transfer characteristic space.
Specifically, the computational methods of cosine similarity measurement are:With a depth characteristic in depth characteristic vector space S
The cosine value of the angle of a depth characteristic vector y in vector x and depth characteristic vector space D is measured as cosine similarity
dcos(x, y),
Step 4 calculates the distance metric for needing two facial images to be tested and cosine similarity measurement.In step 1
To in four, describes to obtain distance metric and the process of cosine similarity measurement obtains need according to the above process in this step
The distance metric and cosine similarity of two facial images to be tested are measured.
Step 5, it is similar between two facial images to judge in conjunction with the distance metric and cosine similarity measurement
Degree.Specifically:The distance matrix d (x, y) after distance metric and cosine similarity measurement fusion is calculated,
D (x, y)=dW(x,y)+dcos(x,y)
Given threshold T, if the result of calculation of distance matrix d (x, y) is more than threshold value T, two facial image dissmilarities are
Different people;Otherwise, it means that two facial images are similar, it is same person.
Cosine similarity measurement is to use two depth characteristic vectorial angle cosine values as poor between two images of measurement
Different size more focuses on difference of two vectors on direction, while correcting that module that may be present is skimble-scamble to ask
Topic.Distance metric and cosine similarity measurement are combined together, considered on common distance metric angle because
Element, therefore result is more accurate.
By using the present invention, following effect may be implemented:
1. the facial image of different angle shooting and different light shootings is trained by convolutional neural networks,
When identifying the facial image that different angle shooting and different light are shot, the accuracy of discrimination is improved;
2. distance metric and cosine similarity measurement are combined between can more fully judging two width different images
Similarity, judge it is more accurate.
Those skilled in the art can make various modifications to described specific embodiment
Or supplement or substitute by a similar method, however, it does not deviate from the spirit of the invention or surmounts the appended claims determines
The range of justice.
Claims (4)
1. a kind of recognition methods again of face, which is characterized in that include the following steps:
Obtain training set A, B, wherein the sample in training set A be same person by different angle under different light environments
The facial image of shooting, the sample in training set B are the face that different people is shot by different angle under different light environments
Image;
Using training set A, B training convolutional neural networks, depth characteristic vector set M, N are extracted respectively, establish depth characteristic vector
Collect corresponding depth characteristic vector space S, D of M, N;
Learn to obtain subspace W by depth characteristic vector set M, N, it is special by depth characteristic vector set M, N, subspace W, depth
It levies vector space S, D and calculates distance metric and cosine similarity measurement;
Calculate the distance metric for needing two facial images to be tested and cosine similarity measurement;
Judge the similarity between two facial images in conjunction with the distance metric and cosine similarity measurement.
2. face according to claim 1 recognition methods again, which is characterized in that the calculating distance metric includes:Distance
The calculation formula of measurement is:
Wherein,xi,yjIt is the depth characteristic vector of image i and j, i ∈
A, j ∈ B, xi∈ N, yj∈M。
3. face according to claim 1 recognition methods again, which is characterized in that the calculating cosine similarity measurement packet
It includes:With a depth spy in the depth characteristic vector x and depth characteristic vector space D in depth characteristic vector space S
The cosine value for levying the angle of vector y measures d as cosine similaritycos(x,y)。
4. face according to claim 2 or 3 recognition methods again, which is characterized in that distance metric described in the combination and
Cosine similarity measurement judges that the similarity between two facial images includes:
The distance matrix d (x, y) after distance metric and cosine similarity measurement fusion is calculated,
D (x, y)=dW(x,y)+dcos(x,y)
Given threshold T, if the result of calculation of distance matrix d (x, y) is more than threshold value T, two facial image dissmilarities, for difference
People;Otherwise, it means that two facial images are similar, it is same person.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810486584.6A CN108549883A (en) | 2018-08-06 | 2018-08-06 | A kind of face recognition methods again |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810486584.6A CN108549883A (en) | 2018-08-06 | 2018-08-06 | A kind of face recognition methods again |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108549883A true CN108549883A (en) | 2018-09-18 |
Family
ID=63495325
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810486584.6A Pending CN108549883A (en) | 2018-08-06 | 2018-08-06 | A kind of face recognition methods again |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108549883A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109543611A (en) * | 2018-11-22 | 2019-03-29 | 珠海市蓝云科技有限公司 | A method of the images match based on artificial intelligence |
CN109726756A (en) * | 2018-12-25 | 2019-05-07 | 北京旷视科技有限公司 | Image processing method, device, electronic equipment and storage medium |
CN111626567A (en) * | 2020-04-30 | 2020-09-04 | 中国直升机设计研究所 | Identification and calculation method for guaranteeing resource similarity |
CN111652260A (en) * | 2019-04-30 | 2020-09-11 | 上海铼锶信息技术有限公司 | Method and system for selecting number of face clustering samples |
CN113361301A (en) * | 2020-03-04 | 2021-09-07 | 上海分众软件技术有限公司 | Advertisement video identification method based on deep learning |
CN113362096A (en) * | 2020-03-04 | 2021-09-07 | 驰众信息技术(上海)有限公司 | Frame advertisement image matching method based on deep learning |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102592148A (en) * | 2011-12-29 | 2012-07-18 | 华南师范大学 | Face identification method based on non-negative matrix factorization and a plurality of distance functions |
US9129148B1 (en) * | 2012-11-09 | 2015-09-08 | Orbeus Inc. | System, method and apparatus for scene recognition |
CN106599807A (en) * | 2016-12-01 | 2017-04-26 | 中科唯实科技(北京)有限公司 | Auto-encoding-based pedestrian retrieval method |
CN106845397A (en) * | 2017-01-18 | 2017-06-13 | 湘潭大学 | A kind of confirming face method based on measuring similarity |
CN106919909A (en) * | 2017-02-10 | 2017-07-04 | 华中科技大学 | The metric learning method and system that a kind of pedestrian recognizes again |
CN107133601A (en) * | 2017-05-13 | 2017-09-05 | 五邑大学 | A kind of pedestrian's recognition methods again that network image super-resolution technique is resisted based on production |
CN107330397A (en) * | 2017-06-28 | 2017-11-07 | 苏州经贸职业技术学院 | A kind of pedestrian's recognition methods again based on large-spacing relative distance metric learning |
CN107423690A (en) * | 2017-06-26 | 2017-12-01 | 广东工业大学 | A kind of face identification method and device |
CN107729835A (en) * | 2017-10-10 | 2018-02-23 | 浙江大学 | A kind of expression recognition method based on face key point region traditional characteristic and face global depth Fusion Features |
CN108133192A (en) * | 2017-12-26 | 2018-06-08 | 武汉大学 | A kind of pedestrian based on Gauss-Laplace distribution statistics identifies again |
-
2018
- 2018-08-06 CN CN201810486584.6A patent/CN108549883A/en active Pending
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102592148A (en) * | 2011-12-29 | 2012-07-18 | 华南师范大学 | Face identification method based on non-negative matrix factorization and a plurality of distance functions |
US9129148B1 (en) * | 2012-11-09 | 2015-09-08 | Orbeus Inc. | System, method and apparatus for scene recognition |
CN106599807A (en) * | 2016-12-01 | 2017-04-26 | 中科唯实科技(北京)有限公司 | Auto-encoding-based pedestrian retrieval method |
CN106845397A (en) * | 2017-01-18 | 2017-06-13 | 湘潭大学 | A kind of confirming face method based on measuring similarity |
CN106919909A (en) * | 2017-02-10 | 2017-07-04 | 华中科技大学 | The metric learning method and system that a kind of pedestrian recognizes again |
CN107133601A (en) * | 2017-05-13 | 2017-09-05 | 五邑大学 | A kind of pedestrian's recognition methods again that network image super-resolution technique is resisted based on production |
CN107423690A (en) * | 2017-06-26 | 2017-12-01 | 广东工业大学 | A kind of face identification method and device |
CN107330397A (en) * | 2017-06-28 | 2017-11-07 | 苏州经贸职业技术学院 | A kind of pedestrian's recognition methods again based on large-spacing relative distance metric learning |
CN107729835A (en) * | 2017-10-10 | 2018-02-23 | 浙江大学 | A kind of expression recognition method based on face key point region traditional characteristic and face global depth Fusion Features |
CN108133192A (en) * | 2017-12-26 | 2018-06-08 | 武汉大学 | A kind of pedestrian based on Gauss-Laplace distribution statistics identifies again |
Non-Patent Citations (3)
Title |
---|
HONG ZHANG ET AL: "Unconstrained Face Verification by Subspace Similarity Metric Learning", 《 2017 12TH IEEE CONFERENCE ON INDUSTRIAL ELECTRONICS AND APPLICATIONS (ICIEA)》 * |
施彦等: "《神经网络设计方法与实例分析》", 31 December 2009 * |
李勇等: "基于特征脸的主成分分析人脸识别", 《计算计技术与自动化》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109543611A (en) * | 2018-11-22 | 2019-03-29 | 珠海市蓝云科技有限公司 | A method of the images match based on artificial intelligence |
CN109726756A (en) * | 2018-12-25 | 2019-05-07 | 北京旷视科技有限公司 | Image processing method, device, electronic equipment and storage medium |
CN111652260A (en) * | 2019-04-30 | 2020-09-11 | 上海铼锶信息技术有限公司 | Method and system for selecting number of face clustering samples |
CN111652260B (en) * | 2019-04-30 | 2023-06-20 | 上海铼锶信息技术有限公司 | Face clustering sample number selection method and system |
CN113361301A (en) * | 2020-03-04 | 2021-09-07 | 上海分众软件技术有限公司 | Advertisement video identification method based on deep learning |
CN113362096A (en) * | 2020-03-04 | 2021-09-07 | 驰众信息技术(上海)有限公司 | Frame advertisement image matching method based on deep learning |
CN111626567A (en) * | 2020-04-30 | 2020-09-04 | 中国直升机设计研究所 | Identification and calculation method for guaranteeing resource similarity |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108549883A (en) | A kind of face recognition methods again | |
Bhattacharya et al. | Smart attendance monitoring system (SAMS): a face recognition based attendance system for classroom environment | |
CN107506702B (en) | Multi-angle-based face recognition model training and testing system and method | |
CN106096538B (en) | Face identification method and device based on sequencing neural network model | |
CN105719188B (en) | The anti-method cheated of settlement of insurance claim and server are realized based on plurality of pictures uniformity | |
Rathod et al. | Automated attendance system using machine learning approach | |
CN108647583B (en) | Face recognition algorithm training method based on multi-target learning | |
CN109711366B (en) | Pedestrian re-identification method based on group information loss function | |
CN109101865A (en) | A kind of recognition methods again of the pedestrian based on deep learning | |
TWI439951B (en) | Facial gender identification system and method and computer program products thereof | |
CN106203242A (en) | A kind of similar image recognition methods and equipment | |
CN110516616A (en) | A kind of double authentication face method for anti-counterfeit based on extensive RGB and near-infrared data set | |
CN105989369B (en) | Pedestrian re-identification method based on metric learning | |
CN111914761A (en) | Thermal infrared face recognition method and system | |
CN105138951B (en) | Human face portrait-photo array the method represented based on graph model | |
CN116311549A (en) | Living body object identification method, apparatus, and computer-readable storage medium | |
CN104700089A (en) | Face identification method based on Gabor wavelet and SB2DLPP | |
CN109635634A (en) | A kind of pedestrian based on stochastic linear interpolation identifies data enhancement methods again | |
Liu et al. | Detecting presentation attacks from 3d face masks under multispectral imaging | |
CN107145841A (en) | A kind of low-rank sparse face identification method and its system based on matrix | |
CN112668557A (en) | Method for defending image noise attack in pedestrian re-identification system | |
CN108268839A (en) | A kind of live body verification method and its system | |
CN107944340B (en) | Pedestrian re-identification method combining direct measurement and indirect measurement | |
Damer et al. | Deep learning-based face recognition and the robustness to perspective distortion | |
TW201828156A (en) | Image identification method, measurement learning method, and image source identification method and device capable of effectively dealing with the problem of asymmetric object image identification so as to possess better robustness and higher accuracy |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180918 |
|
RJ01 | Rejection of invention patent application after publication |