CN101751551B - Method, device, system and device for identifying face based on image - Google Patents
Method, device, system and device for identifying face based on image Download PDFInfo
- Publication number
- CN101751551B CN101751551B CN 200810217988 CN200810217988A CN101751551B CN 101751551 B CN101751551 B CN 101751551B CN 200810217988 CN200810217988 CN 200810217988 CN 200810217988 A CN200810217988 A CN 200810217988A CN 101751551 B CN101751551 B CN 101751551B
- Authority
- CN
- China
- Prior art keywords
- face
- facial image
- image
- people
- sample
- 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.)
- Active
Links
Images
Abstract
The invention is applicable to the field of pattern recognition, and provides a method, a device, a system and a device for identifying face based on image. The method comprises the following steps: collecting face images from multiple points; extracting an optimal face image from the collected face images from multiple points; extracting a face feature from the optimal face image, and identifying the class of the face from the face feature library according to the extracted face feature. In the invention, the method comprises judging the optimal face image from the face images collected from multiple points, extracting the face feature, and identifying the face class from the face feature library according to the extracted face feature so as to realize the method for identifying face based on image; and the front image of the face still can be obtained with high identification accuracy and low cost when the face swings greatly.
Description
Technical field
The invention belongs to area of pattern recognition, relate in particular to a kind of image-based face identification method, device, system and equipment.
Background technology
Since the nineties in 20th century, along with the sharp increase of needs, face recognition technology becomes the research topic of a hot topic.Although research has in this respect obtained some gratifying achievements, some application have also been obtained in a lot of fields, for example criminal's identification of public security system, driving license, I.D. and passport etc. are checked with identity verification in driving license, the security system of bank, the fields such as the supervisory system of customs and automatic gatekeeper system.In actual applications, a lot of situations need to be used the constant face recognition algorithms of attitude, and present most face identification system is all for front face, but swinging at people's face, existing face identification system can not guarantee in the larger situation that people's face of inputting is accurate positive, can cause certain influence to the accuracy of identification like this, how guarantee to photograph as far as possible good problem to study of front face image.
Summary of the invention
The purpose of the embodiment of the invention is to provide a kind of image-based face identification method, is intended to solve existing face identification system and can not guarantees that in the larger situation of people's face swing people's face of inputting is accurate positive problem.
The embodiment of the invention is achieved in that a kind of image-based face identification method, and described method comprises the steps:
At the multipoint acquisition facial image;
From multipoint acquisition to facial image extract best facial image;
From described best facial image, extract face characteristic, from the face characteristic storehouse, identify classification under people's face according to the face characteristic that extracts;
Described from multipoint acquisition to facial image extract best facial image step be specially:
In the facial image that respectively collects, detect people's face position, and abandon the facial image that can't detect people's face position;
In each facial image that detects people's face position, detect position of human eye, and abandon the facial image that can't detect position of human eye;
Relatively detect the difference of two oculopupillary distances and interpupillary distance gauged distance in each facial image of position of human eye, the facial image of difference minimum is defined as best facial image;
Described detection people face position and detection position of human eye adopt ADABOOST people's face detection algorithm of optimizing to form strong classifier, and ADABOOST people's face detection algorithm of described optimization comprises the step of the error weight of following more new samples:
Wherein, w
T, iBe the error weight of i sample in the t time circulation, x
iBe i sample data, yi=0 represents that i sample is false, and yi=1 represents that i sample is true, h
t(x
i) be Weak Classifier h
tTo sample x
iClassification results, ε
tBe Weak Classifier h
tCorresponding error rate, β
t=ε
t/ (1-ε
t).
Another purpose of the embodiment of the invention is to provide a kind of image-based face identification device, and described device comprises:
Image acquisition units is used at the multipoint acquisition facial image;
Image extraction unit, be used for from described image acquisition units multipoint acquisition to facial image extract best facial image; And
Recognition unit is used for extracting face characteristic from described best facial image, identifies classification under people's face according to the face characteristic that extracts from the face characteristic storehouse;
Described image extraction unit comprises:
People's face detection module is used for detecting people's face position at the described facial image that respectively collects, and abandons the facial image that can't detect people's face position;
The human eye detection module is used for detecting position of human eye at each the described facial image that detects people's face position, and abandons the facial image that can't detect position of human eye; And
The image confirming module is used for relatively detecting each described facial image two oculopupillary distance of position of human eye and the difference of interpupillary distance gauged distance, and the facial image of difference minimum is defined as best facial image;
Described people's face detection module and human eye detection module adopt ADABOOST people's face detection algorithm of optimizing to form strong classifier, and ADABOOST people's face detection algorithm of described optimization comprises the lower more step of the error weight of new samples:
Wherein, w
T, iBe the error weight of i sample in the t time circulation, x
iBe i sample data, yi=0 represents that i sample is false, and yi=1 represents that i sample is true, h
t(x
i) be Weak Classifier h
tTo sample x
iClassification results, ε
tBe Weak Classifier h
tCorresponding error rate, β
t=ε
t/ (1-ε
t).
Another purpose of the embodiment of the invention is to provide a kind of pattern recognition system that comprises above-mentioned image-based face identification device.
Another purpose of the embodiment of the invention is to provide a kind of electronic equipment that comprises above-mentioned pattern recognition system.
In embodiments of the present invention, by from multipoint acquisition to facial image judge and to draw best facial image, and extraction face characteristic, from the face characteristic storehouse, identify classification under people's face according to the face characteristic that extracts, realized a kind of image-based face identification method, it is high to swing direct picture, the recognition correct rate that still can obtain people's face in the larger situation at people's face, and realizes that cost is low.
Description of drawings
Fig. 1 is the realization flow figure of the image-based face identification method that provides of the embodiment of the invention;
Fig. 2 is the arrangement synoptic diagram of the camera that provides of the embodiment of the invention;
Fig. 3 is the synoptic diagram of selecting rectangular characteristic that the embodiment of the invention provides;
Fig. 4 is the synoptic diagram of two interpupillary distance providing of the embodiment of the invention;
Fig. 5 is that the facial image with the best that the embodiment of the invention provides is divided into 5 from top to bottom to extract the synoptic diagram of LDA feature;
Fig. 6 is the structural drawing of the image-based face identification device that provides of the embodiment of the invention.
Embodiment
In order to make purpose of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, is not intended to limit the present invention.
In embodiments of the present invention, by from multipoint acquisition to facial image judge and to draw best facial image, and extraction face characteristic, identify classification under people's face according to the face characteristic that extracts from the face characteristic storehouse, what namely confirm to collect according to the face characteristic that extracts is whose facial image in the eigenface storehouse.
Fig. 1 shows the realization flow of the image-based face identification method that the embodiment of the invention provides, and details are as follows:
In step S101, at the multipoint acquisition facial image;
In step S102, from multipoint acquisition to facial image extract best facial image;
In step S103, from best facial image, extract face characteristic, from the face characteristic storehouse, identify classification under people's face according to the face characteristic that extracts.
In embodiments of the present invention, be in same plane by 5, and lay respectively at the mid point on square four limits and the first-class image capture module of shooting of foursquare central point and gather facial image, the distance on people's face distance plane, 5 camera places is 100CM during shooting, and the line at the square center that people's face and camera consist of is perpendicular to this plane, square place, and the synoptic diagram of 5 cameras and people's face position as shown in Figure 2.In general recognition of face, the angle that the front face image allows to tilt can be calculated the mounting distance L of each stylus by following formula within 5 degree:
L≤100×tan(15°)＝26.8(cm)；
The mounting distance L that calculates two cameras needs only at 26.8CM with interior.Here, we round to get the mounting distance of two cameras, and namely square 1/2 length of side is 20CM, obtains the arrangement synoptic diagram of 5 cameras as shown in Figure 2.
Then, judge best facial image, the realization flow of step S102, details are as follows:
Here, to the detection of people's face position, carry out people's face position probing based on the MITEX face database in ADABOOST people's face detection algorithm completing steps 1 that employing is optimized, altogether positive attitude facial image 2706 width of cloth, non-face image 3841 width of cloth, image size 20 * 20 pixels.Because identification is front face in the present invention, so optimize the simple rectangular characteristic of part of selecting as shown in Figure 3, can improve training speed, detection speed and accuracy rate that people's face detects sorter.Certainly, except adopting rectangular characteristic shown in Figure 3, also can select the rectangular characteristic of other types.
Wherein, details are as follows for ADABOOST people's face detection algorithm of optimization:
1. choose N training sample (x
1, y
1) ..., (x
n, y
n) sample set that forms;
X wherein
iBe sample data (being facial image), y
i=0,1}, and i=1,2 ..., n, y
i=1 this sample of expression is true, y
i=0 this sample of expression is false, wherein positive sample (target object that namely will detect is facial image in embodiments of the present invention) k, and negative sample m.
2. establish w
T, jIt is the error weight of j sample in the t time circulation;
Error weight in the training sample is pressed following formula initialization: for y
j=0 sample, w
T, j=1/2m; For sample y
j=1, w
T, j=1/2k.
3. by interative computation, select optimum Weak Classifier;
If iterations is T time, T can get 15,20 etc., and iterations T generally begins to get 20, then revises according to the accuracy of identification of the strong classifier that obtains below, and is inadequate such as the strong classifier accuracy of identification, then will increase the T value, otherwise can reduce.For the t time iteration, proceed as follows:
(1) carries out first weight normalization, use
Be w
T, jAssignment;
(2) for each feature j, train its Weak Classifier h
j, the expression formula of each Weak Classifier is:
Wherein, f
jBe character pair j, the value of the rectangular characteristic of sample x.Definite threshold θ
j, partially be worth p
jAnd f
j, make the device objective function
Reach minimum.
(3) from top definite Weak Classifier, find out and have minimal error rate ε
tWeak Classifier h
t
(4) the error weight of new samples more:
Wherein, h
t(x
i) be Weak Classifier h
tTo sample x
iClassification results, ε
tBe Weak Classifier h
tCorresponding error rate, β
t=ε
t/ (1-ε
t).
With respect to general ADABOOST algorithm, problems such as here by increase adopting following formula to upgrade the step of sample weights, can effectively preventing endless loop, crossing training, precision is not high.
4. formation strong classifier:
Adopting ADABOOST people's face detection algorithm of optimizing to form on the basis of strong classifier, make up cascade classifier, details are as follows for detailed process:
1. train the sorter of equivalent layer with the positive and negative samples set, so that the misclassification rate of this layer sorter is less than default maximum misclassification rate, percent of pass is greater than default percent of pass, and with the product of the misclassification rate that obtains and this layer target misclassification rate target misclassification rate as corresponding lower floor;
2. whether judge the target misclassification rate of corresponding lower floor greater than the target misclassification rate, and during greater than the target misclassification rate, scan non-face image with current cascade detectors at it, collect the wrong knowledge of institute to the negative sample set.
The method of the structure cascade classifier that the embodiment of the invention provides can improve the training speed that people's face detects sorter, and details are as follows for its corresponding algorithm:
1. set every layer of maximum misclassification rate f, the target misclassification rate F of every layer of minimum percent of pass d and whole detecting device
Target, known positive sample set Pos and negative sample set Neg;
2. initialization F
1=1, i=1;
3. judge F
i＞F
TargetThe time, execution in step 4, otherwise finish;
4 usefulness Pos and Neg train i layer and setting threshold b, so that misclassification rate f
iLess than f, percent of pass is greater than d;
5 use f
iF
iBe F
I+1Assignment is the i assignment with i+1, uses
Be the Neg assignment;
If 6 F
i＞F
Target, then scanning non-face image with current cascade detectors, and collect all wrongheaded samples to set Neg, execution 4 after finishing; Otherwise, finish.
In above-mentioned steps 2, also ADABOOST people's face detection algorithm of employing optimization is finished the detection to position of human eye, carries out position of human eye based on the MIT-CBCL face database and detects.This MIT-CBCL database has 10 people, 3200 training images, and image size 200 * 200 is by hand-making out front human eye image 1000 width of cloth, 2700 of background pictures, image size 20 * 20 pixels.Then adopt above-mentioned training method to train.
By experiment, the above-mentioned algorithm that the embodiment of the invention provides has obtained good effect, people's face is no more than the scope of 30 degree (comprising the inclination of upper and lower, left and right direction) to stylus tilt matrix angle in, all can capture fast human eye, real-time is very high.
According to the statistics to Asia and European interpupillary distance, Asian interpupillary distance mean value is 65CM, and European interpupillary distance mean value is 63CM.In the embodiment of the invention, as shown in Figure 4, with the distance of the central point of two human eyes that the detect interpupillary distance as two, according to for the crowd, the interpupillary distance gauged distance adopts 65CM, the interpupillary distance of people's face image of detecting human eye sorted, and be best facial image near 65CM.
After drawing best facial image, determine classification under people's face according to this best facial image, above-mentioned steps S103 specifically comprises three steps, and details are as follows:
Facial image piecemeal after step 2. will be proofreaied and correct extracts linear discriminant analysis (Linear Discrirninant Analysis, LDA) feature;
Here, the facial image of the best is divided into as shown in Figure 55 from top to bottom, then piecemeal projects to projection matrix and extracts the LDA feature.
Three rank nearest neighbour classification algorithms are specially:
(1) with said extracted to LDA feature and the people's face in the eigenface storehouse calculate Euclidean distance, obtain three minimum eigenfaces of distance;
(2) judge whether to have at least two eigenfaces that obtain to belong to same classification in the eigenface storehouse;
(3) if wherein at least two eigenfaces that obtain belong to same classification, then people's face to be identified belongs to this classification;
(4) if three eigenfaces that obtain belong to respectively different classifications, then people's face to be identified belongs to the affiliated classification of that minimum eigenface of distance.
Above-mentioned eigenface storehouse obtains according to following algorithm, and details are as follows:
(1) with the training sample picture matrix according to the corresponding manner piecemeal that extracts the LDA feature and adopt;
If training sample is divided into into the C class, the quantity of every class sample is respectively: n
1, n
2..., n
c, training sample picture matrix A then
IjBe divided into 5, can be expressed as follows:
A
ij＝[(ξ
ij)
1?(ξ
ij)
2?(ξ
ij)
3?(ξ
ij)
4?(ξ
ij)
5]
T；
Wherein, j depends on the kind C of training sample, i≤C; J depends on the data of Different categories of samples, j≤n
1, n
2... n
c(ξ
Ij)
1Be first view data, (ξ
Ij)
2Be second view data ..., (ξ
Ij)
5It is the 5th view data.
(2) the feature space matrix of calculation training sample;
All image subsections in the training sample image are all seen the training sample vector, and then scatter matrix is between the class of the subimage matrix of all training image samples:
Scatter matrix is in the class of the subimage matrix of all training image samples:
Wherein,
Be the average of all subimage matrixes of i class training sample,
Be the average of all training sample subimage matrixes, M is all image subsection matrix numbers.
Then, find out
Front r the corresponding proper vector u of eigenvalue of maximum
1, u
2..., u
rWherein, r determines according to the parameters such as size of concrete picture.
If projection matrix is U=[u
1, u
2..., u
r], training sample A then
IjThe feature space matrix be:
A
ij＝[U
T(ξ
ij)
1?U
T(ξ
ij)
2?U
T(ξ
ij)
3?U
T(ξ
ij)
4?U
T(ξ
ij)
5]
T。
(3) all training sample piecemeals are projected to projection matrix U=[u
1, u
2..., u
r], obtain the eigenface storehouse.
When reducing the camera collection facial image, because the parameters such as focal length, aperture, gain, white balance arrange the variation that difference affects the facial image data that collect, the stability of the facial image data that assurance collects, as a preferred embodiment of the present invention, gather facial image by infrared camera.
Certainly, also can adopt the image capture module collection facial images such as digital camera, can also revise as required the arrangement mode of image capture module.When extracting the LDA feature of facial image, also facial image can be divided into polylith according to other modes.But, the facial image after proofreading and correct is divided into 5 is conducive to carry out the LDA feature extraction.
One of ordinary skill in the art will appreciate that, realize that all or part of step in above-described embodiment method is to come the relevant hardware of instruction to finish by program, described program can be in being stored in a computer read/write memory medium, described storage medium, such as ROM/RAM, disk, CD etc., this program is used for carrying out following steps:
1. at the multipoint acquisition facial image;
From multipoint acquisition to facial image extract best facial image;
3. from best facial image, extract face characteristic, from the face characteristic storehouse, identify classification under people's face according to the face characteristic that extracts.
Fig. 6 shows the structure of the image-based face identification device that the embodiment of the invention provides, and only shows for convenience of explanation the part relevant with the embodiment of the invention.
This device can be used for various pattern recognition system, these pattern recognition system can be loaded into electronic equipment, for example computing machine, personal digital assistant (Personal Digital Assistant, PDA) etc., this image-based face identification device can be to run on the unit that software unit, hardware cell or software and hardware in these electronic equipments combine, also can be used as independently, suspension member is integrated in these electronic equipments or runs in the application system of these electronic equipments, wherein:
In embodiments of the present invention, image extraction unit 602 comprises:
People's face detection module 6021 detects people's face position, and abandons the facial image that can't detect people's face position in the facial image that respectively collects, its implementation repeats no more as mentioned above.
Human eye detection module 6022 detects position of human eye in each facial image that detects people's face position, and abandons the facial image that can't detect position of human eye, and its implementation repeats no more as mentioned above.
In embodiments of the present invention, recognition unit 603 comprises:
Characteristic extracting module 6032 is extracted the LDA feature with the facial image piecemeal after proofreading and correct, and specific implementation repeats no more as mentioned above.
Classification is confirmed module 6033, identifies classification under people's face according to the LDA feature of extracting and three rank nearest neighbour classification algorithms from the face characteristic storehouse, and specific implementation repeats no more as mentioned above.
When reducing the camera collection facial image, because the parameters such as focal length, aperture, gain, white balance arrange the variation that difference affects the facial image data that collect, the stability of the facial image data that assurance collects, as a preferred embodiment of the present invention, gather facial image by infrared camera.
In embodiments of the present invention, by from multipoint acquisition to facial image judge and to draw best facial image, and extraction face characteristic, from the face characteristic storehouse, identify classification under people's face according to the face characteristic that extracts, realized a kind of image-based face identification method, it is high to swing direct picture, the recognition correct rate that still can obtain people's face in the larger situation at people's face, and realizes that cost is low.And, adopting infrared camera and LDA to calculate facial image is processed, the face characteristic that statistics is extracted adopts three rank nearest neighbour classification algorithms that people's face is classified, and efficiently solves the problem of uneven illumination.
The above only is preferred embodiment of the present invention, not in order to limiting the present invention, all any modifications of doing within the spirit and principles in the present invention, is equal to and replaces and improvement etc., all should be included within protection scope of the present invention.
Claims (13)
1. an image-based face identification method is characterized in that, described method comprises the steps:
At the multipoint acquisition facial image;
From multipoint acquisition to facial image extract best facial image;
From described best facial image, extract face characteristic, from the face characteristic storehouse, identify classification under people's face according to the face characteristic that extracts;
Described from multipoint acquisition to facial image extract best facial image step be specially:
In the facial image that respectively collects, detect people's face position, and abandon the facial image that can't detect people's face position;
In each facial image that detects people's face position, detect position of human eye, and abandon the facial image that can't detect position of human eye;
Relatively detect the difference of two oculopupillary distances and interpupillary distance gauged distance in each facial image of position of human eye, the facial image of difference minimum is defined as best facial image;
Described detection people face position and detection position of human eye adopt ADABOOST people's face detection algorithm of optimizing to form strong classifier, and ADABOOST people's face detection algorithm of described optimization comprises the step of the error weight of following more new samples:
Wherein, w
T, iBe the error weight of i sample in the t time circulation, x
iBe i sample data, y
iI sample of=0 expression is false, y
iI sample of=1 expression is true, h
t(x
i) be Weak Classifier h
tTo sample x
iClassification results, ε
tBe Weak Classifier h
tCorresponding error rate, β
t=ε
t/ (1-ε
t).
2. the method for claim 1 is characterized in that, described multiple spot is 5 and is in same plane, and lays respectively at the mid point on square four limits, and this foursquare central point.
3. the method for claim 1 is characterized in that, makes up cascade classifier on the basis of described strong classifier, and the process that makes up cascade classifier is specially:
Sorter with positive and negative samples set training equivalent layer, so that the misclassification rate of this layer sorter is less than default maximum misclassification rate, percent of pass is greater than default percent of pass, and with the product of the misclassification rate that obtains and this layer target misclassification rate target misclassification rate as corresponding lower floor;
Whether judge the target misclassification rate of corresponding lower floor greater than the target misclassification rate, and during greater than the target misclassification rate, scan non-face image with current cascade detectors at it, collect the wrong knowledge of institute to the negative sample set.
4. the method for claim 1 is characterized in that, describedly extracts face characteristic from described best facial image, identifies from the face characteristic storehouse according to the face characteristic that extracts that the step of classification is specially under people's face:
According to the angle of inclination of two human eyes the facial image of the best is carried out slant correction;
Facial image piecemeal after proofreading and correct is extracted the linear discriminant analysis feature;
From the face characteristic storehouse, identify classification under people's face according to the linear discriminant analysis feature of extracting and three rank nearest neighbour classification algorithms.
5. method as claimed in claim 4 is characterized in that, described three rank nearest neighbour classification algorithms are specially:
The described linear discriminant analysis feature and the people's face in the eigenface storehouse that extract are calculated Euclidean distance, obtain three minimum eigenfaces of distance;
Judge whether to have at least two eigenfaces that obtain to belong to same classification in the described eigenface storehouse;
If wherein at least two eigenfaces that obtain belong to same classification, then people's face to be identified belongs to this classification;
If three eigenfaces that obtain belong to respectively different classifications, then people's face to be identified belongs to the affiliated classification of that minimum eigenface of distance.
6. the method for claim 1 is characterized in that, described eigenface storehouse obtains according to following algorithm:
With the training sample picture matrix according to the corresponding manner piecemeal that extracts the linear discriminant analysis feature and adopt;
The feature space matrix of calculation training sample;
All training sample piecemeals are projected to projection matrix, obtain the eigenface storehouse.
7. method as claimed in claim 2 is characterized in that, described facial image is by the infrared camera collection.
8. an image-based face identification device is characterized in that, described device comprises:
Image acquisition units is used at the multipoint acquisition facial image;
Image extraction unit, be used for from described image acquisition units multipoint acquisition to facial image extract best facial image; And
Recognition unit is used for extracting face characteristic from described best facial image, identifies classification under people's face according to the face characteristic that extracts from the face characteristic storehouse;
Described image extraction unit comprises:
People's face detection module is used for detecting people's face position at the described facial image that respectively collects, and abandons the facial image that can't detect people's face position;
The human eye detection module is used for detecting position of human eye at each the described facial image that detects people's face position, and abandons the facial image that can't detect position of human eye; And
The image confirming module is used for relatively detecting each described facial image two oculopupillary distance of position of human eye and the difference of interpupillary distance gauged distance, and the facial image of difference minimum is defined as best facial image;
Described people's face detection module and human eye detection module adopt ADABOOST people's face detection algorithm of optimizing to form strong classifier, and ADABOOST people's face detection algorithm of described optimization comprises the lower more step of the error weight of new samples:
Wherein, w
T, iBe the error weight of i sample in the t time circulation, x
iBe i sample data, y
iI sample of=0 expression is false, y
iI sample of=1 expression is true, h
t(x
i) be Weak Classifier h
tTo sample x
iClassification results, ε
tBe Weak Classifier h
tCorresponding error rate, β
t=ε
t/ (1-ε
t).
9. device as claimed in claim 8 is characterized in that, described image acquisition units comprises that 5 are in same plane, and lays respectively at the mid point on square four limits and the image capture module of foursquare central point.
10. device as claimed in claim 8 is characterized in that, described recognition unit comprises:
Correction module carries out slant correction according to the angle of inclination of two human eyes to the facial image of the best;
Characteristic extracting module, the described facial image piecemeal after being used for proofreading and correct extracts the linear discriminant analysis feature;
And
Classification is confirmed module, is used for identifying classification under people's face from the face characteristic storehouse according to the described linear discriminant analysis feature of extracting and three rank nearest neighbour classification algorithms.
11. device as claimed in claim 9 is characterized in that, described image capture module is infrared camera.
12. pattern recognition system that comprises each described image-based face identification device of claim 8 to 11.
13. electronic equipment that comprises the described pattern recognition system of claim 12.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200810217988 CN101751551B (en) | 2008-12-05 | 2008-12-05 | Method, device, system and device for identifying face based on image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200810217988 CN101751551B (en) | 2008-12-05 | 2008-12-05 | Method, device, system and device for identifying face based on image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101751551A CN101751551A (en) | 2010-06-23 |
| CN101751551B true CN101751551B (en) | 2013-03-20 |
Family
ID=42478516
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200810217988 Active CN101751551B (en) | 2008-12-05 | 2008-12-05 | Method, device, system and device for identifying face based on image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101751551B (en) |
Families Citing this family (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8705812B2 (en) * | 2011-06-10 | 2014-04-22 | Amazon Technologies, Inc. | Enhanced face recognition in video |
| JP5231685B1 (en) | 2011-07-07 | 2013-07-10 | 花王株式会社 | Facial impression analysis method, beauty counseling method and face image generation method |
| US8971574B2 (en) * | 2011-11-22 | 2015-03-03 | Ulsee Inc. | Orientation correction method for electronic device used to perform facial recognition and electronic device thereof |
| CN103186233B (en) * | 2011-12-31 | 2016-03-09 | 上海飞来飞去新媒体展示设计有限公司 | Panoramic interaction control method for eye location |
| CN103035049A (en) * | 2012-12-12 | 2013-04-10 | 山东神思电子技术股份有限公司 | FPGA (Field Programmable Gate Array)-based face recognition entrance guard device and FPGA-based face recognition entrance guard method |
| CN103440476A (en) * | 2013-08-26 | 2013-12-11 | 大连理工大学 | Locating method for pupil in face video |
| CN105303149B (en) * | 2014-05-29 | 2019-11-05 | 腾讯科技（深圳）有限公司 | The methods of exhibiting and device of character image |
| CN104091176B (en) * | 2014-07-18 | 2015-10-14 | 吴建忠 | Portrait comparison application technology in video |
| CN106355138A (en) * | 2016-08-18 | 2017-01-25 | 电子科技大学 | Face recognition method based on deep learning and key features extraction |
| CN106778559A (en) * | 2016-12-01 | 2017-05-31 | 北京旷视科技有限公司 | The method and device of In vivo detection |
| CN107145873A (en) * | 2017-05-12 | 2017-09-08 | 江苏鸿信系统集成有限公司 | ID Card Image identification checking method and system based on recognition of face and OCR |
| CN107633244A (en) * | 2017-11-13 | 2018-01-26 | 姜茂清 | Human body direct picture harvester, the image capturing system comprising it and its application |
| CN108171207A (en) * | 2018-01-17 | 2018-06-15 | 百度在线网络技术（北京）有限公司 | Face identification method and device based on video sequence |
| CN108921026A (en) * | 2018-06-01 | 2018-11-30 | 平安科技（深圳）有限公司 | Recognition methods, device, computer equipment and the storage medium of animal identification |
| CN109635625B (en) * | 2018-10-16 | 2023-08-18 | 平安科技（深圳）有限公司 | Intelligent identity verification method, equipment, storage medium and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1618079A (en) * | 2002-01-16 | 2005-05-18 | 虹膜技术公司 | System and method for iris identification using stereoscopic face recognition |
| CN1731417A (en) * | 2005-08-19 | 2006-02-08 | 清华大学 | Method of robust human face detection in complicated background image |
| CN1908960A (en) * | 2005-08-02 | 2007-02-07 | 中国科学院计算技术研究所 | Feature classification based multiple classifiers combined people face recognition method |
| CN101236599A (en) * | 2007-12-29 | 2008-08-06 | 浙江工业大学 | Human face recognition detection device based on multi- video camera information integration |
-
2008
- 2008-12-05 CN CN 200810217988 patent/CN101751551B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1618079A (en) * | 2002-01-16 | 2005-05-18 | 虹膜技术公司 | System and method for iris identification using stereoscopic face recognition |
| CN1908960A (en) * | 2005-08-02 | 2007-02-07 | 中国科学院计算技术研究所 | Feature classification based multiple classifiers combined people face recognition method |
| CN1731417A (en) * | 2005-08-19 | 2006-02-08 | 清华大学 | Method of robust human face detection in complicated background image |
| CN101236599A (en) * | 2007-12-29 | 2008-08-06 | 浙江工业大学 | Human face recognition detection device based on multi- video camera information integration |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101751551A (en) | 2010-06-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101751551B (en) | Method, device, system and device for identifying face based on image | |
| King | Max-margin object detection | |
| Jin et al. | Face detection using improved LBP under Bayesian framework | |
| CN100568262C (en) | Human face recognition detection device based on the multi-video camera information fusion | |
| CN101339609B (en) | Image processing apparatus and image processing method | |
| Gualdi et al. | Multistage particle windows for fast and accurate object detection | |
| Yao et al. | Fast human detection from videos using covariance features | |
| Sahbi et al. | A Hierarchy of Support Vector Machines for Pattern Detection. | |
| CN102855496A (en) | Method and system for authenticating shielded face | |
| CN108563999A (en) | A kind of piece identity's recognition methods and device towards low quality video image | |
| CN102930300B (en) | Method and system for identifying airplane target | |
| Kaâniche et al. | Recognizing gestures by learning local motion signatures of HOG descriptors | |
| CN103279768A (en) | Method for identifying faces in videos based on incremental learning of face partitioning visual representations | |
| Amato et al. | A comparison of face verification with facial landmarks and deep features | |
| CN104573672B (en) | A kind of discriminating kept based on neighborhood is embedded in face identification method | |
| US9147130B2 (en) | Information processing method, information processing apparatus, and recording medium for identifying a class of an object by using a plurality of discriminators | |
| CN112149538A (en) | Pedestrian re-identification method based on multi-task learning | |
| Imran et al. | Face recognition using eigenfaces | |
| CN102915435A (en) | Multi-pose face recognition method based on face energy diagram | |
| Yeung et al. | Improved performance of face recognition using CNN with constrained triplet loss layer | |
| Subburaman et al. | Fast bounding box estimation based face detection | |
| CN103218606A (en) | Multi-pose face recognition method based on face mean and variance energy images | |
| CN100573549C (en) | Special object is surveyed method and apparatus | |
| Ngxande et al. | Detecting inter-sectional accuracy differences in driver drowsiness detection algorithms | |
| Andiani et al. | Face recognition for work attendance using multitask convolutional neural network (MTCNN) and pre-trained facenet |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |