CN105320921B - Eyes localization method and eyes positioning device - Google Patents
Eyes localization method and eyes positioning device Download PDFInfo
- Publication number
- CN105320921B CN105320921B CN201410374902.1A CN201410374902A CN105320921B CN 105320921 B CN105320921 B CN 105320921B CN 201410374902 A CN201410374902 A CN 201410374902A CN 105320921 B CN105320921 B CN 105320921B
- Authority
- CN
- China
- Prior art keywords
- points
- face
- frame
- encirclement
- point
- 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
- 238000000034 method Methods 0.000 title claims abstract description 42
- 230000004807 localization Effects 0.000 title claims abstract description 16
- 210000001747 pupil Anatomy 0.000 claims abstract description 34
- 238000012545 processing Methods 0.000 claims abstract description 20
- 238000006073 displacement reaction Methods 0.000 claims description 33
- 238000006243 chemical reaction Methods 0.000 claims description 15
- 238000005070 sampling Methods 0.000 claims description 13
- 230000001815 facial effect Effects 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 10
- 238000001514 detection method Methods 0.000 claims description 9
- 238000003860 storage Methods 0.000 claims description 9
- 230000002093 peripheral effect Effects 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 6
- 239000000203 mixture Substances 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 3
- 238000012804 iterative process Methods 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 18
- 238000004422 calculation algorithm Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000013507 mapping Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 230000014759 maintenance of location Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000010267 cellular communication Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 210000000887 face Anatomy 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Collating Specific Patterns (AREA)
Abstract
A kind of eyes localization method and device obtain face location this method comprises: carrying out Face datection to image to be detected, determine that the face to be detected comprising the face location surrounds frame;According to 5 points of corresponding relationships for surrounding frame and face encirclement frame are prestored, determining estimation 5 points of encirclements frame to be detected corresponding with face to be detected encirclement frame surrounds the initial position of five key points in frame including described image to be detected according to described 5 points to be detected;Processing is corrected to the initial position of five key points according to default display shape regression model, obtains the final position comprising five key points;The position coordinates of left pupil, right pupil are extracted from the final position of five key points.The present invention still can be accurately located out eyes position in the case where face artwork is pasted or eyes are blocked, and improve the Stability and veracity of eyes positioning.
Description
Technical field
The present invention relates to technical field of image processing, in particular to a kind of eyes localization method, a kind of eyes positioning device.
Background technique
Eyes positioning is a very important basic technology of Digital Image Processing and area of pattern recognition, many faces
Alignment algorithm, facial feature localization algorithm, eye tracking algorithm etc. all rely on the result of eyes positioning.Existing eyes location technology
Eyes are all directly positioned, in face artwork paste, face there is expression or posture, sunglasses to be easy to appear missing inspection when blocking
And the case where erroneous detection.
The prior art provides another and is based on class Ha Er (Haar) feature and adaptive enhancing (Adaboost) algorithm
Human-eye positioning method.This method is based on class Lis Hartel sign and is trained to a large amount of human eye sample and non-human eye sample, utilizes
Self-adaptive enhancement algorithm extracts some Weak Classifiers of better performances, and the Weak Classifier extracted is cascaded, composition
Final strong classifier, and the human eye in images to be recognized is detected and positioned using the strong classifier.This method phase
For the human-eye positioning method based on Gray Projection method, detection effect is good, locating speed is fast, but still unresolved video camera with
The limitation of the distance between face, when between video camera and face distance farther out when, since the human eye feature that detects is relatively fuzzy, hold
Easily there is the case where missing inspection, erroneous detection.
Summary of the invention
Based on this, for above-mentioned problems of the prior art, the purpose of the present invention is to provide a kind of positioning of eyes
Method and a kind of eyes positioning device, can face artwork paste or eyes be blocked in the case where still can be accurately
Eyes position is oriented, the Stability and veracity of eyes positioning is improved.
In order to achieve the above objectives, the embodiment of the present invention uses following technical scheme:
A kind of eyes localization method, comprising steps of
Face datection is carried out to image to be detected, obtains face location, determines the face to be detected comprising the face location
Surround frame;
According to 5 points of corresponding relationships for surrounding frame and face encirclement frame are prestored, the determining and described face to be detected surrounds frame pair
The estimation answered 5 points of encirclements frame to be detected, five passes surrounded in frame including described image to be detected according to described 5 points to be detected
The initial position of key point, five key points include: left pupil, right pupil, prenasale, left corners of the mouth point, right corners of the mouth point,
It is the encirclement frame in facial image comprising five key points that 5 points, which are surrounded frame, and it includes face position in facial image that face, which surrounds frame to be,
The encirclement frame set;
Processing is corrected to the initial position of five key points according to default display shape regression model, is wrapped
Final position containing five key points;
The position coordinates of left pupil, right pupil are extracted from the final position of five key points.
A kind of eyes positioning device, comprising:
Face detection module obtains face location, determines to include the face for carrying out Face datection to image to be detected
The face to be detected of position surrounds frame;
Initial alignment module, for surrounding the corresponding relationships that frames and face surround frame according to prestoring at 5 points, it is determining with it is described
Face to be detected surrounds the corresponding estimation 5 points of encirclements frame to be detected of frame, according to including in 5 points of encirclements frame to be detected
The initial position of five key points of image to be detected, five key points include: left pupil, right pupil, prenasale,
Left corners of the mouth point, right corners of the mouth point, it is the encirclement frame in facial image comprising five key points that 5 points, which are surrounded frame, and face surrounds frame and behaves
It include the encirclement frame of face location in face image;
Positioning correcting module, for according to default display shape regression model to the initial positions of five key points into
Row correction process obtains the final position comprising five key points;
Eyes position extraction module, for extracting left pupil, right pupil from the final position of five key points
The position coordinates of point.
According to the scheme of embodiment present invention as described above, based on nose, two corners of the mouth positions to the several of eyes position
What the constraint relationship utilizes pupil left in face, right pupil, prenasale, left corners of the mouth point, right corners of the mouth point this five key points pair
Eyes are positioned, and it is unrelated whether this geometrical-restriction relation is blocked with eyes, thus, even if face artwork paste or it is double
Eye still can be accurately located out eyes position in the case where being blocked, and improve the Stability and veracity of eyes positioning.
Detailed description of the invention
Fig. 1 is the flow diagram of eyes localization method embodiment of the invention;
Fig. 2 is the flow diagram that corresponding relationship is determined in a specific example;
Fig. 3 is the flow diagram that display shape regression model is determined in a specific example;
Fig. 4 be a specific example based on showing that shape regression model method of determination positions eyes in Fig. 3 when
Flow diagram;
Fig. 5 is the modular structure schematic diagram of eyes positioning device embodiment of the invention;
Fig. 6 is the structural schematic diagram of the corresponding relationship determining module in an example in Fig. 5;
Fig. 7 is the structural schematic diagram of the model determining module in an example in Fig. 5;
Fig. 8 is the structural schematic diagram of the positioning correcting module in an example in Fig. 5;
Fig. 9 is the modular structure schematic diagram for being able to achieve the computer system of the present invention program.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention more comprehensible, with reference to the accompanying drawings and embodiments, to this
Invention is described in further detail.It should be appreciated that the specific embodiments described herein are only used to explain the present invention,
And the scope of protection of the present invention is not limited.
The flow diagram of eyes localization method embodiment of the invention is shown in Fig. 1, as shown in Figure 1, the present embodiment
In eyes localization method comprising steps of
Step S101: carrying out Face datection to image to be detected, obtain face location, determine comprising the face location to
It detects face and surrounds frame;
Step S102: according to the corresponding relationships for prestoring 5 points of encirclement frames and face encirclement frame, the determining and people to be detected
Face surrounds the corresponding estimation 5 points of encirclements frame to be detected of frame, and it includes described image to be detected in frame that described 5 points to be detected, which are surrounded,
The initial position of five key points, five key points include: left pupil, right pupil, prenasale, left corners of the mouth point, the right side
Corners of the mouth point, it is the encirclement frame in facial image comprising five key points that 5 points, which are surrounded frame, and it is to wrap in facial image that face, which surrounds frame,
Encirclement frame containing face location;
Step S103: place is corrected to the initial position of five key points according to default display shape regression model
Reason obtains the final position comprising five key points;
Step S104: the position coordinates of left pupil, right pupil are extracted from the final position of five key points.
According to the scheme of embodiment present invention as described above, based on nose, two corners of the mouth positions to the several of eyes position
What the constraint relationship utilizes pupil left in face, right pupil, prenasale, left corners of the mouth point, right corners of the mouth point this five key points pair
Eyes are positioned, and it is unrelated whether this geometrical-restriction relation is blocked with eyes, thus, even if face artwork paste or it is double
Eye still can be accurately located out eyes position in the case where being blocked, and improve the Stability and veracity of eyes positioning.
Wherein, it is contemplated that when current human-face detector detects face, the Face datection image of output is generally
Square or rectangle, therefore, the face to be detected in embodiments of the present invention scheme surround frame, encirclement frames, face at 5 points of
Encirclement frame, estimation face encirclement frame to be detected can be square or rectangle, in following exemplary explanations, and combines packet
Peripheral frame be square or rectangle for be illustrated.Certainly, as known to those skilled in the art, the face obtained in detection
In the case that image is other shapes, for example, it is round, eyes can also be positioned based on the scheme of the embodiment of the present invention.
Wherein, corresponding relationship, the above-mentioned default display shape regression model that above-mentioned 5 points of encirclement frames and face surround frame can
Being determined before specifically being positioned to eyes.Below in conjunction in a specific example 5 points of encirclement frames of determination with
Face surrounds the corresponding relationship of frame, shows that the process of shape regression model is illustrated.
As shown in connection with fig. 2, the process for the corresponding relationship that surround frame and face encirclement frame is determined in a specific example at 5 points
It can be discussed further below:
M (the first preset number) face figure is collected, the occurrence of the M can be set according to actual needs, such as
2000;
Five key points in this M face figures are marked, determine five comprising five key points in the face figure respectively
Point surrounds frame, that is, is directed to each face figure, all obtains corresponding 5 points of encirclements frame, wherein marks in this M face figures
The processes of five key points can be carried out by way of marking by hand;
The face location in this M face figures is detected, determines that the face of the face location comprising the face figure surrounds respectively
Frame is directed to each face figure, all obtain a corresponding face and surround frame;
According to the 5 points of encirclement frames and face encirclement frame of M face figures, counts and determine that 5 points of encirclement frames and face surround frame
Corresponding relationship.
Wherein, in a specific example, the corresponding relationship of above-mentioned 5 points of encirclement frames and face encirclement frame can be at 5 points
The width ratio between frame and face encirclement frame is surrounded, such as 5 points of abscissas for surrounding frames surround the horizontal seat of frame relative to face
Target deviation ratio, the ordinate of 5 points of encirclement frames surround the deviation ratio of the ordinate of frame, 5 points counted relative to face
It surrounds frame and face surrounds the corresponding relationship of frame, can be the average value of the deviation ratio of abscissa and the deviation ratio of ordinate,
It can be the deviation ratio of the final abscissa obtained by other means and the deviation ratio of ordinate, in the present invention program
It is not specifically limited in this embodiment.Certainly, based on actual needs, 5 points of encirclement frames and face packet can also be determined using other modes
The corresponding relationship of peripheral frame, as long as the coordinate for the shape that the shape and face that can surround frame for 5 points surround frame is interrelated i.e.
Can, it is not specifically limited in this embodiment in the present invention program.
Surrounded after frame and face surround the corresponding relationship of frame obtaining at above-mentioned 5 points, it can based on this 5 points encirclements frames with
Face surrounds the corresponding relationship of frame, opens 5 points of encirclement frames of face figure in conjunction with above-mentioned M and face surrounds frame and determines above-mentioned display shape
Shape regression model.
The schematic diagram that display shape regression model is determined in a specific example is shown in Fig. 3.As shown in figure 3, at this
In example, determine that the process of display shape regression model can be discussed further below.
Step S301: 5 points of encirclement frames of above-mentioned M face figures are zoomed to center point alignment after same size, according to
Each 5 points after alignment are surrounded the average coordinates position that frame determines five key points, and determine the mean place of this five key points
Corresponding average 5 points of encirclement frames;
Step S302: frame is surrounded according to above-mentioned 5 points and face surrounds the corresponding relationship of frame, determines M face figures respectively
It estimates 5 points of encirclement frames, that is, is directed to each face figure, all obtains 5 points of encirclement frames of a corresponding estimation;
Step S303: the average coordinates position of five key points is zoomed into the size with 5 points of encirclement frames of each estimation respectively
Unanimously, center is aligned, and obtains five key point positions of 5 points of encirclement frames of M estimation respectively;
Step S304: N (the second preset number) a sampled point is randomly selected in the range of average 5 points of encirclement frames, and will
Each sampled point and average 5 points of encirclement frames are interior, corresponding with the nearest key point of the sampled point;
Step S305: in the range of determining current 5 points of encirclement frames of an estimation, opposite with above-mentioned N number of sampling point position
The N number of sampled point answered, and for each sampled point in the range of 5 points of encirclement frames of estimation, it will the sampled point and sampled point place
It is interior, corresponding with the nearest key point of the sampled point to estimate 5 points of encirclement frames, and each sampled point is matched to composition point pair two-by-two, by this
Point constitutes a feature to plus a random number, by taking N value is 500 as an example, then obtains within the scope of 5 points of encirclement frames of the estimation
After 500 sampled points, this 500 sampled points are matched two-by-two, obtains amounting to 124750 points pair;
Step S306: the image pixel intensities for calculating separately the point pair in each feature are poor, calculate in 5 points of encirclement frames of each estimation
The coordinate displacement of five key point positions and the average coordinates position is poor, calculates the image pixel intensities difference of each feature and corresponding
Coordinate displacement difference between the degree of correlation;
Step S307: judging whether face figure is processed finishes to M, if it is not, then return step S305, returns pair
5 points of encirclement frames of estimation of next face figure, in the range of determining 5 points of encirclement frames of the estimation, with average 5 points of encirclement frame models
The interior corresponding N number of sampled point of N number of sampling point position is enclosed, if so, 5 points of encirclement frames of estimation of each face figure are directed to,
The degree of correlation for all having calculated its feature, then enter step S308;
Step S308: from the M corresponding features of face figure, the highest K of the degree of correlation (third preset number) a spy is chosen
Sign write-in display shape regression model;
Step S309: the image pixel intensities for calculating separately the point pair in L (the 4th preset number) a feature of each face figure are poor,
It determines the size relation between the random number in image pixel intensities difference and this feature, and the size relation is mapped to corresponding spy
Value indicative, wherein this feature value specifically can be the difference between the random number in image pixel intensities difference and this feature;Carry out
When mapping, mapping relations can be preset with, a size relation is corresponding with a characteristic value in the mapping relations,
To after size relation, corresponding characteristic value is mapped directly to;
Step S3010: according to each characteristic value of M face figures, classify to this M face figures, and calculate all kinds of
In face figure corresponding 5 points of five key point coordinates surrounded in frames of estimation and above-mentioned M face figure 5 points of encirclement frames
The coordinate displacement of the average coordinates position of five key points is poor, and calculates the average value of all kinds of coordinate displacement differences, obtains all kinds of
Average coordinates displacement difference, and by all kinds of average coordinates displacement difference write-in display shape regression models;Wherein, to this M
When face figure is classified, it can be classified according to features described above value, i.e. the corresponding classification of a characteristic value;
Step S3011: for each face figure, it will estimate that surround frame plus the average seat of class where the face figure at 5 points
Marker displacement is poor, obtains five key point positions of new 5 points of encirclement frames of each estimation;
Step S3012: after current iteration number is added 1, judge whether to reach default the number of iterations, if it is not, return step
S304 randomly selects N number of sampled point in the range of average 5 points of encirclement frames again, no to be, then terminates iterative process, obtain most
Whole display shape regression model.
It wherein, can be using existing at present and be likely to occur later when calculating the degree of correlation in above-mentioned steps S309
Any mode carries out, calculate for example, by using covariance matrix etc..
It should be noted that above-mentioned M (the first preset number), N (the second preset number), K (third preset number), L (
Four preset numbers) specific value can be set according to actual needs, the size of specific M, N, K, L can be according to specific
Treatment effeciency, accuracy in treatment process etc. are balanced.As described above, when the value of M is 2000, it is specific at one
In example, it is 500 that the value of N, which can choose,.In a specific implementation, the value of above-mentioned K, L can be identical, and for example, 5.This
When, since the permutation and combination of 5 features shares possibility in 32, the range of above-mentioned characteristic value corresponding with size relation be can be
[0,31].
Frame is surrounded based on 5 points shown in Fig. 2, Fig. 3 and face surrounds the corresponding relationship of frame, default display shape returns
The method of determination of model shows the flow diagram positioned in a specific example to eyes in Fig. 4.
As shown in figure 4, the eyes position fixing process in the example includes:
Step S401: carrying out Face datection to image to be detected, obtain face location, determine comprising the face location to
It detects face and surrounds frame;
Step S402: according to the corresponding relationships for prestoring 5 points of encirclement frames and face encirclement frame, the determining and people to be detected
Face surrounds the corresponding estimation 5 points of encirclements frame to be detected of frame;
Step S403: the average coordinates position of five key points of above-mentioned M face figures is zoomed to be checked with estimation
It surveys that 5 points of encirclement frames are in the same size, center alignment, determines 5 initial positions of image to be detected;
Step S404: five point encirclement frame to be detected to above-mentioned estimation is normalized, by above-mentioned estimation to be detected five
Point surrounds frame and is transformed into average 5 points of encirclement frame coordinate systems, 5 points of encirclement frames after being converted;
Step S405: the corresponding feature of current iteration number is read from above-mentioned display shape regression model, and calculates and turns
Change in rear 5 points of encirclements frame, the image pixel intensities between two pixels corresponding to position with the point in this feature of position it is poor;
Step S406: the image pixel intensities difference is compared with the random number in corresponding this feature, is obtained between the two
Size relation, and the size relation is mapped to corresponding characteristic value;
Step S407: it is read from the display shape regression model according to this feature value corresponding average with this feature value
Coordinate displacement is poor, and is updated according to the average coordinates displacement difference to 5 points of encirclement frames after conversion, obtains at updated 5 points
Surround frame;
Step S408: after current iteration number is added 1, judge whether to reach the default the number of iterations;If it is not, then returning
Step S405 reads the new corresponding feature of current iteration number from the display shape regression model, carries out subsequent processing,
If so, entering step S409;
Step S409: five key points surrounded in frame to 5 points after updated conversion carry out inverse conversion, obtain described
The final position of five key points.
Step S4010: reading the coordinate of left pupil, right pupil from the final position of five key points, completes double
Eye positioning.
Based on thought identical with the eyes localization method of aforementioned present invention, it is fixed that the embodiment of the present invention also provides a kind of eyes
Position device.The structural schematic diagram of eyes positioning device embodiment of the invention is shown in Fig. 5.
As shown in figure 5, the eyes positioning device in the present embodiment includes:
Face detection module 501 is used to carry out Face datection to image to be detected, obtains face location, determines comprising being somebody's turn to do
The face to be detected of face location surrounds frame;
Initial alignment module 502, for according to the corresponding relationships for prestoring 5 points of encirclement frames and face encirclement frame, determining and institute
State face to be detected and surround corresponding estimation 5 points of encirclement frame to be detected of frame, in 5 points of encirclement frame to be detected including described in
The initial position of five key points of detection image, five key points include: left pupil, right pupil, prenasale, a left side
Corners of the mouth point, right corners of the mouth point, it is the encirclement frame in facial image comprising five key points that 5 points, which are surrounded frame, and it is face that face, which surrounds frame,
It include the encirclement frame of face location in image;
Positioning correcting module 503, for the initial bit according to default display shape regression model to five key points
It sets and is corrected processing, obtain the final position comprising five key points;
Eyes position extraction module 504, for extracting left pupil, right pupil from the final position of five key points
The position coordinates of hole point.
According to the scheme of embodiment present invention as described above, based on nose, two corners of the mouth positions to the several of eyes position
What the constraint relationship utilizes pupil left in face, right pupil, prenasale, left corners of the mouth point, right corners of the mouth point this five key points pair
Eyes are positioned, and it is unrelated whether this geometrical-restriction relation is blocked with eyes, thus, even if face artwork paste or it is double
Eye still can be accurately located out eyes position in the case where being blocked, and improve the Stability and veracity of eyes positioning.
Wherein, it is contemplated that when current human-face detector detects face, the Face datection image of output is generally
Square or rectangle, therefore, the face to be detected in embodiments of the present invention scheme surround frame, encirclement frames, face at 5 points of
Encirclement frame, estimation face encirclement frame to be detected can be square or rectangle, in following exemplary explanations, and combines packet
Peripheral frame be square or rectangle for be illustrated.Certainly, as known to those skilled in the art, the face obtained in detection
In the case that image is other shapes, for example, it is round, eyes can also be positioned based on the scheme of the embodiment of the present invention.
Wherein, corresponding relationship, the above-mentioned default display shape regression model that above-mentioned 5 points of encirclement frames and face surround frame can
Being determined before specifically being positioned to eyes.
Accordingly, shown in Fig. 5, in a specific example of the invention, the eyes positioning device in the present embodiment further includes
There is corresponding relationship determining module 510.
Wherein, which is used to determine that surround the corresponding pass that frame surrounds frame with face at described 5 points
System.
The structural schematic diagram of the corresponding relationship determining module 510 in an example is shown in Fig. 6.As shown in fig. 6, right
Relationship determination module 510 is answered to specifically include:
5 points of encirclement frame determining modules 5101, five in the first preset number face figure for marking collection respectively
Key point determines 5 points of encirclement frames comprising five key points in the face figure respectively;
Face surrounds frame determining module 5102, for detecting the face position in first preset number face figure respectively
It sets, determines that the face of the face location comprising the face figure surrounds frame respectively;
Determining module 5103 is counted, for the 5 points of encirclement frames and face packet according to first preset number face figure
Peripheral frame, statistics determine the described 5 points corresponding relationships for surrounding frame and face encirclement frame.
As shown in figure 5, in another specific example, which further includes having model determining module 520, use
In 5 points of encirclements of the corresponding relationship, first preset number face figure for surrounding frame and face encirclement frame according to above-mentioned 5 points
Frame and face surround frame and determine the display shape regression model.
The structural schematic diagram of the model determining module 520 in an example is shown in Fig. 7.As shown in fig. 7, model is true
Cover half block 520 specifically includes:
It is homogenized processing module 5201, for 5 points of encirclement frames of the first preset number face figure to be zoomed to same size
Afterwards by center point alignment, frame is surrounded according to each 5 points after alignment and determines the average coordinates position of five key points, at averagely 5 points
Surround frame;
It estimates processing module 5202, for surrounding the corresponding relationship that frames and face surround frame according to 5 points, determines the respectively
5 points of encirclements frames of estimation of one preset number face figure, and the average coordinates position of five key points zoomed to respectively and respectively
In the same size, the center alignment for estimating 5 points of encirclement frames, obtain five key point positions of 5 points of encirclement frames of each estimation;
Sampling module 5203, for randomly selecting the sampling of the second preset number in the range of average 5 points of encirclement frames
Point, and by each sampled point in average 5 points of encirclement frames, it is corresponding with the nearest key point of the sampled point;
Feature generation module 5204, in the range of determining 5 points of encirclement frames of each estimation respectively, with it is described second default
The corresponding second preset number sampled point of number sampling point position, and it is directed to each sampled point, by the sampled point and the sampling
It is interior, corresponding with the nearest key point of the sampled point that 5 points of encirclement frames are estimated where point, and each sampled point is matched two-by-two and constitutes point
It is right, which is constituted into a feature to plus a random number;
Feature Selection module 5205, the image pixel intensities for calculating separately the point pair in each feature are poor, calculate each estimation five
The coordinate displacement that point surrounds the five key point positions and the average coordinates position in frame is poor, calculates the image pixel intensities of each feature
The degree of correlation between poor and corresponding coordinate displacement difference;From the first preset number corresponding feature of face figure, choose
The highest third preset number feature write-in display shape regression model of the degree of correlation;
The First Eigenvalue determining module 5206, the point in the 4th preset number feature for calculating separately each face figure
Pair image pixel intensities it is poor, determine the size relation between the random number in image pixel intensities difference and this feature, and the size closed
System is mapped to corresponding characteristic value;
Categorization module 5207, for the characteristic value according to the first preset number face figure, to first preset number
Face figure is classified, and calculates the average value of the coordinate displacement difference of face figure in all kinds of, obtains all kinds of average coordinates
Displacement difference, and all kinds of average coordinates displacement difference write-ins is shown into shape regression model;
First update module 5208, for for each face figure, 5 points of encirclement frames of estimation to be added the face figure institute
In the average coordinates displacement difference of class, five key point positions of new 5 points of encirclement frames of each estimation are obtained;
First iterative processing module 5209 judges whether to reach default iteration time after current iteration number is added 1
Number, and in not up to default the number of iterations, Xiang Suoshu sampling module come back for the information sampled next time, and by each face
Five key point positions of new 5 points of encirclement frames of each estimation of figure return to the sampling module and are handled.
5 points of pairs for surrounding frame and face encirclement frame are determined based on Fig. 6, corresponding relationship determining module 510 shown in fig. 7
Mode, the model determining module 520 that should be related to determine the method for determination of default display shape regression model, show one in Fig. 8
The structural schematic diagram of positioning correcting module 503 in a specific example.
As shown in figure 8, the positioning correcting module 503 in the specific example includes:
Alignment module 5031, the average coordinates position of five key points for first preset number to be opened to face figure
Zoom to and surround that frame is in the same size, center is aligned with face to be detected is estimated, determines 5 initial positions of image to be detected;
Normalized module 5032, for estimation 5 points of encirclements frame to be detected to be normalized, by institute
It states estimation 5 points of encirclements frame to be detected and is transformed into average 5 points of encirclement frame coordinate systems, 5 points of encirclement frames after being converted;
Second Eigenvalue determining module 5033, for reading current iteration number pair from the display shape regression model
The feature answered, and calculate after conversion in 5 points of encirclement frames, position two pixels corresponding to position with the point in this feature it
Between image pixel intensities it is poor, determine image pixel intensities difference and the size relation between the random number in corresponding feature, and this is big
Small relationship map is to corresponding characteristic value;
Second update module 5034, the characteristic value for being determined according to the characteristic value calculating module is from the display shape
Average coordinates displacement difference corresponding with this feature value is read in regression model, and according to the average coordinates displacement difference to after conversion five
Point surrounds frame and is updated;
Secondary iteration processing module 5035 judges whether to reach the default iteration after current iteration number is added 1
Number, and in not up to default the number of iterations, it is corresponding that current iteration number is read from the display shape regression model
After feature, this feature is returned into the alignment module and is handled;
Inverse transform block 5036, for the iterative processing module judgement reach default the number of iterations when, after update
Conversion after 5 points surround frames in five key points carry out inverse conversions, obtain the final position of five key points.
The eyes localization method and eyes positioning device of aforementioned present invention, can apply can be positioned various based on human eye
As a result the field handled, such as eyes are amplified automatically after obtaining eyes positioning result, it is positioned and is tied according to eyes
Fruit detects whether user matches glasses, and finally determines whether to need to add eyeglasses assembly etc. when synthesizing final image.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a computer-readable storage medium
In, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the storage medium can be magnetic
Dish, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random Access
Memory, RAM) etc..
Fig. 9 is the module map for being able to achieve a computer system 1000 of the embodiment of the present invention.The computer system 1000
An only example for being suitable for the invention computer environment is not construed as proposing appointing to use scope of the invention
What is limited.Computer system 1000 can not be construed to need to rely on or the illustrative computer system 1000 with diagram
In one or more components combination.The computer system 1000 is in specific implementation, it can be understood as the embodiment of the present invention
In server.
Computer system 1000 shown in Fig. 9 is the example for being suitable for computer system of the invention.Have
Other frameworks of different sub-systems configuration also can be used.Such as there are the similar devices such as big well known desktop computer, notebook
It can be adapted for some embodiments of the present invention.But it is not limited to equipment enumerated above.
As shown in figure 9, computer system 1000 includes processor 1010, memory 1020 and system bus 1022.Including
Various system components including memory 1020 and processor 1010 are connected on system bus 1022.Processor 1010 is one
For executing the hardware of computer program instructions by arithmetic sum logical operation basic in computer system.Memory 1020
It is one for temporarily or permanently storing the physical equipment of calculation procedure or data (for example, program state information).System is total
Line 1020 can be any one in the bus structures of following several types, including memory bus or storage control, outer
If bus and local bus.Processor 1010 and memory 1020 can carry out data communication by system bus 1022.Wherein
Memory 1020 includes read-only memory (ROM) or flash memory (being all not shown in figure) and random access memory (RAM), RAM
Typically refer to the main memory for being loaded with operating system and application program.
Computer system 1000 further includes display interface 1030 (for example, graphics processing unit), display 1040 (example of equipment
Such as, liquid crystal display), audio interface 1050 (for example, sound card) and audio frequency apparatus 1060 (for example, loudspeaker).Show equipment
1040 and audio frequency apparatus 1060 be media device for experiencing multimedia content.
Computer system 1000 generally comprises a storage equipment 1070.Storing equipment 1070 can from a variety of computers
It reads to select in medium, computer-readable medium refers to any available medium that can be accessed by computer system 1000,
Including mobile and fixed two media.For example, computer-readable medium includes but is not limited to, flash memory (miniature SD
Card), CD-ROM, digital versatile disc (DVD) or other optical disc storages, cassette, tape, disk storage or other magnetic storages are set
Any other medium that is standby, or can be used for storing information needed and can be accessed by computer system 1000.
Computer system 1000 further includes input unit 1080 and input interface 1090 (for example, I/O controller).User can
With by input unit 1080, such as the touch panel equipment in keyboard, mouse, display device 1040, input instruction and information are arrived
In computer system 1000.Input unit 1080 is usually connected on system bus 1022 by input interface 1090, but
It can also be connected by other interfaces or bus structures, such as universal serial bus (USB).
Computer system 1000 can carry out logical connection with one or more network equipment in a network environment.Network is set
It is standby to can be PC, server, router, smart phone, tablet computer or other common network nodes.Department of computer science
System 1000 is connected by local area network (LAN) interface 1100 or mobile comm unit 1110 with the network equipment.Local area network (LAN)
Refer in finite region, such as family, school, computer laboratory or the office building using the network media, interconnection composition
Computer network.WiFi and twisted pair wiring Ethernet are two kinds of technologies of most common building local area network.WiFi is a kind of
It can make 1000 swapping data of computer system or be connected to the technology of wireless network by radio wave.Mobile comm unit
1110 are answered and are made a phone call by radio communication diagram while capable of moving in a wide geographic area.In addition to logical
Other than words, mobile comm unit 1110 is also supported to carry out in 2G, 3G or the 4G cellular communication system for providing mobile data service
Internet access.
It should be pointed out that other includes than the computer system of the more or fewer subsystems of computer system 1000
It can be suitably used for inventing.
As detailed above, the specified of eyes localization method can be executed by being suitable for the invention computer system 1000
Operation.Computer system 1000 is executed by way of the software instruction that processor 1010 is run in computer-readable medium
These operations.These software instructions can be read into from storage equipment 1070 or by lan interfaces 1100 from another equipment
In memory 1020.The software instruction being stored in memory 1020 makes processor 1010 execute above-mentioned eyes positioning side
Method.In addition, also can equally realize the present invention by hardware circuit or hardware circuit combination software instruction.Therefore, this hair is realized
The bright combination for being not limited to any specific hardware circuit and software.
The embodiments described above only express several embodiments of the present invention, and the description thereof is more specific and detailed, but simultaneously
Limitations on the scope of the patent of the present invention therefore cannot be interpreted as.It should be pointed out that for those of ordinary skill in the art
For, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to guarantor of the invention
Protect range.Therefore, the scope of protection of the patent of the invention shall be subject to the appended claims.
Claims (10)
1. a kind of eyes localization method, which is characterized in that comprising steps of
Face datection is carried out to image to be detected, obtains face location, determines that the face to be detected comprising the face location surrounds
Frame;
According to 5 points of corresponding relationships for surrounding frame and face encirclement frame are prestored, determination is corresponding with the face encirclement frame to be detected
Estimate 5 points of encirclements frame to be detected, five key points in frame including described image to be detected are surrounded in 5 points to be detected of the estimation
Initial position, five key points include: left pupil, right pupil, prenasale, left corners of the mouth point, right corners of the mouth point, 5 points
Surrounding frame is the encirclement frame in facial image comprising five key points, and it is in facial image comprising face location that face, which surrounds frame,
Surround frame;
Processing is corrected to the initial position of five key points according to default display shape regression model, obtaining includes five
The final position of a key point;
The position coordinates of left pupil, right pupil are extracted from the final position of five key points;
Wherein, the display shape regression model surrounds frame according to described 5 points and face surrounds the corresponding relationship of frame, first in advance
If the 5 points of encirclement frames and face of number face figure surround frame determination and obtain, the display shape regression model was determined
Journey specifically includes:
5 points of encirclement frames of the first preset number face figure are zoomed to after same size by center point alignment, after alignment
Each 5 points surround frame and determine the average coordinates position of five key points, average 5 points of encirclement frames;
Frame is surrounded according to 5 points and face surrounds the corresponding relationship of frame, determines the estimation five of the first preset number face figure respectively
Point surrounds frame;
The average coordinates position of five key points is zoomed into the same size, the center pair with 5 points of encirclement frames of each estimation respectively
Together, five key point positions of 5 points of encirclement frames of each estimation are obtained;
The second preset number sampled point is randomly selected in the range of average 5 points of encirclement frames, and each sampled point is averaged with this
5 points of encirclement frames are interior, corresponding with the nearest key point of the sampled point;
In the range of determining 5 points of encirclement frames of each estimation respectively, with the second preset number sampling point position corresponding second
Preset number sampled point, and for each sampled point, which is estimated in 5 points of encirclement frames and be somebody's turn to do with sampled point place
The nearest key point of sampled point is corresponding, and each sampled point is matched correspondence two-by-two, and each sampled point is matched to composition point pair two-by-two,
The point is constituted into a feature to plus a random number;
The image pixel intensities for calculating separately the point pair in each feature are poor, calculate 5 points of five key point positions surrounded in frame of each estimation
It is with the coordinate displacement of the average coordinates position poor, calculate the image pixel intensities difference and corresponding coordinate displacement difference of each feature
Between the degree of correlation;
From the first preset number corresponding feature of face figure, the highest third preset number feature write-in of the degree of correlation is chosen
Show shape regression model;
The image pixel intensities for calculating separately the point pair in the 4th preset number feature of each face figure are poor, determine that the image pixel intensities are poor
With the size relation between the random number in this feature, and the size relation is mapped to corresponding characteristic value;
According to the characteristic value of the first preset number face figure, classify to first preset number face figure, and counts
The average value for calculating the coordinate displacement difference of the face figure in all kinds of obtains all kinds of average coordinates displacement differences, and all kinds of is averaged
Coordinate displacement difference write-in display shape regression model;
For each face figure, it will estimate that surround frame plus the average coordinates displacement difference of class where the face figure, obtains at 5 points
Five key point positions of new 5 points of encirclement frames of each estimation;
After current iteration number is added 1, judge whether to reach default the number of iterations;
If it is not, returning to the step of randomly selecting the second preset number sampled point in the range of average 5 points of encirclement frames;
If so, terminating iterative process.
2. eyes localization method according to claim 1, which is characterized in that the face to be detected surrounds frame, five point packets
Peripheral frame, face encirclement frame, estimation 5 points of encirclements frame to be detected is square or rectangle.
3. eyes localization method according to claim 1 or 2, which is characterized in that 5 points of encirclement frames and face surround
The corresponding relationship of frame is determined by following manner:
Collect the first preset number face figure;
Five key points in each face figure are marked, determine the five point packets comprising five key points in the face figure respectively
Peripheral frame;
The face location in each face figure is detected, determines that the face of the face location comprising the face figure surrounds frame respectively;
Frame is surrounded according to 5 points of encirclement frames of first preset number face figure and face, statistics determines 5 points of encirclements
Frame and face surround the corresponding relationship of frame.
4. eyes localization method according to claim 1, which is characterized in that according to default display shape regression model to institute
The initial position for stating five key points is corrected processing, and the process for obtaining the final position comprising five key points includes:
The average coordinates position of five key points of first preset number face figure is zoomed to and estimates people to be detected
Face encirclement frame is in the same size, center is aligned;
Estimation 5 points of encirclements frame to be detected is normalized, estimation 5 points of encirclements frame to be detected is transformed into
Average 5 points of encirclement frame coordinate systems, 5 points of encirclement frames after being converted;
The corresponding feature of current iteration number is read from the display shape regression model, and calculates 5 points of encirclement frames after conversion
In, the image pixel intensities between two pixels corresponding to position with the point in this feature of position it is poor;
It determines image pixel intensities difference and the size relation between the random number in corresponding feature, and the size relation is mapped to
Corresponding characteristic value;
Average coordinates displacement difference corresponding with this feature value is read from the display shape regression model according to this feature value, and
5 points of encirclement frames after conversion are updated according to the average coordinates displacement difference;
After current iteration number is added 1, judge whether to reach the default the number of iterations;
If it is not, the step of returning to the reading current iteration number corresponding feature from the display shape regression model;
If so, five key points surrounded in frame to 5 points after updated conversion carry out inverse conversion, five keys are obtained
The final position of point.
5. a kind of eyes positioning device characterized by comprising
Face detection module obtains face location, determines to include the face location for carrying out Face datection to image to be detected
Face to be detected surround frame;
Initial alignment module, for surrounding the corresponding relationships that frames and face surround frame according to prestoring at 5 points, it is determining with it is described to be checked
It surveys face and surrounds the corresponding estimation 5 points of encirclements frame to be detected of frame, include described to be checked in estimation 5 points of encirclements frame to be detected
The initial position of five key points of altimetric image, five key points include: left pupil, right pupil, prenasale, Zuo Zui
Angle point, right corners of the mouth point, it is the encirclement frame in facial image comprising five key points that 5 points, which are surrounded frame, and it is face figure that face, which surrounds frame,
Include the encirclement frame of face location as in;
Positioning correcting module, for carrying out school to the initial position of five key points according to default display shape regression model
Positive processing, obtains the final position comprising five key points;
Eyes position extraction module, for extracting from the final position of five key points left pupil, right pupil
Position coordinates;
Model determining module, for surrounding corresponding relationship, the first preset number of frame and face encirclement frame according to described 5 points
The 5 points of encirclement frames and face of face figure surround frame and determine the display shape regression model;
The model determining module includes:
It is homogenized processing module, for zooming to center 5 points of encirclement frames of the first preset number face figure after same size
Point alignment surrounds frame according to each 5 points after alignment and determines the average coordinates position of five key points, average 5 points of encirclement frames;
It estimates processing module, for surrounding the corresponding relationship of frame and face encirclement frame according to 5 points, determines the first present count respectively
5 points of encirclement frames of estimation of mesh face figure, and the average coordinates position of five key points is zoomed to and each 5 points of estimation respectively
The in the same size of frame, center alignment are surrounded, five key point positions of 5 points of encirclement frames of each estimation are obtained;
Sampling module, for randomly selecting the second preset number sampled point in the range of average 5 points of encirclement frames, and will be each
Sampled point and average 5 points of encirclement frames are interior, corresponding with the nearest key point of the sampled point;
Feature generation module, for being adopted in the range of determining 5 points of encirclement frames of each estimation respectively, with second preset number
The corresponding second preset number sampled point of sampling point position, and it is directed to each sampled point, it will estimate where the sampled point and the sampled point
It is interior, corresponding with the nearest key point of the sampled point to calculate 5 points of encirclement frames, and each sampled point is matched to composition point pair two-by-two, by the point
A feature is constituted to plus a random number;
Feature Selection module, the image pixel intensities for calculating separately the point pair in each feature are poor, calculate 5 points of encirclement frames of each estimation
The coordinate displacement of interior five key point positions and the average coordinates position is poor, calculate each feature image pixel intensities difference and and its
The degree of correlation between corresponding coordinate displacement difference;From the first preset number corresponding feature of face figure, the degree of correlation is chosen most
High third preset number feature write-in display shape regression model;
The First Eigenvalue determining module, the pixel of the point pair in the 4th preset number feature for calculating separately each face figure
Intensity difference determines the size relation between the random number in image pixel intensities difference and this feature, and the size relation is mapped to
Corresponding characteristic value;
Categorization module, for the characteristic value according to the first preset number face figure, to first preset number face figure
Classify, and calculate the average value of the coordinate displacement difference of face figure in all kinds of, obtains all kinds of average coordinates displacement differences, and
By all kinds of average coordinates displacement difference write-in display shape regression models;
First update module, for will estimate that surround frame plus the flat of class where the face figure at 5 points for each face figure
Equal coordinate displacement is poor, obtains five key point positions of new 5 points of encirclement frames of each estimation;
First iterative processing module judges whether to reach default the number of iterations, and not after current iteration number is added 1
When reaching default the number of iterations, Xiang Suoshu sampling module comes back for the information sampled next time, and will be described in each face figure
Five key point positions of new 5 points of encirclement frames of each estimation return to the sampling module and are handled.
6. eyes positioning device according to claim 5, which is characterized in that the face to be detected surrounds frame, five point packets
Peripheral frame, face encirclement frame, estimation 5 points of encirclements frame to be detected is square or rectangle.
7. eyes positioning device according to claim 5 or 6, which is characterized in that further include for determining the 5 points packet
Peripheral frame and face surround the corresponding relationship determining module of the corresponding relationship of frame, and the corresponding relationship determining module includes:
5 points of encirclement frame determining modules, five key points in the first preset number face figure for marking collection respectively,
5 points of encirclement frames comprising five key points in the face figure are determined respectively;
Face surrounds frame determining module, for detecting the face location in first preset number face figure respectively, respectively
Determine that the face of the face location comprising the face figure surrounds frame;
Determining module is counted, for surrounding frame, system according to the 5 points of encirclement frames and face of first preset number face figure
Meter determines the described 5 points corresponding relationships for surrounding frame and face encirclement frame.
8. eyes positioning device according to claim 5, which is characterized in that the positioning correcting module includes:
Alignment module, for by the average coordinates position of five key points of first preset number face figure zoom to
Estimate that face to be detected surrounds that frame is in the same size, center alignment, determines 5 initial positions of image to be detected;
Normalized module, for the estimation 5 points of encirclement frame to be detected to be normalized, by it is described estimate to
It detects 5 points of encirclement frames and is transformed into average 5 points of encirclement frame coordinate systems, 5 points of encirclement frames after being converted;
Second Eigenvalue determining module, for reading the corresponding spy of current iteration number from the display shape regression model
Sign, and calculate after conversion in 5 points of encirclement frames, the picture between two pixels corresponding to position with the point in this feature of position
Plain intensity difference determines image pixel intensities difference and the size relation between the random number in corresponding feature, and by the size relation
It is mapped to corresponding characteristic value;
Second update module, the characteristic value for being determined according to the characteristic value calculating module is from the display shape regression model
It is middle to read average coordinates displacement difference corresponding with this feature value, and according to the average coordinates displacement difference to 5 points of encirclement frames after conversion
It is updated;
Secondary iteration processing module judges whether to reach the default the number of iterations after current iteration number is added 1, and
In not up to default the number of iterations, after reading the corresponding feature of current iteration number in the display shape regression model,
This feature is returned to the alignment module to handle;
Inverse transform block, for the iterative processing module judgement reach default the number of iterations when, after updated conversion
5 points of five key points surrounded in frame carry out inverse conversion, obtain the final position of five key points.
9. a kind of computer equipment, including memory and processor, the memory are stored with computer program, feature exists
In the step of processor realizes any one of claims 1 to 4 the method when executing the computer program.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The step of method described in any one of Claims 1-4 is realized when being executed by processor.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410374902.1A CN105320921B (en) | 2014-07-31 | 2014-07-31 | Eyes localization method and eyes positioning device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410374902.1A CN105320921B (en) | 2014-07-31 | 2014-07-31 | Eyes localization method and eyes positioning device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105320921A CN105320921A (en) | 2016-02-10 |
CN105320921B true CN105320921B (en) | 2019-07-19 |
Family
ID=55248283
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410374902.1A Active CN105320921B (en) | 2014-07-31 | 2014-07-31 | Eyes localization method and eyes positioning device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105320921B (en) |
Families Citing this family (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107169397B (en) * | 2016-03-07 | 2022-03-01 | 佳能株式会社 | Feature point detection method and device, image processing system and monitoring system |
CN107463865B (en) * | 2016-06-02 | 2020-11-13 | 北京陌上花科技有限公司 | Face detection model training method, face detection method and device |
CN106295567B (en) * | 2016-08-10 | 2019-04-12 | 腾讯科技(深圳)有限公司 | A kind of localization method and terminal of key point |
CN107622252B (en) * | 2017-09-29 | 2022-02-22 | 百度在线网络技术(北京)有限公司 | Information generation method and device |
CN109063679A (en) * | 2018-08-24 | 2018-12-21 | 广州多益网络股份有限公司 | A kind of human face expression detection method, device, equipment, system and medium |
CN109657652A (en) * | 2019-01-16 | 2019-04-19 | 平安科技(深圳)有限公司 | A kind of face identification method and device |
CN109934112B (en) * | 2019-02-14 | 2021-07-13 | 青岛小鸟看看科技有限公司 | Face alignment method and camera |
CN111079164B (en) * | 2019-12-18 | 2021-09-07 | 深圳前海微众银行股份有限公司 | Feature correlation calculation method, device, equipment and computer-readable storage medium |
CN111667518B (en) * | 2020-06-24 | 2023-10-31 | 北京百度网讯科技有限公司 | Face image display method and device, electronic equipment and storage medium |
CN112308011B (en) * | 2020-11-12 | 2024-03-19 | 湖北九感科技有限公司 | Multi-feature combined target detection method and device |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101276408A (en) * | 2008-04-24 | 2008-10-01 | 长春供电公司 | Method for recognizing human face based on electrical power system network safety |
CN102622613A (en) * | 2011-12-16 | 2012-08-01 | 彭强 | Hair style design method based on eyes location and face recognition |
CN103605965A (en) * | 2013-11-25 | 2014-02-26 | 苏州大学 | Multi-pose face recognition method and device |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070047761A1 (en) * | 2005-06-10 | 2007-03-01 | Wasilunas Elizabeth A | Methods Of Analyzing Human Facial Symmetry And Balance To Provide Beauty Advice |
US8238604B2 (en) * | 2008-08-18 | 2012-08-07 | Kabushiki Kaisha Toshiba | System and method for validation of face detection in electronic images |
-
2014
- 2014-07-31 CN CN201410374902.1A patent/CN105320921B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101276408A (en) * | 2008-04-24 | 2008-10-01 | 长春供电公司 | Method for recognizing human face based on electrical power system network safety |
CN102622613A (en) * | 2011-12-16 | 2012-08-01 | 彭强 | Hair style design method based on eyes location and face recognition |
CN103605965A (en) * | 2013-11-25 | 2014-02-26 | 苏州大学 | Multi-pose face recognition method and device |
Also Published As
Publication number | Publication date |
---|---|
CN105320921A (en) | 2016-02-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105320921B (en) | Eyes localization method and eyes positioning device | |
US10373244B2 (en) | System and method for virtual clothes fitting based on video augmented reality in mobile phone | |
CN107590482A (en) | information generating method and device | |
CN108985199A (en) | Detection method, device and the storage medium of commodity loading or unloading operation | |
CN111832447B (en) | Building drawing component identification method, electronic equipment and related product | |
Cuevas et al. | Improved background modeling for real-time spatio-temporal non-parametric moving object detection strategies | |
CN109345553A (en) | A kind of palm and its critical point detection method, apparatus and terminal device | |
CN106154301B (en) | A kind of implementation method and device of more net positioning | |
CN110287810B (en) | Vehicle door motion detection method, device and computer readable storage medium | |
CN109409241A (en) | Video checking method, device, equipment and readable storage medium storing program for executing | |
CN108491142B (en) | Control method of mobile terminal, mobile terminal and storage medium | |
CN113705362B (en) | Training method and device of image detection model, electronic equipment and storage medium | |
US20230021338A1 (en) | Conditionally independent data generation for training machine learning systems | |
CN110555798A (en) | Image deformation method and device, electronic equipment and computer readable storage medium | |
US20220270352A1 (en) | Methods, apparatuses, devices, storage media and program products for determining performance parameters | |
CN108628442A (en) | A kind of information cuing method, device and electronic equipment | |
CN110334650A (en) | Object detecting method, device, electronic equipment and storage medium | |
CN112732553A (en) | Image testing method and device, electronic equipment and storage medium | |
CN106537460B (en) | For the automatic dividing technique of multistage pixel shader | |
CN110781809A (en) | Identification method and device based on registration feature update and electronic equipment | |
CN113269730B (en) | Image processing method, image processing device, computer equipment and storage medium | |
CN109993767A (en) | Image processing method and system | |
CN110059576A (en) | Screening technique, device and the electronic equipment of picture | |
CN115082202A (en) | Information processing method, device, equipment and storage medium for house mortgage | |
CN116109531A (en) | Image processing method, device, computer equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20240103 Address after: 518057 Tencent Building, No. 1 High-tech Zone, Nanshan District, Shenzhen City, Guangdong Province, 35 floors Patentee after: TENCENT TECHNOLOGY (SHENZHEN) Co.,Ltd. Patentee after: TENCENT CLOUD COMPUTING (BEIJING) Co.,Ltd. Address before: 2, 518000, East 403 room, SEG science and Technology Park, Zhenxing Road, Shenzhen, Guangdong, Futian District Patentee before: TENCENT TECHNOLOGY (SHENZHEN) Co.,Ltd. |