CN109858466A - A kind of face critical point detection method and device based on convolutional neural networks - Google Patents
A kind of face critical point detection method and device based on convolutional neural networks Download PDFInfo
- Publication number
- CN109858466A CN109858466A CN201910157356.9A CN201910157356A CN109858466A CN 109858466 A CN109858466 A CN 109858466A CN 201910157356 A CN201910157356 A CN 201910157356A CN 109858466 A CN109858466 A CN 109858466A
- Authority
- CN
- China
- Prior art keywords
- neural networks
- convolutional neural
- training set
- training
- face
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The face critical point detection method and device based on convolutional neural networks that the embodiment of the invention discloses a kind of, which comprises establish training set, include multiple facial images in the training set, key point is marked out to the facial image in training set；Construct convolutional neural networks model；Use the training set training convolutional neural networks model, the convolutional neural networks model is trained as loss function using improved L1loss, by way of to loss function weighted, change learning strategy using a variety of weights, reject influence of the invalid characteristic point to network, enable the network to the part for being intended to pay close attention to the habit that finds it difficult to learn, reduce the average influence for being easy the part of study to network losses, so that whole network changes human face posture, key point at facial detail, situations such as illumination background, is more robust, detection effect is more accurate, realize the precise positioning of face key point.
Description
Technical field
The present embodiments relate to computer vision processing technology fields, and in particular to a kind of based on convolutional neural networks
Face critical point detection method and device.
Background technique
Traditional face critical point detection method is mainly based upon the homing method of image design feature, first using artificial
The characteristic information of the mode statistical picture of design feature (hand-crafted feature) extracts the feature in facial image,
For preferably describing image, then pass through the recurrence device of high quality, recurrence calculating directly is carried out to the key point position of face.Object
The image feature information of body includes gray scale, texture, shape of image etc., carries out statistical calculation by the characteristic information to image,
To better describe to object, mainly there are the characteristics of image such as HIG, SIFT, HAAR to describe operator.SIFT is in image
Clutter or partial occlusion have very strong robustness because SIFT feature describes operator in uniform scaling, direction, illumination change
There is invariance in the case where changing with fractional distortion, SIFT detector is by key point (point-of-interest) come what is used, is utilized
The Gaussian filter of different scale carries out convolution to image, then carries out difference processing, key point to continuous Gaussian blurred picture
It is identified as Local Minimum/maximum value across scale image, this is by by the same scale of each pixel in image
Eight neighbours are compared, and in adjacent each scale compare nine corresponding adjacent pixels to complete, if picture
Plain value is the maximum value or minimum value between all compared pixels, then is chosen to candidate key point；HOG is another common
Characteristics of image operator is described, similar with SIFT feature, HOG counts the appearance of the gradient orientations in the Part portions of image
Number, HOG descriptor have the advantages that several keys than other descriptors, and since it works on local grid, it is for geometry
All there is very strong robustness with luminosity transformation.
Before deep learning rise, this traditional artificial design features method is to solve computer vision processing task
Mainstream, but traditional image design feature method scope of application compares limitation, and detection effect is also relatively rough, it is difficult to people
The key point of details is accurately detected at the face eyes corners of the mouth, in addition, this artificial design features are for appearances such as side face, new lines
State changes bigger face and does not also have preferable robustness, and illumination background, face location have whole detection effect
Large effect.
Summary of the invention
For this purpose, the embodiment of the present invention provides a kind of face critical point detection method based on convolutional neural networks, to solve
Traditional face critical point detection method is difficult to realize precisely detect to the key point at facial detail, for side face, new line, low
The first-class biggish Face datection of attitudes vibration is ineffective, and illumination background, face location have whole detection effect larger
Influence.
To achieve the goals above, the embodiment of the present invention provides the following technical solutions:
According to a first aspect of the embodiments of the present invention, a kind of face critical point detection based on convolutional neural networks is proposed
Method, which comprises
Training set is established, includes multiple facial images in the training set, pass is marked out to the facial image in training set
Key point；
Construct convolutional neural networks model；
Using the training set training convolutional neural networks model, using improved L1 loss as loss function to institute
It states convolutional neural networks model to be trained, the form of loss function is as follows:
Wherein, M is the keypoint quantity marked in every facial image, wiFor weight, ytIndicate the true tag of supervision,
ypThe predicted value for indicating convolutional neural networks output indicates that N indicates the dimension of feature vector by key point feature vector form
Number.
Further, described to include: using the training set training convolutional neural networks model
The training set includes the first training set and the second training set, includes that multiple include lift in second training set
Head bows, blinks, opening one's mouth, left shaking the head and the facial image of right six kinds of human face actions of shaking the head；
Pre-training is carried out to the convolutional neural networks model using the first training set；
Training is optimized to the convolutional neural networks model after pre-training using the second training set.
Further, the method also includes:
Calculate separately the equalization point for the key point that face images are marked in the first training set or the second training set；
Using the first training set or the second training set training convolutional neural networks model learning face key point to putting down
Offset between point.
Further, the convolutional neural networks model carries out the spy of facial image using improved ResNet18 network
Sign is extracted, and it is logical that the output channel number of every layer of convolutional layer of improved ResNet18 network is reduced to ResNet18 network convolutional layer
The 1/10 of road quantity.
Further, the convolutional neural networks model by the crucial point feature of a full articulamentum output multidimensional to
Amount, the odd term of described eigenvector correspond to the abscissa offset of multiple key points, and even item corresponds to the vertical of multiple key points
Coordinate shift amount.
Further, the method also includes:
For the invalid key point lost in facial image, weight wiIt is set to 0.
Further, the method also includes:
Eye key point in facial image is arranged compared to the higher weight of other key points.
Further, the method also includes:
The convolutional neural networks model is trained using batch gradient descent method and back-propagation algorithm.
Further, the method also includes:
Gradient passback is carried out for the loss for being more than preset threshold；
For being less than the loss of preset threshold for weight wiIt is set to 0, is returned without gradient.
According to a second aspect of the embodiments of the present invention, a kind of face critical point detection based on convolutional neural networks is proposed
Device, described device include:
Training set establishes module, marks out key point for establishing training set, and to the face images in training set；
Network establishes module, for constructing convolutional neural networks model；
Training module, for using the training set training convolutional neural networks model.
The embodiment of the present invention has the advantages that
A kind of face critical point detection method and device based on convolutional neural networks that the embodiment of the present invention proposes, passes through
The depth information for building multilayer convolutional neural networks study face obtains people using the depth ability to express of convolutional neural networks
The depth characteristic of face, the position of face key point is returned by returning device from the face characteristic learnt, and loss function uses
The mode of weighted can change learning strategy by a variety of weights, reject influence of the invalid characteristic point to network, enable network
Enough it is intended to pay close attention to the part for the habit that finds it difficult to learn, the average influence for being easy the part of study to network losses is reduced, so that whole network
Human face posture is changed, is more robust situations such as key point, illumination background at facial detail, detection effect is more accurate, real
The precise positioning of existing face key point.
Detailed description of the invention
It, below will be to embodiment party in order to illustrate more clearly of embodiments of the present invention or technical solution in the prior art
Formula or attached drawing needed to be used in the description of the prior art are briefly described.It should be evident that the accompanying drawings in the following description is only
It is merely exemplary, it for those of ordinary skill in the art, without creative efforts, can also basis
The attached drawing of offer, which is extended, obtains other implementation attached drawings.
Fig. 1 is a kind of stream for face critical point detection method based on convolutional neural networks that the embodiment of the present invention 1 provides
Journey schematic diagram；
Fig. 2 is a kind of people for face critical point detection method based on convolutional neural networks that the embodiment of the present invention 1 provides
Face image key point marks schematic diagram；
Fig. 3 is the short cut connection schematic diagram in ResNet network；
Fig. 4 is a kind of net for face critical point detection method based on convolutional neural networks that the embodiment of the present invention 1 provides
Network processing flow schematic diagram.
Specific embodiment
Embodiments of the present invention are illustrated by particular specific embodiment below, those skilled in the art can be by this explanation
Content disclosed by book is understood other advantages and efficacy of the present invention easily, it is clear that described embodiment is the present invention one
Section Example, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not doing
Every other embodiment obtained under the premise of creative work out, shall fall within the protection scope of the present invention.
Embodiment 1
As shown in Figure 1, the present embodiment proposes a kind of face critical point detection method based on convolutional neural networks, the party
Method includes:
S100, training set is established, includes multiple facial images in training set, pass is marked out to the facial image in training set
Key point.Training set includes the first training set and the second training set, include in second training set multiple include come back, bow,
It blinks, open one's mouth, a left side is shaken the head and the facial image of right six kinds of human face actions of shaking the head.In first training set and the second training set, such as Fig. 2
Shown, every facial image is marked out all in accordance with the position of the key areas such as face mask, eyebrow, eyes, nose, mouth most can
Embody 70 key points of face characteristic.
In order to allow network to learn the key point at details as far as possible, network needs training set sample as much as possible, this
600,000 facial images have been collected in embodiment as the first training set.Further, in order to enhance network for face difference appearance
The robustness of state movement, the first training set that the present embodiment first marks out key point using 60w are trained network, training
After the completion, it using obtained model as pre-training model, is then carried out using 10,000 the second training sets comprising various movements micro-
It adjusts, robustness of the enhancing network for various movements.
Further, this method further include:
Calculate separately the equalization point for the key point that face images are marked in the first training set or the second training set；
Use the first training set or the second training set training convolutional neural networks model learning face key point to equalization point
Between offset.
Pre-training and optimization training constantly, calculate the equalization point of the first training set and all pictures of the second training set, as
Initial Face key point allows all faces to go to return the offset with equalization point, than directly returning 70 key points in a network
Position want much simpler, therefore allow network go study face key point arrive equalization point offset, rather than directly recurrence 70
The position of a point.
S200, building convolutional neural networks model.
Convolutional neural networks model carries out the feature of facial image using improved ResNet18 network in the present embodiment
It extracts.As shown in figure 3, ResNet network passes through the connection of a short cut, allow network than common convolutional Neural net
Network (VGG, GoogleNet etc.) is deeper, therefore the ability in feature extraction of network, learning ability and network capability of fitting are all significantly
Reinforce.
A facial image is given, first by facial image using long side as standard, equal proportion scaling to 96x96, short side is not
Foot benefit is black, and face is avoided to generate deformation, and then by treated, facial image is input to convolutional neural networks model, passes through improvement
ResNet18 network afterwards extracts the feature of face.
In the present embodiment, in order to reduce the parameter amount at network, avoid parameter redundancy, improved ResNet18 network it is every
The output channel number of layer convolutional layer is reduced to the 1/10 of former ResNet18 network convolutional layer number of channels, and speed is than conventionally used
ResNet18 network greatly promotes, and is more applicable for practical application.
The key point feature vector finally by complete 140 dimension of articulamentum output of convolutional neural networks model, feature
Vertical (y) coordinate of cross (x) the coordinate shift amount of corresponding 70 key points of the odd term of vector, corresponding 70 key points of even item is inclined
Shifting amount.
In the present embodiment, convolutional neural networks model is instructed using batch gradient descent method and back-propagation algorithm
Practice.As shown in figure 4, learning the depth information of face by building multilayer convolutional neural networks, the depth of ResNet18 network is utilized
Ability to express is spent, the depth characteristic of face is extracted, face is then returned from the face characteristic learnt by a recurrence device
The position of key point.
S300, using training set training convolutional neural networks model, using improved L1loss as loss function to volume
Product neural network model is trained, and the form of loss function is as follows:
Wherein, M is the keypoint quantity marked in every facial image, and in this implementations, the value of M is 70, wiFor weight, yt
Indicate the true tag of supervision, ypIt indicates the predicted value that convolutional neural networks export, is indicated by key point feature vector form,
The dimension of N expression feature vector, as 140.
Specifically, including: using the training set training convolutional neural networks model
Pre-training is carried out to the convolutional neural networks model using the first training set；
Training is optimized to the convolutional neural networks model after pre-training using the second training set.
What L1_loss was measured is the L1_ norm between predicted value and actual value, and form is as follows:
Wherein ytIndicate the true tag of supervision, ypIndicate network predicted value, be herein feature vector, N indicate feature to
The dimension of amount.
In order to make full use of the information of facial image in training set, while avoiding the mark sample of some mistakes to network
Weight a wi, w is added in trained influence, the present embodiment in the corresponding trained loss function of networkiIt is that mark key point is
No effective label.
In the application, for the invalid key point lost in facial image, for example, in side face some loss key point, power
Weight wiIt is set to 0, corresponding key point does not lose loss at this time, i.e., no gradient passback may be implemented to have made full use of in this way
The key point information of effect rejects influence of the invalid key point to network, while network also can be enhanced to the robustness of side face.
The training study of convolutional neural networks model in the present embodiment changes learning strategy using more weights:
(1) eye key point loss weight reinforces strategy
Eye key point in facial image is arranged compared to the higher weight of other key points.For complicated eye
Portion's key point, feature is often unobvious, and eye key point is often inaccurate, however to identify that eye closing acts in practical application, institute
It is most important with accurately identifying for eye key point, in order to allow network to be intended to learn eye key point, make model for eye
Feature learning obtains more preferably, devises a kind of eye weight enhancing strategy, i.e., for the part other than eye key point, is supervising
The loss function superintended and directed adds 1 weight；And for eye key point, using 5 weight, so that network increases eye key point
Loss, it is more preferable in the effect in study of eye, it realizes and the key point at human eye is learnt emphatically.
(2) weight dynamic change strategy is lost
Since the loss of different key points is different, the complexity of study is different, it is more difficult to which the key point loss of study is big, holds
The key point loss easily learnt is small, when gradient passback, is averaged to the loss of all key points, so that the sample for the habit that finds it difficult to learn damages
It loses and reduces, and the sample losses for being easy study increase, so that network convergence effect is worse, find it difficult to learn to allow network to be intended to concern
Sample is practised, the present embodiment proposes a kind of dynamic loss weight variation strategy: carrying out gradient for the loss for being more than preset threshold
Passback, for being less than the loss of preset threshold for weight wiIt is set to 0, is returned without gradient, can thus network be made to exist
The effect in study of difficult learning sample is more preferable.
In the present embodiment, the training of convolutional neural networks model uses following parameter: using Caffe deep learning frame；
GPU performance: 1 picture 1080Ti；Gradient declines mode: batch gradient descent method (SGD)；Batch size: 64；Learning rate is set
It sets: initial 0.01, every 5w wheel learning rate of crossing falls to original 1/10；Maximum exercise wheel number: 120,000.
The depth information for learning face by building multilayer convolutional neural networks is expressed using the depth of convolutional neural networks
Ability obtains the depth characteristic of face, and the position of key point is returned from the face characteristic learnt, by loss function
The mode of weighted changes learning strategy using a variety of weights, rejects influence of the invalid characteristic point to network, enable the network to
Find it difficult to learn the part of habit in concern, reduce the average influence for being easy the part of study to network losses so that whole network for
Situations such as key point, illumination background, is more robust at human face posture variation, facial detail, and detection effect is more accurate, realizes people
The precise positioning of face key point.
Embodiment 2
Corresponding with above-described embodiment 1, the present embodiment proposes a kind of face key point based on convolutional neural networks
Detection device, device include:
Training set establishes module, marks out key point for establishing training set, and to the face images in training set；
Network establishes module, for constructing convolutional neural networks model；
Training module, for using training set training convolutional neural networks model.
Although above having used general explanation and specific embodiment, the present invention is described in detail, at this
On the basis of invention, it can be made some modifications or improvements, this will be apparent to those skilled in the art.Therefore,
These modifications or improvements without departing from theon the basis of the spirit of the present invention are fallen within the scope of the claimed invention.
Claims (10)
1. a kind of face critical point detection method based on convolutional neural networks, which is characterized in that the described method includes:
Training set is established, includes multiple facial images in the training set, key point is marked out to the facial image in training set；
Construct convolutional neural networks model；
Using the training set training convolutional neural networks model, using improved L1 loss as loss function to the volume
Product neural network model is trained, and the form of loss function is as follows:
Wherein, M is the keypoint quantity marked in every facial image, wiFor weight, ytIndicate the true tag of supervision, ypTable
The predicted value for showing convolutional neural networks output indicates that N indicates the dimension of feature vector by key point feature vector form.
2. a kind of face critical point detection method based on convolutional neural networks according to claim 1, which is characterized in that
It is described to include: using the training set training convolutional neural networks model
The training set includes the first training set and the second training set, include multiple in second training set includes to come back, is low
Head blinks, opens one's mouth, left shaking the head and the facial image of right six kinds of human face actions of shaking the head；
Pre-training is carried out to the convolutional neural networks model using the first training set；
Training is optimized to the convolutional neural networks model after pre-training using the second training set.
3. a kind of face critical point detection method based on convolutional neural networks according to claim 2, which is characterized in that
The method also includes:
Calculate separately the equalization point for the key point that face images are marked in the first training set or the second training set；
Train the convolutional neural networks model learning face key point to equalization point using the first training set or the second training set
Between offset.
4. a kind of face critical point detection method based on convolutional neural networks according to claim 1, which is characterized in that
The convolutional neural networks model carries out the feature extraction of facial image using improved ResNet18 network, improved
The output channel number of every layer of convolutional layer of ResNet18 network is reduced to the 1/10 of ResNet18 network convolutional layer number of channels.
5. a kind of face critical point detection method based on convolutional neural networks according to claim 4, which is characterized in that
The convolutional neural networks model exports the key point feature vector of multidimensional, the surprise of described eigenvector by a full articulamentum
The abscissa offset of the several multiple key points of correspondence, even item correspond to the ordinate offset of multiple key points.
6. a kind of face critical point detection method based on convolutional neural networks according to claim 1, which is characterized in that
The method also includes:
For the invalid key point lost in facial image, weight wiIt is set to 0.
7. a kind of face critical point detection method based on convolutional neural networks according to claim 1, which is characterized in that
The method also includes:
Eye key point in facial image is arranged compared to the higher weight of other key points.
8. a kind of face critical point detection method based on convolutional neural networks according to claim 1, which is characterized in that
The method also includes:
The convolutional neural networks model is trained using batch gradient descent method and back-propagation algorithm.
9. a kind of face critical point detection method based on convolutional neural networks according to claim 7, which is characterized in that
The method also includes:
Gradient passback is carried out for the loss for being more than preset threshold；
For being less than the loss of preset threshold for weight wiIt is set to 0, is returned without gradient.
10. a kind of face critical point detection device based on convolutional neural networks, which is characterized in that described device includes:
Training set establishes module, marks out key point for establishing training set, and to the face images in training set；
Network establishes module, for constructing convolutional neural networks model；
Training module, for using the training set training convolutional neural networks model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910157356.9A CN109858466A (en) | 2019-03-01 | 2019-03-01 | A kind of face critical point detection method and device based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910157356.9A CN109858466A (en) | 2019-03-01 | 2019-03-01 | A kind of face critical point detection method and device based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109858466A true CN109858466A (en) | 2019-06-07 |
Family
ID=66899591
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910157356.9A Pending CN109858466A (en) | 2019-03-01 | 2019-03-01 | A kind of face critical point detection method and device based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109858466A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110263774A (en) * | 2019-08-19 | 2019-09-20 | 珠海亿智电子科技有限公司 | A kind of method for detecting human face |
| CN110414350A (en) * | 2019-06-26 | 2019-11-05 | 浙江大学 | The face false-proof detection method of two-way convolutional neural networks based on attention model |
| CN110569809A (en) * | 2019-09-11 | 2019-12-13 | 淄博矿业集团有限责任公司 | coal mine dynamic face recognition attendance checking method and system based on deep learning |
| CN110598727A (en) * | 2019-07-19 | 2019-12-20 | 深圳力维智联技术有限公司 | Model construction method based on transfer learning, image identification method and device thereof |
| CN110633711A (en) * | 2019-09-09 | 2019-12-31 | 长沙理工大学 | Computer device and method for training feature point detector and feature point detection method |
| CN110659596A (en) * | 2019-09-11 | 2020-01-07 | 高新兴科技集团股份有限公司 | Face key point positioning method under case and management scene, computer storage medium and equipment |
| CN110659585A (en) * | 2019-08-31 | 2020-01-07 | 电子科技大学 | Pedestrian detection method based on interactive attribute supervision |
| CN110738110A (en) * | 2019-09-11 | 2020-01-31 | 北京迈格威科技有限公司 | Human face key point detection method, device, system and storage medium based on anchor point |
| CN110874587A (en) * | 2019-12-26 | 2020-03-10 | 浙江大学 | Face characteristic parameter extraction system |
| CN111652240A (en) * | 2019-12-18 | 2020-09-11 | 南京航空航天大学 | Image local feature detection and description method based on CNN |
| CN112052843A (en) * | 2020-10-14 | 2020-12-08 | 福建天晴在线互动科技有限公司 | Method for detecting key points of human face from coarse to fine |
| CN112070227A (en) * | 2020-09-08 | 2020-12-11 | 厦门真景科技有限公司 | Neural network training method, device and equipment |
| CN112287765A (en) * | 2020-09-30 | 2021-01-29 | 新大陆数字技术股份有限公司 | Face living body detection method, device and equipment and readable storage medium |
| CN112380978A (en) * | 2020-11-12 | 2021-02-19 | 平安科技（深圳）有限公司 | Multi-face detection method, system and storage medium based on key point positioning |
| CN112784800A (en) * | 2021-02-02 | 2021-05-11 | 浙江大学 | Face key point detection method based on neural network and shape constraint |
| CN112862095A (en) * | 2021-02-02 | 2021-05-28 | 浙江大华技术股份有限公司 | Self-distillation learning method and device based on characteristic analysis and readable storage medium |
| CN113221698A (en) * | 2021-04-29 | 2021-08-06 | 北京科技大学 | Facial key point positioning method based on deep learning and expression recognition |
| CN113313010A (en) * | 2021-05-26 | 2021-08-27 | 广州织点智能科技有限公司 | Face key point detection model training method, device and equipment |
| CN114419105A (en) * | 2022-03-14 | 2022-04-29 | 深圳市海清视讯科技有限公司 | Multi-target pedestrian trajectory prediction model training method, prediction method and device |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160275341A1 (en) * | 2015-03-18 | 2016-09-22 | Adobe Systems Incorporated | Facial Expression Capture for Character Animation |
| CN106127170A (en) * | 2016-07-01 | 2016-11-16 | 重庆中科云丛科技有限公司 | A kind of merge the training method of key feature points, recognition methods and system |
| CN106909879A (en) * | 2017-01-11 | 2017-06-30 | 开易（北京）科技有限公司 | A kind of method for detecting fatigue driving and system |
| CN107748858A (en) * | 2017-06-15 | 2018-03-02 | 华南理工大学 | A kind of multi-pose eye locating method based on concatenated convolutional neutral net |
| CN107784263A (en) * | 2017-04-28 | 2018-03-09 | 新疆大学 | Based on the method for improving the Plane Rotation Face datection for accelerating robust features |
| CN108229301A (en) * | 2017-11-03 | 2018-06-29 | 北京市商汤科技开发有限公司 | Eyelid line detecting method, device and electronic equipment |
| CN108304357A (en) * | 2018-01-31 | 2018-07-20 | 北京大学 | A kind of Chinese word library automatic generation method based on font manifold |
| CN108399373A (en) * | 2018-02-06 | 2018-08-14 | 北京达佳互联信息技术有限公司 | The model training and its detection method and device of face key point |
| CN108734078A (en) * | 2017-12-14 | 2018-11-02 | 北京市商汤科技开发有限公司 | Image processing method, device, electronic equipment, storage medium and program |
-
2019
- 2019-03-01 CN CN201910157356.9A patent/CN109858466A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160275341A1 (en) * | 2015-03-18 | 2016-09-22 | Adobe Systems Incorporated | Facial Expression Capture for Character Animation |
| CN106127170A (en) * | 2016-07-01 | 2016-11-16 | 重庆中科云丛科技有限公司 | A kind of merge the training method of key feature points, recognition methods and system |
| CN106909879A (en) * | 2017-01-11 | 2017-06-30 | 开易（北京）科技有限公司 | A kind of method for detecting fatigue driving and system |
| CN107784263A (en) * | 2017-04-28 | 2018-03-09 | 新疆大学 | Based on the method for improving the Plane Rotation Face datection for accelerating robust features |
| CN107748858A (en) * | 2017-06-15 | 2018-03-02 | 华南理工大学 | A kind of multi-pose eye locating method based on concatenated convolutional neutral net |
| CN108229301A (en) * | 2017-11-03 | 2018-06-29 | 北京市商汤科技开发有限公司 | Eyelid line detecting method, device and electronic equipment |
| CN108734078A (en) * | 2017-12-14 | 2018-11-02 | 北京市商汤科技开发有限公司 | Image processing method, device, electronic equipment, storage medium and program |
| CN108304357A (en) * | 2018-01-31 | 2018-07-20 | 北京大学 | A kind of Chinese word library automatic generation method based on font manifold |
| CN108399373A (en) * | 2018-02-06 | 2018-08-14 | 北京达佳互联信息技术有限公司 | The model training and its detection method and device of face key point |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414350A (en) * | 2019-06-26 | 2019-11-05 | 浙江大学 | The face false-proof detection method of two-way convolutional neural networks based on attention model |
| CN110598727A (en) * | 2019-07-19 | 2019-12-20 | 深圳力维智联技术有限公司 | Model construction method based on transfer learning, image identification method and device thereof |
| CN110263774A (en) * | 2019-08-19 | 2019-09-20 | 珠海亿智电子科技有限公司 | A kind of method for detecting human face |
| CN110659585B (en) * | 2019-08-31 | 2022-03-15 | 电子科技大学 | Pedestrian detection method based on interactive attribute supervision |
| CN110659585A (en) * | 2019-08-31 | 2020-01-07 | 电子科技大学 | Pedestrian detection method based on interactive attribute supervision |
| CN110633711A (en) * | 2019-09-09 | 2019-12-31 | 长沙理工大学 | Computer device and method for training feature point detector and feature point detection method |
| CN110633711B (en) * | 2019-09-09 | 2022-02-11 | 长沙理工大学 | Computer device and method for training feature point detector and feature point detection method |
| CN110738110A (en) * | 2019-09-11 | 2020-01-31 | 北京迈格威科技有限公司 | Human face key point detection method, device, system and storage medium based on anchor point |
| CN110569809A (en) * | 2019-09-11 | 2019-12-13 | 淄博矿业集团有限责任公司 | coal mine dynamic face recognition attendance checking method and system based on deep learning |
| CN110659596A (en) * | 2019-09-11 | 2020-01-07 | 高新兴科技集团股份有限公司 | Face key point positioning method under case and management scene, computer storage medium and equipment |
| CN111652240B (en) * | 2019-12-18 | 2023-06-27 | 南京航空航天大学 | CNN-based image local feature detection and description method |
| CN111652240A (en) * | 2019-12-18 | 2020-09-11 | 南京航空航天大学 | Image local feature detection and description method based on CNN |
| CN110874587B (en) * | 2019-12-26 | 2020-07-28 | 浙江大学 | Face characteristic parameter extraction system |
| CN110874587A (en) * | 2019-12-26 | 2020-03-10 | 浙江大学 | Face characteristic parameter extraction system |
| CN112070227A (en) * | 2020-09-08 | 2020-12-11 | 厦门真景科技有限公司 | Neural network training method, device and equipment |
| CN112070227B (en) * | 2020-09-08 | 2023-08-18 | 厦门真景科技有限公司 | Neural network training method, device and equipment |
| CN112287765A (en) * | 2020-09-30 | 2021-01-29 | 新大陆数字技术股份有限公司 | Face living body detection method, device and equipment and readable storage medium |
| CN112287765B (en) * | 2020-09-30 | 2024-06-04 | 新大陆数字技术股份有限公司 | Face living body detection method, device, equipment and readable storage medium |
| CN112052843A (en) * | 2020-10-14 | 2020-12-08 | 福建天晴在线互动科技有限公司 | Method for detecting key points of human face from coarse to fine |
| CN112052843B (en) * | 2020-10-14 | 2023-06-06 | 福建天晴在线互动科技有限公司 | Face key point detection method from coarse face to fine face |
| WO2021190664A1 (en) * | 2020-11-12 | 2021-09-30 | 平安科技（深圳）有限公司 | Multi-face detection method and system based on key point positioning, and storage medium |
| CN112380978B (en) * | 2020-11-12 | 2024-05-07 | 平安科技（深圳）有限公司 | Multi-face detection method, system and storage medium based on key point positioning |
| CN112380978A (en) * | 2020-11-12 | 2021-02-19 | 平安科技（深圳）有限公司 | Multi-face detection method, system and storage medium based on key point positioning |
| CN112784800B (en) * | 2021-02-02 | 2022-05-10 | 浙江大学 | Face key point detection method based on neural network and shape constraint |
| CN112862095B (en) * | 2021-02-02 | 2023-09-29 | 浙江大华技术股份有限公司 | Self-distillation learning method and device based on feature analysis and readable storage medium |
| CN112862095A (en) * | 2021-02-02 | 2021-05-28 | 浙江大华技术股份有限公司 | Self-distillation learning method and device based on characteristic analysis and readable storage medium |
| CN112784800A (en) * | 2021-02-02 | 2021-05-11 | 浙江大学 | Face key point detection method based on neural network and shape constraint |
| CN113221698B (en) * | 2021-04-29 | 2023-08-15 | 北京科技大学 | Facial key point positioning method based on deep learning and expression recognition |
| CN113221698A (en) * | 2021-04-29 | 2021-08-06 | 北京科技大学 | Facial key point positioning method based on deep learning and expression recognition |
| CN113313010A (en) * | 2021-05-26 | 2021-08-27 | 广州织点智能科技有限公司 | Face key point detection model training method, device and equipment |
| CN114419105A (en) * | 2022-03-14 | 2022-04-29 | 深圳市海清视讯科技有限公司 | Multi-target pedestrian trajectory prediction model training method, prediction method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109858466A (en) | A kind of face critical point detection method and device based on convolutional neural networks | |
| CN106096538B (en) | Face identification method and device based on sequencing neural network model | |
| CN106997597B (en) | It is a kind of based on have supervision conspicuousness detection method for tracking target | |
| CN105069400B (en) | Facial image gender identifying system based on the sparse own coding of stack | |
| CN109657612B (en) | Quality sorting system based on facial image features and application method thereof | |
| CN108268859A (en) | A kind of facial expression recognizing method based on deep learning | |
| CN107153816A (en) | A kind of data enhancement methods recognized for robust human face | |
| CN108470354A (en) | Video target tracking method, device and realization device | |
| CN106599789A (en) | Video class identification method and device, data processing device and electronic device | |
| CN107657249A (en) | Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again | |
| CN107316307A (en) | A kind of Chinese medicine tongue image automatic segmentation method based on depth convolutional neural networks | |
| CN108961675A (en) | Fall detection method based on convolutional neural networks | |
| CN107808132A (en) | A kind of scene image classification method for merging topic model | |
| CN109410168A (en) | For determining the modeling method of the convolutional neural networks model of the classification of the subgraph block in image | |
| CN109800629A (en) | A kind of Remote Sensing Target detection method based on convolutional neural networks | |
| CN104850825A (en) | Facial image face score calculating method based on convolutional neural network | |
| CN108052884A (en) | A kind of gesture identification method based on improvement residual error neutral net | |
| CN110348322A (en) | Human face in-vivo detection method and equipment based on multi-feature fusion | |
| CN105205804A (en) | Caryon-cytolymph separation method and apparatus of white blood cells in blood cell image, and classification method and apparatus of white blood cells in blood cell image | |
| CN104794693B (en) | A kind of portrait optimization method of face key area automatic detection masking-out | |
| CN106778852A (en) | A kind of picture material recognition methods for correcting erroneous judgement | |
| CN110288555A (en) | A kind of low-light (level) Enhancement Method based on improved capsule network | |
| CN106557750A (en) | It is a kind of based on the colour of skin and the method for detecting human face of depth y-bend characteristics tree | |
| CN107944437B (en) | A kind of Face detection method based on neural network and integral image | |
| CN104298974A (en) | Human body behavior recognition method based on depth video sequence |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190607 |
|
| RJ01 | Rejection of invention patent application after publication |