CN106909909A - A kind of Face datection and alignment schemes based on shared convolution feature - Google Patents
A kind of Face datection and alignment schemes based on shared convolution feature Download PDFInfo
- Publication number
- CN106909909A CN106909909A CN201710135007.8A CN201710135007A CN106909909A CN 106909909 A CN106909909 A CN 106909909A CN 201710135007 A CN201710135007 A CN 201710135007A CN 106909909 A CN106909909 A CN 106909909A
- Authority
- CN
- China
- Prior art keywords
- face
- feature
- convolution
- output
- image
- 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.)
- Granted
Links
- 230000001815 facial effect Effects 0.000 claims abstract description 43
- 210000004709 eyebrow Anatomy 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 20
- 239000000284 extract Substances 0.000 claims description 13
- 238000000605 extraction Methods 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 8
- 238000013527 convolutional neural network Methods 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 3
- 241000406668 Loxodonta cyclotis Species 0.000 claims 1
- 238000005516 engineering process Methods 0.000 abstract description 11
- 238000001514 detection method Methods 0.000 abstract description 9
- 238000012795 verification Methods 0.000 abstract description 5
- 238000000034 method Methods 0.000 description 14
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000011478 gradient descent method Methods 0.000 description 2
- 210000004209 hair Anatomy 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/164—Detection; Localisation; Normalisation using holistic features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
- G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention provides a kind of Face datection and alignment schemes based on shared convolution feature, comprising four steps:Multilayer convolution and pondization operation are carried out to input picture first, convolution feature is extracted;Then Face datection is carried out according to the convolution feature extracted, the face location information in output image;Then according to the face location information of previous step output, the convolution feature of correspondence face location is extracted, facial characteristics point location is carried out to face, that is, export human face characteristic point position such as eyebrow, canthus, nose, the corners of the mouth etc. in facial image;Facial image and its corresponding human face characteristic point position finally according to the output of facial feature points detection step, are rotated and scaling treatment, the facial image after output alignment to face.The present invention is capable of achieving to the automatic detection and automatic aligning of face in image, with speed it is fast, accuracy rate is high the characteristics of, be favorably improved the accuracy rate of face verification and face recognition technology.
Description
Technical field
The present invention relates to technical field of computer vision, and in particular to a kind of Face datection based on shared convolution feature with
Alignment schemes.
Background technology
With continuing to develop for computer science, man-machine interaction has turned into more and more valued technology.As computer
The recognition of face of visual field and face verification technology start to be applied in industrial quarters, and face is known in the past few decades
It is not always the hot research problem of computer vision field with verification technique.And Face datection with align be in recognition of face extremely
Close an important step.
First, currently used relatively common method for detecting human face is based on Haar-like features and AdaBoost technologies
Method for detecting human face, it is by extracting the human-face detector of the features training based on AdaBoost technologies of engineer.But
It is, because Haar-like features are a kind of low level abstract characteristics of engineer, not have complete face information, so
Cause the detector accuracy rate for training not high.
Secondly, currently used relatively common man face characteristic point positioning method be adaptive shape model method (ASM,
adaptive shape model).This method does not possess very strong robustness for off-note point and attitudes vibration, because
This hardly results in accurate human face characteristic point, and this will directly affect face alignment effect, further results in recognition of face performance
Degradation.
Again, although the method based on convolutional neural networks is higher than conventional method in terms of accuracy rate, convolution god
Big through network calculations amount, time-consuming for treatment single picture, is extremely difficult to real-time detection face and the requirement alignd.So needing hair
A kind of bright new Face datection and alignment schemes, can be determined using the Face datection based on shared convolution feature and face feature point
Position method.
In order to solve the above problems, the present invention provides a kind of Face datection and alignment schemes based on shared convolution feature,
The method is input arbitrarily to include the image of face, can rapidly and accurately be detected and the face that aligns.
The content of the invention
The technical problem to be solved in the present invention is:Overcome existing based on traditional characteristic and based on convolutional neural networks
The deficiency of Face datection and alignment schemes, there is provided a kind of Face datection and alignment schemes based on shared convolution feature.
The technical solution adopted by the present invention is:A kind of Face datection and alignment schemes based on shared convolution feature, this hair
Bright schematic flow sheet is as shown in fig. 7, comprises following four step:
Step (1), convolution feature extraction, convolutional layer are carried out to input picture first by a deep layer convolutional network structure
Activation primitive uses RELU activation primitives, pond layer to use maximum pond.The input picture of given arbitrary dimension size, passes through
Multilayer convolution pondization is operated, and the convolution feature of output is fixed size 512 on depth axle, and size spatially is schemed with input
The proportional relation of size of piece.Convolution characteristic extraction part has 16 layers, including 12 layers of convolutional layer and 4 layers of pond layer, volume
The convolution kernel size of lamination is 3x3, and convolution stride is 1, and zero padding number is 2, and the pond core size of pond layer is big
Small to be 2x2, pond layer stride is 2;
After step (2), input picture are operated by multilayer convolution pondization, convolution feature is extracted, trains human-face detector,
Advise network and area-of-interest pond layer including region;Region advises that network is characterized as input with convolution, and output is probably mesh
The candidate region of object is marked, area-of-interest pond layer is then introduced and is extracted the corresponding convolution feature in candidate region, to each
Candidate region carries out two classification and is returned with bounding box.The output valve of classified part represents whether candidate region is face, if
Face then exports 1, otherwise exports 0;Bounding box returns part output human face region position in the input image, if classified part
Output 0, then ignore the output that bounding box returns part.
Step (3), input picture extract convolution feature after being operated by multilayer convolution pondization, then by region of interest
Domain pond layer extracts human face region feature, and training face feature point returns device, will the feature one-dimensional full articulamentum of connection, opposite
Portion's characteristic point position is returned, and exports the human face characteristic point position in the facial image, including eyebrow, canthus, nose, mouth
Angle.
Step (4), the facial image finally according to step (3) output and its corresponding human face characteristic point position, choose two
Individual fixed human face characteristic point (such as left and right canthus) is rotated and scaling treatment, after output alignment as fixed point to facial image
Facial image, the position of selected fixed characteristic points will be constant in the facial image after alignment.
Further, the deep layer convolutional network model described in step (1), is input with original image, not to input picture
Carry out size change over, it is to avoid anamorphose, lose information.
Further, the human-face detector described in step (2) is by extracting the feature of image for input training is obtained
, the feature of image is extracted by the convolutional neural networks shared and obtained.
Further, it is by extracting the feature of image for input is instructed that the face feature point described in step (3) returns device
Get, the feature of image is extracted by the convolutional neural networks shared and obtained.
Further, being rotated to facial image described in step (4) and scaled, be based on ensureing that face or so is outer
2 points of canthus is constant to be processed.
Principle of the invention is:
The present invention provides a kind of Face datection and alignment schemes based on shared convolution feature, and the method is with arbitrarily comprising people
The image of face is input, face that can rapidly and accurately in detection image, and face to detecting aligns.This method
Comprising four steps:Input picture is passed into deep layer convolutional network model first, convolution feature is extracted;Then will extract
Shared convolution feature passes to human-face detector, human-face detector output face window;Then the face for being exported with previous step
The shared convolution of image and correspondence position is characterized as that input passes to face feature point and returns device, and face feature point returns device output
Human face characteristic point position (eyebrow, canthus, nose, corners of the mouth etc.) in the facial image;Finally according to facial feature points detection step
The facial image of rapid output and its corresponding human face characteristic point position, are rotated and scaling treatment, after output alignment to face
Facial image.The present invention is capable of achieving to the automatic detection and automatic aligning of face in image, with speed is fast, accuracy rate is high
Feature, is favorably improved the accuracy rate of face verification and face recognition technology.
Present disclosure mainly includes following four step:
(1) convolution characteristic extraction step:Convolution is carried out to input picture first by a deep layer convolutional network structure special
Extraction is levied, convolutional layer activation primitive uses RELU activation primitives, pond layer to use maximum pond.Given arbitrary dimension size
Input picture, operates by multilayer convolution pondization, and the convolution feature of output is fixed size 512 on depth axle, spatially
Size and input picture the proportional relation of size.Convolution characteristic extraction part has 16 layers, including 12 layers of convolutional layer
With 4 layers of pond layer, the convolution kernel size of convolutional layer is 3x3, and convolution stride is 1, and zero padding number is 2, pond layer
Pond core size be 2x2, pond layer stride is 2.
(2) Face datection step:This step needs that training in advance what a human-face detector, input picture are rolled up by multilayer
After product pondization operation, convolution feature is extracted, pass to human-face detector, including region suggestion network and area-of-interest pond
Layer;Region advises that network is characterized as input with convolution, and output is probably the candidate region of destination object, then introduces region of interest
Domain pond layer extracts the corresponding convolution feature in candidate region, two classification is carried out to each candidate region and is returned with bounding box.Point
The output valve of class part represents whether candidate region is face, if face then exports 1, otherwise exports 0;Bounding box recurrence portion
Divide output human face region position in the input image, if classified part output 0, ignore the output that bounding box returns part.
(3) face feature point positioning step:This step needs also exist for one face feature point of training in advance and returns device, input
Picture extracts convolution feature after being operated by multilayer convolution pondization, then extracts human face region by area-of-interest pond layer
Feature, passes to face feature point and returns device, will feature one-dimensional connect full articulamentum, facial characteristic point position is returned
Return, export the human face characteristic point position in the facial image, including eyebrow, canthus, nose, the corners of the mouth.
(4) face alignment step:The input of this step is the facial image and its right of face feature point positioning step output
The human face characteristic point position answered, the facial image after output alignment.Face alignment step according to two fixed human face characteristic points (such as
Left and right canthus) alignd, facial image is rotated and scaling treatment, the facial image after output alignment, after alignment
The position of fixed characteristic points selected by previous step will be constant in facial image.
Present invention advantage compared with prior art is:
1st, for the problem that Face datection effect precision in open scene hypograph is low, present invention employs based on convolution god
Through the Face datection algorithm of network, the algorithm can carry out self adaptation to various sizes of input picture, can be prevented effectively from input figure
As information loss caused by normalization.
2nd, the present invention proposes a kind of new face alignment algorithm based on shared convolution feature, and the algorithm is used and face
The strategy of the shared convolution feature of detection model, and model is trained using a kind of three stage training methods, can be fast
Speed efficiently locates face feature point (eyes, eyebrow, nose, face etc.), face shape information is preferably presented.
3rd, Face datection proposed by the present invention and alignment schemes to illumination, attitude, stronger robustness is blocked, to follow-up
The performance of recognition of face and face verification has lifting higher.
Brief description of the drawings
Fig. 1 is based on flow chart of the present invention;
Fig. 2 is face detection part training pattern figure;
Fig. 3 is face feature point position portion training pattern figure;
Fig. 4 is overall fine setting illustraton of model;
Fig. 5 is that human face characteristic point returns device result schematic diagram;
Fig. 6 is face alignment result schematic diagram;
Fig. 7 is schematic flow sheet of the present invention.
Specific embodiment
Fig. 1 gives the overall process flow of Face datection and alignment schemes based on shared convolution feature, with reference to
Other accompanying drawings and specific embodiment further illustrate the present invention.
The present invention provides Face datection and alignment schemes based on shared convolution feature, and key step is described below:
1st, off-line training step
1) Face datection
The first step is first trained to Face datection part, and Fig. 2 institutes representation model is instructed using stochastic gradient descent method
Practice, it is 0.01 to set initial learning rate α, and total iterations is 80000 times, 50000 10000 adjustment of every iteration later of iteration
Learning rate α is original 0.1 times, and the method for this progressively regularized learning algorithm rate is conducive to model to converge to more excellent solution.
2) facial Feature Localization
Second step is trained to facial positioning feature point part, as shown in figure 3, before training is started, retaining first
The extraction convolution characteristic parameter of Face datection training pattern is walked, convolution feature afterbody continues to connect facial positioning feature point portion
Point.Similarly, input picture extracts convolution feature after being operated by multilayer convolution pondization, then by area-of-interest pond
Layer extracts human face region feature, and feature one-dimensional then is connected into full articulamentum, and finally facial characteristic point position is returned.
For the regression model of face feature point, this problem is equally trained using gradient descent method, and initial learning rate α is 0.001
(0.1 times in first step Face datection), total iterations is 80000 times, and iteration is 50000 times often adjust for 10000 times later by iteration
Whole learning rate α is original 0.1 times, and finally 68 characteristic points to face are returned.
3) overall fine setting
3rd step is finely adjusted to block mold, as shown in figure 4, before training is started, extracting convolution features code insurance
The parameter after second step face feature point regression training is stayed, Face datection part retains the ginseng after the training of first step Face datection
Number, face feature point returns the parameter after part retains second step face feature point regression training.On this basis, continue to mould
Type is trained, and initial learning rate α is 0.001, and total iterations is 80000 times, and iteration 50000 times is later per iteration 10000
Secondary regularized learning algorithm rate α is original 0.1 times.
2nd, characteristic extraction step
Convolution feature extraction, convolutional layer activation primitive are carried out to input picture first by a deep layer convolutional network structure
Using RELU activation primitives, pond layer uses maximum pond.The input picture of given arbitrary dimension size, by multilayer convolution
Pondization is operated, and the convolution feature of output is fixed size 512 on depth axle, size and the size of input picture spatially
Proportional relation.Convolution characteristic extraction part has 16 layers, including 12 layers of convolutional layer and 4 layers of pond layer, the volume of convolutional layer
Product core size is 3x3, and convolution stride is 1, and zero padding number is 2, and the pond core size of pond layer is
2x2, pond layer stride is 2.
3rd, Face datection step
This step needs training in advance what a human-face detector, after input picture is by the operation of multilayer convolution pondization,
Convolution feature is extracted, human-face detector, including region suggestion network and area-of-interest pond layer is passed to;Advise network in region
Input is characterized as with convolution, output is probably the candidate region of destination object, then introduces area-of-interest pond layer and extract
The corresponding convolution feature in candidate region, two classification is carried out to each candidate region and is returned with bounding box.The output valve of classified part
Represent whether candidate region is face, if face then exports 1, otherwise export 0;Bounding box returns part output human face region
Position in the input image, if classified part output 0, ignores the output that bounding box returns part.
4th, face feature point positioning step
Facial feature points detection step extracts traditional characteristic (the local binary pattern features of facial image first
With Haar-like features) and convolutional neural networks feature, then train a human face characteristic point to return device;The recurrence device is received
Fixed size facial image, exports human face characteristic point position (eyebrow, canthus, nose, the corners of the mouth etc., such as Fig. 5 in the facial image
It is shown).
5th, face alignment step
The input of this step is facial image and its corresponding human face characteristic point position of facial feature points detection step output
Put, the facial image after output alignment.It is right that face alignment step is carried out according to two fixed human face characteristic points (such as left and right canthus)
Together, facial image is rotated and scaling treatment, the facial image after output alignment, back in the facial image after alignment
The position of rapid selected fixed characteristic points is by constant (as shown in Figure 6).
The technology contents that the present invention is not elaborated belong to the known technology of those skilled in the art.
Although being described to illustrative specific embodiment of the invention above, in order to the technology people of this technology neck
Member understands the present invention, it should be apparent that the invention is not restricted to the scope of specific embodiment, to the ordinary skill of the art
For personnel, as long as various change is in appended claim restriction and the spirit and scope of the present invention for determining, these changes
Change is it will be apparent that all utilize the innovation and creation of present inventive concept in the row of protection.
Claims (5)
1. a kind of Face datection and alignment schemes based on shared convolution feature, it is characterised in that including following four step:
Step (1), convolution feature extraction is carried out to input picture first by a deep layer convolutional network structure, convolutional layer activation
Function uses RELU activation primitives, pond layer to use maximum pond;The input picture of given arbitrary dimension size, by multilayer
Convolution pondization is operated, and the convolution feature of output is fixed size 512 on depth axle, size spatially and input picture
The proportional relation of size;
After step (2), input picture are operated by multilayer convolution pondization, convolution feature is extracted, train human-face detector, including
Advise network and area-of-interest pond layer in region;Region advises that network is characterized as input with convolution, and output is probably target pair
The candidate region of elephant, then introduces area-of-interest pond layer and extracts the corresponding convolution feature in candidate region, to each candidate
Region carries out two classification and is returned with bounding box;The output valve of classified part represents whether candidate region is face, if face
1 is then exported, 0 is otherwise exported;Bounding box returns part output human face region position in the input image, if classified part is exported
0, then ignore the output that bounding box returns part;
Step (3), input picture extract convolution feature after being operated by multilayer convolution pondization, then by area-of-interest pond
Change layer and extract human face region feature, training face feature point returns device, will feature one-dimensional connect full articulamentum, to facial special
Levy a position to be returned, export the human face characteristic point position in the facial image, including eyebrow, canthus, nose, the corners of the mouth;
Step (4), the facial image finally according to step (3) output and its corresponding human face characteristic point position, choose two admittedly
Human face characteristic point is determined as fixed point, facial image is rotated and scaling treatment, the facial image after output alignment, after alignment
Facial image in the position of selected fixed characteristic points will be constant.
2. Face datection and alignment schemes based on shared convolution feature according to claim 1, it is characterised in that:Step
(1) convolution feature described in for step (2) and step (3) while use, both solved face feature point location algorithm excessively according to
Rely the problem of Face datection effect, reduce again and compute repeatedly, improve the time efficiency of Face datection and alignment algorithm.
3. Face datection and alignment schemes based on shared convolution feature according to claim 1, it is characterised in that:Step
(2) human-face detector described in is to be input into training by extracting the feature of image to obtain, and the feature of image is by sharing
Convolutional neural networks are extracted and obtained.
4. Face datection and alignment schemes based on shared convolution feature according to claim 1, it is characterised in that:Step
(3) it is to be input into training by extracting the feature of image to obtain that face feature point described in returns device, the feature of image by
Shared convolutional neural networks are extracted and obtained.
5. Face datection and alignment schemes based on shared convolution feature according to claim 1, it is characterised in that:Step
(4) being rotated to facial image described in and scaled, located based on ensureing that face or so 2 points of the tail of the eye is constant
Reason.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710135007.8A CN106909909B (en) | 2017-03-08 | 2017-03-08 | Face detection and alignment method based on shared convolution characteristics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710135007.8A CN106909909B (en) | 2017-03-08 | 2017-03-08 | Face detection and alignment method based on shared convolution characteristics |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106909909A true CN106909909A (en) | 2017-06-30 |
CN106909909B CN106909909B (en) | 2021-02-02 |
Family
ID=59186826
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710135007.8A Active CN106909909B (en) | 2017-03-08 | 2017-03-08 | Face detection and alignment method based on shared convolution characteristics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106909909B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107808129A (en) * | 2017-10-17 | 2018-03-16 | 南京理工大学 | A kind of facial multi-characteristic points localization method based on single convolutional neural networks |
CN108182384A (en) * | 2017-12-07 | 2018-06-19 | 浙江大华技术股份有限公司 | A kind of man face characteristic point positioning method and device |
CN108460787A (en) * | 2018-03-06 | 2018-08-28 | 北京市商汤科技开发有限公司 | Method for tracking target and device, electronic equipment, program, storage medium |
CN108564022A (en) * | 2018-04-10 | 2018-09-21 | 深圳市唯特视科技有限公司 | A kind of more personage's pose detection methods based on positioning classification Recurrent networks |
CN109145783A (en) * | 2018-08-03 | 2019-01-04 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating information |
CN109271974A (en) * | 2018-11-16 | 2019-01-25 | 中山大学 | A kind of lightweight face joint-detection and recognition methods and its system |
CN109635674A (en) * | 2018-11-22 | 2019-04-16 | 深圳市唯特视科技有限公司 | A kind of face alignment method of the dendron shape convolutional neural networks adapted to based on posture |
CN110837785A (en) * | 2019-10-28 | 2020-02-25 | 北京影谱科技股份有限公司 | Face image detection method and device, computing equipment and storage medium |
US11699240B2 (en) | 2018-03-06 | 2023-07-11 | Beijing Sensetime Technology Development Co., Ltd. | Target tracking method and apparatus, and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104504362A (en) * | 2014-11-19 | 2015-04-08 | 南京艾柯勒斯网络科技有限公司 | Face detection method based on convolutional neural network |
CN104992167A (en) * | 2015-07-28 | 2015-10-21 | 中国科学院自动化研究所 | Convolution neural network based face detection method and apparatus |
CN105512638A (en) * | 2015-12-24 | 2016-04-20 | 黄江 | Fused featured-based face detection and alignment method |
CN105550658A (en) * | 2015-12-24 | 2016-05-04 | 蔡叶荷 | Face comparison method based on high-dimensional LBP (Local Binary Patterns) and convolutional neural network feature fusion |
CN105760836A (en) * | 2016-02-17 | 2016-07-13 | 厦门美图之家科技有限公司 | Multi-angle face alignment method based on deep learning and system thereof and photographing terminal |
CN106372630A (en) * | 2016-11-23 | 2017-02-01 | 华南理工大学 | Face direction detection method based on deep learning |
-
2017
- 2017-03-08 CN CN201710135007.8A patent/CN106909909B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104504362A (en) * | 2014-11-19 | 2015-04-08 | 南京艾柯勒斯网络科技有限公司 | Face detection method based on convolutional neural network |
CN104992167A (en) * | 2015-07-28 | 2015-10-21 | 中国科学院自动化研究所 | Convolution neural network based face detection method and apparatus |
CN105512638A (en) * | 2015-12-24 | 2016-04-20 | 黄江 | Fused featured-based face detection and alignment method |
CN105550658A (en) * | 2015-12-24 | 2016-05-04 | 蔡叶荷 | Face comparison method based on high-dimensional LBP (Local Binary Patterns) and convolutional neural network feature fusion |
CN105760836A (en) * | 2016-02-17 | 2016-07-13 | 厦门美图之家科技有限公司 | Multi-angle face alignment method based on deep learning and system thereof and photographing terminal |
CN106372630A (en) * | 2016-11-23 | 2017-02-01 | 华南理工大学 | Face direction detection method based on deep learning |
Non-Patent Citations (5)
Title |
---|
HUAIZU JIANG 等: "Face Detection with the Faster R-CNN", 《ARXIV》 * |
SHAOQING REN 等: "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks", 《ARXIV》 * |
SHARTOO: "RCNN,Fast RCNN,Faster RCNN总结", 《HTTP_SHARTOO.GITHUB.IO_RCNN-SERIES》 * |
YI SUN 等: "Deep Convolutional Network Cascade for Facial Point Detection", 《CVPR 2013》 * |
王华锋 等: "视频中适应光照可变情况下的人脸识别方法", 《模式识别与人工智能》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107808129A (en) * | 2017-10-17 | 2018-03-16 | 南京理工大学 | A kind of facial multi-characteristic points localization method based on single convolutional neural networks |
CN108182384A (en) * | 2017-12-07 | 2018-06-19 | 浙江大华技术股份有限公司 | A kind of man face characteristic point positioning method and device |
CN108182384B (en) * | 2017-12-07 | 2020-09-29 | 浙江大华技术股份有限公司 | Face feature point positioning method and device |
WO2019170023A1 (en) * | 2018-03-06 | 2019-09-12 | 北京市商汤科技开发有限公司 | Target tracking method and apparatus, and electronic device and storage medium |
CN108460787A (en) * | 2018-03-06 | 2018-08-28 | 北京市商汤科技开发有限公司 | Method for tracking target and device, electronic equipment, program, storage medium |
US11699240B2 (en) | 2018-03-06 | 2023-07-11 | Beijing Sensetime Technology Development Co., Ltd. | Target tracking method and apparatus, and storage medium |
US11216955B2 (en) | 2018-03-06 | 2022-01-04 | Beijing Sensetime Technology Development Co., Ltd. | Target tracking methods and apparatuses, electronic devices, and storage media |
CN108564022A (en) * | 2018-04-10 | 2018-09-21 | 深圳市唯特视科技有限公司 | A kind of more personage's pose detection methods based on positioning classification Recurrent networks |
CN109145783B (en) * | 2018-08-03 | 2022-03-25 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating information |
CN109145783A (en) * | 2018-08-03 | 2019-01-04 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating information |
CN109271974A (en) * | 2018-11-16 | 2019-01-25 | 中山大学 | A kind of lightweight face joint-detection and recognition methods and its system |
CN109635674A (en) * | 2018-11-22 | 2019-04-16 | 深圳市唯特视科技有限公司 | A kind of face alignment method of the dendron shape convolutional neural networks adapted to based on posture |
CN110837785A (en) * | 2019-10-28 | 2020-02-25 | 北京影谱科技股份有限公司 | Face image detection method and device, computing equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN106909909B (en) | 2021-02-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106909909A (en) | A kind of Face datection and alignment schemes based on shared convolution feature | |
CN105512638B (en) | A kind of Face datection and alignment schemes based on fusion feature | |
CN104992167B (en) | A kind of method for detecting human face and device based on convolutional neural networks | |
CN104573731B (en) | Fast target detection method based on convolutional neural networks | |
CN107844795B (en) | Convolutional neural network feature extraction method based on principal component analysis | |
CN105512624B (en) | A kind of smiling face's recognition methods of facial image and its device | |
CN104978580B (en) | A kind of insulator recognition methods for unmanned plane inspection transmission line of electricity | |
CN107977609A (en) | A kind of finger vein identity verification method based on CNN | |
CN106778506A (en) | A kind of expression recognition method for merging depth image and multi-channel feature | |
CN104123545B (en) | A kind of real-time human facial feature extraction and expression recognition method | |
CN109145939A (en) | A kind of binary channels convolutional neural networks semantic segmentation method of Small object sensitivity | |
CN105868785A (en) | Image identification method based on convolutional neural network and image identification system thereof | |
CN107886064A (en) | A kind of method that recognition of face scene based on convolutional neural networks adapts to | |
CN107679491A (en) | A kind of 3D convolutional neural networks sign Language Recognition Methods for merging multi-modal data | |
CN109359526A (en) | A kind of face pose estimation, device and equipment | |
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 | |
CN108921019A (en) | A kind of gait recognition method based on GEI and TripletLoss-DenseNet | |
CN108182397A (en) | A kind of multiple dimensioned face verification method of multi-pose | |
CN107844760A (en) | Three-dimensional face identification method based on curved surface normal direction component map Neural Networks Representation | |
CN106778785A (en) | Build the method for image characteristics extraction model and method, the device of image recognition | |
CN106446773A (en) | Automatic robust three-dimensional face detection method | |
CN108009472A (en) | A kind of finger back arthrosis line recognition methods based on convolutional neural networks and Bayes classifier | |
CN107169954A (en) | A kind of image significance detection method based on parallel-convolution neutral net | |
CN104636732A (en) | Sequence deeply convinced network-based pedestrian identifying method |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20210728 Address after: 310052 room 1202, block a, No. 66, Yueming Road, Xixing street, Binjiang District, Hangzhou City, Zhejiang Province Patentee after: Ruishi netcloud (Hangzhou) Technology Co.,Ltd. Address before: 100083 room 1010, Xueyuan international building, No.1 Zhichun Road, Haidian District, Beijing Patentee before: Wang Huafeng |