CN108986094A - For the recognition of face data automatic update method in training image library - Google Patents
For the recognition of face data automatic update method in training image library Download PDFInfo
- Publication number
- CN108986094A CN108986094A CN201810805051.XA CN201810805051A CN108986094A CN 108986094 A CN108986094 A CN 108986094A CN 201810805051 A CN201810805051 A CN 201810805051A CN 108986094 A CN108986094 A CN 108986094A
- Authority
- CN
- China
- Prior art keywords
- face
- image data
- data
- training
- picture
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/50—Maintenance of biometric data or enrolment thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30168—Image quality inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
- G06T2207/30201—Face
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of for the recognition of face data automatic update method in training image library, comprising the following steps: Step 1: obtaining face picture data using crawler technology；Step 2: being judged with the facial angle that three-dimension altitude angle comes in picture, the picture too big with positive face deviation is filtered out Step 3: extracting five characteristic points in picture using MTCNN technology, affine transformation is carried out between five characteristic points initially set, to achieve the effect that face is aligned；Step 4: the image data after face is aligned is automatically added in training image library.The present invention improves rich, the accuracy of training data in recognition of face training process, ensure that deep neural network model training accuracy rate and training speed, and shorten the cycle of training of deep neural network model.
Description
Technical field
The present invention relates to a kind of for the recognition of face data automatic update method in training image library, belongs to recognition of face
Technical field.
Background technique
In the face recognition technology based on deep neural network model, the acquisition of image data has model training
Very important effect.For our existing high-accuracies and efficient model, need the data of a collection of high quality as defeated
Enter, to play the maximum efficiency of the model.Meanwhile we are also required to database itself powerful anti-interference ability, and
The ability for automatically updating and filtering, to cope with the characteristics of current artificial intelligence field data information amount is exploded.
In the prior art for the data of recognition of face substantially from online public data collection or data with existing, these
Data can not all carry out updating again automatically after acquisition.In addition, these data are to the clarity of picture and rectifying for face
Property description be also at mostly qualitatively within the scope of.Since there are the above limitations for these data, so that existing model
Maximum performance is unable to get in terms of predictablity rate and predetermined speed.
Summary of the invention
The invention solves technical problems to be: providing and a kind of improves the rich, accuracy for people of training data
Face identification use training image library data automatic update method, this method ensure that deep neural network model training accuracy rate and
Training speed, and shorten the cycle of training of deep neural network model.
In order to solve the above-mentioned technical problem, technical solution proposed by the present invention is: one kind is for recognition of face training figure
As the data automatic update method in library, comprising the following steps:
Step 1: obtaining image data
Obtain online face picture data automatically according to preset rules using crawler technology；
Step 2: filtering face picture data
Its left eye, right eye, nose, the crown and mouth totally five features are extracted with MTCNN algorithm to each width image data
Point calculates the distance between eyes midpoint and the crown s in each width image data1With the distance s at eyes midpoint to mouth midpoint, from
Nose makees vertical line to straight line where eyes, and the distance between intersection point and eyes are respectively l1、l2, then the distance between eyes is l=
l1+l2, meet 0.4s < s when simultaneously1< 0.6s and 0.4l < l1When < 0.6l, retain the face image data；
Step 3: being aligned to the face in image data
Assuming that the initial coordinate for five characteristic points that each width image data is extracted with MTCNN algorithm is respectively (x1',
y1')、(x2',y2')、(x3',y3')、(x4',y4') and (x5',y5')；
Five initial coordinate point (x are determined in painting canvas1,y1)、(x2,y2)、(x3,y3)、(x4,y4) and (x5,y5), so that
This five points have symmetry in painting canvas；
It is built between the five initial coordinate points determined in five characteristic points extracted in each width image data and painting canvas
Vertical affine transformation equation, carrying out posture correction to each width image data according to affine transformation equation is aligned face；
Step 4: the image data after face is aligned is automatically added in training image library.
It should be noted that MTCNN algorithm is the prior art, can refer to by Kaipeng Zhang, Zhanpeng
" the Joint Face Detection and Alignment that Zhang, Zhifeng Li, Yu Qiao were delivered in 2016
using Multi-task Cascaded Convolutional Networks》。
The present invention is aligned to the automatic crawler technology of network, picture quality detection algorithm, face critical point detection and face
An integrative solution is designed and is integrated on basis, and by above-mentioned dispersion and many and diverse process is comprehensive as efficiently convenient easy
Data processing tools.Present invention encompasses the whole flow process of training deep learning neural network, and can be deployed in this
In ground computing resource, both solves the problems, such as Information Security, also perfect whole flow process.
The present invention uses crawler technology in terms of obtaining data, to make the available real-time update of facial image；Right
Using pitch angle (Pitch), yaw angle (Yaw), roll angle (Roll) three-dimension altitude angle come to figure in terms of the quality testing of picture
The angle of face is judged as in；In the context of detection to image definition, detected using Laplce's ambiguity to image
It is analyzed, in addition, we are by the coordinate to facial key point and target position, it is affine to carry out facial image progress
Transformation, to reach the alignment effect of face.It is improved by these, we make model not predictablity rate and to test the speed in advance
Degree has the stronger adaptability to data and faster cycle of training as under the premise of cost.These pass through modified figure
Sector-meeting makes the effect of model training higher, and recognition speed and accuracy rate also have certain guarantee.Thereafter further according to us
Needs, select different loss functions to be trained, the speed or accuracy rate identified to it does further promotion.
The further improvement of above-mentioned technical proposal is: before executing step 2, carrying out root using Laplce's variance algorithm
Image data is screened according to clarity, i.e., each width face picture data is accorded with by Laplace's operationInto
Then row convolution algorithm calculates its variance to the result after convolution algorithm again, if the variance is less than preset threshold 100, sentence
The image data that breaks is judged as blurred picture, and rejects the blurred picture.
The further improvement of above-mentioned technical proposal is: the method for affine transformation equation is established in step 3 are as follows:
Mathematic(al) representation according to two dimensional affine transformation is x=ax'+by'+c, y=dx'+ey'+f, and wherein x', y' are to become
The coordinate value of preceding pixel is changed, x, y are the coordinate values of pixel after transformation, and a, b, c, d, e, f are affine transformation parameters；
When there are 5 groups of coordinates, mathematic(al) representation x is obtainedi=axi'+byi'+c,yi=dxi'+eyi'+f, wherein i=1,2,
3,4,5。
This 5 groups of mathematic(al) representations can be merged in a matrix equation:
A, b, c, d, e, this six unknown numbers of f, to obtain affine transformation side can be acquired by arbitrarily choosing three characteristic points
Journey.
Specific embodiment
Embodiment
The data automatic update method for recognition of face training image library of the present embodiment, comprising the following steps:
Step 1: obtaining image data
Obtain online face picture data automatically according to preset rules using crawler technology.Crawler technology is existing skill
Art is the program or script of a kind of information that acquisition automatic according to certain rules is online.Its principle is by webpage
The interception of source code, to obtain the source code field information of our required parts.In the present embodiment, face picture is for we
Required information, we find feature code corresponding to picture in web page source code, such as comprising ' image ' and ' face '
The row of printed words intercepts it as our required objects.Other feature codes can certainly be chosen.
Step 2: filtering face picture data
In the new image data obtained by crawler, it is understood that there may be the problems such as picture angle and too big positive face deviation, need
It is screened, be filtered out and picture that positive face deviation is too big.
Using 3 d pose pitch angle (pitch), yaw angle (yaw), roll angle (roll) to face figure in the present embodiment
Sheet data is judged, because the angular deviation on rolling angular direction is planar, it is possible to pass through imitating later
Transformation is penetrated to be modified.The definition of 3 d pose is the prior art, can refer to related data, repeats no more.
Its left eye, right eye, nose, the crown and mouth totally five features are extracted with MTCNN algorithm to each width image data
Point calculates the distance between eyes midpoint and the crown s in each width image data1With the distance s at eyes midpoint to mouth midpoint, from
Nose makees vertical line to straight line where eyes, and the distance between intersection point and eyes are respectively l1、l2, then the distance between eyes is l=
l1+l2, meet 0.4s < s when simultaneously1< 0.6s (pitching angular direction) and 0.4l < l1When (the yaw angular direction) < 0.6l, protect
Stay the face image data.
Step 3: being aligned to the face in image data
Assuming that the initial coordinate for five characteristic points that each width image data is extracted with MTCNN algorithm is respectively (x1',
y1')、(x2',y2')、(x3',y3')、(x4',y4') and (x5',y5')；
Five initial coordinate point (x are determined in painting canvas1,y1)、(x2,y2)、(x3,y3)、(x4,y4) and (x5,y5), so that
This five points have symmetry in painting canvas；
It is built between the five initial coordinate points determined in five characteristic points extracted in each width image data and painting canvas
Vertical affine transformation equation, carrying out posture correction to each width image data according to affine transformation equation is aligned face.
The method of affine transformation equation is established in the present embodiment are as follows:
Mathematic(al) representation according to two dimensional affine transformation is x=ax'+by'+c, y=dx'+ey'+f, and wherein x', y' are to become
The coordinate value of preceding pixel is changed, x, y are the coordinate values of pixel after transformation, and a, b, c, d, e, f are affine transformation parameters；
When there are 5 groups of coordinates, mathematic(al) representation x is obtainedi=axi'+byi'+c,yi=dxi'+eyi'+f, wherein i=1,2,
3,4,5。
This 5 groups of mathematic(al) representations can be merged in a matrix equation:
A, b, c, d, e, this six unknown numbers of f, to obtain affine transformation side can be acquired by arbitrarily choosing three characteristic points
Journey.
But more characteristic points can be generally taken in the practical solution procedure of affine transformation equation, and (there are five special altogether in this example
Point is levied, thus ten equations can be obtained), equation number is allowed in this way more than unknown number number, to obtain an overdetermination side
Least square method can be used at this time and solve a, b, c for journey, and d, e, f is to reduce influence caused by error.
Step 4: the image data after face is aligned is automatically added in training image library.
The present embodiment can also make following improve: before executing step 2, fuzzy picture be rejected first, for picture
The judgement of clarity generally all relatively directly, is carried out in the present embodiment using Laplce's variance algorithm come the clarity to picture
Detection.A threshold value 100 is given first, and for each given picture, we are accorded with Laplace's operation
Convolution algorithm is carried out to the picture, variance calculating then is carried out to it again, if the variance less than 100, judges the picture quilt
It is judged as fuzzy, is otherwise judged as clearly.
After above-mentioned sequence of operations, the excessive and excessively fuzzy picture of misalignment angle may filter that, and to surplus
Remaining picture carries out affine transformation, obtains a series of picture of alignment of face as input and carries out recognition of face.These warps
Cross that the effect of modified picture can make model training is higher, and recognition speed and accuracy rate also have certain guarantee.Thereafter
Further according to our needs, different loss functions is selected to be trained, the speed or accuracy rate identified to it is done further
It is promoted.
The present invention is not limited to the above embodiment the specific technical solution, and in addition to the implementation, the present invention may be used also
To there is other embodiments.For those skilled in the art, all within the spirits and principles of the present invention, made
The technical solution of the formation such as what modification, equivalent replacement, improvement, should all be included in the protection scope of the present invention.
Claims (3)
1. a kind of for the recognition of face data automatic update method in training image library, comprising the following steps:
Step 1: obtaining image data
Obtain online face picture data automatically according to preset rules using crawler technology；
Step 2: filtering face picture data
Its left eye, right eye, nose, the crown and mouth totally five characteristic points are extracted with MTCNN algorithm to each width image data, are counted
Calculate the distance between eyes midpoint and the crown s in each width image data1With the distance s at eyes midpoint to mouth midpoint, from nose
Straight line where to eyes makees vertical line, and the distance between intersection point and eyes are respectively l1、l2, then the distance between eyes is l=l1+
l2, meet 0.4s < s when simultaneously1< 0.6s and 0.4l < l1When < 0.6l, retain the face image data；
Step 3: being aligned to the face in image data
Assuming that the initial coordinate for five characteristic points that each width image data is extracted with MTCNN algorithm is respectively (x1',y1')、
(x2',y2')、(x3',y3')、(x4',y4') and (x5',y5')；
Five initial coordinate point (x are determined in painting canvas1,y1)、(x2,y2)、(x3,y3)、(x4,y4) and (x5,y5), so that this five
Point has symmetry in painting canvas；
It is imitative to being established between the five initial coordinate points determined in five characteristic points extracted in each width image data and painting canvas
Transformation equation is penetrated, carrying out posture correction to each width image data according to affine transformation equation is aligned face；
Step 4: the image data after face is aligned is automatically added in training image library.
2. the data automatic update method according to claim 1 for recognition of face training image library, feature exist
In: before executing step 2, image data is screened according to clarity using Laplce's variance algorithm, i.e., for each
Width face picture data are accorded with by Laplace's operationCarry out convolution algorithm, then again to convolution algorithm after
As a result its variance is calculated, if the variance is less than preset threshold 100, judges that the image data is judged as blurred picture, and
Reject the blurred picture.
3. the data automatic update method according to claim 1 for recognition of face training image library, feature exist
In establishing the method for affine transformation equation in step 3 are as follows:
Mathematic(al) representation according to two dimensional affine transformation is x=ax'+by'+c, y=dx'+ey'+f, before wherein x', y' are transformation
The coordinate value of pixel, x, y are the coordinate values of pixel after transformation, and a, b, c, d, e, f are affine transformation parameters；
When there are 5 groups of coordinates, mathematic(al) representation x is obtainedi=axi'+byi'+c,yi=dxi'+eyi'+f, wherein i=1,2,3,4,
5.This 5 groups of mathematic(al) representations can be merged in a matrix equation:
A, b, c, d, e, this six unknown numbers of f, to obtain affine transformation equation can be acquired by arbitrarily choosing three characteristic points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810805051.XA CN108986094B (en) | 2018-07-20 | 2018-07-20 | Automatic data updating method for training image library for face recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810805051.XA CN108986094B (en) | 2018-07-20 | 2018-07-20 | Automatic data updating method for training image library for face recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108986094A true CN108986094A (en) | 2018-12-11 |
| CN108986094B CN108986094B (en) | 2021-08-17 |
Family
ID=64549170
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810805051.XA Active CN108986094B (en) | 2018-07-20 | 2018-07-20 | Automatic data updating method for training image library for face recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108986094B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635755A (en) * | 2018-12-17 | 2019-04-16 | 苏州市科远软件技术开发有限公司 | Face extraction method, apparatus and storage medium |
| CN110032970A (en) * | 2019-04-11 | 2019-07-19 | 深圳市华付信息技术有限公司 | Biopsy method, device, computer equipment and the storage medium of high-accuracy |
| CN110046554A (en) * | 2019-03-26 | 2019-07-23 | 青岛小鸟看看科技有限公司 | A kind of face alignment method and camera |
| CN110059607A (en) * | 2019-04-11 | 2019-07-26 | 深圳市华付信息技术有限公司 | Living body multiple detection method, device, computer equipment and storage medium |
| CN110059576A (en) * | 2019-03-26 | 2019-07-26 | 北京字节跳动网络技术有限公司 | Screening technique, device and the electronic equipment of picture |
| CN110096965A (en) * | 2019-04-09 | 2019-08-06 | 华东师范大学 | A kind of face identification method based on head pose |
| CN110176039A (en) * | 2019-04-23 | 2019-08-27 | 苏宁易购集团股份有限公司 | A kind of video camera adjusting process and system for recognition of face |
| CN110245577A (en) * | 2019-05-23 | 2019-09-17 | 复钧智能科技(苏州)有限公司 | Target vehicle recognition methods, device and Vehicular real time monitoring system |
| CN111079659A (en) * | 2019-12-19 | 2020-04-28 | 武汉水象电子科技有限公司 | Face feature point positioning method |
| CN113486858A (en) * | 2021-08-03 | 2021-10-08 | 济南博观智能科技有限公司 | Face recognition model training method and device, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104036276A (en) * | 2014-05-29 | 2014-09-10 | 无锡天脉聚源传媒科技有限公司 | Face recognition method and device |
| CN106503669A (en) * | 2016-11-02 | 2017-03-15 | 重庆中科云丛科技有限公司 | A kind of based on the training of multitask deep learning network, recognition methods and system |
| CN107832700A (en) * | 2017-11-03 | 2018-03-23 | 全悉科技（北京）有限公司 | A kind of face identification method and system |
-
2018
- 2018-07-20 CN CN201810805051.XA patent/CN108986094B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104036276A (en) * | 2014-05-29 | 2014-09-10 | 无锡天脉聚源传媒科技有限公司 | Face recognition method and device |
| CN106503669A (en) * | 2016-11-02 | 2017-03-15 | 重庆中科云丛科技有限公司 | A kind of based on the training of multitask deep learning network, recognition methods and system |
| CN107832700A (en) * | 2017-11-03 | 2018-03-23 | 全悉科技（北京）有限公司 | A kind of face identification method and system |
Non-Patent Citations (2)
| Title |
|---|
| 夏洋洋 等: "人脸识别背后的数据清理问题研究", 《智能系统学报》 * |
| 钱程: "基于深度学习的人脸识别技术研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635755A (en) * | 2018-12-17 | 2019-04-16 | 苏州市科远软件技术开发有限公司 | Face extraction method, apparatus and storage medium |
| CN110046554A (en) * | 2019-03-26 | 2019-07-23 | 青岛小鸟看看科技有限公司 | A kind of face alignment method and camera |
| CN110059576A (en) * | 2019-03-26 | 2019-07-26 | 北京字节跳动网络技术有限公司 | Screening technique, device and the electronic equipment of picture |
| CN110046554B (en) * | 2019-03-26 | 2022-07-12 | 青岛小鸟看看科技有限公司 | Face alignment method and camera |
| CN110096965A (en) * | 2019-04-09 | 2019-08-06 | 华东师范大学 | A kind of face identification method based on head pose |
| CN110032970A (en) * | 2019-04-11 | 2019-07-19 | 深圳市华付信息技术有限公司 | Biopsy method, device, computer equipment and the storage medium of high-accuracy |
| CN110059607A (en) * | 2019-04-11 | 2019-07-26 | 深圳市华付信息技术有限公司 | Living body multiple detection method, device, computer equipment and storage medium |
| CN110176039A (en) * | 2019-04-23 | 2019-08-27 | 苏宁易购集团股份有限公司 | A kind of video camera adjusting process and system for recognition of face |
| CN110245577A (en) * | 2019-05-23 | 2019-09-17 | 复钧智能科技(苏州)有限公司 | Target vehicle recognition methods, device and Vehicular real time monitoring system |
| CN111079659A (en) * | 2019-12-19 | 2020-04-28 | 武汉水象电子科技有限公司 | Face feature point positioning method |
| CN113486858A (en) * | 2021-08-03 | 2021-10-08 | 济南博观智能科技有限公司 | Face recognition model training method and device, electronic equipment and storage medium |
| CN113486858B (en) * | 2021-08-03 | 2024-01-23 | 济南博观智能科技有限公司 | Face recognition model training method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108986094B (en) | 2021-08-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108986094A (en) | For the recognition of face data automatic update method in training image library | |
| CN104008370B (en) | A kind of video face identification method | |
| CN107784294B (en) | Face detection and tracking method based on deep learning | |
| CN108805078A (en) | Video pedestrian based on pedestrian's average state recognition methods and system again | |
| CN109934195A (en) | A kind of anti-spoofing three-dimensional face identification method based on information fusion | |
| CN110135282B (en) | Examinee return plagiarism cheating detection method based on deep convolutional neural network model | |
| CN109858466A (en) | A kind of face critical point detection method and device based on convolutional neural networks | |
| CN102270308B (en) | Facial feature location method based on five sense organs related AAM (Active Appearance Model) | |
| CN105930767A (en) | Human body skeleton-based action recognition method | |
| CN109657612B (en) | Quality sorting system based on facial image features and application method thereof | |
| CN111274916A (en) | Face recognition method and face recognition device | |
| CN107153816A (en) | A kind of data enhancement methods recognized for robust human face | |
| CN107403168A (en) | A kind of facial-recognition security systems | |
| CN103473564B (en) | A kind of obverse face detection method based on sensitizing range | |
| CN105117692A (en) | Real-time face identification method and system based on deep learning | |
| CN106778796A (en) | Human motion recognition method and system based on hybrid cooperative model training | |
| CN109359526A (en) | A kind of face pose estimation, device and equipment | |
| CN106570491A (en) | Robot intelligent interaction method and intelligent robot | |
| CN108537143B (en) | A kind of face identification method and system based on key area aspect ratio pair | |
| CN110111316A (en) | Method and system based on eyes image identification amblyopia | |
| CN107301406A (en) | Fast face angle recognition method based on deep learning | |
| CN106127193B (en) | A kind of facial image recognition method | |
| CN107679450A (en) | Obstruction conditions servant's face recognition method based on deep learning | |
| Perveen et al. | Facial expression recognition using facial characteristic points and Gini index | |
| CN105893981A (en) | Human face posture correction 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 |