CN105046202B - Adaptive recognition of face lighting process method - Google Patents
Adaptive recognition of face lighting process method Download PDFInfo
- Publication number
- CN105046202B CN105046202B CN201510349830.XA CN201510349830A CN105046202B CN 105046202 B CN105046202 B CN 105046202B CN 201510349830 A CN201510349830 A CN 201510349830A CN 105046202 B CN105046202 B CN 105046202B
- Authority
- CN
- China
- Prior art keywords
- face
- illumination
- image
- classifier
- lighting process
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
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
-
- 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
- G06V40/10—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
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of adaptive recognition of face lighting process methods.This method passes through Adaboost method first and carries out Face detection, obtains the face part in picture；Illumination complicated dynamic behaviour is carried out to face after carrying out Face detection；Different lighting process methods is selected to carry out lighting process to face by setting complexity threshold.The present invention is handled input picture using above three step, finally obtains the facial image of removal illumination effect.The present invention can not only effectively inhibit light to change the influence generated to later period recognition of face, it avoids and is lacked under varying environment using robustness caused by the algorithm of same lighting process simultaneously, and can rapidly and accurately obtain normalized facial image after lighting process.
Description
Technical field
The invention belongs to recognition of face preconditioning technique, specifically a kind of adaptive recognition of face lighting process method.
Background technique
The research of recent years, the identification of controllable environment human face make great progress, but under natural situation, at
The variation of various uncontrollable factors always causes same face to generate very big variation as during.The wherein influence nothing of illumination
It doubts and has been known as one of bottleneck of face identification system.
Difference of the same facial image under different illumination conditions may be greater than different faces image and shine in same light
Under difference.Therefore, the solution of researcher's growing interest lighting issues is constantly combined existing method and improves, mentions
Go out that many discriminations are higher, the stronger method of universality.These methods or using digital image processing techniques to illumination into
Row pretreatment, or image is converted using mathematical theory combination image technique, to obtain the good image of illumination, again
Or modeling processing is carried out to facial image, it is intended to restore the information such as shape, the posture of face；On this basis again to face into
Row identification, mostly obtains preferable recognition performance.
Three kinds of recognition of face lighting process methods of mainstream have at present: histogram equalization, gradient face and illumination pretreatment chain.
Histogram equalization method is intensity profile a kind of for statistical analysis in the overall situation, in adjustment histogram, then regenerates
The method of histogram.This method primarily to selectively inhibit certain information (the less gray scale of pixel proportion),
And enhance the certain information (the more gray scale of pixel proportion) for additionally needing performance, that is, achieve the purpose that equalization, in light
In the case where all right, to the recognition effect robust that the later period carries out, and very simple it is quick (1.Stephen M.Pizer,
E.Philip Amburn,John D.Austin,Robert Cromartie,Ari Geselowitz.Adaptive
histogram equalization and its variations,Computer Vision,Graphics,and Image
Processing,1987,vol.39,pp.355-368.).Illumination pretreatment chain (TT) is by filtering gamma transformation, difference of Gaussian
The technologies such as wave combine one illumination pretreatment chain of composition, can eliminate light change bring adverse effect well, achieve non-
Normal high discrimination, the face image processing more serious suitable for illumination effect, but process is complex.Gradient face
(GRF) treatment process are as follows: the gradient component for seeking image X and Y-direction first does convolution by Gauss first derivative and image
It can be in the hope of；It is a kind of illumination secondly, obtaining the gradient component of image in X and the respective gradient component of Y-direction according to image
The method of robust, face identification rate in the extremely complex situation of illumination condition it is higher (2.Stephen M.Pizer,
E.Philip Amburn,John D.Austin,Robert Cromartie,Ari Geselowitz.Face
Recognition Under Varying Illumination Using Gradientfaces,Image Processing,
IEEE Transactions on,2009,vol.18,pp.2599-2606).Three kinds of methods have their own specific aim,
Each light conditions cannot all be carried out with the processing of robust.
Summary of the invention
The purpose of the present invention is to provide one kind to provide the judgement of different faces complexity, so that different algorithms be selected to take
Match, reaches the method that can realize the lighting process effect of robust under any light environment.
The technical solution for realizing the aim of the invention is as follows: a kind of adaptive recognition of face lighting process method, step
It is as follows:
The first step, locating human face carry out the positioning of face by Adaboost to input picture.A sample is organized first
This collection finds out the Haar feature (face and non-face) of sample set, trains classifier with feature: according in each layer of classifier
Whether sample classification correctly modifies the weight of sample, and is sent into next layer of classifier training, then by each layer of Multiple Classifier Fusion
It rises as final Adaboost Decision Classfication device.Then facial image is detected with classifier, realizes Face detection.
Second step establishes illumination subspace, for three kinds of possible light conditions, organizes respectively multiple by Gaussian Blur
The sample set of processing is trained strong classifier made of obtaining the cascade of several Weak Classifiers to sample set.It will pass through
The facial image of first step processing is put into classifier, and the illumination complexity of face is obtained.
Third step, it is serious if illumination effect according to the judgement that second step is classified, then select gradient face to carry out at illumination
Reason selects illumination pretreatment chain if moderate, and histogram equalization well can be then directly selected if Irradiance and carries out light
According to processing.
Compared with prior art, the present invention its remarkable advantage: (1) at the human face light for having merged a variety of existing excellents
Reason method can normalize the lighting process that robust is carried out containing facial image.(2) by sentencing to different light conditions
Disconnected, the different targetedly lighting process methods of selection carry out unitary of illumination processing to containing facial image.So that lighting process has
There are very strong robustness and anti-noise ability.(3) concept for introducing human face light complexity has merged Face datection cascade point
The thought of class device can select lighting process algorithm according to different faces light conditions.
Detailed description of the invention
Fig. 1 is the flow chart of the adaptive recognition of face lighting process method of the present invention.
Fig. 2 is common Haar feature appearance schematic diagram.
Fig. 3 is Gaussian Blur smoothed image schematic diagram.
Fig. 4 is cascade illumination complexity classifier schematic diagram.
Fig. 5 is gradient face lighting process method schematic diagram.
Fig. 6 is illumination pretreatment chain method schematic diagram.
Fig. 7 is histogram equalization schematic diagram.
Specific embodiment
Present invention is further described in detail with reference to the accompanying drawing.
In conjunction with Fig. 1, the adaptive recognition of face lighting process method of the present invention, steps are as follows:
The first step carries out Face detection processing to input picture: one sample set of tissue extracts the Haar feature of picture,
Suitable Haar feature is then picked out by Adaboost algorithm to form detector, people is carried out containing facial image to input
Face detection, obtains facial image.
(1) Haar feature appearance as shown in Figure 2 is common Haar feature extraction, these feature are in face and non-
The position upper value of face characterizes different.By the pixel in white area and subtracts the pixel in black region and obtain a mould
The value of version.The integrated value ii (x, y) of certain point A (x, y) is defined as in imageDefining s (x, y) is
The sum of upward all pixels of A (x, y) and its direction y, in this way, each pixel calculates ii (x, y) and s in the image of m*n size
Entire integrogram matrix can be obtained in (x, y), m*n*2 calculating.
(2) Weak Classifier is chosen.
Wherein f is characterized, and θ is threshold value, and p is the direction of the sign of inequality, and x is a detection child window.
Wherein wt,iFor the weight of t-th of classifier ith feature, qt,iIt is the weight after normalization.
The weighting fault rate ε of all features is calculated for each feature f, one Weak Classifier h of trainingf,
εf=∑iqi|h(xi,f,p,θ)-yi|
Choosing has minimal error rate εfWeak Classifier hi, and adjust the weight of sample
Wherein ei=0 indicates xiCorrectly classified, it is on the contrary then for mistake classification,
(3) it is cascaded into strong classifier
Wherein
Image is detected, adjusts position and the ratio of detection window, constantly to look for face and be cut.
Second step, human face light complicated dynamic behaviour.By the foundation of illumination subspace, to facial image illumination complexity into
Row measurement classification, selects according to the recognition of face lighting process that different illumination complexities enters third step.Detailed process is such as
Under:
(1) sample set is established, which includes the facial image under multiple various illumination, and the light conditions of image are carried out
Classification: 1. almost without illumination effect, face in uniform light；2. a small amount of illumination effect, such as a small amount of polarisation, facial contour and edge can
See；3. illumination effect is serious, if face shadow occlusion range is big, overexposure etc..
(2) Gaussian Blur is carried out to sample set picture:
Wherein x, y respectively represent the position of coordinate points, and σ indicates smooth degree, this method selected parameter σ=10.0.It is flat
Sliding facial image, hides its apparent feature, obtains the face matrix of significant face's comparison of light and shade.Smooth effect is as shown in Figure 3.
(3) Haar feature classifiers thought used in first step Face datection is used for reference, to the sample set of a large amount of Gaussian Blurs
It is trained, obtains many Weak Classifiers, multiple Weak Classifiers are connected, obtain a strong classifier and come to light conditions
Classify.
1. choosing Weak Classifier.
Wherein f is characterized, and θ is threshold value, and p is the direction of the sign of inequality, and x is a detection child window.
Wherein wt,iFor the weight of t-th of classifier ith feature, qt,iIt is the weight after normalization.
2. calculating the weighting fault of all images for each feature f, one Weak Classifier h of training in every one kind
Rate εf,
εf=∑iqi|h(xi,f,p,θ)-yi|
Choosing has minimal error rate εfWeak Classifier hi, and adjust the weight of sample
Wherein ei=0 indicates xiCorrectly classified, it is on the contrary then for mistake classification,
3. being cascaded into three groups of strong classifiers
Wherein
(4) cutting facial image to be sorted is put into the strong classifier of third step composition, if first layer classifier obtains
The result arrived is 0, then enters next layer of judgement, if obtained result is 1, obtains the classification of the image, enters next
Step.
Third step, recognition of face lighting process:
(1) if image classification is that illumination effect is serious, Systematic selection gradient face.Detailed process is as follows:
1. seeking the gradient component of image X and Y-direction first, gradient transform definition is
WhereinWherein σ is standard deviation, x direction gradient and y direction gradient meter
It is
WhereinIt is expressed as x, the derivative in the direction y.
2. the calculating of eigenface are as follows:
(2) if image classification is by a small amount of illumination effect, Systematic selection illumination pretreatment chain handles illumination effect, it
It is by integrating the light irradiation preprocess methods such as Gamma correction, contrast equalization, to have obtained an illumination pretreatment
Chain.Detailed process is as follows:
1. using Iγ(γ > 0) or log (I) (γ=0) replace image grayscale I, and as γ ∈ [0,1] and γ is a use
Family custom parameter.One index is that the power law of γ and γ ∈ [0,0.5] range is a good compromise.Present invention selection
γ=0.2 is used as default setting.
2. in order to reach illumination robustness, this method readjusts image ash using contrast equalization after Gamma correction
It spends within the scope of an expectation, while can guarantee overall contrast of the image in the overall situation again.Process is as follows:
Wherein, a is a strength cake compressibility, and can be reduced individual larger gray value brings influences；τ is a threshold value,
It is used to intercept larger gray value in equalization；Mean () is the averaging operation operator to whole image.The present invention is here
Take α=0.1 and τ=10 as default value.
3. finally, this method completes the inhibition to gray value of image maximum using a kind of strategy of Nonlinear Mapping,
That is I (x, y) ← τ tanh (I (x, y)/τ), wherein tanh () is hyperbolic tangent function, the value range of I be limited in (- τ,
τ)。
(3) if almost without illumination effect, this system selection histogram equalization method is handled.Steps are as follows:
1. calculating the probability density function of image, i.e., tonal range is become 0~1 from 0~255, normalized.Use Pr
(r) probability density function of original image, P are indicateds(S) it indicates to equalize later probability density function, r, s are respectively represented directly
The gray value of side figure equalization front and back, and belong to [0,1].It is available by the knowledge of probability theoryWherein T-1(s) inverse transform function for being T (r).And have
2. calculating Cumulative Distribution Function.The probability occurred for some gray-level pixels are as follows:
Pr(rk)=nk/N
Wherein Pr(rk) it is the probability that the k-th gray-level pixels of original image occur, rkIt is k-th of gray level, i.e. current color
Rank, range are [0,1], and N is total number of image pixels, nkFor rkPixel quantity.
3. gray-level histogram equalization formula are as follows:
Wherein, T (rk) indicates the transfer function of k-th of gray level of original image.∑ Pr (rk) indicates the ash of 0~k
Spend grade probability of occurrence cumulative addition.Because s is normalized numerical value (s ∈ [0,1]), 0~255 color value is converted to, is needed
It will be multiplied by 255, i.e.,
S=∑ Pr (rk) * 255.
In conclusion human face light processing method of the present invention by a variety of existing excellents of fusion, it can be to containing people
Face image carries out adaptive lighting process normalization.And can be by judging different light conditions, selection is different
Targetedly lighting process method carries out unitary of illumination processing to containing facial image, so that lighting process has very strong robust
Property and anti-noise ability.
Claims (3)
1. a kind of adaptive recognition of face lighting process method, it is characterised in that steps are as follows:
The first step, training Haar feature classifiers handle the image containing face, realize Face detection, face part is cut out
Come；
Second step establishes illumination subspace, for three kinds of possible light conditions, sample set of the tissue by Gaussian Blur processing
It closes, strong classifier made of obtaining the cascade of several Weak Classifiers is trained to sample set；
Third step will be put into classifier by the facial image of first step processing, obtain the illumination complexity of face；
4th step, it is serious if illumination effect according to the illumination complexity for the face that third step obtains, then select gradient face into
Row lighting process then selects illumination pretreatment chain to carry out lighting process, if almost no light shadow if a small amount of illumination effect
It rings, then histogram equalization is selected to carry out lighting process；
It establishes illumination subspace and carries out the judgement of illumination complexity specific step is as follows:
2.1 establish sample set, which includes the facial image under multiple various illumination, according to criterion by sample image
Light conditions are classified: 1. almost without illumination effect, i.e. face in uniform light；2. a small amount of illumination effect, i.e., a small amount of polarisation, people
Face profile and edge are visible；3. serious illumination effect, i.e. face shadow occlusion range are big, overexposure；
2.2 pairs of sample set pictures carry out Gaussian Blur:
Wherein, x, y respectively represent the position of coordinate points, and σ indicates smooth degree, hidden using the smooth facial image of Gaussian Blur
Its facial five features is hidden, the image of face's comparison of light and shade is only obtained；
2.3 are trained using the Adaboost method based on Haar feature to by the sample set of Gaussian Blur, are obtained many
Weak Classifier connects multiple Weak Classifiers, obtains a strong classifier；The specific method is as follows:
2.3.1 initializing weight:
Wherein f is light image feature, and θ is threshold value, and p is the direction of the sign of inequality, and x is a detection child window；
Wherein, wt,iFor the weight of t-th of classifier ith feature, qt,iIt is the weight after normalization；
2.3.2 is calculated for the light image feature f of every multiple angles of one kind, one Weak Classifier h of training adding for all images
Weigh error rate εf,
εf=∑iqi|h(xi,f,p,θ)-yi|
Choosing has minimal error rate εfWeak Classifier hi, and adjust the weight of sample
Wherein ei=0 indicates xiCorrectly classified, it is on the contrary then for mistake classification；
Wherein εtFor the weighting fault rate of t-th of classifier,For Dynamic gene；
2.3.3 Weak Classifier obtained in 2.3.2 is cascaded into three groups of strong classifiers
Wherein
2. adaptive recognition of face lighting process method according to claim 1, it is characterised in that: third step will pass through
The facial image of processing is put into the strong classifier of second step composition, if the result that first layer classifier obtains is 1, image category
It is serious in illumination effect；If the result that first layer classifier obtains is 0, next layer of judgement is entered, if classifier obtains at this time
The result arrived is 1, then it is moderate to belong to illumination effect for image；Otherwise judge it for almost without illumination effect.
3. adaptive recognition of face lighting process method according to claim 1, it is characterised in that the 4th step
The specific method is as follows:
(1) when being classified as normal illumination, i.e., the image array dark pixel after Gaussian Blur accounts for 5%~30%, then selects histogram
Equalization handles picture；
(2) when being classified as slightly illumination effect, i.e., the image array dark pixel after Gaussian Blur accounts for 30%~70%, then selects light
Picture is handled according to pretreatment chain；
(3) when being classified as serious illumination effect, i.e., the image array dark pixel after Gaussian Blur accounts for 70%~100%, then selects
Gradient face handles picture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510349830.XA CN105046202B (en) | 2015-06-23 | 2015-06-23 | Adaptive recognition of face lighting process method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510349830.XA CN105046202B (en) | 2015-06-23 | 2015-06-23 | Adaptive recognition of face lighting process method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105046202A CN105046202A (en) | 2015-11-11 |
| CN105046202B true CN105046202B (en) | 2019-04-16 |
Family
ID=54452731
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510349830.XA Active CN105046202B (en) | 2015-06-23 | 2015-06-23 | Adaptive recognition of face lighting process method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105046202B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106022225B (en) * | 2016-05-10 | 2019-03-05 | 中科天网（广东）科技有限公司 | A kind of Face datection classifier building method based on AdaBoost |
| CN106407966B (en) * | 2016-11-28 | 2019-10-18 | 南京理工大学 | A kind of face identification method applied to attendance |
| CN107545251A (en) * | 2017-08-31 | 2018-01-05 | 北京图铭视界科技有限公司 | Face quality discrimination and the method and device of picture enhancing |
| CN108985222B (en) * | 2018-07-12 | 2024-02-20 | 天津艾思科尔科技有限公司 | Deep learning network model and system for recognition of incoming calls |
| CN109431045B (en) * | 2018-12-08 | 2022-04-22 | 余姚市恒正金属制品有限公司 | Leg self-adaptive telescopic system |
| CN110910320B (en) * | 2019-11-04 | 2022-09-06 | 南京邮电大学 | Face image illumination restoration method based on singular value decomposition |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102214292A (en) * | 2010-04-09 | 2011-10-12 | 汉王科技股份有限公司 | Illumination processing method for human face images |
| CN103208012A (en) * | 2013-05-08 | 2013-07-17 | 重庆邮电大学 | Light face recognition method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7747079B2 (en) * | 2006-09-27 | 2010-06-29 | Tandent Vision Science, Inc. | Method and system for learning spatio-spectral features in an image |
-
2015
- 2015-06-23 CN CN201510349830.XA patent/CN105046202B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102214292A (en) * | 2010-04-09 | 2011-10-12 | 汉王科技股份有限公司 | Illumination processing method for human face images |
| CN103208012A (en) * | 2013-05-08 | 2013-07-17 | 重庆邮电大学 | Light face recognition method |
Non-Patent Citations (2)
| Title |
|---|
| Facial-Lighting归一化方法研究;周枫 等;《软件》;20150615;第36卷(第6期);第58-65页 |
| 基于光照分类的可变光照下人脸识别方法;崔瑞 等;《计算机工程与应用》;20101231;第46卷(第28期);第185~188页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105046202A (en) | 2015-11-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105046202B (en) | Adaptive recognition of face lighting process method | |
| Anila et al. | Preprocessing technique for face recognition applications under varying illumination conditions | |
| CN108734138B (en) | Melanoma skin disease image classification method based on ensemble learning | |
| CN109684922B (en) | Multi-model finished dish identification method based on convolutional neural network | |
| CN108268859A (en) | A kind of facial expression recognizing method based on deep learning | |
| CN102306384B (en) | Color constancy processing method based on single image | |
| CN109766898A (en) | Image character recognition method, device, computer equipment and storage medium | |
| CN109859181A (en) | A kind of PCB welding point defect detection method | |
| CN104318262A (en) | Method and system for replacing skin through human face photos | |
| CN1975759A (en) | Human face identifying method based on structural principal element analysis | |
| CN105488456A (en) | Adaptive rejection threshold adjustment subspace learning based human face detection method | |
| CN110458792B (en) | Method and device for evaluating quality of face image | |
| CN105678245A (en) | Target position identification method based on Haar features | |
| Smith et al. | Effect of pre-processing on binarization | |
| CN108664839A (en) | A kind of image processing method and equipment | |
| CN107516083A (en) | A kind of remote facial image Enhancement Method towards identification | |
| CN109522865A (en) | A kind of characteristic weighing fusion face identification method based on deep neural network | |
| CN114463269A (en) | Chip defect detection method based on deep learning method | |
| CN104966271B (en) | Image de-noising method based on biological vision receptive field mechanism | |
| Karamizadeh et al. | Race classification using gaussian-based weight K-nn algorithm for face recognition | |
| US8311358B2 (en) | Method and system for image extraction and identification | |
| CN110490053B (en) | Human face attribute identification method based on trinocular camera depth estimation | |
| Pathak et al. | Entropy based CNN for segmentation of noisy color eye images using color, texture and brightness contour features | |
| CN111339856A (en) | Deep learning-based face recognition method and recognition system under complex illumination condition | |
| Giap et al. | Adaptive multiple layer retinex-enabled color face enhancement for deep learning-based recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into 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: 20210107 Address after: 210014 Guanghua Road 129-3, Qinhuai District, Nanjing City, Jiangsu Province Patentee after: NUST SCIENCE AND TECHNOLOGY PARK Co.,Ltd. Address before: 210094 No. 200, Xiaolingwei, Jiangsu, Nanjing Patentee before: NANJING University OF SCIENCE AND TECHNOLOGY |