CN106250877A - Near-infrared face identification method and device - Google Patents

Near-infrared face identification method and device Download PDF

Info

Publication number
CN106250877A
CN106250877A CN201610695263.8A CN201610695263A CN106250877A CN 106250877 A CN106250877 A CN 106250877A CN 201610695263 A CN201610695263 A CN 201610695263A CN 106250877 A CN106250877 A CN 106250877A
Authority
CN
China
Prior art keywords
face
infrared
characteristic
picture
characteristic vector
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
Application number
CN201610695263.8A
Other languages
Chinese (zh)
Other versions
CN106250877B (en
Inventor
陈雁
郭开
徐勇
梁子正
姜安刘
封其国
崔华
谭静
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Sunwin Intelligent Co Ltd
Original Assignee
Shenzhen Sunwin Intelligent Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shenzhen Sunwin Intelligent Co Ltd filed Critical Shenzhen Sunwin Intelligent Co Ltd
Priority to CN201610695263.8A priority Critical patent/CN106250877B/en
Publication of CN106250877A publication Critical patent/CN106250877A/en
Application granted granted Critical
Publication of CN106250877B publication Critical patent/CN106250877B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168Feature extraction; Face representation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172Classification, 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 Analysis (AREA)
  • Image Processing (AREA)

Abstract

The present invention provides a kind of near-infrared face identification method and device, automatically a kind of method detected by face is caught in frame of video based near infrared face picture, the face picture captured is processed by the local binary algorithm utilizing a kind of improvement, then the convolutional neural networks framework in degree of depth learning algorithm is utilized to extract the characteristic vector of face picture, it is fused to face characteristic, finally the face characteristic of reacquisition is carried out classification and matching and identification by grader, export recognition result.Near-infrared face identification method and the device of the present invention are relatively strong to the robustness of illumination, and solve existing recognition of face is affected relatively big by ambient lighting, and the technical problem that recognition performance is poor, recognition accuracy is high, the suitability is strong.

Description

Near-infrared face identification method and device
Technical field
The present invention relates to a kind of near-infrared face identification method and device.
Background technology
The research of present stage recognition of face has been achieved for great progress, defines the most outstanding many recognizer, But under complicated photoenvironment, most of algorithms all also exist bigger defect and deficiency, greatly limit its application model Enclose.
In order to overcome the impact of ambient lighting, academia and relevant enterprise to do substantial amounts of research and technological development, mainly Improve for existing visible ray face identification system, to alleviate the impact of ambient lighting, but produce little effect.
The method processed image based on thermal infrared or far infrared, easily by ambient temperature, the emotion of people and healthy shape The impact of state, the facial image got is unstable, in actual applications, poor-performing.
Summary of the invention
Present invention is primarily targeted at a kind of near-infrared face identification method of offer, it is intended to solve existing recognition of face Affected relatively big by ambient lighting, the technical problem that recognition performance is poor.
For achieving the above object, the present invention proposes a kind of near-infrared face identification method, comprises the following steps:
Catch in frame of video based near infrared face picture;
Utilize the local binary algorithm improved that described face picture is processed;
Utilize convolutional neural networks to extract the characteristic vector of described face picture, be fused to face characteristic and preserve;
Again extract face characteristic, carry out match cognization with the face characteristic preserved.
Further, described " catching in frame of video based near infrared face picture " step includes:
Load the tag file of Face datection;
Frame of video is obtained from photographic head according to described tag file;
Detect face from described frame of video and show picture;
Face Status Flag is set, exists and be set to True, be otherwise False;
Judge whether face Status Flag exists, if it is not, the most again obtain video according to described tag file from photographic head Frame, the most then quit a program.
Further, described " utilizing the local binary algorithm improved that described face picture is processed " step bag Include:
Pixel centered by face pixel in described face picture, selected threshold compares with neighbor;
If the gray value of neighbor is less than or equal to the gray value of center pixel, then neighbor is labeled as 0, on the contrary labelling It is 1;
Utilize center pixel described in binary number representation;
Would indicate that the binary number of center pixel is converted to decimal number.
Further, described convolutional neural networks includes multiple convolutional layer and a Softmax layer for classification, described The step " utilizing convolutional neural networks to extract the characteristic vector of described face picture, be fused to face characteristic and preserve " includes:
Utilize the human face recognition model that visible ray picture training is original, obtain its model parameter;
Utilize described model parameter to initialize the parameter of near-infrared human face recognition model, and use near-infrared face picture pair Described model parameter is finely adjusted, and obtains based near infrared human face recognition model;
Described based on the near infrared multiple nonlinear function of human face recognition model matching, it is distributed in the multiple of described convolutional layer Neurode, extracts respectively to the characteristic vector of same face picture;
The characteristic vector extracted respectively is fused at described Softmax layer the final expression of face characteristic, and protects Deposit.
Further, described nonlinear function isWherein θiRepresenting i-th parameter, α represents Habit rate, J (θ) represents the cost function of degree of depth study.
Further, described " again extracting face characteristic, carry out match cognization with the face characteristic preserved " step bag Include:
Grader is trained according to described characteristic vector;
Utilize described grader that the characteristic vector of face characteristic is classified, calculate similarity;
According to described similarity, the recognition result of output existing object.
The present invention also provides for a kind of near-infrared face identification device, including:
Face acquisition module, catches in frame of video based near infrared face picture;
Facial pretreatment module, utilizes the local binary algorithm improved to process described face picture;
Face characteristic extraction module, utilizes convolutional neural networks to extract the characteristic vector of described face picture, merges as people Face feature also preserves;And
Face recognition module, extracts face characteristic again, carries out match cognization with the face characteristic preserved.
Further, described convolutional neural networks includes multiple convolutional layer and a Softmax layer for classification, described Face characteristic extraction module includes:
First training unit, utilizes the human face recognition model that visible ray picture training is original, obtains its model parameter;
Initialization unit, utilizes described model parameter to initialize the parameter of near-infrared human face recognition model;
Fine-adjusting unit, uses near-infrared face picture to be finely adjusted described model parameter, obtains based near infrared people Face identification model;
Extraction unit, described based on the near infrared multiple nonlinear function of human face recognition model matching, it is distributed in described volume Multiple neurodes of lamination, extract respectively to the characteristic vector of same face picture;And
Integrated unit, is fused to the final expression of face characteristic by the characteristic vector extracted respectively at described Softmax layer, And preserve.
Further, described nonlinear function isWherein θiRepresenting i-th parameter, α represents Habit rate, J (θ) represents the cost function of degree of depth study.
Further, described face recognition module includes:
Second training unit, trains grader according to described characteristic vector;
Computing unit, utilizes described grader to classify the characteristic vector of face characteristic, calculates similarity;And
Result output unit, according to described similarity, the recognition result of output existing object.
The near-infrared face identification method of the present invention, a kind of method automatically detected by face catch in frame of video based on Near infrared face picture, utilizes the local binary algorithm of a kind of improvement to process the face picture captured, then profit Extract the characteristic vector of face picture with the convolutional neural networks framework in degree of depth learning algorithm, be fused to face characteristic, finally By grader, the face characteristic of reacquisition being carried out classification and matching and identification, exports recognition result, the present invention is to illumination Robustness is relatively strong, and solve existing recognition of face is affected relatively big by ambient lighting, and the technical problem that recognition performance is poor identifies Accuracy rate is high, the suitability is strong.
Accompanying drawing explanation
In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing In having technology to describe, the required accompanying drawing used is briefly described, it should be apparent that, the accompanying drawing in describing below is only this Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to Other accompanying drawing is obtained according to the structure shown in these accompanying drawings.
Fig. 1 is the flow chart of near-infrared face identification method one embodiment of the present invention;
Fig. 2 is the particular flow sheet of step S100 in Fig. 1;
Fig. 3 is the particular flow sheet of step S200 in Fig. 1;
Fig. 4 is the particular flow sheet of step S300 in Fig. 1;
Fig. 5 is the particular flow sheet of step S400 in Fig. 1;
Fig. 6 is the functional block diagram of near-infrared face identification device one embodiment of the present invention;
Fig. 7 is the structured flowchart of face characteristic extraction module in Fig. 6;
Fig. 8 is the structured flowchart of face recognition module in Fig. 6.
Drawing reference numeral illustrates:
Label Title Label Title
100 Near-infrared face identification device 134 Extraction unit
11 Face acquisition module 135 Integrated unit
12 Facial pretreatment module 14 Face recognition module
13 Face characteristic extraction module 141 Second training unit
131 First training unit 142 Computing unit
132 Initialization unit 143 Result output unit
133 Fine-adjusting unit
The realization of the object of the invention, functional characteristics and advantage will in conjunction with the embodiments, are described further referring to the drawings.
Detailed description of the invention
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Describe, it is clear that described embodiment is only a part of embodiment of the present invention rather than whole embodiments wholely.Base Embodiment in the present invention, those of ordinary skill in the art obtained under not paying creative work premise all its His embodiment, broadly falls into the scope of protection of the invention.
It is to be appreciated that directional instruction in the embodiment of the present invention (such as up, down, left, right, before and after ...) is only used In explaining relative position relation between each parts, motion conditions etc. under a certain particular pose (as shown in drawings), if should When particular pose changes, then directionality instruction changes the most therewith.
It addition, the description relating to " first ", " second " etc. in the present invention is only used for describing purpose, and it is not intended that refer to Show or imply its relative importance or the implicit quantity indicating indicated technical characteristic.Thus, define " first ", " Two " feature can express or implicitly include at least one this feature.It addition, the technical scheme between each embodiment can To be combined with each other, but must be based on those of ordinary skill in the art are capable of, when the combination of technical scheme occurs Conflicting will be understood that the combination of this technical scheme does not exists, the most not at the protection model of application claims when maybe cannot realize Within enclosing.
With reference to Fig. 1, in a kind of embodiment that the present invention proposes, this near-infrared face identification method, comprise the following steps:
S100: catch in frame of video based near infrared face picture;
S200: utilize the local binary algorithm improved that described face picture is processed;
S300: utilize convolutional neural networks to extract the characteristic vector of described face picture, be fused to face characteristic and preserve;
S400: again extract face characteristic, carries out match cognization with the face characteristic preserved.
The near-infrared face identification method of the present embodiment, is divided into registration and identifies two stages, at registration phase, first holding Row step S100, catches and occurs in the video frame based near infrared face picture, and intercept appropriately sized face picture work For the input of successive depths learning algorithm model, the present embodiment uses near-infrared photographic head to obtain frame of video, reduces ambient light According to impact, in order to eliminate the impact near infrared imaging system of the intensity of near infrared light further, perform step S200, utilize The face picture captured from frame of video is processed by the local binary algorithm improved, and then performs step S300, utilizes The characteristic vector of the face picture that the convolutional neural networks algorithm extraction step S200 in degree of depth study processes, is fused to further Face characteristic also preserves, and completes to identify the registration of object;At cognitive phase, perform step S400, repeat step S100, S200 and S300, extracts the face characteristic of existing object, by classifying the face characteristic of the existing object extracted, calculates itself and note The similarity of the face characteristic that the volume stage extracts, exports recognition result finally according to described similarity, completes based near infrared Recognition of face.
Further, reference Fig. 2, step S100, including:
S11: load the tag file of Face datection;
S12: obtain frame of video from photographic head according to described tag file;
S13: detect face from described frame of video and show picture;
S14: arrange face Status Flag, exist and be set to True, is otherwise False;
S15: judge whether face Status Flag exists, if it is not, then perform step S12, the most then quits a program.
The near-infrared face identification method of the present embodiment, the step of " catching in frame of video based near infrared face picture " Specifically include: step S11, load the tag file of Face datection, to reduce detection range, improve the speed of detection with accurate Rate;Step S12, obtains frame of video according to described tag file from photographic head, reduce further detection object scope, described in take the photograph As head uses near infrared spectrum light emitting diode between 780~1100nm to be specifically designed to face as what active light source formed The near-infrared photographic head identified, selects the optical filter of 850nm further, the near infrared light of 850nm so can be allowed to pass through, and filters Fall the visible ray between 350~770nm;Step S13, detects face from described frame of video and shows picture, catching from video flowing Catch the picture with face characteristic, then through step S14, face Status Flag is set, exists and be set to True, there is not setting For False, the picture with face characteristic captured is cached or stores;Finally perform step S15, determine whether Whether face Status Flag exists, and without caching successfully or storing successfully, is then repeated execution step S12, the most right Face picture carries out catching and caching or store, if fruit has cached successfully or stored successfully, then terminates this program step, carries out Follow-up operation.
Further, reference Fig. 3, step S200, including:
S21: pixel centered by the face pixel in described face picture, selected threshold compares with neighbor;
S22: if the gray value of neighbor is less than or equal to the gray value of center pixel, then neighbor is labeled as 0, otherwise It is labeled as 1;
S23: utilize center pixel described in binary number representation;
S24: would indicate that the binary number of center pixel is converted to decimal number.
The near-infrared face identification method of the present embodiment, the impact of the incident direction of intensity of illumination and light source is bigger, although In indoor environment, near-infrared photographic head can filter out visible ray to greatest extent by wave filter, reduces because of different Time period and the intensity of illumination that causes and the change in light source incidence direction, and then reduce visible ray face recognition algorithms is caused Impact;But, the distance change between face and LED also can cause the change of intensity of illumination, accordingly, it would be desirable to utilize improvement Near-infrared face picture is further processed by local binary algorithm.The ultimate principle of local binary algorithm is: to image Pixel and the pixel around it are compared, and constitute a binary pattern, are reconverted into decimal number, then this decimal number is just It it is the local binary description value of preceding pixel point.Concrete steps are exactly: first, with the near-infrared face figure in the frame of video that captures Pixel centered by face pixel in sheet, selected threshold compares with neighbor, if the gray value of neighbor is less than In the gray value of center pixel, then neighbor is labeled as 0, otherwise is then labeled as 1;Then by described center pixel with a string two System number represents, such as 01000011, finally would indicate that that string binary number of center pixel is converted to decimal number, it is simply that center The local binary code of pixel.
Further, described convolutional neural networks includes multiple convolutional layer and a Softmax layer for classification, step S300, including:
S31: utilize the human face recognition model that visible ray picture training is original, obtain its model parameter;
S32: utilize described model parameter to initialize the parameter of near-infrared human face recognition model, and use near-infrared face figure Described model parameter is finely adjusted by sheet, obtains based near infrared human face recognition model;
S33: described based on the near infrared multiple nonlinear function of human face recognition model matching, is distributed in described convolutional layer Multiple neurodes, extract respectively to the characteristic vector of same face picture;
S34: the characteristic vector extracted respectively is fused at described Softmax layer the final expression of face characteristic, goes forward side by side Row preserves.
The near-infrared face identification method of the present embodiment, uses the convolutional neural networks framework in degree of depth learning method, In the present embodiment, this convolutional neural networks includes 10 convolutional layers and last is for the Softmax layer classified, neutral net Input time one two-dimentional image, by Pooling layer and Normalization layer between convolutional layer.
Owing to not currently existing the extensive near-infrared face database that can be used for training degree of depth learning model, thus utilize volume The characteristic vector that long-pending neutral net framework extracts near-infrared face picture is divided into two stages, and the first stage is that utilization is the most visible The human face recognition model that light picture training is original, obtains its model parameter, second stage according to the feature of degree of depth learning algorithm, For identical identification mission, the original human face recognition model parameter initialization near-infrared face utilizing visible ray to train is known The parameter of other model, finally uses a small amount of near-infrared people to connect picture and enters the parameter of the near-infrared human face recognition model after initializing Row fine setting, obtains based near infrared human face recognition model, specially directly utilizes the degree of depth study mould that the first stage trains The parameter of type initializes the parameter of near-infrared human face recognition model, and the model after initialization can extract visible ray face picture Feature, owing to near-infrared face picture and visible ray picture there are differences, but this species diversity is the least, so using near-infrared figure The parameter of model is finely adjusted by sheet, it is possible to obtains one and can extract learning based on the degree of depth of near-infrared face picture feature Near infrared human face recognition model.
Described convolutional neural networks framework is when extracting the characteristic vector of near-infrared face picture, by based near infrared The multiple nonlinear function of human face recognition model matching, is distributed in multiple neurodes of described convolutional layer, to same face figure The characteristic vector of sheet is extracted respectively, by some nonlinear transformations, finds the face representation of a kind of low-dimensional, nonlinear function Parameter use stochastic gradient descent algorithm study obtain, formula is Wherein θiRepresenting i-th parameter, α represents learning rate, and J (θ) represents the cost function of degree of depth study.
The characteristic vector of same face picture is extracted respectively, namely uses identical network structure to train one Multiple Patch of face picture, say, that train multiple near-infrared human face recognition model based on degree of depth study, namely pass through The multiple nonlinear functions being distributed in different convolutional layer of matching extract corresponding characteristic vector respectively, the most all import Softmax layer is classified, and is finally fused to the final expression of face characteristic.Convolutional neural networks framework details is defined as follows Shown in:
During identification, improved local two is referred to that the near-infrared face picture after algorithm process is as log on by us Input, last full articulamentum as the feature of network, uses PCA to carry out dimensionality reduction the characteristic vector obtained, finally makes Classify with associating bayesian algorithm.
Further, reference Fig. 5, step S400, including:
S41: train grader according to described characteristic vector;
S42: utilize described grader to classify the characteristic vector of face characteristic, calculates similarity;
S43: according to described similarity, the recognition result of output existing object.
The near-infrared face identification method of the present embodiment, depend on grader to the feature of near-infrared face picture to Amount carries out comparison of classifying, and then output comparison result is recognition result, specific as follows: first according to the face figure of registration phase The characteristic vector training grader of sheet, then utilizes the near-infrared people of the identified object that cognitive phase extracts by grader again The characteristic vector of face picture is classified, and calculates the similarity of the characteristic vector in two stages, finally according to described similarity, defeated Go out the current recognition result identifying object.
With reference to Fig. 6, the present invention also provides for a kind of near-infrared face identification device 100, including:
Face acquisition module 11, catches in frame of video based near infrared face picture;
Facial pretreatment module 12, utilizes the local binary algorithm improved to process described face picture;
Face characteristic extraction module 13, utilizes convolutional neural networks to extract the characteristic vector of described face picture, is fused to Face characteristic also preserves;And
Face recognition module 14, extracts face characteristic again, carries out match cognization with the face characteristic preserved.
The near-infrared face identification device of the present embodiment, including face acquisition module 11, facial pretreatment module 12, face Characteristic extracting module 13 and face recognition module 14, at registration phase, first occurred in by face acquisition module 11 seizure and regard Frequently based near infrared face picture in frame, and appropriately sized face picture is intercepted as successive depths learning algorithm model Input, the present embodiment uses near-infrared photographic head to obtain frame of video, reduces the impact of ambient lighting, near in order to eliminate further The impact near infrared imaging system of the intensity of infrared light, by the facial pretreatment module 12 people to capturing from frame of video Face picture processes, and then extracts the face picture after facial pretreatment module 12 processes by face characteristic extraction module 13 Characteristic vector, be fused to face characteristic further and preserve, complete identify object registration;In cognitive phase, recognition of face Module 14 extracts the face characteristic of existing object, by the face characteristic of existing object extracted is classified, calculate its with The similarity of the face characteristic that registration phase extracts, exports recognition result finally according to described similarity, completes based on near-infrared Recognition of face.
Further, with reference to Fig. 7, convolutional neural networks includes multiple convolutional layer and a Softmax layer for classification, Described face characteristic extraction module 13 includes:
First training unit 131, utilizes the human face recognition model that visible ray picture training is original, obtains its model parameter;
Initialization unit 132, utilizes described model parameter to initialize the parameter of near-infrared human face recognition model;
Fine-adjusting unit 133, uses near-infrared face picture to be finely adjusted described model parameter, obtains based near infrared Human face recognition model;
Extraction unit 134, described based on the near infrared multiple nonlinear function of human face recognition model matching, it is distributed in described Multiple neurodes of convolutional layer, extract respectively to the characteristic vector of same face picture;And
Integrated unit 135, is fused to the final table of face characteristic by the characteristic vector extracted respectively at described Softmax layer Show, and preserve.
The near-infrared face identification device of the present embodiment, uses the convolutional neural networks framework in degree of depth learning method, In the present embodiment, this convolutional neural networks includes 10 convolutional layers and last is for the Softmax layer classified, neutral net Input time one two-dimentional image, by Pooling layer and Normalization layer between convolutional layer.This face characteristic extracts Module 13, when extracting the characteristic vector of near-infrared face picture, is first trained original face to know by the first training unit 131 Other model, obtains its model parameter, then according to the feature of degree of depth learning algorithm, for identical identification mission, initializes single Unit 132 utilizes the parameter of the original human face recognition model parameter initialization near-infrared human face recognition model that visible ray trains, Used a small amount of near-infrared people to connect picture by fine-adjusting unit 133 parameter of the near-infrared human face recognition model after initializing is carried out Fine setting, obtains based near infrared human face recognition model, specially directly utilizes the degree of depth learning model that the first stage trains Parameter initialize the parameter of near-infrared human face recognition model, it is special that the model after initialization can extract visible ray face picture Levying, owing to near-infrared face picture and visible ray picture there are differences, but this species diversity is the least, so using near-infrared picture The parameter of model is finely adjusted, it is possible to obtain one and can extract the based on degree of depth study of near-infrared face picture feature Near infrared human face recognition model, then by extraction unit 134 by multiple non-based near infrared human face recognition model matching Linear function, is distributed in multiple neurodes of described convolutional layer, puies forward the characteristic vector of same face picture respectively Taking, by some nonlinear transformations, find the face representation of a kind of low-dimensional, the parameter of nonlinear function uses stochastic gradient descent Algorithm Learning obtains, and formula isWherein θiRepresent I-th parameter, α represents learning rate, and J (θ) represents the cost function of degree of depth study, finally by integrated unit 135 by matching The multiple nonlinear functions being distributed in different convolutional layer extract corresponding characteristic vector respectively, the most all import Softmax layer and enter Row classification, is finally fused to the final expression of face characteristic.
Further, with reference to Fig. 8, face recognition module 14 includes:
Second training unit 141, trains grader according to described characteristic vector;
Computing unit 142, utilizes described grader to classify the characteristic vector of face characteristic, calculates similarity;And
Result output unit 143, according to described similarity, the recognition result of output existing object.
The near-infrared face identification device of the present embodiment, face recognition module 14 includes the second training unit 141, calculates list Unit 142 and result output unit 143, face recognition module 14 depend on grader to the feature of near-infrared face picture to Amount carries out comparison of classifying, and then output comparison result is recognition result, specific as follows: first by the second training unit 141 According to the characteristic vector training grader of the face picture of registration phase, then utilized grader to identifying rank by computing unit 142 The characteristic vector of near-infrared face picture of the identified object that Duan Chongxin extracts is classified, calculate the feature in two stages to The similarity of amount, finally by result output unit 143 according to described similarity, the current recognition result identifying object of output.
The near-infrared face identification method of the present embodiment and device, a kind of method automatically detected by face catches video Based near infrared face picture in frame, utilize at the local binary algorithm of a kind of improvement face picture to capturing Reason, then utilizes the convolutional neural networks framework in degree of depth learning algorithm to extract the characteristic vector of face picture, is fused to face Feature, finally carries out classification and matching and identification by the face characteristic of reacquisition by grader, exports recognition result, the present invention Relatively strong to the robustness of illumination, currently used the inventive method, on many illumination face database that match provides for intelligence, we Use visible ray face recognition algorithms accuracy rate on the face database of same light photograph can reach 99%, but, in difference Under illumination condition, accuracy rate have decreased to 81%, after using near-infrared face recognition algorithms, and accuracy rate under the conditions of same light is shone 98% can be reached, but under different illumination conditions, accuracy rate can rise to 92%, solve existing recognition of face and be subject to Ambient lighting impact is relatively big, the technical problem that recognition performance is poor, and recognition accuracy is high, the suitability is strong.
The foregoing is only the preferred embodiments of the present invention, not thereby limit the scope of the claims of the present invention, every at this Under the inventive concept of invention, utilize the equivalent structure transformation that description of the invention and accompanying drawing content are made, or directly/indirectly use The technical field relevant at other is included in the scope of patent protection of the present invention.

Claims (10)

1. a near-infrared face identification method, it is characterised in that comprise the following steps:
Catch in frame of video based near infrared face picture;
Utilize the local binary algorithm improved that described face picture is processed;
Utilize convolutional neural networks to extract the characteristic vector of described face picture, be fused to face characteristic and preserve;
Again extract face characteristic, carry out match cognization with the face characteristic preserved.
Near-infrared face identification method the most according to claim 1, it is characterised in that described " catch in frame of video based on Near infrared face picture " step include:
Load the tag file of Face datection;
Frame of video is obtained from photographic head according to described tag file;
Detect face from described frame of video and show picture;
Face Status Flag is set, exists and be set to True, be otherwise False;
Judge whether face Status Flag exists, if it is not, the most again obtain frame of video according to described tag file from photographic head, if It is then to quit a program.
Near-infrared face identification method the most according to claim 1, it is characterised in that the described " local two that utilization improves Described face picture is processed by value-based algorithm " step include:
Pixel centered by face pixel in described face picture, selected threshold compares with neighbor;
If the gray value of neighbor is less than or equal to the gray value of center pixel, then neighbor is labeled as 0, otherwise is labeled as 1;
Utilize center pixel described in binary number representation;
Would indicate that the binary number of center pixel is converted to decimal number.
Near-infrared face identification method the most according to claim 1, it is characterised in that described convolutional neural networks includes many Individual convolutional layer and one, for the Softmax layer of classification, described " utilize the feature that convolutional neural networks extracts described face picture Vector, is fused to face characteristic and preserves " step include:
Utilize the human face recognition model that visible ray picture training is original, obtain its model parameter;
Utilize described model parameter to initialize the parameter of near-infrared human face recognition model, and use near-infrared face picture to described Model parameter is finely adjusted, and obtains based near infrared human face recognition model;
Described based on the near infrared multiple nonlinear function of human face recognition model matching, it is distributed in multiple nerves of described convolutional layer Node, extracts respectively to the characteristic vector of same face picture;
The characteristic vector extracted respectively is fused at described Softmax layer the final expression of face characteristic, and preserves.
Near-infrared face identification method the most according to claim 4, it is characterised in that described nonlinear function isWherein θiRepresenting i-th parameter, α represents learning rate, and J (θ) represents the cost function of degree of depth study.
Near-infrared face identification method the most according to claim 1, it is characterised in that described " again extract face characteristic, With preserve face characteristic carry out match cognization " step include:
Grader is trained according to described characteristic vector;
Utilize described grader that the characteristic vector of face characteristic is classified, calculate similarity;
According to described similarity, the recognition result of output existing object.
7. a near-infrared face identification device, it is characterised in that including:
Face acquisition module, catches in frame of video based near infrared face picture;
Facial pretreatment module, utilizes the local binary algorithm improved to process described face picture;
Face characteristic extraction module, utilizes convolutional neural networks to extract the characteristic vector of described face picture, is fused to face special Levy and preserve;And
Face recognition module, extracts face characteristic again, carries out match cognization with the face characteristic preserved.
Near-infrared face identification device the most according to claim 7, it is characterised in that described convolutional neural networks includes many Individual convolutional layer and a Softmax layer for classification, described face characteristic extraction module includes:
First training unit, utilizes the human face recognition model that visible ray picture training is original, obtains its model parameter;
Initialization unit, utilizes described model parameter to initialize the parameter of near-infrared human face recognition model;
Fine-adjusting unit, uses near-infrared face picture to be finely adjusted described model parameter, obtains knowing based near infrared face Other model;
Extraction unit, described based on the near infrared multiple nonlinear function of human face recognition model matching, it is distributed in described convolutional layer Multiple neurodes, the characteristic vector of same face picture is extracted respectively;And
Integrated unit, is fused to the characteristic vector extracted respectively the final expression of face characteristic, goes forward side by side at described Softmax layer Row preserves.
Near-infrared face identification device the most according to claim 8, it is characterised in that described nonlinear function isWherein θiRepresenting i-th parameter, α represents learning rate, and J (θ) represents the cost function of degree of depth study.
Near-infrared face identification device the most according to claim 7, it is characterised in that described face recognition module includes:
Second training unit, trains grader according to described characteristic vector;
Computing unit, utilizes described grader to classify the characteristic vector of face characteristic, calculates similarity;And
Result output unit, according to described similarity, the recognition result of output existing object.
CN201610695263.8A 2016-08-19 2016-08-19 Near-infrared face identification method and device Active CN106250877B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610695263.8A CN106250877B (en) 2016-08-19 2016-08-19 Near-infrared face identification method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610695263.8A CN106250877B (en) 2016-08-19 2016-08-19 Near-infrared face identification method and device

Publications (2)

Publication Number Publication Date
CN106250877A true CN106250877A (en) 2016-12-21
CN106250877B CN106250877B (en) 2019-05-21

Family

ID=57593359

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610695263.8A Active CN106250877B (en) 2016-08-19 2016-08-19 Near-infrared face identification method and device

Country Status (1)

Country Link
CN (1) CN106250877B (en)

Cited By (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106709477A (en) * 2017-02-23 2017-05-24 哈尔滨工业大学深圳研究生院 Face recognition method and system based on adaptive score fusion and deep learning
CN106909882A (en) * 2017-01-16 2017-06-30 广东工业大学 A kind of face identification system and method for being applied to security robot
CN107832735A (en) * 2017-11-24 2018-03-23 百度在线网络技术(北京)有限公司 Method and apparatus for identifying face
CN108491805A (en) * 2018-03-27 2018-09-04 百度在线网络技术(北京)有限公司 Identity identifying method and device
CN108491809A (en) * 2018-03-28 2018-09-04 百度在线网络技术(北京)有限公司 The method and apparatus for generating model for generating near-infrared image
CN108710838A (en) * 2018-05-08 2018-10-26 河南工程学院 Thermal infrared facial image recognition method under a kind of overnight sight
CN108764057A (en) * 2018-05-03 2018-11-06 武汉高德智感科技有限公司 A kind of far infrared human type of face detection method and system based on deep learning
CN108932459A (en) * 2017-05-26 2018-12-04 富士通株式会社 Face recognition model training method and device and recognition algorithms
CN109034064A (en) * 2018-07-26 2018-12-18 长沙舍同智能科技有限责任公司 Near-infrared face identification method, device and realization device
CN109145567A (en) * 2018-09-13 2019-01-04 深圳阜时科技有限公司 mobile device
CN109460792A (en) * 2018-11-14 2019-03-12 深圳市威富视界有限公司 A kind of artificial intelligence model training method and device based on image recognition
CN109528217A (en) * 2018-10-16 2019-03-29 北京入思技术有限公司 A kind of mood detection and method for early warning based on physiological vibrations analysis
CN109583266A (en) * 2017-09-28 2019-04-05 杭州海康威视数字技术股份有限公司 A kind of object detection method, device, computer equipment and storage medium
CN109800707A (en) * 2019-01-17 2019-05-24 中控智慧科技股份有限公司 A kind of face identification method based on CNN model, device and storage medium
CN110084134A (en) * 2019-04-03 2019-08-02 东华大学 A kind of face attendance checking system based on cascade neural network and Fusion Features
CN110298212A (en) * 2018-03-21 2019-10-01 腾讯科技(深圳)有限公司 Model training method, Emotion identification method, expression display methods and relevant device
CN110619200A (en) * 2018-06-19 2019-12-27 Oppo广东移动通信有限公司 Verification system and electronic device
CN111401107A (en) * 2019-01-02 2020-07-10 上海大学 Multi-mode face recognition method based on feature fusion neural network
CN111579083A (en) * 2020-05-13 2020-08-25 芋头科技(杭州)有限公司 Body temperature measurement method and device based on infrared image face detection
CN111623883A (en) * 2020-05-26 2020-09-04 成都电科崇实科技有限公司 Target temperature measurement method combining visible light image and infrared image
CN112069895A (en) * 2020-08-03 2020-12-11 广州杰赛科技股份有限公司 Small target face recognition method and device
CN112101186A (en) * 2020-09-11 2020-12-18 广州小鹏自动驾驶科技有限公司 Device and method for identifying a vehicle driver and use thereof
CN112380938A (en) * 2020-11-04 2021-02-19 浙江大华技术股份有限公司 Face recognition and temperature measurement method, device, equipment and medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103116749A (en) * 2013-03-12 2013-05-22 上海洪剑智能科技有限公司 Near-infrared face identification method based on self-built image library
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
WO2016112630A1 (en) * 2015-01-12 2016-07-21 芋头科技(杭州)有限公司 Image recognition system and method

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103116749A (en) * 2013-03-12 2013-05-22 上海洪剑智能科技有限公司 Near-infrared face identification method based on self-built image library
WO2016112630A1 (en) * 2015-01-12 2016-07-21 芋头科技(杭州)有限公司 Image recognition system and 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

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
徐佳杰: "双模态人脸识别***的研究与实现", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
王剑邦: "实用近红外人脸识别方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
秦连杰: "基于近红外图片的人脸检测识别算法的研究与应用", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *

Cited By (32)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106909882A (en) * 2017-01-16 2017-06-30 广东工业大学 A kind of face identification system and method for being applied to security robot
CN106709477A (en) * 2017-02-23 2017-05-24 哈尔滨工业大学深圳研究生院 Face recognition method and system based on adaptive score fusion and deep learning
CN108932459A (en) * 2017-05-26 2018-12-04 富士通株式会社 Face recognition model training method and device and recognition algorithms
CN108932459B (en) * 2017-05-26 2021-12-10 富士通株式会社 Face recognition model training method and device and face recognition method
CN109583266A (en) * 2017-09-28 2019-04-05 杭州海康威视数字技术股份有限公司 A kind of object detection method, device, computer equipment and storage medium
CN107832735A (en) * 2017-11-24 2018-03-23 百度在线网络技术(北京)有限公司 Method and apparatus for identifying face
CN110298212B (en) * 2018-03-21 2023-04-07 腾讯科技(深圳)有限公司 Model training method, emotion recognition method, expression display method and related equipment
CN110298212A (en) * 2018-03-21 2019-10-01 腾讯科技(深圳)有限公司 Model training method, Emotion identification method, expression display methods and relevant device
CN108491805A (en) * 2018-03-27 2018-09-04 百度在线网络技术(北京)有限公司 Identity identifying method and device
CN108491805B (en) * 2018-03-27 2022-02-22 百度在线网络技术(北京)有限公司 Identity authentication method and device
CN108491809A (en) * 2018-03-28 2018-09-04 百度在线网络技术(北京)有限公司 The method and apparatus for generating model for generating near-infrared image
CN108491809B (en) * 2018-03-28 2023-09-22 百度在线网络技术(北京)有限公司 Method and apparatus for generating near infrared image generation model
CN108764057A (en) * 2018-05-03 2018-11-06 武汉高德智感科技有限公司 A kind of far infrared human type of face detection method and system based on deep learning
CN108710838A (en) * 2018-05-08 2018-10-26 河南工程学院 Thermal infrared facial image recognition method under a kind of overnight sight
CN108710838B (en) * 2018-05-08 2019-11-08 河南工程学院 Thermal infrared facial image recognition method under a kind of overnight sight
CN110619200A (en) * 2018-06-19 2019-12-27 Oppo广东移动通信有限公司 Verification system and electronic device
CN109034064A (en) * 2018-07-26 2018-12-18 长沙舍同智能科技有限责任公司 Near-infrared face identification method, device and realization device
CN109034064B (en) * 2018-07-26 2021-01-08 长沙舍同智能科技有限责任公司 Near-infrared face recognition method, device and implementation device
CN109145567A (en) * 2018-09-13 2019-01-04 深圳阜时科技有限公司 mobile device
CN109528217A (en) * 2018-10-16 2019-03-29 北京入思技术有限公司 A kind of mood detection and method for early warning based on physiological vibrations analysis
CN109460792A (en) * 2018-11-14 2019-03-12 深圳市威富视界有限公司 A kind of artificial intelligence model training method and device based on image recognition
CN111401107B (en) * 2019-01-02 2023-08-18 上海大学 Multi-mode face recognition method based on feature fusion neural network
CN111401107A (en) * 2019-01-02 2020-07-10 上海大学 Multi-mode face recognition method based on feature fusion neural network
CN109800707A (en) * 2019-01-17 2019-05-24 中控智慧科技股份有限公司 A kind of face identification method based on CNN model, device and storage medium
CN110084134A (en) * 2019-04-03 2019-08-02 东华大学 A kind of face attendance checking system based on cascade neural network and Fusion Features
CN111579083A (en) * 2020-05-13 2020-08-25 芋头科技(杭州)有限公司 Body temperature measurement method and device based on infrared image face detection
CN111579083B (en) * 2020-05-13 2022-06-07 芋头科技(杭州)有限公司 Body temperature measuring method and device based on infrared image face detection
CN111623883A (en) * 2020-05-26 2020-09-04 成都电科崇实科技有限公司 Target temperature measurement method combining visible light image and infrared image
CN112069895A (en) * 2020-08-03 2020-12-11 广州杰赛科技股份有限公司 Small target face recognition method and device
CN112101186A (en) * 2020-09-11 2020-12-18 广州小鹏自动驾驶科技有限公司 Device and method for identifying a vehicle driver and use thereof
CN112380938A (en) * 2020-11-04 2021-02-19 浙江大华技术股份有限公司 Face recognition and temperature measurement method, device, equipment and medium
CN112380938B (en) * 2020-11-04 2024-05-03 浙江大华技术股份有限公司 Face recognition and temperature measurement method, device, equipment and medium

Also Published As

Publication number Publication date
CN106250877B (en) 2019-05-21

Similar Documents

Publication Publication Date Title
CN106250877A (en) Near-infrared face identification method and device
Tu et al. Passion fruit detection and counting based on multiple scale faster R-CNN using RGB-D images
CN109685115B (en) Fine-grained conceptual model with bilinear feature fusion and learning method
Singh et al. Face detection and recognition system using digital image processing
CN107194341B (en) Face recognition method and system based on fusion of Maxout multi-convolution neural network
Kim et al. Pedestrian detection based on faster R-CNN in nighttime by fusing deep convolutional features of successive images
CN107330397B (en) Pedestrian re-identification method based on large-interval relative distance measurement learning
AU2014368997B2 (en) System and method for identifying faces in unconstrained media
CN111767882A (en) Multi-mode pedestrian detection method based on improved YOLO model
CN102332092B (en) Flame detection method based on video analysis
CN106682578B (en) Weak light face recognition method based on blink detection
CN106909882A (en) A kind of face identification system and method for being applied to security robot
CN111597955A (en) Smart home control method and device based on expression emotion recognition of deep learning
US20180349716A1 (en) Apparatus and method for recognizing traffic signs
Kang et al. Deep learning-based weather image recognition
CN103116749A (en) Near-infrared face identification method based on self-built image library
Liu et al. The method of insulator recognition based on deep learning
CN109145742A (en) A kind of pedestrian recognition method and system
CN103136516A (en) Face recognition method and system fusing visible light and near-infrared information
CN108764096B (en) Pedestrian re-identification system and method
CN113298158B (en) Data detection method, device, equipment and storage medium
CN112287770A (en) Face quality sensing method and system for identity recognition
CN111401145A (en) Visible light iris recognition method based on deep learning and DS evidence theory
CN109344700A (en) A kind of pedestrian's posture attribute recognition approach based on deep neural network
Gawande et al. SIRA: Scale illumination rotation affine invariant mask R-CNN for pedestrian detection

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