CN109409322A - Biopsy method, device and face identification method and face detection system - Google Patents
Biopsy method, device and face identification method and face detection system Download PDFInfo
- Publication number
- CN109409322A CN109409322A CN201811329008.7A CN201811329008A CN109409322A CN 109409322 A CN109409322 A CN 109409322A CN 201811329008 A CN201811329008 A CN 201811329008A CN 109409322 A CN109409322 A CN 109409322A
- Authority
- CN
- China
- Prior art keywords
- face
- face picture
- characteristic
- feature
- picture characteristic
- 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
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06V40/40—Spoof detection, e.g. liveness detection
- G06V40/45—Detection of the body part being alive
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The disclosure proposes a kind of biopsy method, device and face identification method and face detection system, is related to technical field of face recognition.A kind of biopsy method of the disclosure includes: the face picture characteristic obtained in facial image by deep learning；Face picture characteristic is handled by the attention mechanism based on channel region, determines vivo identification as a result, it is living body image or non-living body image that vivo identification result, which includes facial image,.By such method, the face picture characteristic extracted in deep learning can be applied to In vivo detection, and pass through the attention mechanism processing feature data based on channel region, identify that facial image is living body or non-living body image, required movement is made without user's cooperation, convenience and efficiency are improved, and improves the accuracy of In vivo detection.
Description
Technical field
This disclosure relates to technical field of face recognition, especially a kind of biopsy method, device and face identification method
And face detection system.
Background technique
With the development of biological identification technology, face recognition technology has had preferable progress, in good illumination item
In the case where part and posture, face identification system can accurately carry out human face detection and tracing.
Relevant face identification system can confirm in image in the case where not distinguishing the non-living bodies information such as photo, video
The identity of user simultaneously passes through safety verification.In today of network information high speed development, the acquisition of the information such as photograph video of user
Cost is lower and lower, this causes the safety of face identification system to reduce.
In order to enhance the safety of face identification system, need to increase before recognition the security system of vivo identification.
Summary of the invention
Specified movement is made inventors have found that generally requiring user in relevant In vivo detection technology and cooperating on one's own initiative, is grasped
Make cumbersome and screen low efficiency, or be merely able to screen the non-living body image of still photo, accuracy is lower.
One purpose of the disclosure be guarantee In vivo detection convenience under the premise of, improve In vivo detection efficiency and
Accuracy.
According to the one aspect of some embodiments of the present disclosure, a kind of biopsy method is proposed, comprising: by depth
Practise the face picture characteristic obtained in facial image；Face picture feature is handled by the attention mechanism based on channel region
Data determine vivo identification as a result, it is living body image or non-living body image that vivo identification result, which includes facial image,.
In some embodiments, face picture characteristic is by the neural network model of face identification system from face
It is extracted in image.
In some embodiments, handling face picture characteristic by the attention mechanism based on channel region includes: root
According to face picture characteristic, biological characteristic and non-biometric are obtained by neural network convolutional layer；It is obtained by pond layer
Pond feature；According to pond feature, biological characteristic and non-biometric are associated with by full Connection Neural Network, obtain processing knot
Fruit.
In some embodiments, biological characteristic and non-biometric are associated with by full Connection Neural Network, obtain processing knot
Fruit include: pond feature is inputted into the first full articulamentum, and pass sequentially through ReLU (Rectified Linear Unit, it is linear whole
Stream function), the second full articulamentum and S sigmoid growth curve sigmoid function, processing result is obtained, so as to true according to processing result
Determine vivo identification result.
In some embodiments, by attention mechanism based on channel region handle face picture characteristic include: by
Face picture characteristic is handled by the attention mechanism based on channel region, obtains single treatment data；By single treatment number
According to by handling after convolutional neural networks again by the attention mechanism based on channel region, vivo identification result is obtained.
In some embodiments, by attention mechanism based on channel region handle face picture characteristic include: by
Face picture characteristic is handled by the attention mechanism based on channel region；By processing result by being followed after convolutional neural networks
Ring is handled by the attention mechanism based on channel region, until being reached by the attention mechanism number of processing based on channel region
When predetermined cycle-index, vivo identification result is determined according to determining processing result.
In some embodiments, biopsy method further include: using the output of sigmoid function to face picture feature
Each channel carries out scaling processing in data, obtains optimization face picture characteristic, so as to according to optimization face picture
Characteristic executes recognition of face.
In some embodiments, biopsy method further include: handled using by the attention mechanism based on channel region
The processing result of face picture characteristic reinforces the biological characteristic in face picture characteristic, and it is special to obtain optimization face picture
Levy data；Recognition of face is executed according to optimization face picture characteristic.
By such method, the face picture characteristic extracted in deep learning can be applied to In vivo detection,
And by the attention mechanism processing feature data based on channel region, identifies that facial image is living body or non-living body image, be not necessarily to
Required movement is made in user's cooperation, improves convenience and efficiency, and improve the accuracy of In vivo detection.
According to the one aspect of other embodiments of the disclosure, a kind of face identification method is proposed, comprising: pass through nerve
Network model extracts face picture characteristic from facial image；It is determined and is lived by above any one biopsy method
Body recognition result；Reinforce people using the processing result for handling face picture characteristic by the attention mechanism based on channel region
Biological characteristic in face picture feature data obtains optimization face picture characteristic；According to optimization face picture characteristic
Execute recognition of face.
By such method, the face picture characteristic application that deep learning can will be utilized to extract in recognition of face
In In vivo detection, and by the attention mechanism processing feature data based on channel region, identification facial image is living body or non-live
Body image makes required movement without user's cooperation, improves convenience and efficiency, and improve the accuracy of In vivo detection；
It can reinforce the biological characteristic in face picture characteristic, improve the accuracy of recognition of face.
According to the one aspect of the other embodiment of the disclosure, a kind of living body detection device is proposed, comprising: feature obtains
Module is configured as obtaining the face picture characteristic in facial image by deep learning；Feature processing block is configured
To handle face picture characteristic by the attention mechanism based on channel region；Vivo identification module, is configured as according to spy
The processing result of sign processing module determines vivo identification as a result, it is living body image or non-live that vivo identification result, which includes facial image,
Body image.
In some embodiments, feature obtains the neural network model that module is face identification system.
In some embodiments, feature processing block includes: convolutional layer, is configured as being obtained according to face picture characteristic
Take biological characteristic and non-biometric；Pond layer is configured as obtaining pond feature；Full connection processing unit, is configured as root
According to pond feature, biological characteristic and non-biometric are associated with by full Connection Neural Network, obtain processing result.
In some embodiments, connection processing unit is configured as entirely: by pond feature the first full articulamentum of input, and according to
It is secondary that processing result is obtained by ReLU, the second full articulamentum and sigmoid function, to determine that living body is known according to processing result
Other result.
In some embodiments, include in living body detection device by convolutional neural networks separately, serial connection two
A above feature processing block；Vivo identification module is configured as the processing according to the last one concatenated feature processing block
As a result vivo identification result is determined.
In some embodiments, feature processing block is additionally configured to the output using sigmoid function to face picture
Each channel carries out scaling processing in characteristic, obtains optimization face picture characteristic, so as to according to optimization face
Picture feature data execute recognition of face.
In some embodiments, feature processing block is additionally configured to reinforce people using the processing result of feature processing block
Biological characteristic in face picture feature data obtains optimization face picture characteristic；Living body detection device further include: face is known
Other module is configured as executing recognition of face according to optimization face picture characteristic.
According to the one aspect of the still other embodiments of the disclosure, a kind of living body detection device is proposed, comprising: memory；
And it is coupled to the processor of memory, processor is configured as above any one based on the instruction execution for being stored in memory
Kind biopsy method.
Such living body detection device can examine the face picture characteristic extracted in deep learning applied to living body
It surveys, and by the attention mechanism processing feature data based on channel region, identifies that facial image is living body or non-living body image, nothing
It needs user's cooperation to make required movement, improves convenience and efficiency, and improve the accuracy of In vivo detection.
According to the one aspect of some of embodiments of the disclosure, a kind of face detection system is proposed, comprising: above
Any one living body detection device；With face identification device is configured as: from facial image extract face picture characteristic
According to；Added using living body detection device by the processing result that the attention mechanism based on channel region handles face picture characteristic
Biological characteristic in strong man's face picture feature data obtains optimization face picture characteristic；According to optimization face picture feature
Data execute recognition of face.
According to the one aspect of some of embodiments of the disclosure, a kind of face detection system is proposed, comprising: storage
Device；And it is coupled to the processor of memory, processor is configured as the knowledge of the face based on the instruction execution for being stored in memory
Other method.
The face picture characteristic that such In vivo detection system can will utilize deep learning to extract in recognition of face
Applied to In vivo detection, and by the attention mechanism processing feature data based on channel region, identify facial image be living body or
Non-living body image makes required movement without user's cooperation, improves convenience and efficiency, and improve the accurate of In vivo detection
Degree；It can reinforce the biological characteristic in face picture characteristic, improve the accuracy of recognition of face.
In addition, proposing a kind of computer readable storage medium, thereon according to the one aspect of some embodiments of the present disclosure
The step of being stored with computer program instructions, above any one method realized when which is executed by processor.
By executing the instruction on such computer readable storage medium, the face figure that can will be extracted in deep learning
Piece characteristic is applied to In vivo detection, and by the attention mechanism processing feature data based on channel region, identifies face figure
As being living body or non-living body image, required movement is made without user's cooperation, improves convenience and efficiency, and improve living body
The accuracy of detection.
Detailed description of the invention
Attached drawing described herein is used to provide further understanding of the disclosure, constitutes a part of this disclosure, this public affairs
The illustrative embodiments and their description opened do not constitute the improper restriction to the disclosure for explaining the disclosure.In the accompanying drawings:
Fig. 1 is the flow chart of one embodiment of the biopsy method of the disclosure.
Fig. 2 is the stream of one embodiment of the attention mechanism processing in the biopsy method of the disclosure based on channel region
Cheng Tu.
Fig. 3 is the flow chart of another embodiment of the biopsy method of the disclosure.
Fig. 4 is the flow chart of one embodiment of the face identification method of the disclosure.
Fig. 5 is the schematic diagram of one embodiment of the living body detection device of the disclosure.
Fig. 6 is the schematic diagram of one embodiment of feature processing block in the living body detection device of the disclosure.
Fig. 7 is the schematic diagram of another embodiment of the treatment process of the living body detection device of the disclosure.
Fig. 8 is the schematic diagram of one embodiment of the face detection system of the disclosure.
Fig. 9 is the schematic diagram of the living body detection device of the disclosure or one embodiment of face detection system.
Figure 10 is the schematic diagram of the living body detection device of the disclosure or another embodiment of face detection system.
Specific embodiment
Below by drawings and examples, the technical solution of the disclosure is described in further detail.
Inventors have found that relevant biopsy method has several thinkings:
1, living body faces detection is carried out using co-occurrence matrix and wavelet analysis.The gray level image of human face region is subjected to gray scale
Compression, counts co-occurrence matrix respectively later, then extracts four texture characteristic amounts again on the basis of gray level co-occurrence matrixes and average
And variance；Second level decomposition is carried out using Haar wavelet basis to original image simultaneously, is averaged after extracting sub-band coefficients matrix and square
Difference；SVM (Support Vector Machine, branch after all characteristic values are finally sent into training as sample to be detected
Hold vector machine) in detected, Classification and Identification really and personation facial image.But such mode can only screen photo and take advantage of
The case where deceiving, cheating to video is helpless.
2, input continuous facial image (abandoned if not being same state if adjacent two width facial image, it is again more
The continuous facial image of width), pupil position is determined to every width facial image and determines human eye area；Pass through support vector machines training
Method and iterative algorithm AdaBoost are trained eye opening and eye closing sample, finally judge that eyeball opens closed state, blink if it exists
Eye process then passes through living body determination, but such mode needs user to cooperate on one's own initiative, uses cumbersome and low efficiency.
3, it has pre-defined a behavior aggregate (including blink, raise one's eyebrows, close one's eyes, stare, smile), user is carrying out living body
When detection, system is concentrated from movement every time and selects a kind of or several movement, is randomly assigned the number of execution, it is desirable that use
Them are completed before the deadline in family.Such mode equally needs user to cooperate on one's own initiative, and use is cumbersome, low efficiency, and is easy
By external environment influence, it is low to detect the success rate passed through.
The flow chart of one embodiment of the biopsy method of the disclosure is as shown in Figure 1.
In a step 101, the face picture characteristic in facial image is obtained by deep learning.In one embodiment
In, face picture characteristic can be extracted from facial image by the neural network model of face identification system.
Assuming thatFor face picture feature, H indicates that the height of picture, W indicate the length of picture, and C indicates figure
The number of channels (such as the number of channels of the RGB picture of a standard is 3) of piece.The convolutional layer of deep neural network by using
Convolution kernel and upper one layer of picture feature convolutional calculation obtain new picture feature, each convolution kernel can be on existing channel
Generate a new channel.After face picture feature I process has the convolutional layer of N number of convolution kernel, new feature can be generated, this
Picture feature is expressed asConvolution keeps the original size of picture, and C '=NC is the number of channels of new feature.
After the layer of pond, face picture data characteristicsNew picture feature can be sampled
Therefore, face picture data characteristics is by after several layers deep learning neural network,Meeting
It is extracted featureH ' indicates that the height of the picture after deep learning neural network, W ' indicate depth
The length of picture, C ' indicate the number of channels of deep learning neural network picture after habit neural network.
In a step 102, face picture characteristic is handled by the attention mechanism based on channel region, determines that living body is known
Other result.Vivo identification result includes that facial image is living body image or non-living body image.In one embodiment, using being based on
The attention mechanism of channel region can be by face picture featureIt is Internal reforming at biological characteristicAnd non-biometricWherein, CpFor the number of channels of biological characteristic, CnFor
The number of channels of non-biometric interacts both features, in the case where facial image is non-living body image, biology
Feature can be suppressed identify, it is determined that be non-living body image.
By such method, the face picture characteristic extracted in deep learning can be applied to In vivo detection,
And by the attention mechanism processing feature data based on channel region, identifies that facial image is living body or non-living body image, be not necessarily to
Required movement is made in user's cooperation, improves convenience and efficiency, and improve the accuracy of In vivo detection.
The flow chart of one embodiment of the attention mechanism processing in the biopsy method of the disclosure based on channel region
As shown in Figure 2.
In step 201, according to face picture characteristic, biological characteristic is obtained by neural network convolutional layerAnd non-biometric
In step 202, all feature global pools are obtained by pond feature by pond layer
In step 203, according to pond feature, biological characteristic and non-biometric are associated with by full Connection Neural Network,
Obtain processing result.
In one embodiment, pond feature can be inputted into the first full articulamentum, and passes sequentially through ReLU, second connects entirely
Connect layer.This two layers full articulamentum associates biological characteristic with non-biometric, codetermines whether the picture feature is living
Body characteristics obtain processing result by sigmoid function unitIt can determine figure according to the processing result
Whether the information in piece is biological information.
By such method, biological characteristic and non-biometric can be extracted by convolutional layer, passes through full articulamentum
Biological characteristic is associated with non-biometric, realizes that the attention mechanism based on channel region handles face picture characteristic
According to.
The flow chart of another embodiment of the biopsy method of the disclosure is as shown in Figure 3.
In step 301, the face picture characteristic in facial image is obtained by deep learning.
In step 302, face picture characteristic is handled by the attention mechanism based on channel region, obtains processing knot
Fruit.
In step 303, by processing result by convolutional neural networks handle after again by the note based on channel region
Power mechanism of anticipating processing, and obtain new processing result.
In step 304, judge whether the execution number to step 303 has reached predetermined cycle-index.If reaching pre-
Determine cycle-index, thens follow the steps 305；If not up to predetermined cycle-index, thens follow the steps 303.In one embodiment, often
The parameter of the secondary attention mechanism processing based on channel region can be different, can according to need and the ginseng such as different channel number is arranged
Number.
In step 305, vivo identification result is determined according to the processing result of last time circular treatment.
By such method, after feature is by the processing of attention mechanism based on channel region, picture feature can by into
Line activating, carrying out processing again can be realized the interactive identification repeatedly and detection of feature, improve the accuracy of detection.
In one embodiment, predetermined cycle-index can be 1 time, i.e., handled by the attention mechanism based on channel region
Single treatment data are obtained after face picture characteristic, after handling by convolutional neural networks again by single treatment data
Face picture characteristic is handled by the attention mechanism based on channel region and obtains processing result, and it is true to manage result according to this
Vivo identification is determined as a result, to ensure that recognition efficiency under the premise of improving accuracy.
In one embodiment, as shown in figure 3, can also include step 306: utilizing the processing of last time circular treatment
As a result reinforce the biological characteristic in face picture characteristic, optimization face picture characteristic is obtained, so as to according to optimization people
Face picture feature data execute recognition of face.
In one embodiment, it can use the output of sigmoid functionTo face picture characteristic
Each channel carries out scaling processing in, such as utilizes following formula:
It is a scale operator, the channel information correspondence of each channel i in picture feature can be zoomed in or out
Corresponding scale, it may be assumed that
F ' { i }=F { i } * M { i }
By such method, it can reinforce the biological characteristic in feature in the case where image is living body image, obtain
Optimize face picture characteristic, so as to improve the accuracy of recognition of face.
The flow chart of one embodiment of the face identification method of the disclosure is as shown in Figure 4.
In step 401, face picture characteristic is extracted from facial image by neural network model.In a reality
It applies in example, can use feature extraction functions in relevant face recognition technology and obtain face picture characteristic.
In step 402, vivo identification result is determined using any one face identification method being mentioned above.
In step 403, face picture characteristic is handled by the attention mechanism based on channel region using in step 402
According to processing result reinforce face picture characteristic in biological characteristic, obtain optimization face picture characteristic.
In step 404, recognition of face is executed according to optimization face picture characteristic.
By such method, the face picture characteristic application that deep learning can will be utilized to extract in recognition of face
In In vivo detection, and by the attention mechanism processing feature data based on channel region, identification facial image is living body or non-live
Body image makes required movement without user's cooperation, improves convenience and efficiency, and improve the accuracy of In vivo detection；
It can reinforce the biological characteristic in face picture characteristic, improve the accuracy of recognition of face.
In one embodiment, can be set as needed only determine facial image be living body image in the case where just into
Row recognition of face improves treatment effeciency, reduces operation burden.In another embodiment, can carry out respectively recognition of face and
Living body judgement, and synchronism output is as a result, to also provide vivo identification as a result, it is possible to enrich defeated while realizing recognition of face
Out as a result, being applied to different application scene as needed to facilitate.
The schematic diagram of one embodiment of the living body detection device of the disclosure is as shown in Figure 5.Feature obtains module 502 can
The face picture characteristic in facial image is obtained by deep learning.In one embodiment, face picture characteristic
It can be extracted from facial image by the neural network model of face identification system.Feature processing block 502 can pass through base
Face picture characteristic is handled in the attention mechanism of channel region, obtains processing result.In one embodiment, vivo identification
It as a result is living body image or non-living body image including facial image.Vivo identification module 503 can be according to feature processing block
Processing result determines vivo identification as a result, it is living body image or non-living body image that vivo identification result, which includes facial image,.One
In a embodiment, vivo identification is determined according to the processing result of feature processing block as a result, vivo identification result includes face figure
As being that living body image or non-living body image can will be inside face picture features using the attention mechanism based on channel region
It is converted to biological characteristic and non-biometric, and both features are interacted, obtains processing result.It is non-in facial image
In the case where living body image, biological characteristic can be suppressed identify.Vivo identification module 503 can utilize processing result
Determine that facial image is living body or non-living body image.
Such living body detection device can examine the face picture characteristic extracted in deep learning applied to living body
It surveys, and by the attention mechanism processing feature data based on channel region, identifies that facial image is living body or non-living body image, nothing
It needs user's cooperation to make required movement, improves convenience and efficiency, and improve the accuracy of In vivo detection.
In one embodiment, it in living body detection device may include by convolutional neural networks separately serial connection
More than two feature processing blocks, vivo identification module is according to the processing result of the last one concatenated feature processing block
Determine vivo identification result.Using such device, after feature is handled by the attention mechanism based on channel region, picture is special
Sign can be activated, and carrying out processing again can be realized the interactive identification repeatedly and detection of feature, improve the accuracy of detection.
In one embodiment, the parameter of each feature processing block can be different, can according to need and different channel numbers is arranged
Etc. parameters.In one embodiment, concatenated feature processing block can be two, to protect under the premise of improving accuracy
Recognition efficiency is demonstrate,proved.
In one embodiment, feature processing block can also reinforce face figure according to the processing result of feature processing block
Biological characteristic in piece characteristic obtains optimization face picture characteristic.As shown in figure 5, living body detection device can be with
Including face recognition module 504, vivo identification can be determined according to the processing result of the last one concatenated feature processing block
As a result.Such device can reinforce the biological characteristic in face picture characteristic, to improve the accurate of recognition of face
Degree.
The schematic diagram of one embodiment of feature processing block is as shown in Figure 6 in the living body detection device of the disclosure.Convolution
Layer 601 can obtain biological characteristic according to face picture characteristicAnd non-biometricAll feature global pools can be obtained pond feature by pond layer 602Full connection processing unit 603 is associated with by full Connection Neural Network and is given birth to according to pond feature
Object feature and non-biometric obtain processing result.
Such device can extract biological characteristic and non-biometric by convolutional layer, will be biological by full articulamentum
Feature associates with non-biometric, realizes that the attention mechanism based on channel region handles face picture characteristic.
The schematic diagram of another embodiment of the treatment process of the living body detection device of the disclosure is as shown in Figure 7.Full connection
Processing unit 603 may include the first full articulamentum, ReLU, the second full articulamentum and the sigmoid function list of serial connection
Member, two layers of full articulamentum associate biological characteristic with non-biometric, codetermine whether the picture feature is living body spy
Sign obtains processing result by sigmoid function unitTo can determine figure according to the processing result
Whether the information in piece is biological information.In one embodiment, it can use the output of sigmoid functionScaling processing is carried out to channel each in face picture characteristic, such as utilizes formula:
Scaling is carried out,It is a scale operator, it can be by the channel information pair of each channel i in picture feature
Corresponding scale should be zoomed in or out, it may be assumed that
F ' { i }=F { i } * M { i }
So as to reinforce the biological characteristic in feature in the case where image is living body image, optimization face picture is obtained
Characteristic improves the accuracy of recognition of face.
The schematic diagram of one embodiment of the face detection system of the disclosure is as shown in Figure 8.Living body detection device 81 can be with
For any one living body detection device being mentioned above.Face detection system can also include face identification device 82, can
Face picture characteristic is extracted from facial image, and the face picture feature data are supplied to living body detection device 81；
Reinforced using living body detection device by the processing result that the attention mechanism based on channel region handles face picture characteristic
Biological characteristic in face picture characteristic obtains optimization face picture characteristic；According to optimization face picture characteristic
According to execution recognition of face.
The face picture characteristic that such In vivo detection system can will utilize deep learning to extract in recognition of face
Applied to In vivo detection, and by the attention mechanism processing feature data based on channel region, identify facial image be living body or
Non-living body image makes required movement without user's cooperation, improves convenience and efficiency, and improve the accurate of In vivo detection
Degree；It can reinforce the biological characteristic in face picture characteristic, improve the accuracy of recognition of face.
The structural schematic diagram of one embodiment of disclosure living body detection device is as shown in Figure 9.Living body detection device includes
Memory 901 and processor 902.Wherein: memory 901 can be disk, flash memory or other any non-volatile memory mediums.
Memory is used to store the instruction in the above corresponding embodiment of biopsy method.Processor 902 is coupled to memory
901, it can be used as one or more integrated circuits to implement, such as microprocessor or microcontroller.The processor 902 is for holding
The instruction stored in line storage can be improved the convenience and efficiency of In vivo detection, and improve the accuracy of In vivo detection.
In one embodiment, can also as shown in Figure 10, living body detection device 1000 includes memory 1001 and processing
Device 1002.Processor 1002 is coupled to memory 1001 by BUS bus 1003.The living body detection device 1000 can also pass through
Memory interface 1004 is connected to external memory 1005 to call external data, can also be connected by network interface 1006
To network or an other computer system (not shown).It no longer describes in detail herein.
In this embodiment, it is instructed by memory stores data, then above-metioned instruction is handled by processor, can be not necessarily to
Required movement is made in user's cooperation, improves convenience and efficiency, and improve the accuracy of In vivo detection.
The structural schematic diagram of one embodiment of disclosure face detection system is as shown in Figure 9.Face detection system includes
Memory 901 and processor 902.Wherein: memory 901 can be disk, flash memory or other any non-volatile memory mediums.
Memory is used to store the instruction in the above corresponding embodiment of face identification method.Processor 902 is coupled to memory
901, it can be used as one or more integrated circuits to implement, such as microprocessor or microcontroller.The processor 902 is for holding
The instruction stored in line storage can be improved the efficiency and accuracy of In vivo detection, and improve the accuracy of recognition of face.
In one embodiment, can also as shown in Figure 10, face detection system 1000 includes memory 1001 and processing
Device 1002.Processor 1002 is coupled to memory 1001 by BUS bus 1003.The face detection system 1000 can also pass through
Memory interface 1004 is connected to external memory 1005 to call external data, can also be connected by network interface 1006
To network or an other computer system (not shown).It no longer describes in detail herein.
In this embodiment, it is instructed by memory stores data, then above-metioned instruction is handled by processor, can be improved
The efficiency and accuracy of In vivo detection, and improve the accuracy of recognition of face.
In another embodiment, a kind of computer readable storage medium, is stored thereon with computer program instructions, this refers to
The step of enabling the method realized in biopsy method or face identification method corresponding embodiment when being executed by processor.This field
Interior technical staff is it should be appreciated that embodiment of the disclosure can provide as method, apparatus or computer program product.Therefore, this public affairs
Open the form that complete hardware embodiment, complete software embodiment or embodiment combining software and hardware aspects can be used.And
And the disclosure can be used and can be deposited with non-transient in the computer that one or more wherein includes computer usable program code
The shape for the computer program product implemented on storage media (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.)
Formula.
The disclosure is reference according to the method for the embodiment of the present disclosure, the flow chart of equipment (system) and computer program product
And/or block diagram describes.It should be understood that each process in flowchart and/or the block diagram can be realized by computer program instructions
And/or the combination of the process and/or box in box and flowchart and/or the block diagram.It can provide these computer programs to refer to
Enable the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to generate
One machine so that by the instruction that the processor of computer or other programmable data processing devices executes generate for realizing
The device for the function of being specified in one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
So far, the disclosure is described in detail.In order to avoid covering the design of the disclosure, it is public that this field institute is not described
The some details known.Those skilled in the art as described above, completely it can be appreciated how implementing technology disclosed herein
Scheme.
Disclosed method and device may be achieved in many ways.For example, can by software, hardware, firmware or
Person's software, hardware, firmware any combination realize disclosed method and device.The step of for the method it is above-mentioned
Sequence is merely to be illustrated, and the step of disclosed method is not limited to sequence described in detail above, unless with other sides
Formula illustrates.In addition, in some embodiments, the disclosure can be also embodied as recording program in the recording medium, these
Program includes for realizing according to the machine readable instructions of disclosed method.Thus, the disclosure also covers storage for executing
According to the recording medium of the program of disclosed method.
Finally it should be noted that: above embodiments are only to illustrate the technical solution of the disclosure rather than its limitations；To the greatest extent
Pipe is described in detail the disclosure referring to preferred embodiment, it should be understood by those ordinary skilled in the art that: still
It can modify to the specific embodiment of the disclosure or some technical features can be equivalently replaced；Without departing from this public affairs
The spirit of technical solution is opened, should all be covered in the claimed technical proposal scope of the disclosure.
Claims (19)
1. a kind of biopsy method, comprising:
The face picture characteristic in facial image is obtained by deep learning；
The face picture characteristic is handled by the attention mechanism based on channel region, determines vivo identification as a result, described
Vivo identification result includes that the facial image is living body image or non-living body image.
2. according to the method described in claim 1, wherein, the face picture characteristic is the mind by face identification system
It is extracted from facial image through network model.
It is described to pass through the attention mechanism based on channel region and handle the face 3. according to the method described in claim 1, wherein
Picture feature data include:
According to the face picture characteristic, biological characteristic and non-biometric are obtained by neural network convolutional layer；
Pond feature is obtained by pond layer；
According to the pond feature, the biological characteristic and the non-biometric are associated with by full Connection Neural Network, obtained
Processing result.
It is described to pass through full Connection Neural Network and be associated with the biological characteristic and institute 4. according to the method described in claim 3, wherein
Non-biometric is stated, obtaining processing result includes:
The pond feature is inputted into the first full articulamentum, and it is raw to pass sequentially through line rectification function, the second full articulamentum and S type
Long curve sigmoid function, obtains the processing result, to determine the vivo identification result according to the processing result.
It is described to pass through the attention mechanism based on channel region and handle the face 5. according to the method described in claim 1, wherein
Picture feature data include:
The face picture characteristic is handled by the attention mechanism based on channel region, obtains single treatment data；It will
The single treatment data are handled after passing through convolutional neural networks again by the attention mechanism based on channel region, described in acquisition
Vivo identification result；
Or,
The face picture characteristic is handled by the attention mechanism based on channel region；Processing result is passed through into convolution mind
The attention mechanism processing based on channel region is cycled through after network, until handling by the attention mechanism based on channel region
Number when reaching predetermined cycle-index, according to determining that processing result determines the vivo identification result.
6. according to the method described in claim 4, further include: using the output of the sigmoid function to the face picture
Each channel carries out scaling processing in characteristic, optimization face picture characteristic is obtained, so as to according to the optimization
Face picture characteristic executes recognition of face.
7. method described in any one according to claim 1~5, further includes: utilize and pass through the attention machine based on channel region
The processing result that system handles the face picture characteristic reinforces the biological characteristic in the face picture characteristic, obtains
Optimize face picture characteristic；
Recognition of face is executed according to the optimization face picture characteristic.
8. a kind of face identification method, comprising:
Face picture characteristic is extracted from facial image by neural network model；
Vivo identification result is determined by biopsy method described in Claims 1 to 5 any one；
Using handled by the attention mechanism based on channel region the face picture characteristic processing result reinforce described in
Biological characteristic in face picture characteristic obtains optimization face picture characteristic；
Recognition of face is executed according to the optimization face picture characteristic.
9. a kind of living body detection device, comprising:
Feature obtains module, is configured as obtaining the face picture characteristic in facial image by deep learning；
Feature processing block is configured as handling the face picture characteristic by the attention mechanism based on channel region；
Vivo identification module is configured as determining vivo identification according to the processing result of the feature processing block as a result, described
Vivo identification result includes that the facial image is living body image or non-living body image.
10. device according to claim 9, wherein the feature obtains the neural network that module is face identification system
Model.
11. device according to claim 9, wherein the feature processing block includes:
Convolutional layer is configured as obtaining biological characteristic and non-biometric according to the face picture characteristic；
Pond layer is configured as obtaining pond feature；
Full connection processing unit, is configured as according to the pond feature, and it is special to be associated with the biology by full Connection Neural Network
It seeks peace the non-biometric, obtains processing result.
12. device according to claim 11, wherein the full connection processing unit is configured as:
The pond feature is inputted into the first full articulamentum, and it is raw to pass sequentially through line rectification function, the second full articulamentum and S type
Long curve sigmoid function, obtains the processing result, to determine the vivo identification result according to the processing result.
13. device according to claim 9, wherein include passing through convolutional neural networks phase in the living body detection device
Interval, the more than two feature processing blocks being connected in series；
The vivo identification module is configured as being determined according to the processing result of the last one concatenated feature processing block
Vivo identification result.
14. device according to claim 12, wherein the feature processing block is additionally configured to using described
The output of sigmoid function carries out scaling processing to channel each in the face picture characteristic, obtains optimization people
Face picture feature data, to execute recognition of face according to the optimization face picture characteristic.
15. according to device described in claim 9~13 any one, wherein the feature processing block is additionally configured to benefit
Reinforce the biological characteristic in the face picture characteristic with the processing result of the feature processing block, obtains optimization face
Picture feature data；
Described device further include: face recognition module is configured as executing face according to the optimization face picture characteristic
Identification.
16. a kind of living body detection device, comprising:
Memory；And
It is coupled to the processor of the memory, the processor is configured to based on the instruction execution for being stored in the memory
Method as described in any one of claim 1 to 7.
17. a kind of face detection system, comprising:
Living body detection device described in claim 9~16 any one；With,
Face identification device is configured as:
Face picture characteristic is extracted from facial image；
The face picture characteristic is handled by the attention mechanism based on channel region using the living body detection device
Processing result reinforces the biological characteristic in the face picture characteristic, obtains optimization face picture characteristic；
Recognition of face is executed according to the optimization face picture characteristic.
18. a kind of face detection system, comprising: memory；And
It is coupled to the processor of the memory, the processor is configured to based on the instruction execution for being stored in the memory
Method according to claim 8.
19. a kind of computer readable storage medium, is stored thereon with computer program instructions, real when which is executed by processor
The step of method described in existing claim 1 to 8 any one.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811329008.7A CN109409322B (en) | 2018-11-09 | 2018-11-09 | Living body detection method and device, face recognition method and face detection system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811329008.7A CN109409322B (en) | 2018-11-09 | 2018-11-09 | Living body detection method and device, face recognition method and face detection system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109409322A true CN109409322A (en) | 2019-03-01 |
| CN109409322B CN109409322B (en) | 2020-11-24 |
Family
ID=65472412
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811329008.7A Active CN109409322B (en) | 2018-11-09 | 2018-11-09 | Living body detection method and device, face recognition method and face detection system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109409322B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110866454A (en) * | 2019-10-23 | 2020-03-06 | 智慧眼科技股份有限公司 | Human face living body detection method and system and computer readable storage medium |
| CN112597885A (en) * | 2020-12-22 | 2021-04-02 | 北京华捷艾米科技有限公司 | Face living body detection method and device, electronic equipment and computer storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105956572A (en) * | 2016-05-15 | 2016-09-21 | 北京工业大学 | In vivo face detection method based on convolutional neural network |
| CN107392187A (en) * | 2017-08-30 | 2017-11-24 | 西安建筑科技大学 | A kind of human face in-vivo detection method based on gradient orientation histogram |
| CN107545248A (en) * | 2017-08-24 | 2018-01-05 | 北京小米移动软件有限公司 | Biological characteristic biopsy method, device, equipment and storage medium |
-
2018
- 2018-11-09 CN CN201811329008.7A patent/CN109409322B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105956572A (en) * | 2016-05-15 | 2016-09-21 | 北京工业大学 | In vivo face detection method based on convolutional neural network |
| CN107545248A (en) * | 2017-08-24 | 2018-01-05 | 北京小米移动软件有限公司 | Biological characteristic biopsy method, device, equipment and storage medium |
| CN107392187A (en) * | 2017-08-30 | 2017-11-24 | 西安建筑科技大学 | A kind of human face in-vivo detection method based on gradient orientation histogram |
Non-Patent Citations (2)
| Title |
|---|
| JIE HU ET AL.: "Squeeze-and-Excitation Networks", 《ARXIV》 * |
| MEI WANG ET AL.: "Deep Face Recogniton: A Survey", 《ARXIV》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110866454A (en) * | 2019-10-23 | 2020-03-06 | 智慧眼科技股份有限公司 | Human face living body detection method and system and computer readable storage medium |
| CN112597885A (en) * | 2020-12-22 | 2021-04-02 | 北京华捷艾米科技有限公司 | Face living body detection method and device, electronic equipment and computer storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109409322B (en) | 2020-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105138993B (en) | Establish the method and device of human face recognition model | |
| Han et al. | Demographic estimation from face images: Human vs. machine performance | |
| CN109460734B (en) | Video behavior identification method and system based on hierarchical dynamic depth projection difference image representation | |
| CN114502061A (en) | Image-based automatic skin diagnosis using deep learning | |
| CN108596041B (en) | A kind of human face in-vivo detection method based on video | |
| CN105518710B (en) | Video detecting method, video detection system and computer program product | |
| CN106778785B (en) | Construct the method for image Feature Selection Model and the method, apparatus of image recognition | |
| KR20160101973A (en) | System and method for identifying faces in unconstrained media | |
| CN108875485A (en) | A kind of base map input method, apparatus and system | |
| Rattani et al. | Convolutional neural network for age classification from smart-phone based ocular images | |
| CN104143086A (en) | Application technology of portrait comparison to mobile terminal operating system | |
| US11992331B2 (en) | Neonatal pain identification from neonatal facial expressions | |
| CN102184016B (en) | Noncontact type mouse control method based on video sequence recognition | |
| CN107590473A (en) | A kind of human face in-vivo detection method, medium and relevant apparatus | |
| Güney et al. | Cross-pose facial expression recognition | |
| CN110309767A (en) | In vivo detection equipment, recognition methods, device and storage medium | |
| CN109409322A (en) | Biopsy method, device and face identification method and face detection system | |
| CN108629265A (en) | Method and apparatus for Pupil diameter | |
| TWI357022B (en) | Recognizing apparatus and method for facial expres | |
| Nahar et al. | Twins and Similar Faces Recognition Using Geometric and Photometric Features with Transfer Learning | |
| Shukla et al. | Deep Learning Model to Identify Hide Images using CNN Algorithm | |
| KR101734212B1 (en) | Facial expression training system | |
| JP7270304B2 (en) | Method and mobile device for implementing the method for verifying the identity of a user by identifying an object in an image that has the user's biometric characteristics | |
| Hsu et al. | Extraction of visual facial features for health management | |
| CN110532869A (en) | Expression recognition method, device and computer readable storage medium |
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 |