CN109101899A - A kind of method for detecting human face and system based on convolutional neural networks - Google Patents
A kind of method for detecting human face and system based on convolutional neural networks Download PDFInfo
- Publication number
- CN109101899A CN109101899A CN201810813668.6A CN201810813668A CN109101899A CN 109101899 A CN109101899 A CN 109101899A CN 201810813668 A CN201810813668 A CN 201810813668A CN 109101899 A CN109101899 A CN 109101899A
- Authority
- CN
- China
- Prior art keywords
- esfd
- s3fd
- face
- detected
- enhanced
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The embodiment of the invention provides a kind of method for detecting human face based on convolutional neural networks, it include: in the modified single step multiple scale detecting device ESFD that picture to be detected is input to after training, the ESFD is to joined enhanced pilotaxitic texture and contextual feature fusion structure on the basis of S3FD, change anchor strategy and construct after adjusting loss function；Obtain the Face datection result of picture to be detected described in the detection output layer of institute ESFD.A kind of method for detecting human face and system based on convolutional neural networks provided in an embodiment of the present invention, by the network structure for modifying S3FD, it joined enhanced pilotaxitic texture and contextual feature Fusion Module in the network architecture of original S3FD, change anchor strategy and adjust loss function to constitute new ESFD framework, it enhances the Utilization ability to feature and improves the detectability to small face, to improve the recall rate and accuracy rate of Face datection.
Description
Technical field
The present embodiments relate to human face detection tech field more particularly to a kind of face inspections based on convolutional neural networks
Survey method and system.
Background technique
In recent years, since SSD training frame has detection speed fast, detect the high feature of accuracy rate in terms of big object and
It is paid close attention to by people.
In the prior art, the Face datection frame based on SSD is also mostly used, the core of SSD is on characteristic pattern using volume
Core is accumulated to predict classification score, the offset of a series of default bounding boxes, and in order to improve Detection accuracy,
It is predicted on the characteristic pattern of different scale.
But the Face datection frame of existing SSD is limited to low-level feature abstract deficiency and it is caused to detect small face side
Face performance is relatively poor, lower so as to cause the recall rate and accuracy rate entirely detected, therefore needs a kind of new face now
Detection method solves the above problems.
Summary of the invention
To solve the above-mentioned problems, the embodiment of the present invention provides one kind and overcomes the above problem or at least be partially solved
State the method for detecting human face and system based on convolutional neural networks of problem.
The first aspect embodiment of the present invention provides a kind of method for detecting human face based on convolutional neural networks, comprising:
By picture to be detected be input to training after modified single step multiple scale detecting device ESFD in, the ESFD be
It joined enhanced pilotaxitic texture and contextual feature fusion structure on the basis of S3FD, change anchor strategy and adjust loss
It is constructed after function；
Obtain the Face datection result of picture to be detected described in the detection output layer of institute ESFD.
The embodiment of the invention also provides a kind of face detection systems based on convolutional neural networks for second aspect, comprising:
Input module, for picture to be detected to be input in the modified single step multiple scale detecting device ESFD after training,
The ESFD is to joined enhanced pilotaxitic texture and contextual feature fusion structure on the basis of S3FD, change anchor plan
It omits and adjusts and construct after loss function；
Output module is detected, the Face datection knot of picture to be detected described in the detection output layer of ESFD for obtaining
Fruit.
The embodiment of the invention provides a kind of human-face detection equipments based on convolutional neural networks for the third aspect, comprising:
Processor, memory, communication interface and bus；Wherein, the processor, memory, communication interface pass through described
Bus completes mutual communication；The memory is stored with the program instruction that can be executed by the processor, the processor
Described program instruction is called to be able to carry out a kind of method for detecting human face based on convolutional neural networks described above.
The embodiment of the invention provides a kind of non-transient computer readable storage medium, the non-transient calculating for fourth aspect
Machine readable storage medium storing program for executing stores computer instruction, and it is above-mentioned based on convolutional Neural net that the computer instruction executes the computer
The method for detecting human face of network.
A kind of method for detecting human face and system based on convolutional neural networks provided in an embodiment of the present invention, passes through modification
The network structure of S3FD joined enhanced pilotaxitic texture and contextual feature fusion mould in the network architecture of original S3FD
Block changes anchor strategy and adjusts loss function constituting new ESFD framework, enhancing the Utilization ability to feature and mentioning
The detectability to small face is risen, to improve the recall rate and accuracy rate of Face datection.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is this hair
Bright some embodiments for those of ordinary skill in the art without creative efforts, can be with root
Other attached drawings are obtained according to these attached drawings.
Fig. 1 is a kind of method for detecting human face flow chart based on convolutional neural networks provided in an embodiment of the present invention；
Fig. 2 is the configuration diagram of ESFD model provided in an embodiment of the present invention；
Fig. 3 is enhanced pilotaxitic texture schematic diagram provided in an embodiment of the present invention；
Fig. 4 is the schematic diagram of intermediate merging module provided in an embodiment of the present invention；
Fig. 5 is contextual feature fusion structure schematic diagram provided in an embodiment of the present invention；
Fig. 6 is a kind of face detection system structure chart based on convolutional neural networks provided in an embodiment of the present invention；
Fig. 7 is the structural block diagram of the human-face detection equipment provided in an embodiment of the present invention based on convolutional neural networks.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical solution in the embodiment of the present invention is explicitly described, it is clear that described embodiment is the present invention
A part of the embodiment, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not having
Every other embodiment obtained under the premise of creative work is made, shall fall within the protection scope of the present invention.
Currently, existing SSD (Single Shot MultiBox Detector) training frame have detection speed it is fast,
And it detects the big object aspect high feature of accuracy rate and is paid close attention to by people, but be limited to low-level feature abstract deficiency to lead to it
Performance is relatively poor in terms of detecting small face, therefore how to effectively improve performance of the SSD in terms of detecting small face and mention
The problem of recall rate and accuracy rate of high entire detector are present urgent need to resolve.
In view of the above-mentioned problems, Fig. 1 is a kind of Face datection side based on convolutional neural networks provided in an embodiment of the present invention
Method flow chart, as shown in Figure 1, comprising:
101, picture to be detected is input in the modified single step multiple scale detecting device ESFD after training, the ESFD is
It joined enhanced pilotaxitic texture and contextual feature fusion structure on the basis of S3FD, change anchor strategy and adjust damage
It is constructed after mistake function；
102, the Face datection result of picture to be detected described in the detection output layer of institute ESFD is obtained.
In a step 101, it is to be understood that the Face datection provided in an embodiment of the present invention based on convolutional neural networks
Method is the modified scheme on SSD frame, and SSD is a kind of existing convolutional neural networks deep learning frame.In SSD frame
On the basis of frame, the prior art provides a kind of S3FD (Single Shot Scale-invariant Face Detector)
Frame, thinking is similar with SSD, is to predict target using the anchor of different scale in different layers, although having to SSD frame
It is improved, but still there is a problem that the small face aspect performance of detection is relatively poor.
For the problems in above-mentioned S3FD frame, the embodiment of the invention provides a kind of modified single step multiple scale detecting devices
S3FD (Enhanced SFD) model.Follow-on S3FD is known as ESFD by the embodiment of the present invention.Specifically, the embodiment of the present invention
The ESFD of offer is mainly made of two parts, and a part is by VGG-16 as basic network, and another part is that ESFD is newly added
Feature extraction layer, wherein newly added feature extraction layer mainly include enhanced pilotaxitic texture and contextual feature fusion knot
Structure.It should be noted that causing the detection performance in terms of small face since existing S3FD cannot sufficiently excavate small face characteristic
Poor, small face here refers to size in 50px*50px face below.The embodiment of the present invention utilizes enhanced pilotaxitic texture
The character network of S3FD is improved, it is to be understood that since enhanced pilotaxitic texture has combined multiple convolutional layers
Information can extract more profound feature compared to simple pilotaxitic texture, be reinforced the characterization ability of face,
So same detection can also can be carried out for small face characteristic.On the other hand, upper due to that cannot be utilized in existing S3FD
Feature hereafter carries out synthesis utilization, poor so as to cause detection performance, and the embodiment of the present invention also provides on one accordingly
Following traits fusion structure carries out Fusion Features again to the feature that enhancing pilotaxitic texture is extracted and reinforces, to further mention
High detection performance.Also, anchor in existing S3FD there are aiming at the problem that and loss function there are the problem of, the present invention implement
The ESFD that example provides also has carried out the measures such as corresponding change anchor strategy, adjustment loss function.
In a step 102, it is to be understood that after facial image to be detected is input to ESFD, process is enhanced
After pilotaxitic texture and contextual feature fusion structure Fusion Features, output result is accessed into prediction module, i.e., the present invention is implemented
Face datection is carried out in the detection output layer that example provides.It is understood that detection output layer can be to facial image to be detected
It is detected.The main detection mode for using face frame, Testing index mainly use recall rate and accuracy rate two indices, recall
Rate refers to the ratio of somebody's face number in the face frame number and picture to be detected detected itself；Accuracy rate, which refers to, to be detected
The ratio of the number of face frame and the framed number detected.Implementation of the present invention is able to confirm that by the two Testing index
The detection performance for the method for detecting human face that example provides is better than traditional S3FD.
A kind of method for detecting human face and system based on convolutional neural networks provided in an embodiment of the present invention, passes through modification
The network structure of S3FD joined enhanced pilotaxitic texture and contextual feature fusion mould in the network architecture of original S3FD
Block enhances the Utilization ability to feature and improves the detectability to small face, to mention to constitute new ESFD framework
The high recall rate and accuracy rate of Face datection.
On the basis of the above embodiments, described before in the ESFD after picture to be detected to be input to training described
Method further include:
Enhanced pilotaxitic texture and contextual feature fusion structure is added, in the S3FD to construct the ESFD；
The training ESFD, to obtain the ESFD after the training.
By the content of above-described embodiment it is found that the embodiment of the present invention is substantially to carry out on the basis of S3FD conventional architectures
It improves, it is improved to the effect that joined enhanced pilotaxitic texture and contextual feature fusion structure in S3FD, from
And constitute new ESFD model.Fig. 2 is the configuration diagram of ESFD model provided in an embodiment of the present invention, as shown in Fig. 2, right
This several layers of convolutional layer of Conv3_3, Conv4_3, Conv5_2, Conv_fc7, Conv6_2, Conv7_2 use enhanced single fisherman's knot
Structure, wherein Conv3_3, Conv4_3, Conv5_2 can be normalized by norm layer, the output of these layers and two 3
The convolution nuclear phase of × 3 sizes is rolled up to obtain characteristic value, the probability value of an output category, and each default box generates 2
Probability value, the relative position coordinates of an output regression, 4 relative coordinate values of each default box generation (x, y, w,
h).In addition this 6 convolutional layers also pass through priorBox layers of generation default box (generation is original coordinates).It is described above
6 convolutional layers in each layer of the quantity of default box be 1.Finally the calculated result of front three is merged so respectively
After pass to loss layers of calculating loss and then carry out rear feed, regularized learning algorithm parameter.
On the basis of the above embodiments, described that enhanced pilotaxitic texture and contextual feature are added in the S3FD
Fusion structure is specifically included with constructing the ESFD:
The enhanced pilotaxitic texture is used, in several preset convolutional layers of the S3FD to constitute netted output knot
Structure；
Based on the contextual feature fusion structure, contextual feature fusion is carried out to the netted export structure, with structure
Build the ESFD.
By the content of above-described embodiment it is found that the embodiment of the present invention is will joined in S3FD enhanced pilotaxitic texture and
Contextual feature fusion structure, to constitute new ESFD model, Fig. 3 is enhanced pilotaxitic texture provided in an embodiment of the present invention
Schematic diagram, Fig. 4 are the schematic diagrames of intermediate merging module provided in an embodiment of the present invention, and Fig. 5 is on provided in an embodiment of the present invention
Following traits fusion structure schematic diagram, as shown in Fig. 3, Fig. 4 and Fig. 5, specifically, the embodiment of the present invention is in training and detects rank
Section, to Conv3_3, Conv4_3, Conv5_2, Conv_fc7, Conv6_2, Conv7_2, this several layers of convolutional layer uses multiple layer interleaving
Structure forms reticular structure.As shown in figure 3, wherein adjacent three kinds of different layers are not that simple channel merges, need to illustrate
Multiple layer interleaving structure described in enhancing intertexture framework, that is, embodiment of the present invention in Fig. 3, including synthesis module, that is, it is complete
At the submodule of intermediate pooling function.Low layer and 3 × 3 convolution kernel carry out convolution and then connect pooling layer of max, high-rise to distinguish
By 3 × 3 convolution kernel then reversed convolutional layer, then with middle layer amalgamation result, as shown in Figure 4.Then defeated to its respectively
Result carries out contextual feature fusion out.Wherein three branches use 2,4,63 × 3 convolution kernels to carry out convolution according to this, then make
Result is merged to and utilized 1 × 1 convolution kernel with concat layers and increases nonlinear characteristic, as shown in Figure 5.It is finally respectively that this is several layers of
Corresponding output result accesses prediction module, and prediction module is input to MultiBoxLoss layers of calculating Loss simultaneously for result is exported
Rear feed, the weight of regularized learning algorithm and biasing.
On the basis of the above embodiments, the building ESFD further include:
SoftmaxLoss layer in the S3FD is replaced with FocalLoss layers.
By the content of above-described embodiment it is found that the embodiment of the invention provides a kind of ESFD to improve in original S3FD model
Deficiency.But in original S3FD model, loss function uses SoftmaxLoss layers, and SoftmaxLoss layers have centainly
The classical strength to difficult sample, but intensity or relatively low.
In order to promote classical strength, the embodiment of the present invention preferably used FocalLoss layers as loss function, this
Loss function is that modification obtains on the basis of standard intersects entropy loss.This function can pass through the power of the easy classification samples of reduction
Weight, so that model focuses more on the sample of difficult classification in training.
Original SoftmaxLoss layers of loss is substituted as loss function by FocalLoss layers in the embodiment of the present invention
Function improves network to the study dynamics of difficult sample, improves network inspection so that increasing the classical strength to difficult sample when classification
Survey performance.
On the basis of the above embodiments, the building ESFD further include:
Fixation anchor structure in the S3FD is replaced with into random anchor structure.
The fixation anchor structure by the S3FD replaces with random anchor structure, specifically includes:
Image scaled in the fixed anchor structure is revised as preset ratio；
The position coordinates of characteristic point in preset ratio image are added into random offset, to constitute the random anchor knot
Structure.
By the content of above-described embodiment it is found that the embodiment of the present invention has carried out comprehensive improvement to S3FD frame.As
An important ring in S3FD frame, anchor structure, which also has, is correspondingly improved space.
It should be noted that anchor structure is a kind of structure in S3FD, for generating face candidate frame, anchor and number
This anchor is just noted as positive sample after being greater than certain value according to the frame degree of overlapping of mark, is not matched to the mark of face
For negative sample, anchor structure is mostly used to classify and return task.But the anchor in S3FD frame use it is fixed
Anchor structure, the mode of this fixation make in the limited situation of training data, cannot well mining data information
It is trained.
In view of the above-mentioned problems, the embodiment of the present invention proposes a kind of strategy of random anchor structure to improve.Tool
Body, since face length-width ratio ratio is largely 1:1 or 1:1.5, and the ratio used under original S3FD frame or 1:
2 or 1:3, this ratio and the length-width ratio for mismatching face modify anchor length and width so the embodiment of the present invention is corrected
Than for 1:1 and two kinds of 1:1.5, abandoning 1:2, the anchor of 1:3 ratio, calculation amount is reduced, the matching capacity of face is promoted.
Further, in the original anchor generation strategy of S3FD, for each characteristic pattern (feature map)
Point can all generate the different anchorbox of multiple length-width ratios, the position of each anchor is the position coordinates (x, y) of characteristic pattern
In addition offset (offset), default offset=0.5 is conversed then multiplied by the ratio of current original image and characteristic pattern
Position of the anchor on original image, this position are the centre coordinates of anchorbox, then according to specified length and width
Angle value generates coordinate of the anchorbox in original image, and (length and width is generated according to the ratio of 1:1 and 1:1.5, such characteristic pattern
Each point corresponded to 2 anchor box).In order to excavate training data information as far as possible, this embodiment of the present invention is proposed
A kind of random anchor strategy.Offset value is become into random value, random value range is [- 0.5,0.5], in this way for same
One figure, the position coordinates of anchorbox generated when training every time be it is unfixed, can be easier to be matched to mark
The position of frame is more advantageous to e-learning face characteristic.
On the basis of the above embodiments, the face of picture to be detected described in the detection output layer for obtaining institute ESFD
Testing result, comprising:
Based on preset Soft-NMS algorithm in the detection output layer, the face that classification score is greater than preset threshold is obtained
Frame is as the Face datection result.
By the content of above-described embodiment it is found that the embodiment of the present invention provide ESFD model can to facial image to be detected into
Row detection, detection process carry out output result dependent on the detection output layer of ESFD model.But it is utilized under existing S3FD frame
NMS algorithm can lose face in the case where face has overlapping cases, cause output testing result it is ineffective.
In view of the above-mentioned problems, the embodiment of the invention provides a kind of new Soft-NMS algorithms to change to NMS algorithm
Into.It should be noted that NMS algorithm is alternatively referred to as non-maxima suppression algorithm, thinking mainly searches for local maximum, suppression
Make non-maximum element.And Soft-NMS algorithm provided in an embodiment of the present invention alternatively referred to as softens non-maxima suppression algorithm,
Common practice is to will test frame to sort by score, and then keep score highest frame, while being deleted big with the frame overlapping area
In a certain proportion of other frames, thinking not delete all frames greater than threshold value directly mainly, but reduce its confidence level.
Specifically, being sized the picture to be detected of input for 640 × 640 pixels, it is then input to implementation of the present invention
It is calculated in the ESFD network that example provides, classification score in detection output layer is obtained according to Soft-NMS algorithm and is greater than certain threshold
The face frame of value, and count recall rate and accuracy rate.
Soft-NMS algorithm provided in an embodiment of the present invention can further reduce the loss of testing result, promote detection
Recall rate.
Fig. 6 is a kind of face detection system structure chart based on convolutional neural networks provided in an embodiment of the present invention, such as Fig. 6
It is shown, the system comprises: input module 601 and detection output module 602, in which:
Input module 601 is used for for picture to be detected to be input to the modified single step multiple scale detecting device after training
In ESFD, the ESFD is to joined enhanced pilotaxitic texture and contextual feature fusion structure on the basis of S3FD, change
What anchor strategy and adjusting constructed after loss function；
Detection output module 602 is used to obtain the Face datection knot of picture to be detected described in the detection output layer of institute ESFD
Fruit.
It is specific how by input module 601 and to detect output module 602 and can be used for executing Fig. 1 institute to Face datection
The technical solution for the method for detecting human face embodiment shown, it is similar that the realization principle and technical effect are similar, and details are not described herein again.
A kind of method for detecting human face and system based on convolutional neural networks provided in an embodiment of the present invention, passes through modification
The network structure of S3FD joined enhanced pilotaxitic texture and contextual feature fusion mould in the network architecture of original S3FD
Block changes anchor strategy and adjusts loss function constituting new ESFD framework, enhancing the Utilization ability to feature and mentioning
The detectability to small face is risen, to improve the recall rate and accuracy rate base provided in an embodiment of the present invention of Face datection
In the human-face detection equipment of convolutional neural networks, comprising: at least one processor；And connect with the processor communication to
A few memory, in which:
Fig. 7 is the structural block diagram of the human-face detection equipment provided in an embodiment of the present invention based on convolutional neural networks, reference
Fig. 7, the human-face detection equipment based on convolutional neural networks, comprising: processor (processor) 710, communication interface
(Communications Interface) 720, memory (memory) 730 and bus 740, wherein processor 710, communication
Interface 720, memory 730 complete mutual communication by bus 740.Processor 710 can call patrolling in memory 730
Instruction is collected, to execute following method: picture to be detected is input in the modified single step multiple scale detecting device ESFD after training,
The ESFD is to joined enhanced pilotaxitic texture and contextual feature fusion structure on the basis of S3FD, change anchor plan
It omits and adjusts and construct after loss function；Obtain the Face datection knot of picture to be detected described in the detection output layer of institute ESFD
Fruit.
The embodiment of the present invention discloses a kind of computer program product, and the computer program product is non-transient including being stored in
Computer program on computer readable storage medium, the computer program include program instruction, when described program instructs quilt
When computer executes, computer is able to carry out method provided by above-mentioned each method embodiment, for example, by picture to be detected
Be input to training after modified single step multiple scale detecting device ESFD in, the ESFD be joined on the basis of S3FD it is enhanced
Pilotaxitic texture and contextual feature fusion structure change anchor strategy and adjust and construct after loss function；Obtain institute
The Face datection result of picture to be detected described in the detection output layer of ESFD.
The embodiment of the present invention provides a kind of non-transient computer readable storage medium, the non-transient computer readable storage
Medium storing computer instruction, the computer instruction make the computer execute side provided by above-mentioned each method embodiment
Method, for example, picture to be detected is input in the modified single step multiple scale detecting device ESFD after training, the ESFD is
It joined enhanced pilotaxitic texture and contextual feature fusion structure on the basis of S3FD, change anchor strategy and adjust damage
It is constructed after mistake function；Obtain the Face datection result of picture to be detected described in the detection output layer of institute ESFD.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It realizes by means of software and necessary general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on
Stating technical solution, substantially the part that contributes to existing technology can be embodied in the form of software products in other words, should
Computer software product may be stored in a computer readable storage medium, such as ROM/RAM, magnetic disk, CD, including several fingers
It enables and using so that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation
Method described in certain parts of example or embodiment.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it still may be used
To modify the technical solutions described in the foregoing embodiments or equivalent replacement of some of the technical features；
And these are modified or replaceed, technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (10)
1. a kind of method for detecting human face based on convolutional neural networks characterized by comprising
Picture to be detected is input in the modified single step multiple scale detecting device ESFD after training, the ESFD is in S3FD base
After joined enhanced pilotaxitic texture and contextual feature fusion structure on plinth, changing anchor strategy and adjust loss function
Building；
Obtain the Face datection result of picture to be detected described in the detection output layer of institute ESFD.
2. the method according to claim 1, wherein picture to be detected to be input to the ESFD after training described
In before, the method also includes:
Enhanced pilotaxitic texture and contextual feature fusion structure is added, in the S3FD to construct the ESFD；
The training ESFD, to obtain the ESFD after the training.
3. according to the method described in claim 2, it is characterized in that, it is described be added in the S3FD enhanced pilotaxitic texture with
And contextual feature fusion structure is specifically included with constructing the ESFD:
The enhanced pilotaxitic texture is used, in several preset convolutional layers of the S3FD to constitute netted export structure；
Based on the contextual feature fusion structure, contextual feature fusion is carried out to the netted export structure, to construct
State ESFD.
4. according to the method described in claim 3, it is characterized in that, the building ESFD further include:
SoftmaxLoss layer in the S3FD is replaced with FocalLoss layers.
5. according to the method described in claim 3, it is characterized in that, the building ESFD further include:
Fixation anchor structure in the S3FD is replaced with into random anchor structure.
6. according to the method described in claim 5, it is characterized in that, the fixation anchor structure by the S3FD is replaced
For random anchor structure, specifically include:
Image scaled in the fixed anchor structure is revised as preset ratio；
The position coordinates of characteristic point in preset ratio image are added into random offset, to constitute the random anchor structure.
7. the method according to claim 1, wherein to be checked described in the detection output layer for obtaining institute ESFD
The Face datection result of mapping piece, comprising:
Based on preset Soft-NMS algorithm in the detection output layer, the face frame work that classification score is greater than preset threshold is obtained
For the Face datection result.
8. a kind of face detection system based on convolutional neural networks characterized by comprising
Input module, it is described for picture to be detected to be input in the modified single step multiple scale detecting device ESFD after training
ESFD is to joined enhanced pilotaxitic texture and contextual feature fusion structure on the basis of S3FD, change anchor strategy simultaneously
It is constructed after adjustment loss function；
Output module is detected, the Face datection result of picture to be detected described in the detection output layer of ESFD for obtaining.
9. a kind of human-face detection equipment based on convolutional neural networks, which is characterized in that including memory and processor, the place
Reason device and the memory complete mutual communication by bus；The memory is stored with and can be executed by the processor
Program instruction, the processor call described program instruction to be able to carry out the method as described in claim 1 to 7 is any.
10. a kind of non-transient computer readable storage medium, which is characterized in that the non-transient computer readable storage medium is deposited
Computer instruction is stored up, the computer instruction makes the computer execute method as described in any one of claim 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810813668.6A CN109101899B (en) | 2018-07-23 | 2018-07-23 | Face detection method and system based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810813668.6A CN109101899B (en) | 2018-07-23 | 2018-07-23 | Face detection method and system based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109101899A true CN109101899A (en) | 2018-12-28 |
| CN109101899B CN109101899B (en) | 2020-11-24 |
Family
ID=64847241
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810813668.6A Active CN109101899B (en) | 2018-07-23 | 2018-07-23 | Face detection method and system based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109101899B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110197113A (en) * | 2019-03-28 | 2019-09-03 | 杰创智能科技股份有限公司 | A kind of method for detecting human face of high-precision anchor point matching strategy |
| CN110197217A (en) * | 2019-05-24 | 2019-09-03 | 中国矿业大学 | It is a kind of to be interlocked the image classification method of fused packet convolutional network based on depth |
| CN110263731A (en) * | 2019-06-24 | 2019-09-20 | 电子科技大学 | A kind of single step face detection system |
| CN110490054A (en) * | 2019-07-08 | 2019-11-22 | 北京三快在线科技有限公司 | Detection method, device, electronic equipment and the readable storage medium storing program for executing of target area |
| CN110619460A (en) * | 2019-09-05 | 2019-12-27 | 北京邮电大学 | Classroom quality assessment system and method based on deep learning target detection |
| CN111144310A (en) * | 2019-12-27 | 2020-05-12 | 创新奇智(青岛)科技有限公司 | Face detection method and system based on multi-layer information fusion |
| CN112528851A (en) * | 2020-12-09 | 2021-03-19 | 南京航空航天大学 | Face detection method and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140079297A1 (en) * | 2012-09-17 | 2014-03-20 | Saied Tadayon | Application of Z-Webs and Z-factors to Analytics, Search Engine, Learning, Recognition, Natural Language, and Other Utilities |
| CN107220618A (en) * | 2017-05-25 | 2017-09-29 | 中国科学院自动化研究所 | Method for detecting human face and device, computer-readable recording medium, equipment |
| CN107403141A (en) * | 2017-07-05 | 2017-11-28 | 中国科学院自动化研究所 | Method for detecting human face and device, computer-readable recording medium, equipment |
| CN107871101A (en) * | 2016-09-23 | 2018-04-03 | 北京眼神科技有限公司 | A kind of method for detecting human face and device |
| CN108090417A (en) * | 2017-11-27 | 2018-05-29 | 上海交通大学 | A kind of method for detecting human face based on convolutional neural networks |
-
2018
- 2018-07-23 CN CN201810813668.6A patent/CN109101899B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140079297A1 (en) * | 2012-09-17 | 2014-03-20 | Saied Tadayon | Application of Z-Webs and Z-factors to Analytics, Search Engine, Learning, Recognition, Natural Language, and Other Utilities |
| CN107871101A (en) * | 2016-09-23 | 2018-04-03 | 北京眼神科技有限公司 | A kind of method for detecting human face and device |
| CN107220618A (en) * | 2017-05-25 | 2017-09-29 | 中国科学院自动化研究所 | Method for detecting human face and device, computer-readable recording medium, equipment |
| CN107403141A (en) * | 2017-07-05 | 2017-11-28 | 中国科学院自动化研究所 | Method for detecting human face and device, computer-readable recording medium, equipment |
| CN108090417A (en) * | 2017-11-27 | 2018-05-29 | 上海交通大学 | A kind of method for detecting human face based on convolutional neural networks |
Non-Patent Citations (4)
| Title |
|---|
| CHENCHEN ZHU ET.AL: ""Seeing Small Faces from Robust Anchor"s Perspective"", 《2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| JIANFENG WANG ET.AL: ""Face Attention Network: An Effective Face Detector for the Occluded Faces"", 《RESEARCHGATE》 * |
| SHIFENG ZHANG ET.AL: ""S3FD: Single Shot Scale-invariant Face Detector"", 《2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION》 * |
| XU TANG ET.AL: ""PyramidBox: A Context-assisted Single Shot Face Detector"", 《RESEARCHGATE》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110197113A (en) * | 2019-03-28 | 2019-09-03 | 杰创智能科技股份有限公司 | A kind of method for detecting human face of high-precision anchor point matching strategy |
| CN110197113B (en) * | 2019-03-28 | 2021-06-04 | 杰创智能科技股份有限公司 | Face detection method of high-precision anchor point matching strategy |
| CN110197217A (en) * | 2019-05-24 | 2019-09-03 | 中国矿业大学 | It is a kind of to be interlocked the image classification method of fused packet convolutional network based on depth |
| CN110263731A (en) * | 2019-06-24 | 2019-09-20 | 电子科技大学 | A kind of single step face detection system |
| CN110490054A (en) * | 2019-07-08 | 2019-11-22 | 北京三快在线科技有限公司 | Detection method, device, electronic equipment and the readable storage medium storing program for executing of target area |
| CN110490054B (en) * | 2019-07-08 | 2021-03-09 | 北京三快在线科技有限公司 | Target area detection method and device, electronic equipment and readable storage medium |
| CN110619460A (en) * | 2019-09-05 | 2019-12-27 | 北京邮电大学 | Classroom quality assessment system and method based on deep learning target detection |
| CN111144310A (en) * | 2019-12-27 | 2020-05-12 | 创新奇智(青岛)科技有限公司 | Face detection method and system based on multi-layer information fusion |
| CN112528851A (en) * | 2020-12-09 | 2021-03-19 | 南京航空航天大学 | Face detection method and system |
| CN112528851B (en) * | 2020-12-09 | 2024-03-29 | 南京航空航天大学 | Face detection method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109101899B (en) | 2020-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109101899A (en) | A kind of method for detecting human face and system based on convolutional neural networks | |
| US10769493B2 (en) | Method and apparatus for neural network training and construction and method and apparatus for object detection | |
| CN111401376B (en) | Target detection method, target detection device, electronic equipment and storage medium | |
| CN109815868A (en) | A kind of image object detection method, device and storage medium | |
| CN109815770B (en) | Two-dimensional code detection method, device and system | |
| US10885660B2 (en) | Object detection method, device, system and storage medium | |
| CN106845383A (en) | People's head inspecting method and device | |
| CN112949508A (en) | Model training method, pedestrian detection method, electronic device and readable storage medium | |
| CN110210482B (en) | Target detection method for improving class imbalance | |
| CN111524145A (en) | Intelligent picture clipping method and system, computer equipment and storage medium | |
| CN108664838A (en) | Based on the monitoring scene pedestrian detection method end to end for improving RPN depth networks | |
| CN110807362A (en) | Image detection method and device and computer readable storage medium | |
| CN106650615A (en) | Image processing method and terminal | |
| CN107240087A (en) | Object detection system and method | |
| JP2023530796A (en) | Recognition model training method, recognition method, device, electronic device, storage medium and computer program | |
| Zhang et al. | An improved YOLO network for unopened cotton boll detection in the field | |
| Hu et al. | RGB-D image multi-target detection method based on 3D DSF R-CNN | |
| CN113487610B (en) | Herpes image recognition method and device, computer equipment and storage medium | |
| Jin et al. | Defect identification of adhesive structure based on DCGAN and YOLOv5 | |
| CN112270671B (en) | Image detection method, device, electronic equipment and storage medium | |
| CN113065379B (en) | Image detection method and device integrating image quality and electronic equipment | |
| CN117275086A (en) | Gesture recognition method, gesture recognition device, computer equipment and storage medium | |
| Zhang et al. | JF-YOLO: the jellyfish bloom detector based on deep learning | |
| Hao et al. | General target detection method based on improved SSD | |
| CN117079095A (en) | Deep learning-based high-altitude parabolic detection method, system, medium and equipment |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20201014 Address after: 215000 unit 2-b702, creative industry park, No. 328, Xinghu street, Suzhou Industrial Park, Suzhou City, Jiangsu Province Applicant after: SUZHOU FEISOU TECHNOLOGY Co.,Ltd. Address before: 100876 Beijing, Haidian District, 10 West Road, Beijing, 12 Beijing, North Post Science and technology exchange center, room 1216 Applicant before: BEIJING FEISOU TECHNOLOGY Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |