CN110008819A - A kind of facial expression recognizing method based on figure convolutional neural networks - Google Patents
A kind of facial expression recognizing method based on figure convolutional neural networks Download PDFInfo
- Publication number
- CN110008819A CN110008819A CN201910091261.1A CN201910091261A CN110008819A CN 110008819 A CN110008819 A CN 110008819A CN 201910091261 A CN201910091261 A CN 201910091261A CN 110008819 A CN110008819 A CN 110008819A
- Authority
- CN
- China
- Prior art keywords
- human face
- directed graph
- image
- face expression
- expression
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
-
- 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
- G06V40/174—Facial expression recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of facial expression recognizing methods based on figure convolutional neural networks.Its technical solution is: firstly, obtaining Facial Expression Image set Image and human face expression tag set Label；Then, the undirected set of graphs Graph of human face expression is obtained according to Facial Expression Image set Image；Further according to human face expression undirected set of graphs Graph and human face expression tag set Label, facial expression recognition classifier is established；Finally, input a Facial Expression Image to be identified, carry out gray processing, extract face parts of images, again by the size normalization of face parts of images to m × m, the facial image after normalization is converted into non-directed graph by facial image after being normalized, non-directed graph is input to the classifier of facial expression recognition, obtains the recognition result of human face expression.The present invention has the characteristics that relatively broad characteristics of image, facial expression recognition rate height can be extracted.
Description
Technical field
The invention belongs to facial expression recognition technical fields.More particularly to a kind of face table based on figure convolutional neural networks
Feelings recognition methods.
Background technique
Human face expression is not only the movable physiological performance of human psychology, and has indispensable weight in interpersonal communication
The property wanted.The method of facial expression recognition has two classes: the recognition methods based on machine learning and the recognition methods based on deep learning.
Influence based on the recognition methods of machine learning vulnerable to feature extraction effect superiority and inferiority.And it is based on deep learning identification side
The shortcomings that feature extraction and expression classification are combined together by method, overcome the recognition methods based on machine learning.
In method based on deep learning, convolutional neural networks, which are widely used in, establishes expression classification model.
Matsuguetal(Matsugu M,Mori K,Mitari Y,et al.Subject independent facial
expression recognition with robust face detection using a convolutional
Neural network [J] .Neural Networks, 2003,16 (5-6): 555-559) using convolutional neural networks establish people
Face expression classifier is used for the classification of human face expression.Higher discrimination is although obtained using this method, but uses convolution mind
The classifier established through network can only extract the relevant information of 8 neighborhoods in image, for the phase between pixel apart from each other
The extractability for closing information is poor.Affect accuracy of the classifier to facial expression recognition of foundation.
Summary of the invention
The present invention is directed to overcome prior art defect, and it is an object of the present invention to provide a kind of face table based on figure convolutional neural networks
Feelings recognition methods；This method can extract relevant information between relatively broad pixel, and facial expression recognition rate is high.
To achieve the goals above, the technical solution adopted by the present invention comprises the concrete steps that:
Step 1, the gray level image for obtaining the human face expression that one group of size is m × m, form Facial Expression Image set
Image, by the corresponding human face expression label composition of the gray level image of every human face expression in Facial Expression Image set Image
Human face expression tag set Label；The corresponding human face expression of gray level image of every human face expression therein is labeled as happiness, eats
One of frightened, sad, indignation, detest and fear.
Step 2, by the greyscale image transitions of every human face expression in Facial Expression Image set Image at non-directed graph, group
At the undirected set of graphs Graph of human face expression.
The greyscale image transitions of the human face expression are comprised the concrete steps that at non-directed graph:
Step 2.1, using each of the gray level image of human face expression pixel as a vertex in non-directed graph.
Step 2.2 sets the distance between two pixels adjacent in the gray level image of human face expression as 1.
Step 2.3, by each of the gray level image of human face expression pixel successively centered on pixel, then into
The following operation of row:
First the pixel composition fixation by the Euclidean distance away from central pixel point less than or equal to 2 takes point set Fixed.
The Euclidean distance away from central pixel point is greater than 2 again and the pixel composition less than 4 takes point set Random at random.
Then the vertex of non-directed graph corresponding to central pixel point and fixation are taken into all pixels point institute in point set Fixed
The vertex of corresponding non-directed graph is connected, by the vertex of non-directed graph corresponding to central pixel point with take in point set Random at random
The vertex for randomly selecting non-directed graph corresponding to p pixel is connected.
Step 2.4, the non-directed graph being converted into adjacency matrix W the i-th row jth column numerical value WijAre as follows:
In formula (1): sijIndicate the ith pixel point v in the gray level image of human face expressioniWith j-th of pixel vjBetween
Euclidean distance.
The adjacency matrix W ∈ R of the non-directed graphnxn。
Step 3, according to human face expression undirected set of graphs Graph and human face expression tag set Label, establish human face expression
Recognition classifier:
Step 3.1, a figure convolutional neural networks contain 6 picture scroll laminations, a full articulamentum and a softmax
Layer；Each picture scroll lamination contains the roughened layer of a figure signal wave filtering layer and a figure, determines the input in each picture scroll lamination
The number of non-directed graph, the number for exporting non-directed graph and the size using filter, determine the number of nodes M of full articulamentumf；
Softmax layers have 6 outputs, respectively correspond 6 kinds of basic facial expressions.
Step 3.2, by the 1st picture scroll lamination can training parameter θ1, the 2nd picture scroll lamination can training parameter θ2, the 3rd
A picture scroll lamination can training parameter θ3, the 4th picture scroll lamination can training parameter θ4, the 5th picture scroll lamination can training parameter
θ5, the 6th picture scroll lamination can training parameter θ6, full articulamentum can training parameter θfWith softmax layers can training parameter
θsoftForm figure convolutional neural networks can training parameter set θ.
Step 3.3, by figure convolutional neural networks can training parameter set θ initialize.
Step 3.4, by figure convolutional neural networks can the training parameter set θ and undirected set of graphs Graph of human face expression put
Enter figure convolutional neural networks and carry out propagated forward, obtains human face expression label prediction sets Predict；Human face expression label prediction
Set Predict is the human face expression of the prediction of each non-directed graph in the non-directed graph training set Graph by Facial Expression Image
Mark result composition.
Step 3.5 calculates between human face expression tag set Label and human face expression label prediction sets Predict
Error；Using minimize mean square error cost function method update figure convolutional neural networks can training parameter set θ, then
Step 3.4 is executed, the error between human face expression tag set Label and human face expression label prediction sets Predict
Less than 0.1.
Step 3.6, figure convolutional neural networks can training parameter set θ and picture scroll product neural network group know at human face expression
Other classifier.
One step 4, input Facial Expression Image to be identified, carry out gray processing, extract face parts of images, then by people
Facial image of the size normalization of face's partial image to m × m, after being normalized；Then by the facial image after normalization
It is converted into non-directed graph, non-directed graph is finally input to the classifier of facial expression recognition, obtains the recognition result of human face expression.
The figure convolutional neural networks carry out the specific steps of propagated forward:
The undirected set of graphs Graph of human face expression is inputted figure convolutional neural networks, the input of s-th of picture scroll lamination by step 1
It is with output relation:
In formula (2) and formula (3):
xs,iIndicate i-th of non-directed graph of s-th of picture scroll lamination input of figure convolutional neural networks；
ys,jIndicate j-th of non-directed graph of s-th of picture scroll lamination output of figure convolutional neural networks；
MinIndicate the number of the non-directed graph of s-th of picture scroll lamination input point of figure convolutional neural networks；
K indicates the filter size of s-th of picture scroll lamination of figure convolutional neural networks；
L indicates the Laplacian Matrix of the non-directed graph of s-th of picture scroll lamination input of figure convolutional neural networks；
Indicate the process that i-th of non-directed graph of input is converted into j-th of non-directed graph of output in s-th of picture scroll lamination
In can training parameter vector,
Indicate the mistake that i-th of non-directed graph of input is converted into j-th of non-directed graph of output in s-th of picture scroll lamination
In journey can training parameter vectorIn k-th of number；
F () indicates to carry out roughening operation to the non-directed graph in bracket using the roughening algorithm of the figure based on greedy algorithm.
Step 2, by the vertex value of the non-directed graph obtained after 6 picture scroll laminations vector x arranged in columnsfin, by column vector
xfinFull articulamentum is inputted, full articulamentum is output and input are as follows:
xfout=fReLU(θfxfin) (4)
In formula (4):
xfoutIndicate the output vector of full articulamentum；
θfIndicate full articulamentum can training parameter；
Indicate ReLU activation primitive.
Step 3, the output x by full articulamentumfoutSoftmax layers are input to, softmax layers of intermediate vector xsoftmidWith
Output vector xsoftoutI-th of numberAre as follows:
xsoftmid=fReLU(θsoftxfout) (5)
In formula (5)~(6):
θsoftIndicate softmax layers can training parameter；
xsoftmidIndicate softmax layers of intermediate vector；
Indicate softmax layers of intermediate vector xsoftinI-th of number；
Indicate softmax layers of output vector xsoftoutIn i-th of number.
Step 4, softmax layers of output vector xsoftoutIn number a possibility that respectively indicating 6 kinds of basic facial expressions, will
Softmax layers of output vector xsoftoutHuman face expression mark of the middle maximum corresponding basic facial expression of number as the non-directed graph
Note.
Due to the adoption of the above technical scheme, the present invention has following good effect compared with prior art:
The present invention has used fixed take a little a little to combine with taking at random during facial image is converted into non-directed graph
Method.Using a fixed connection that takes and can establish between closer pixel apart, that is, it is connected to the pixel of 8 neighborhoods, use
The connection that can be established between pixel apart from each other is taken at random.Fixed take a little a little is combined with random take, so that image
In each pixel both had connection with closer pixel, also have connection with farther away pixel so that establish picture scroll product net
Network can extract the relevant information between pixel farther out, can also extract the relevant information between nearlyr pixel.
The present invention can preferably extract pixel apart from each other using picture scroll product neural network facial expression classifier
Relevant information between point provides more effective human face expression features for facial expression classifier, thus with higher
Facial expression recognition rate.
Therefore, the present invention has the characteristics that relevant information between relatively broad pixel, facial expression recognition can be extracted
Rate is high.
Detailed description of the invention
Fig. 1 is a Facial Expression Image to be identified；
Fig. 2 is by Facial Expression Image shown in Fig. 1 using the Facial Expression Image after normalized of the invention.
Specific embodiment
Present invention will be further described below with reference to the accompanying drawings and specific embodiments, not to the limit of its protection scope
System.
Embodiment 1
A kind of facial expression recognizing method based on figure convolutional neural networks.This method comprises the concrete steps that:
Step 1, the gray level image for obtaining the human face expression that one group of size is 64 × 64 form Facial Expression Image set
Image, by the corresponding human face expression label composition of the gray level image of every human face expression in Facial Expression Image set Image
Human face expression tag set Label；The corresponding human face expression of gray level image of every human face expression therein is labeled as happiness, eats
One of frightened, sad, indignation, detest and fear.
Step 2, by the greyscale image transitions of every human face expression in Facial Expression Image set Image at non-directed graph, group
At the undirected set of graphs Graph of human face expression.
The greyscale image transitions of the human face expression are comprised the concrete steps that at non-directed graph:
Step 2.1, using each of the gray level image of human face expression pixel as a vertex in non-directed graph.
Step 2.2 sets the distance between two pixels adjacent in the gray level image of human face expression as 1.
Step 2.3, by each of the gray level image of human face expression pixel successively centered on pixel, then into
The following operation of row:
First the pixel composition fixation by the Euclidean distance away from central pixel point less than or equal to 2 takes point set Fixed.
The Euclidean distance away from central pixel point is greater than 2 again and the pixel composition less than 4 takes point set Random at random.
Then the vertex of non-directed graph corresponding to central pixel point and fixation are taken into all pixels point institute in point set Fixed
The vertex of corresponding non-directed graph is connected, by the vertex of non-directed graph corresponding to central pixel point with take in point set Random at random
The vertex for randomly selecting non-directed graph corresponding to p pixel is connected.
Step 2.4, the non-directed graph being converted into adjacency matrix W the i-th row jth column numerical value WijAre as follows:
In formula (1): sijIndicate the ith pixel point v in the gray level image of human face expressioniWith j-th of pixel vjBetween
Euclidean distance.
The adjacency matrix W ∈ R of the non-directed graphnxn。
Step 3, according to human face expression undirected set of graphs Graph and human face expression tag set Label, establish human face expression
Recognition classifier:
Step 3.1, a figure convolutional neural networks contain 6 picture scroll laminations, a full articulamentum and a softmax
Layer.The roughened layer that each picture scroll lamination contains a figure signal wave filtering layer and a figure: it is inputted in first picture scroll lamination undirected
The number of figure is 1, and the number for exporting non-directed graph is 32, and filter size is 9；Of non-directed graph is inputted in second picture scroll lamination
Number is 32, and the number for exporting non-directed graph is 32, and filter size is 9；The number of input non-directed graph is in third picture scroll lamination
32, the number for exporting non-directed graph is 64, and filter size is 6；The number that non-directed graph is inputted in 4th picture scroll lamination is 64, defeated
The number of non-directed graph is 64 out, and filter size is 6；The number that non-directed graph is inputted in 5th picture scroll lamination is 64, exports nothing
It is 128 to the number of figure, filter size is 4；The number that non-directed graph is inputted in 6th picture scroll lamination is 128, is exported undirected
The number of figure is 128, and filter size is 4.The number of nodes of full articulamentum is 512；Softmax layers have 6 outputs, respectively correspond
6 kinds of basic facial expressions.
Step 3.2, by the 1st picture scroll lamination can training parameter θ1, the 2nd picture scroll lamination can training parameter θ2, the 3rd
A picture scroll lamination can training parameter θ3, the 4th picture scroll lamination can training parameter θ4, the 5th picture scroll lamination can training parameter
θ5, the 6th picture scroll lamination can training parameter θ6, full articulamentum can training parameter θfWith softmax layers can training parameter
θsoftForm figure convolutional neural networks can training parameter set θ.
Step 3.3, by figure convolutional neural networks can training parameter set θ initialize.
Step 3.4, by figure convolutional neural networks can the training parameter set θ and undirected set of graphs Graph of human face expression put
Enter figure convolutional neural networks and carry out propagated forward, obtains human face expression label prediction sets Predict；Human face expression label prediction
Set Predict is the human face expression of the prediction of each non-directed graph in the non-directed graph training set Graph by Facial Expression Image
Mark result composition.
Step 3.5 calculates between human face expression tag set Label and human face expression label prediction sets Predict
Error；Using minimize mean square error cost function method update figure convolutional neural networks can training parameter set θ, then
Step 3.4 is executed, the error between human face expression tag set Label and human face expression label prediction sets Predict
Less than 0.1.
Step 3.6, figure convolutional neural networks can training parameter set θ and picture scroll product neural network group know at human face expression
Other classifier.
One step 4, input Facial Expression Image to be identified as shown in Figure 1, carry out gray processing, extract face part
Image, then by the size normalization of face parts of images to 64 × 64, the facial image after being normalized, as shown in Figure 2；So
The facial image after normalization is converted into non-directed graph afterwards, non-directed graph is finally input to the classifier of facial expression recognition, is obtained
The recognition result of the human face expression arrived is startled.
The figure convolutional neural networks carry out the specific steps of propagated forward:
The undirected set of graphs Graph of human face expression is inputted figure convolutional neural networks, the input of s-th of picture scroll lamination by step 1
It is with output relation:
In formula (2) and formula (3):
xs,iIndicate i-th of non-directed graph of s-th of picture scroll lamination input of figure convolutional neural networks；
ys,jIndicate j-th of non-directed graph of s-th of picture scroll lamination output of figure convolutional neural networks；
MinIndicate the number of the non-directed graph of s-th of picture scroll lamination input point of figure convolutional neural networks；
K indicates the filter size of s-th of picture scroll lamination of figure convolutional neural networks；
L indicates the Laplacian Matrix of the non-directed graph of s-th of picture scroll lamination input of figure convolutional neural networks；
Indicate the process that i-th of non-directed graph of input is converted into j-th of non-directed graph of output in s-th of picture scroll lamination
In can training parameter vector,
Indicate the mistake that i-th of non-directed graph of input is converted into j-th of non-directed graph of output in s-th of picture scroll lamination
In journey can training parameter vectorIn k-th of number；
F () indicates to carry out roughening operation to the non-directed graph in bracket using the roughening algorithm of the figure based on greedy algorithm.
Step 2, by the vertex value of the non-directed graph obtained after 6 picture scroll laminations vector x arranged in columnsfin, by column vector
xfinFull articulamentum is inputted, full articulamentum is output and input are as follows:
xfout=fReLU(θfxfin) (4)
In formula (4):
xfoutIndicate the output vector of full articulamentum；
θfIndicate full articulamentum can training parameter；
Indicate ReLU activation primitive.
Step 3, the output x by full articulamentumfoutSoftmax layers are input to, softmax layers of intermediate vector xsoftmidWith
Output vector xsoftoutI-th of numberAre as follows:
xsoftmid=fReLU(θsoftxfout) (5)
In formula (5)~(6):
θsoftIndicate softmax layers can training parameter；
xsoftmidIndicate softmax layers of intermediate vector；
Indicate softmax layers of intermediate vector xsoftinI-th of number；
Indicate softmax layers of output vector xsoftoutIn i-th of number.
Step 4, softmax layers of output vector xsoftoutIn number a possibility that respectively indicating 6 kinds of basic facial expressions, will
Softmax layers of output vector xsoftoutHuman face expression mark of the middle maximum corresponding basic facial expression of number as the non-directed graph
Note.
Claims (2)
1. a kind of facial expression recognizing method based on figure convolutional neural networks, it is characterised in that the facial expression recognizing method
Comprise the concrete steps that:
Step 1, the gray level image for obtaining the human face expression that one group of size is m × m, form Facial Expression Image set Image, will
The corresponding human face expression mark group of the gray level image of every human face expression in Facial Expression Image set Image is at human face expression
Tag set Label；The corresponding human face expression of gray level image of every human face expression therein labeled as it is glad, startled, sad,
One of indignation, detest and fear；
Step 2, by the greyscale image transitions of every human face expression in Facial Expression Image set Image at non-directed graph, group is adult
The undirected set of graphs Graph of face expression；
The greyscale image transitions of the human face expression are comprised the concrete steps that at non-directed graph:
Step 2.1, using each of the gray level image of human face expression pixel as a vertex in non-directed graph；
Step 2.2 sets the distance between two pixels adjacent in the gray level image of human face expression as 1；
Step 2.3, by each of the gray level image of human face expression pixel successively centered on pixel, then carry out such as
Lower operation:
First the pixel composition fixation by the Euclidean distance away from central pixel point less than or equal to 2 takes point set Fixed；
The Euclidean distance away from central pixel point is greater than 2 again and the pixel composition less than 4 takes point set Random at random；
Then the vertex of non-directed graph corresponding to central pixel point and fixation are taken in point set Fixed corresponding to all pixels point
The vertex of non-directed graph be connected, take in point set Random by the vertex of non-directed graph corresponding to central pixel point and at random random
The vertex for choosing non-directed graph corresponding to p pixel is connected；
Step 2.4, the non-directed graph being converted into adjacency matrix W the i-th row jth column numerical value WijAre as follows:
In formula (1): sijIndicate the ith pixel point v in the gray level image of human face expressioniWith j-th of pixel vjBetween it is European
Distance；
The adjacency matrix W ∈ R of the non-directed graphnxn；
Step 3, according to human face expression undirected set of graphs Graph and human face expression tag set Label, establish facial expression recognition
Classifier:
Step 3.1, a figure convolutional neural networks contain 6 picture scroll laminations, a full articulamentum and one softmax layers；Often
A picture scroll lamination contains the roughened layer of a figure signal wave filtering layer and a figure, determines the input non-directed graph in each picture scroll lamination
Number, export non-directed graph number and using filter size, determine the number of nodes M of full articulamentumf；Softmax layers have 6
A output respectively corresponds 6 kinds of basic facial expressions；
Step 3.2, by the 1st picture scroll lamination can training parameter θ1, the 2nd picture scroll lamination can training parameter θ2, the 3rd figure
Convolutional layer can training parameter θ3, the 4th picture scroll lamination can training parameter θ4, the 5th picture scroll lamination can training parameter θ5、
6th picture scroll lamination can training parameter θ6, full articulamentum can training parameter θfWith softmax layers can training parameter θsoft
Form figure convolutional neural networks can training parameter set θ；
Step 3.3, by figure convolutional neural networks can training parameter set θ initialize；
Step 3.4, by figure convolutional neural networks can the training parameter set θ and undirected set of graphs Graph of human face expression be put into figure
Convolutional neural networks carry out propagated forward, obtain human face expression label prediction sets Predict；Human face expression marks prediction sets
Predict is the human face expression label of the prediction of each non-directed graph in the non-directed graph training set Graph by Facial Expression Image
As a result it forms；
Error between step 3.5, calculating human face expression tag set Label and human face expression label prediction sets Predict；
Using minimize mean square error cost function method update figure convolutional neural networks can training parameter set θ, then execute
Step 3.4, until the error between human face expression tag set Label and human face expression label prediction sets Predict is less than
0.1；
Step 3.6, figure convolutional neural networks can training parameter set θ and picture scroll product neural network group at facial expression recognition
Classifier；
One step 4, input Facial Expression Image to be identified, carry out gray processing, extract face parts of images, then by face
Facial image of the size normalization of partial image to m × m, after being normalized；Then the facial image after normalization is converted
At non-directed graph, non-directed graph is finally input to the classifier of facial expression recognition, obtains the recognition result of human face expression.
2. the facial expression recognizing method according to claim 1 based on figure convolutional neural networks, it is characterised in that described
The specific steps of figure convolutional neural networks progress propagated forward:
The undirected set of graphs Graph of human face expression is inputted figure convolutional neural networks by step 1, the input of s-th picture scroll lamination and defeated
Relationship is out:
In formula (2) and formula (3):
xs,iIndicate i-th of non-directed graph of s-th of picture scroll lamination input of figure convolutional neural networks,
ys,jIndicate j-th of non-directed graph of s-th of picture scroll lamination output of figure convolutional neural networks,
MinIndicate the number of the non-directed graph of s-th of picture scroll lamination input point of figure convolutional neural networks,
K indicates the filter size of s-th of picture scroll lamination of figure convolutional neural networks,
L indicates the Laplacian Matrix of the non-directed graph of s-th of picture scroll lamination input of figure convolutional neural networks,
It indicates during i-th of non-directed graph of input is converted into j-th of the non-directed graph exported in s-th of picture scroll lamination
Can training parameter vector,
It indicates during i-th of non-directed graph of input is converted into j-th of the non-directed graph exported in s-th of picture scroll lamination
Can training parameter vectorIn k-th of number,
F () indicates to carry out roughening operation to the non-directed graph in bracket using the roughening algorithm of the figure based on greedy algorithm；
Step 2, by the vertex value of the non-directed graph obtained after 6 picture scroll laminations vector x arranged in columnsfin, by column vector xfinIt is defeated
Enter full articulamentum, full articulamentum is output and input are as follows:
xfout=fReLU(θfxfin) (4)
In formula (4):
xfoutIndicate the output vector of full articulamentum,
θfIndicate full articulamentum can training parameter,
Indicate ReLU activation primitive；
Step 3, the output x by full articulamentumfoutSoftmax layers are input to, softmax layers of intermediate vector xsoftmidAnd output
Vector xsoftoutI-th of numberAre as follows:
xsoftmid=fReLU(θsoftxfout) (5)
In formula (5)~(6):
θsoftIndicate softmax layers can training parameter,
xsoftmidIndicate softmax layers of intermediate vector,
Indicate softmax layers of intermediate vector xsoftinI-th of number,
Indicate softmax layers of output vector xsoftoutIn i-th of number；
Step 4, softmax layers of output vector xsoftoutIn number a possibility that respectively indicating 6 kinds of basic facial expressions, by softmax
The output vector x of layersoftoutThe middle maximum corresponding basic facial expression of number is marked as the human face expression of the non-directed graph.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910091261.1A CN110008819B (en) | 2019-01-30 | 2019-01-30 | Facial expression recognition method based on graph convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910091261.1A CN110008819B (en) | 2019-01-30 | 2019-01-30 | Facial expression recognition method based on graph convolution neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110008819A true CN110008819A (en) | 2019-07-12 |
| CN110008819B CN110008819B (en) | 2022-11-18 |
Family
ID=67165589
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910091261.1A Active CN110008819B (en) | 2019-01-30 | 2019-01-30 | Facial expression recognition method based on graph convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110008819B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111291212A (en) * | 2020-01-24 | 2020-06-16 | 复旦大学 | Zero sample sketch image retrieval method and system based on graph convolution neural network |
| CN111724289A (en) * | 2020-06-24 | 2020-09-29 | 山东建筑大学 | Environmental protection equipment identification method and system based on time sequence |
| CN112183314A (en) * | 2020-09-27 | 2021-01-05 | 哈尔滨工业大学（深圳） | Expression information acquisition device and expression identification method and system |
| CN112801266A (en) * | 2020-12-24 | 2021-05-14 | 武汉旷视金智科技有限公司 | Neural network construction method, device, equipment and medium |
| CN113255543A (en) * | 2021-06-02 | 2021-08-13 | 西安电子科技大学 | Facial expression recognition method based on graph convolution network |
| CN115686846A (en) * | 2022-10-31 | 2023-02-03 | 重庆理工大学 | Container cluster online deployment method for fusing graph neural network and reinforcement learning in edge computing |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108256426A (en) * | 2017-12-15 | 2018-07-06 | 安徽四创电子股份有限公司 | A kind of facial expression recognizing method based on convolutional neural networks |
| CN108304826A (en) * | 2018-03-01 | 2018-07-20 | 河海大学 | Facial expression recognizing method based on convolutional neural networks |
| CN109033994A (en) * | 2018-07-03 | 2018-12-18 | 辽宁工程技术大学 | A kind of facial expression recognizing method based on convolutional neural networks |
-
2019
- 2019-01-30 CN CN201910091261.1A patent/CN110008819B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108256426A (en) * | 2017-12-15 | 2018-07-06 | 安徽四创电子股份有限公司 | A kind of facial expression recognizing method based on convolutional neural networks |
| CN108304826A (en) * | 2018-03-01 | 2018-07-20 | 河海大学 | Facial expression recognizing method based on convolutional neural networks |
| CN109033994A (en) * | 2018-07-03 | 2018-12-18 | 辽宁工程技术大学 | A kind of facial expression recognizing method based on convolutional neural networks |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111291212A (en) * | 2020-01-24 | 2020-06-16 | 复旦大学 | Zero sample sketch image retrieval method and system based on graph convolution neural network |
| CN111291212B (en) * | 2020-01-24 | 2022-10-11 | 复旦大学 | Zero sample sketch image retrieval method and system based on graph convolution neural network |
| CN111724289A (en) * | 2020-06-24 | 2020-09-29 | 山东建筑大学 | Environmental protection equipment identification method and system based on time sequence |
| CN112183314A (en) * | 2020-09-27 | 2021-01-05 | 哈尔滨工业大学（深圳） | Expression information acquisition device and expression identification method and system |
| CN112183314B (en) * | 2020-09-27 | 2023-12-12 | 哈尔滨工业大学（深圳） | Expression information acquisition device, expression recognition method and system |
| CN112801266A (en) * | 2020-12-24 | 2021-05-14 | 武汉旷视金智科技有限公司 | Neural network construction method, device, equipment and medium |
| CN112801266B (en) * | 2020-12-24 | 2023-10-31 | 武汉旷视金智科技有限公司 | Neural network construction method, device, equipment and medium |
| CN113255543A (en) * | 2021-06-02 | 2021-08-13 | 西安电子科技大学 | Facial expression recognition method based on graph convolution network |
| CN115686846A (en) * | 2022-10-31 | 2023-02-03 | 重庆理工大学 | Container cluster online deployment method for fusing graph neural network and reinforcement learning in edge computing |
| CN115686846B (en) * | 2022-10-31 | 2023-05-02 | 重庆理工大学 | Container cluster online deployment method integrating graph neural network and reinforcement learning in edge calculation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110008819B (en) | 2022-11-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110008819A (en) | A kind of facial expression recognizing method based on figure convolutional neural networks | |
| CN111126472B (en) | SSD (solid State disk) -based improved target detection method | |
| CN104182772B (en) | A kind of gesture identification method based on deep learning | |
| CN104992223B (en) | Intensive Population size estimation method based on deep learning | |
| CN107657249A (en) | Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again | |
| CN109214441A (en) | A kind of fine granularity model recognition system and method | |
| CN107730458A (en) | A kind of fuzzy facial reconstruction method and system based on production confrontation network | |
| CN106960214A (en) | Object identification method based on image | |
| CN107451565B (en) | Semi-supervised small sample deep learning image mode classification and identification method | |
| CN106529499A (en) | Fourier descriptor and gait energy image fusion feature-based gait identification method | |
| CN110378208B (en) | Behavior identification method based on deep residual error network | |
| CN105139004A (en) | Face expression identification method based on video sequences | |
| CN108875907B (en) | Fingerprint identification method and device based on deep learning | |
| CN114582030B (en) | Behavior recognition method based on service robot | |
| CN106875007A (en) | End-to-end deep neural network is remembered based on convolution shot and long term for voice fraud detection | |
| CN111832484A (en) | Loop detection method based on convolution perception hash algorithm | |
| CN104298974A (en) | Human body behavior recognition method based on depth video sequence | |
| CN108009481A (en) | A kind of training method and device of CNN models, face identification method and device | |
| CN106874913A (en) | A kind of vegetable detection method | |
| CN107463881A (en) | A kind of character image searching method based on depth enhancing study | |
| CN107545243A (en) | Yellow race's face identification method based on depth convolution model | |
| CN116052218B (en) | Pedestrian re-identification method | |
| CN112036260A (en) | Expression recognition method and system for multi-scale sub-block aggregation in natural environment | |
| CN111898566B (en) | Attitude estimation method, attitude estimation device, electronic equipment and storage medium | |
| CN114764941A (en) | Expression recognition method and device and electronic 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |