CN107944363B - Face image processing process, system and server - Google Patents
Face image processing process, system and server Download PDFInfo
- Publication number
- CN107944363B CN107944363B CN201711131120.5A CN201711131120A CN107944363B CN 107944363 B CN107944363 B CN 107944363B CN 201711131120 A CN201711131120 A CN 201711131120A CN 107944363 B CN107944363 B CN 107944363B
- Authority
- CN
- China
- Prior art keywords
- classification
- data
- convolutional neural
- neural networks
- sorted
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- 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
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06V40/178—Human faces, e.g. facial parts, sketches or expressions estimating age from face image; using age information for improving recognition
-
- 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
- G06V40/179—Human faces, e.g. facial parts, sketches or expressions metadata assisted face recognition
Abstract
The embodiment of the invention discloses a kind of face image processing process, device and server, include the following steps: to obtain facial image to be sorted；The facial image is input in the convolutional neural networks model for being built with loss function, the loss function carries out the processing of coefficient relaxationization to the data to be sorted that the convolutional neural networks model exports, to increase the classification interface of the data to be sorted；The classification data of the convolutional neural networks model output is obtained, and content understanding is carried out to the facial image according to the classification data.Before classifying to facial image, the data characteristics to be sorted of the facial image of convolutional neural networks model extraction is subjected to the processing of coefficient relaxationization, using coefficient relaxationization handle can under conditions of more harsh training convolutional neural networks model, classification boundaries are significantly increased, therefore greatly improve convolutional neural networks model to content understanding precision.
Description
Technical field
The present embodiments relate to field of image processing, especially a kind of face image processing process, system and server.
Background technique
Recognition of face refers to and is handled facial image, analyzed and understood using computer, to identify various different peoples
The target of face image and technology to picture.Recognition of face can be applied in many fields such as security protection, finance, the process of recognition of face
Be generally divided into three phases: Face datection, face alignment, face characteristic are extracted and are compared, and face characteristic extraction is that face is known
Other key technology.
With the development of depth learning technology, convolutional neural networks have become the powerful for extracting face characteristic, right
For the fixed convolutional neural networks of model, most crucial technology be how allowable loss function, can effectively supervise
The training of convolutional neural networks, to make convolutional neural networks that there is the ability for extracting face characteristic.Mainly make in the prior art
With the cross entropy loss function of Softmax.Wherein, the cross entropy loss function training network of Softmax extracts the ability of feature,
Using the last layer of network as the expression of face, human face data is mapped to cosine spatially, by comparing different faces
Cosine space length judge the similitude of face, same person's cosine space length is more close, different people's cosine spaces
Apart from farther.
But the inventor of the invention has found under study for action, the feature extraction of the cross entropy loss function of Softmax
Method is a kind of method of not end-to-end, is simply easily achieved, but is not optimized to the greatest extent to existing model, i.e.,
Guarantee that the classification boundaries between different classes of maximize, classification boundaries are not obvious enough, cause content understanding accuracy rate that can not improve.
Summary of the invention
The embodiment of the present invention provide a kind of face image processing process for being capable of increasing classification data classification interface, system and
Server.
In order to solve the above technical problems, the technical solution that the embodiment of the invention uses is: providing a kind of people
Face image processing method, includes the following steps:
Obtain facial image to be sorted；
The facial image is input in the convolutional neural networks model for being built with loss function, the loss function pair
The data to be sorted of the convolutional neural networks model output carry out the processing of coefficient relaxationization, to increase the data to be sorted
Classification interface, wherein the coefficient relaxationization processing is specifically included to the output of the convolutional neural networks model full articulamentum
Data to be sorted carry out the processing of diminution in proportion, to increase the classification interface of the data to be sorted, the convolutional neural networks
The loss function that model uses is the cross entropy loss function of Softmax；
The classification data of the convolutional neural networks model output is obtained, and according to the classification data to the face figure
As carrying out content understanding.
Specifically, coefficient relaxationization processing specifically include the following steps:
The processing of diminution in proportion is carried out to the data to be sorted of the convolutional neural networks model full articulamentum output, to increase
The classification interface of the big data to be sorted.
Specifically, the feature description of the forward-propagating of the convolutional neural networks model are as follows:
L=log (pi)
Defined function:
Specifically, the feature description of the backpropagation of the convolutional neural networks model are as follows:
Defined function:
Wherein, i indicates classification belonging to input picture itself, and j is indicated and i different classes of class categories, t expression and i
Different classes of class categories, k indicate coefficient relaxationization parameter, and f (x) indicates that the face of convolutional neural networks model extraction is special
Sign, wiIndicate the weight of the i-th classification, wjIndicate the weight of jth classification, wtIndicate that the weight of t classification, N are expressed as the class of classification
Shuo not.
Specifically, the convolutional neural networks model is formed by following step training:
It obtains and is marked with the training sample data that classification judges information；
Training sample data input convolutional neural networks model is obtained to the category of model of the training sample data
Information；
The category of model information of samples different in the training sample data is sentenced with the classification by stopping loss function ratio
Whether disconnected information is consistent；
When the category of model information and the classification judge that information is inconsistent, the update of the iterative cycles iteration volume
Weight in product neural network model, until the comparison result terminates when judging that information is consistent with the classification.
Specifically, it is described by stop loss function ratio to the category of model information of samples different in the training sample data with
The classification judges the whether consistent step of information, specifically include the following steps:
Figure parameters processing is carried out to the data to be sorted of the convolutional neural networks model full articulamentum output, makes institute
It states data to be sorted and synchronizes diminution；
By the boundary value in the data to be sorted handled by figure parametersization and default first classification value interval
It is compared, determines section of the data to be sorted handled by figure parametersization in the first classification value interval
Position；
According to the corresponding classification results in the section position, the model point of different samples in the training sample data is determined
Category information；
Judge the classification judges whether information is consistent referring to information and the classification.
Specifically, described the step of obtaining facial image to be sorted, further include following step later:
The facial image is input in the convolutional neural networks model, the convolutional neural networks model is to described
The characteristics of image of facial image extracts to form the data to be sorted；
The data to be sorted are subjected to the processing of coefficient relaxationization, when the data to be sorted handled through coefficient relaxationization
When greater than preset classification thresholds, classify to the data to be sorted.
Specifically, the facial image carry out content understanding include: to facial image carry out gender identification, Age estimation,
The marking of face value or human face similarity degree compare.
In order to solve the above technical problems, the embodiment of the present invention also provides a kind of face image processing system, the face figure
As processing system includes:
Module is obtained, for obtaining facial image to be sorted；
Processing module, for the facial image to be input in the convolutional neural networks model for being built with loss function,
The loss function carries out the processing of coefficient relaxationization to the data to be sorted that the convolutional neural networks model exports, to increase
The classification interface of data to be sorted is stated, the loss function that the convolutional neural networks model uses damages for the cross entropy of Softmax
Lose function；
Categorization module, for obtaining the classification data of the convolutional neural networks model output, and according to the classification number
Content understanding is carried out according to the facial image.
Specifically, the face image processing system further include:
First processing submodule, the data to be sorted for exporting to the full articulamentum of the convolutional neural networks model carry out
The processing of diminution in proportion, to increase the classification interface of the data to be sorted.
Specifically, the feature description of the forward-propagating of the convolutional neural networks model are as follows:
L=log (pi)
Defined function:
Specifically, the feature description of the backpropagation of the convolutional neural networks model are as follows:
Defined function:
Wherein, i indicates classification belonging to input picture itself, and j is indicated and i different classes of class categories, t expression and i
Different classes of class categories, k indicate coefficient relaxationization parameter, and f (x) indicates that the face of convolutional neural networks model extraction is special
Sign, wiIndicate the weight of the i-th classification, wjIndicate the weight of jth classification, wtIndicate that the weight of t classification, N are expressed as the class of classification
Shuo not.
Specifically, the face image processing system further include:
First acquisition submodule, for obtaining the training sample data for being marked with classification and judging information；
First classification submodule, for training sample data input convolutional neural networks model to be obtained the training
The category of model information of sample data；
First compares submodule, for by stopping loss function ratio to the model point of samples different in the training sample data
Category information judges whether information is consistent with the classification；
Second processing submodule is used for when the category of model information judges that information is inconsistent with the classification, repeatedly
Weight in the update convolutional neural networks model of loop iteration, until the comparison result and the classification judge information one
Terminate when cause.
Specifically, the face image processing system further include:
First computational submodule, the data to be sorted for exporting to the full articulamentum of the convolutional neural networks model carry out
Figure parametersization processing, makes the data to be sorted synchronize diminution；
Second compares submodule, for the data to be sorted handled by figure parametersization and default first to be classified
Boundary value in value interval is compared, and determines the data to be sorted handled by figure parametersization at described first point
Section position in class value interval；
Second classification submodule, for determining the number of training according to the corresponding classification results in the section position
According to the category of model information of interior different samples；
First judging submodule, for judging the classification judges whether information is consistent referring to information and the classification.
Specifically, the face image processing system further include:
Third classification submodule, for the facial image to be input in the convolutional neural networks model, the volume
Product neural network model extracts the characteristics of image of the facial image to form the data to be sorted；
Third handles submodule, for the data to be sorted to be carried out the processing of coefficient relaxationization, when described through coefficient pine
When the data to be sorted of relaxationization processing are greater than preset classification thresholds, classify to the data to be sorted.
Specifically, described includes: to sentence to facial image progress gender identification, age to facial image progress content understanding
Disconnected, face value marking or human face similarity degree compare.
In order to solve the above technical problems, the embodiment of the present invention also provides a kind of server characterized by comprising
One or more processors；
Memory；
One or more application program, wherein one or more of application programs are stored in the memory and quilt
It is configured to be executed by one or more of processors, one or more of programs are configured to carry out face described above
Image processing method.
The beneficial effect of the embodiment of the present invention is: before classifying to facial image, by convolutional neural networks model
The data characteristics to be sorted of the facial image of extraction carries out the processing of coefficient relaxationization, that is, treats classification data and contracted in proportion
It puts, increases the classification interface of classification data, realization principle is, the data after scaling can satisfy the phase of loss function
Prestige demand then also can satisfy the expectation demand of loss function certainly without the classification data of scaling.Using coefficient relaxationization
Processing can under conditions of more harsh training convolutional neural networks model, significantly increase classification boundaries, therefore make convolution
Neural network model greatly improves content understanding precision.
Detailed description of the invention
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for
For those skilled in the art, without creative efforts, it can also be obtained according to these attached drawings other attached
Figure.
Fig. 1 is face image processing process of embodiment of the present invention basic procedure schematic diagram；
Fig. 2 is convolutional neural networks of embodiment of the present invention model training method basic procedure schematic diagram；
The method that the processing of coefficient relaxationization is carried out to parameter when Fig. 3 is convolutional neural networks of embodiment of the present invention model training
Flow chart；
Fig. 4 is a kind of concrete application implementation method of convolutional neural networks of embodiment of the present invention model；
Fig. 5 is face image processing of embodiment of the present invention system basic structure block diagram；
Fig. 6 is server of embodiment of the present invention basic structure block diagram.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, below in conjunction in the embodiment of the present invention
Attached drawing, technical scheme in the embodiment of the invention is clearly and completely described.
In some processes of the description in description and claims of this specification and above-mentioned attached drawing, contain according to
Multiple operations that particular order occurs, but it should be clearly understood that these operations can not be what appears in this article suitable according to its
Sequence is executed or is executed parallel, and serial number of operation such as 101,102 etc. is only used for distinguishing each different operation, serial number
It itself does not represent and any executes sequence.In addition, these processes may include more or fewer operations, and these operations can
To execute or execute parallel in order.It should be noted that the description such as " first " herein, " second ", is for distinguishing not
Same message, equipment, module etc., does not represent sequencing, does not also limit " first " and " second " and be different type.
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those skilled in the art's every other implementation obtained without creative efforts
Example, shall fall within the protection scope of the present invention.
Embodiment
It is to be noted that the basic structure of convolutional neural networks includes two layers, one is characterized extract layer, each nerve
The input of member is connected with the local acceptance region of preceding layer, and extracts the feature of the part.After the local feature is extracted, it
Positional relationship between other feature is also decided therewith；The second is Feature Mapping layer, each computation layer of network is by multiple
Feature Mapping composition, each Feature Mapping is a plane, and the weight of all neurons is equal in plane.Feature Mapping structure is adopted
The sigmoid function for using influence function core small as convolutional network activation primitive so that Feature Mapping have shift invariant.
Further, since the neuron on a mapping face shares weight, thus reduce the number of network freedom parameter.Convolutional Neural net
Each of network convolutional layer all followed by one is used to ask the computation layer of local average and second extraction, it is this it is distinctive twice
Feature extraction structure reduces feature resolution.
Convolutional neural networks are mainly used to the X-Y scheme of identification displacement, scaling and other forms distortion invariance.Due to
The feature detection layer of convolutional neural networks is learnt by training data, so avoiding when using convolutional neural networks
The feature extraction of display, and implicitly learnt from training data；Furthermore due to the neuron on same Feature Mapping face
Weight is identical, so network can be with collateral learning, this is also convolutional network is connected with each other the one big excellent of network relative to neuron
Gesture.
Convolutional neural networks model uses the Inception_v2 model of GoogleNet in the present embodiment.But it is not limited to
This, according to the difference of concrete application scene, convolutional neural networks model can also use Inception_v3 or Inception_
V4 model.
Referring to Fig. 1, Fig. 1 is the present embodiment face image processing process basic procedure schematic diagram.
As shown in Figure 1, a kind of face image processing process, includes the following steps:
S1100, facial image to be sorted is obtained；
The method for obtaining facial image includes two methods of acquisition in real time and extraction storage image/video data.Acquisition in real time
It is mainly used for the real-time application of intelligent terminal (mobile phone, tablet computer and monitoring device) (such as: judging age of user, gender, face value
With similarity etc.).Storage image/video data is extracted to be mainly used for further locating the image and video data of storage
Reason, also can be used in intelligent terminal and applies to historical photograph.
S1200, the facial image is input in the convolutional neural networks model for being built with loss function, the loss
Function carries out the processing of coefficient relaxationization to the data to be sorted that the convolutional neural networks model exports, described to be sorted to increase
The classification interface of data, wherein the coefficient relaxationization processing is specifically included to the full articulamentum of convolutional neural networks model
The data to be sorted of output carry out the processing of diminution in proportion, to increase the classification interface of the data to be sorted；
Convolutional neural networks model has been trained to when carrying out face image processing to convergence, and by specifically instructing
The mode of white silk, has been able to that convolutional neural networks model is enable as expected to handle facial image.
In present embodiment, convolutional neural networks model carries out feature extraction to the facial image of input, and acquisition being capable of table
The feature of traveller on a long journey's face image most expression power, and data to be sorted are formed in the full articulamentum of convolutional neural networks model.
In the present embodiment, classification data is treated using the cross entropy loss function of Softmax and is pre-processed, it is pretreated
Mode is to treat classification data to carry out the processing of coefficient relaxationization, i.e., to the to be sorted of convolutional neural networks model full articulamentum output
Data carry out the processing of diminution in proportion, to increase the classification interface of data to be sorted.Specific operating method is, in number to be processed
According to 0 coefficient of relaxation less than 1 is greater than multiplied by one before, coefficient of relaxation is verified by test of many times and is obtained, can use one
Kind scheme are as follows: the classification precision of setting convolutional neural networks model, by selecting different coefficient of relaxation, to convolutional Neural net
Network model is trained, and is recorded the time that different convolutional neural networks categories of model reach the precision, is taken the training time most short
Coefficient used by convolutional neural networks model, as coefficient of relaxation.
It when a data are after zooming in and out, is compared with classification boundaries value, remains to be greater than or be located at the classification boundaries
Within value, then centainly it is greater than without the initial data of scaling or is located within the classification boundaries value.Data to be sorted are carried out
The processing of coefficient relaxationization, but classification boundaries value does not change, then passes through coefficient relaxationization place for classification boundaries value
The data to be sorted of reason are reduced, and covert increases the classification interface of data to be sorted.Classification data is treated simultaneously
Processing, with more convergent training condition training convolutional neural networks model, makes the classification of convolutional neural networks model more
Accurately.
S1300, the classification data for obtaining the convolutional neural networks model output, and according to the classification data to described
Facial image carries out content understanding.
Data to be sorted are after the screening of loss function, and data (handling by coefficient relaxationization) to be sorted are in convolution mind
Classification layer through network model classifies to data.
Classification layer treats classification data and classifies according to preset classification standard, and classification data is exported.Point
The classification data of class layer output is that one or more numerical value are realized by the way that above-mentioned classification data to be compared with classification thresholds
The content understanding of facial image.For example, similarity threshold is preset when the content understanding of facial image is that human face similarity degree matches,
By classification data export numerical value be compared with similarity threshold, comparison result be greater than the threshold value when, then facial image with join
According to compare image be it is homologous, otherwise to compare image different with reference for facial image.
Content understanding including but not limited to carries out gender identification, Age estimation, the marking of face value or human face similarity degree and compares.Point
Class tables of data, which is leted others have a look at, mainly can recognize feature in face image, this feature is compared with preset classification standard, it will be able to right
Gender, age and the face value of facial image judge.And according to the cos of two facial image classification data (cosine space) away from
From comparison, it will be able to calculate the similarity between two facial images.
Above embodiment is before classifying to facial image, by the facial image of convolutional neural networks model extraction
Data characteristics to be sorted carry out coefficient relaxationization processing, that is, treat classification data and scaled in proportion, increase classification data
Classification interface, realization principle is, the data after scaling can satisfy the expectation demand of loss function, then without
The classification data of scaling also can satisfy the expectation demand of loss function certainly.Being handled using coefficient relaxationization can be more tight
Training convolutional neural networks model under conditions of severe, significantly increases classification boundaries, therefore keeps convolutional neural networks model internal
Appearance understands that precision greatly improves.
Specifically, the loss function that convolutional neural networks model uses in the present embodiment is the intersection entropy loss of Softmax
Function, after carrying out figure parameters processing to the loss function, the loss function forward-propagating formula is as follows:
The feature of the forward-propagating of convolutional neural networks model describes are as follows:
L=log (pi)
Defined function:
Wherein, i indicates classification belonging to input picture itself, and j is indicated and i different classes of class categories, t expression and i
Different classes of class categories, k indicate coefficient relaxationization parameter, and f (x) indicates that the face of convolutional neural networks model extraction is special
Sign, wiIndicate the weight of the i-th classification, wjIndicate the weight of jth classification, wtIndicate that the weight of t classification, N are expressed as the class of classification
Shuo not.
Convolutional neural networks model in the training process, passes through the classification data exported to it and the ratio being expected between expectation
Convolutional neural networks model weight is finely adjusted compared with triggering, i.e. convolutional neural networks model backpropagation, by seeking local derviation
Several modes is finely adjusted the weight of convolutional neural networks model, and concrete operations formula is as follows:
The feature of the backpropagation of convolutional neural networks model describes are as follows:
Defined function:
Wherein, i indicates classification belonging to input picture itself, and j is indicated and i different classes of class categories, t expression and i
Different classes of class categories, k indicate coefficient relaxationization parameter, and f (x) indicates that the face of convolutional neural networks model extraction is special
Sign, wiIndicate the weight of the i-th classification, wjIndicate the weight of jth classification, wtIndicate that the weight of t classification, N are expressed as the class of classification
Shuo not.
In some preferred embodiments, the value of k is 0.5.The value of k is determined according to experimental data, realizes knot
Fruit shows as k=0.5, convolutional neural networks model separation accuracy rate highest, and the time required for training pattern is most short.
Citing is illustrated coefficient relaxationization, with f (x) * wiIndicate data to be sorted, f (x) * wjPresentation class threshold value,
And if only if f (x) * wi> f (x) * wjWhen, data to be sorted can pass through the screening of loss function.
Cause are as follows:
f(x)*wi> f (x) * wj
If making:
f(x)*wi> f (x) * wj
Might as well so it make
kf(x)*wi> f (x) * wj(0 < k < 1.0)
Wherein, f (x) indicates deep learning network to the feature of image zooming-out, and this feature will be fed to Softmax loss
In function category device, the image that this feature represents belongs to the i-th class, so should have f (x) * wi> f (x) * wjIf our loss
Function is able to satisfy k*f (x) * wi> f (x) * wj(0 < k < 1.0), then must have f (x) * wi> f (x) * wj, that is, complete classification and appoint
Business, this method are to carry out coefficient relaxation to Softmax.
Because coefficient (0 < k < 1.0), i.e., relax to the result of Softmax, make the classification between different classes of
Interface becomes larger, and increases model in this way to the classification accuracy for being not easy classification samples, that is, improves the robustness of model.
Specifically, the training method of convolutional neural networks model is as follows in the present embodiment:
Referring to Fig. 2, Fig. 2 is the present embodiment convolutional neural networks model training method basic procedure schematic diagram.
As shown in Fig. 2, convolutional neural networks model is formed by following step training:
S2100, acquisition are marked with the training sample data that classification judges information；
Training sample data are the component units of entire training set, and training set is by several training sample training data groups
At.
Training sample data judge what information formed to the classification being marked by human face data and to human face data.
Classification judges that information refers to that people according to the training direction of input convolutional neural networks model, pass through sentencing for universality
The artificial judgement that disconnected standard and true state make training sample data, that is, people are defeated to convolutional neural networks model
The expectation target of numerical value out.Such as, in a training sample data, manual identified goes out the face image data and pre-stored mesh
Mark facial image be the same person, then demarcate the facial image classification judge information for pre-stored target facial image phase
Together.
S2200, the mould that training sample data input convolutional neural networks model is obtained to the training sample data
Type classification information；
Training sample set is sequentially inputted in convolutional neural networks model, and obtains convolutional neural networks model inverse
The category of model information of one full articulamentum output.
Category of model information is the excited data that convolutional neural networks model is exported according to the facial image of input, is being rolled up
Product neural network model is not trained to before convergence, and classification is the biggish numerical value of discreteness referring to information, when convolutional Neural net
Network model is not trained to convergence, and classification is metastable data referring to information.
S2300, by stop loss function ratio to the category of model information of samples different in the training sample data with it is described
Classification judges whether information is consistent；
Stopping loss function is to judge information with desired classification for detecting category of model information in convolutional neural networks model
Whether consistent detection function.When the output result of convolutional neural networks model and classification judge the expected result of information
It when inconsistent, needs to be corrected the weight in convolutional neural networks model, so that the output knot of convolutional neural networks model
Fruit judges that the expected result of information is identical with classification.
S2400, when the category of model information and it is described classification judge that information is inconsistent when, the update of iterative cycles iteration
Weight in the convolutional neural networks model, until the comparison result terminates when judging that information is consistent with the classification.
When the output result of convolutional neural networks model and classification judge information expected result it is inconsistent when, need to volume
Weight in product neural network model is corrected, so that the output result of convolutional neural networks model and classification judge information
Expected result is identical.
Specifically, in the training process, it needs to stop loss function and treats classification data progress coefficient relaxationization processing.
Referring to Fig. 3, carrying out the processing of coefficient relaxationization to parameter when Fig. 3 is the present embodiment convolutional neural networks model training
Method flow diagram.
As shown in figure 3, step S2300 includes the following steps:
S2310, the data to be sorted of the convolutional neural networks model full articulamentum output are carried out at figure parameters
Reason, makes the data to be sorted synchronize diminution；
Training sample data are sequentially inputted in convolutional neural networks model, and obtains convolutional neural networks model and connects entirely
The data to be sorted of layer output, and the synchronization scaling data to be sorted are connect, i.e., are greater than zero multiplied by one before data to be sorted
Coefficient less than 1.
S2320, the data to be sorted handled by figure parametersization and default first are classified in value interval
Boundary value is compared, and determines the data to be sorted handled by figure parametersization in the first classification value interval
Section position；
Data to be sorted by the processing of coefficient relaxationization are compared with preset first classification number value interval.
Wherein, the first classification value interval is according to expected classification results setting, when classification results are that similarity compares, first
Classification value interval is single classification thresholds, and when classification results are that race divides, the first classification value interval is 3 different
Value interval.When classification results are that gender divides, the first classification value interval is 2 different value intervals.Classification results are
When face value divides, the first classification value interval is multiple continuous value intervals.
Data to be sorted by the processing of coefficient relaxationization are compared with preset first classification number value interval, are obtained
Specific location of the data to be sorted for taking coefficient relaxationization to handle in the first classification number value interval.
S2330, according to the corresponding classification results in the section position, determine in the training sample data different samples
Category of model information；
Specific location acquisition pair of the data to be sorted handled according to coefficient relaxationization in the first classification number value interval
The classification results answered.For example, the data to be sorted of coefficient relaxationization processing, which are greater than, divides when classification results are that similarity compares
When class threshold value, then corresponding classification results are that the training sample data are similar to referring to image, and otherwise the two result is dissmilarity.
Category of model information, that is, model classification results.
S2340, judge the classification judges whether information is consistent referring to information and the classification.
The category of model information of acquisition judge that information is compared with artificial expected classification, compare the two whether one
It causes, for example, category of model information judges to practice sample data similar to referring to image, and manually preset is contemplated to be experienced sample
Data then assert that the classification results do not meet default expectation with referring to image dissmilarity.
In some embodiments, the convolutional neural networks model of the embodiment of the present invention is used for human face similarity degree ratio
It is right.
Referring specifically to Fig. 4, Fig. 4 is a kind of concrete application implementation method of the present embodiment convolutional neural networks model.
As shown in figure 4, further including following step after step S1100:
S1110, the facial image is input in the convolutional neural networks model, the convolutional neural networks model
The characteristics of image of the facial image is extracted to form the data to be sorted；
Facial image is used to compare with referring to image progress similarity in this step, confirms facial image to be sorted and reference
Whether whether image is homologous, i.e. be same people in two photos.
Facial image to be sorted is input in convolutional neural networks model, the data to be sorted of facial image are obtained.
S1120, the data to be sorted are carried out to the processing of coefficient relaxationization, when it is described through the processing of coefficient relaxationization wait divide
When class data are greater than preset classification thresholds, classify to the data to be sorted.
Data to be sorted are subjected to the processing of coefficient relaxationization, and will treated classification data and preset classification thresholds into
Row compares, and classification thresholds are obtained according to experimental data, use to identify facial image to be sorted and whether identical referring to image
In the specific threshold value of contrast judgement.For example, setting data to be sorted carries out coefficient relaxationization treated value interval as 0
To between 1, if classification thresholds are 0.5, then when data to be sorted carry out coefficient relaxationization treated value greater than 0.5, to
Survey facial image with referring to image be it is homologous；When data to be sorted carry out coefficient relaxationization treated value less than 0.5, then
Facial image to be measured with referring to the not identical or different source of image.
The processing of coefficient relaxationization is carried out by treating classification data, the classification interface of data to be sorted can be made to increase, subtracted
Accidental error caused by limit judgement when small data are near classification thresholds, improves classification accuracy.
In order to solve the above technical problems, the embodiment of the present invention also provides a kind of face image processing system.Referring specifically to
Fig. 5, Fig. 5 are the present embodiment face image processing system basic structure block diagram.
As shown in figure 5, face image processing system includes: to obtain module 2100, processing module 2200 and categorization module
2300.Wherein, module 2100 is obtained for obtaining facial image to be sorted；Processing module 2200 is for inputting facial image
Into the convolutional neural networks model for being built with loss function, number to be sorted that loss function exports convolutional neural networks model
According to the processing of coefficient relaxationization is carried out, to increase the classification interface of data to be sorted；Categorization module 2300 is for obtaining convolutional Neural
The classification data of network model output, and content understanding is carried out to facial image according to classification data；In some embodiments,
Face image processing system further include: first processing submodule, for the full articulamentum of convolutional neural networks model export to
Classification data carries out the processing of diminution in proportion, to increase the classification interface of data to be sorted.
In some embodiments, the feature description of the forward-propagating of convolutional neural networks model are as follows:
L=log (pi)
Defined function:
Wherein, i indicates classification belonging to input picture itself, and j is indicated and i different classes of class categories, t expression and i
Different classes of class categories, k indicate coefficient relaxationization parameter, and f (x) indicates that the face of convolutional neural networks model extraction is special
Sign, wiIndicate the weight of the i-th classification, wjIndicate the weight of jth classification, wtIndicate that the weight of t classification, N are expressed as the class of classification
Shuo not.
In some embodiments, the feature description of the backpropagation of convolutional neural networks model are as follows:
Defined function:
Wherein, i indicates classification belonging to input picture itself, and j is indicated and i different classes of class categories, t expression and i
Different classes of class categories, k indicate coefficient relaxationization parameter, and f (x) indicates that the face of convolutional neural networks model extraction is special
Sign, wiIndicate the weight of the i-th classification, wjIndicate the weight of jth classification, wtIndicate that the weight of t classification, N are expressed as the class of classification
Shuo not.
In some embodiments, face image processing system further include: the first acquisition submodule, the first classification submodule
Block, first compare submodule and second processing submodule.Wherein, the first acquisition submodule is marked with classification judgement letter for obtaining
The training sample data of breath；First classification submodule is used to training sample data input convolutional neural networks model obtaining training
The category of model information of sample data；First comparison submodule is used for by stopping loss function ratio to not same in training sample data
This category of model information judges whether information is consistent with classification；Second processing submodule is used for when category of model information and classification
When judging that information is inconsistent, the weight of iterative cycles iteration updated in convolutional neural networks model, until comparison result and classification
Judge to terminate when information is consistent.
In some embodiments, face image processing system further include: the first computational submodule, second compare submodule
Block, the second classification submodule and the first judging submodule.Wherein, the first computational submodule is for complete to convolutional neural networks model
The data to be sorted of articulamentum output carry out figure parameters processing, and data to be sorted is made to synchronize diminution；Second comparer
Module is for carrying out the boundary value in the data to be sorted by figure parametersization processing and default first classification value interval
Compare, determines section position of the data to be sorted in the first classification value interval by figure parametersization processing；Second point
Class submodule is used to determine the category of model letter of different samples in training sample data according to the corresponding classification results in section position
Breath；First judging submodule is for judging classification judges whether information is consistent referring to information and classification.
In some embodiments, face image processing system further include: third classification submodule and third handle submodule
Block.Wherein, third classification submodule is for facial image to be input in convolutional neural networks model, convolutional neural networks model
The characteristics of image of facial image is extracted to form data to be sorted；Third processing submodule is for carrying out data to be sorted
Coefficient relaxationization processing, when the data to be sorted handled through coefficient relaxationization are greater than preset classification thresholds, to number to be sorted
According to classifying.
In some embodiments, it includes: to carry out gender identification, age to facial image that facial image, which carries out content understanding,
Judgement, the marking of face value or human face similarity degree compare.
The present embodiment also provides a kind of server.Referring specifically to Fig. 6, Fig. 6 is that the present embodiment server basic structure is shown
It is intended to.
As shown in fig. 6, server includes: one or more processors 3110 and memory 3120；One or more application
Program, wherein one or more application programs are stored in memory and are configured as being performed by one or more processors,
One or more programs are configured to:
Obtain facial image to be sorted；
The facial image is input in the convolutional neural networks model for being built with loss function, the loss function pair
The data to be sorted of the convolutional neural networks model output carry out the processing of coefficient relaxationization, to increase the data to be sorted
Classification interface；
The classification data of the convolutional neural networks model output is obtained, and according to the classification data to the face figure
As carrying out content understanding.
Server is before classifying to facial image, by the facial image of convolutional neural networks model extraction wait divide
Class data characteristics carries out the processing of coefficient relaxationization, that is, treats classification data and scaled in proportion, increase the classification of classification data
Interface, realization principle are that the data after scaling can satisfy the expectation demand of loss function, then without scaling
Classification data also can satisfy the expectation demand of loss function certainly.Being handled using coefficient relaxationization can be in more harsh item
Training convolutional neural networks model under part, significantly increases classification boundaries, therefore makes convolutional neural networks model to content understanding
Precision greatly improves.
It is to be noted that storage is in the memory of server for realizing facial image in the present embodiment in this implementation column
All programs in processing method, processor can call the program in the memory, execute above-mentioned face image processing process
Cited institute is functional.Since the function face image processing process in the present embodiment that server is realized has carried out in detail
It states, is no longer repeated herein.
It should be noted that specification of the invention and its a better embodiment of the invention is given in the attached drawing, still,
The present invention can be realized by many different forms, however it is not limited to this specification described embodiment, these embodiments
Not as the additional limitation to the content of present invention, purpose of providing these embodiments is makes understanding to the disclosure
It is more thorough and comprehensive.Also, above-mentioned each technical characteristic continues to be combined with each other, and forms the various embodiments not being enumerated above,
It is considered as the range of description of the invention record；It further, for those of ordinary skills, can be according to the above description
It is improved or converted, and all these modifications and variations should all belong to the protection domain of appended claims of the present invention.
Claims (11)
1. a kind of face image processing process, which is characterized in that include the following steps:
Obtain facial image to be sorted；
The facial image is input in the convolutional neural networks model for being built with loss function, the loss function is to described
The data to be sorted of convolutional neural networks model output carry out the processing of coefficient relaxationization, to increase the classification of the data to be sorted
Interface, wherein the coefficient relaxationization processing is specifically included to the output of the convolutional neural networks model full articulamentum wait divide
Class data carry out the processing of diminution in proportion, to increase the classification interface of the data to be sorted, the convolutional neural networks model
The loss function used is the cross entropy loss function of Softmax；
Obtain the classification data of convolutional neural networks model output, and according to the classification data to the facial image into
Row content understanding；
The feature of the forward-propagating of the convolutional neural networks model describes are as follows:
L=log (pi)
Defined function:
The feature of the backpropagation of the convolutional neural networks model describes are as follows:
Defined function:
Wherein, i indicates classification belonging to input picture itself, and j indicates to indicate different from i with i different classes of class categories, t
The class categories of classification, k indicate coefficient relaxationization parameter, and f (x) indicates the face characteristic of convolutional neural networks model extraction, wi
Indicate the weight of the i-th classification, wjIndicate the weight of jth classification, wtIndicate that the weight of t classification, N are expressed as the classification number of classification.
2. face image processing process according to claim 1, which is characterized in that the convolutional neural networks model passes through
Following step training is formed:
It obtains and is marked with the training sample data that classification judges information；
Training sample data input convolutional neural networks model is obtained to the category of model information of the training sample data；
The category of model information of different samples and classification judgement in the training sample data is compared by loss function to believe
It whether consistent ceases；
When the category of model information and the classification judge that information is inconsistent, the update of the iterative cycles iteration convolution mind
Through the weight in network model, until the comparison result terminates when judging that information is consistent with the classification.
3. face image processing process according to claim 2, which is characterized in that it is described compared by loss function described in
The category of model information of different samples and the classification judge the whether consistent step of information in training sample data, specifically include
Following step:
Figure parameters processing is carried out to the data to be sorted of the convolutional neural networks model full articulamentum output, make it is described to
Classification data synchronizes diminution；
Boundary value in the data to be sorted handled by figure parametersization and default first classification value interval is carried out
Compare, determines section position of the data to be sorted handled by figure parametersization in the first classification value interval
It sets；
According to the corresponding classification results in the section position, the category of model letter of different samples in the training sample data is determined
Breath；
Judge the category of model information judges whether information is consistent with the classification.
4. image processing method according to claim 1, which is characterized in that the step for obtaining facial image to be sorted
Suddenly, later further include following step:
The facial image is input in the convolutional neural networks model, the convolutional neural networks model is to the face
The characteristics of image of image extracts to form the data to be sorted；
The data to be sorted are subjected to the processing of coefficient relaxationization, when the data to be sorted handled through coefficient relaxationization are greater than
When preset classification thresholds, classify to the data to be sorted.
5. face image processing process described in any one according to claim 1~4, which is characterized in that the facial image
Carrying out content understanding includes: to carry out gender identification, Age estimation, the marking of face value or human face similarity degree to facial image to compare.
6. a kind of face image processing system, which is characterized in that the face image processing system includes:
Module is obtained, for obtaining facial image to be sorted；
Processing module, it is described for the facial image to be input in the convolutional neural networks model for being built with loss function
The data to be sorted that loss function exports the convolutional neural networks model carry out the processing of coefficient relaxationization, with described in increasing to
The classification interface of classification data, the loss function that the convolutional neural networks model uses is the intersection entropy loss letter of Softmax
Number；
Categorization module, for obtaining the classification data of the convolutional neural networks model output, and according to the classification data pair
The facial image carries out content understanding；
First processing submodule, the data to be sorted for exporting to the full articulamentum of the convolutional neural networks model carry out year-on-year
Example diminution processing, to increase the classification interface of the data to be sorted；
The feature of the forward-propagating of the convolutional neural networks model describes are as follows:
L=log (pi)
Defined function:
The feature of the backpropagation of the convolutional neural networks model describes are as follows:
Defined function:
Wherein, i indicates classification belonging to input picture itself, and j indicates to indicate different from i with i different classes of class categories, t
The class categories of classification, k indicate coefficient relaxationization parameter, and f (x) indicates the face characteristic of convolutional neural networks model extraction, wi
Indicate the weight of the i-th classification, wjIndicate the weight of jth classification, wtIndicate that the weight of t classification, N are expressed as the classification number of classification.
7. face image processing system according to claim 6, which is characterized in that the face image processing system is also wrapped
It includes:
First acquisition submodule, for obtaining the training sample data for being marked with classification and judging information；
First classification submodule, for training sample data input convolutional neural networks model to be obtained the training sample
The category of model information of data；
First compares submodule, and the category of model for comparing different samples in the training sample data by loss function is believed
Breath judges whether information is consistent with the classification；
Second processing submodule is used for when the category of model information judges that information is inconsistent with the classification, iterative cycles
Weight in the update convolutional neural networks model of iteration, until when the comparison result judges that information is consistent with the classification
Terminate.
8. face image processing system according to claim 7, which is characterized in that the face image processing system is also wrapped
It includes:
First computational submodule, the data to be sorted for exporting to the full articulamentum of the convolutional neural networks model carry out coefficient
Parameterized treatment makes the data to be sorted synchronize diminution；
Second compares submodule, for by the data to be sorted handled by figure parametersization and default first classification value
Boundary value in section is compared, and determines that the data to be sorted handled by figure parametersization take in first classification
It is worth the section position in section；
Second classification submodule, for determining in the training sample data according to the corresponding classification results in the section position
The category of model information of different samples；
First judging submodule, for judging the category of model information judges whether information is consistent with the classification.
9. face image processing system according to claim 6, which is characterized in that the face image processing system is also wrapped
It includes:
Third classification submodule, for the facial image to be input in the convolutional neural networks model, the convolution mind
It extracts to form the data to be sorted through characteristics of image of the network model to the facial image；
Third handles submodule, for the data to be sorted to be carried out the processing of coefficient relaxationization, when described through coefficient relaxationization
When the data to be sorted of processing are greater than preset classification thresholds, classify to the data to be sorted.
10. according to face image processing system described in claim 6~9 any one, which is characterized in that described to face figure
It include: to carry out gender identification, Age estimation, the marking of face value or human face similarity degree to facial image to compare as carrying out content understanding.
11. a kind of server characterized by comprising
One or more processors；
Memory；
One or more application program, wherein one or more of application programs are stored in the memory and are configured
To be executed by one or more of processors, one or more of application programs are configured to carry out claim 1-5 and appoint
Face image processing process described in meaning one.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711131120.5A CN107944363B (en) | 2017-11-15 | 2017-11-15 | Face image processing process, system and server |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711131120.5A CN107944363B (en) | 2017-11-15 | 2017-11-15 | Face image processing process, system and server |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107944363A CN107944363A (en) | 2018-04-20 |
| CN107944363B true CN107944363B (en) | 2019-04-26 |
Family
ID=61931294
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711131120.5A Active CN107944363B (en) | 2017-11-15 | 2017-11-15 | Face image processing process, system and server |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107944363B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109299690B (en) * | 2018-09-21 | 2020-12-29 | 浙江中正智能科技有限公司 | Method capable of improving video real-time face recognition precision |
| CN110009059B (en) * | 2019-04-16 | 2022-03-29 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating a model |
| CN110490242B (en) * | 2019-08-12 | 2024-03-29 | 腾讯医疗健康(深圳)有限公司 | Training method of image classification network, fundus image classification method and related equipment |
| CN116702014A (en) * | 2023-08-03 | 2023-09-05 | 中电科新型智慧城市研究院有限公司 | Population identification method, device, terminal equipment and storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106022317A (en) * | 2016-06-27 | 2016-10-12 | 北京小米移动软件有限公司 | Face identification method and apparatus |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104346622A (en) * | 2013-07-31 | 2015-02-11 | 富士通株式会社 | Convolutional neural network classifier, and classifying method and training method thereof |
| CN104408470B (en) * | 2014-12-01 | 2017-07-25 | 中科创达软件股份有限公司 | The sex-screening method learnt in advance based on average face |
| US10074041B2 (en) * | 2015-04-17 | 2018-09-11 | Nec Corporation | Fine-grained image classification by exploring bipartite-graph labels |
| US10002313B2 (en) * | 2015-12-15 | 2018-06-19 | Sighthound, Inc. | Deeply learned convolutional neural networks (CNNS) for object localization and classification |
| CN106096538B (en) * | 2016-06-08 | 2019-08-23 | 中国科学院自动化研究所 | Face identification method and device based on sequencing neural network model |
| CN106384080A (en) * | 2016-08-31 | 2017-02-08 | 广州精点计算机科技有限公司 | Apparent age estimating method and device based on convolutional neural network |
| CN106649886A (en) * | 2017-01-13 | 2017-05-10 | 深圳市唯特视科技有限公司 | Method for searching for images by utilizing depth monitoring hash of triple label |
| CN107301640A (en) * | 2017-06-19 | 2017-10-27 | 太原理工大学 | A kind of method that target detection based on convolutional neural networks realizes small pulmonary nodules detection |
-
2017
- 2017-11-15 CN CN201711131120.5A patent/CN107944363B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106022317A (en) * | 2016-06-27 | 2016-10-12 | 北京小米移动软件有限公司 | Face identification method and apparatus |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107944363A (en) | 2018-04-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yuan et al. | Fingerprint liveness detection using an improved CNN with image scale equalization | |
| CN107944363B (en) | Face image processing process, system and server | |
| Hosseini et al. | Age and gender classification using wide convolutional neural network and Gabor filter | |
| Hossain et al. | An emotion recognition system for mobile applications | |
| Lee et al. | Intra-class variation reduction using training expression images for sparse representation based facial expression recognition | |
| CN111523462B (en) | Video sequence expression recognition system and method based on self-attention enhanced CNN | |
| CN105426850A (en) | Human face identification based related information pushing device and method | |
| Lovato et al. | Faved! biometrics: Tell me which image you like and I'll tell you who you are | |
| CN107886062B (en) | Image processing method, system and server | |
| CN108108807A (en) | Learning-oriented image processing method, system and server | |
| CN105721936B (en) | A kind of intelligent television program recommendation system based on context aware | |
| CN110503076A (en) | Video classification methods, device, equipment and medium based on artificial intelligence | |
| Zhang et al. | Short and long range relation based spatio-temporal transformer for micro-expression recognition | |
| CN109829448A (en) | Face identification method, device and storage medium | |
| Anand et al. | An improved local binary patterns histograms techniques for face recognition for real time application | |
| Dzhangarov et al. | Face recognition methods | |
| CN109271930A (en) | Micro- expression recognition method, device and storage medium | |
| CN110427881A (en) | The micro- expression recognition method of integration across database and device based on the study of face local features | |
| Adnan et al. | Facial feature extraction for face recognition | |
| KR101174103B1 (en) | A face recognition method of Mathematics pattern analysis for muscloskeletal in basics | |
| CN113205045A (en) | Pedestrian re-identification method and device, electronic equipment and readable storage medium | |
| Bharadi et al. | Multi-instance iris recognition | |
| Kadambari et al. | Automation of attendance system using facial recognition | |
| Zheng et al. | A normalized light CNN for face recognition | |
| Kekre et al. | Improved Face Recognition with Multilevel BTC using Kekre’s LUV Color Space |
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 |