CN110516603A - Information processing method and device - Google Patents
Information processing method and device Download PDFInfo
- Publication number
- CN110516603A CN110516603A CN201910799755.5A CN201910799755A CN110516603A CN 110516603 A CN110516603 A CN 110516603A CN 201910799755 A CN201910799755 A CN 201910799755A CN 110516603 A CN110516603 A CN 110516603A
- Authority
- CN
- China
- Prior art keywords
- human body
- human
- result
- component
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the present application discloses information processing method and device.One specific embodiment of this method includes: to obtain target body image, by the human testing network and component segmentation network in target body image input human testing model trained in advance, wherein, human body detection model is used to extract the characteristics of human body of the component feature including all parts；Obtain the characteristics of human body for the human body for being included from the target body image that human body detection model exports, wherein, the characteristics of human body of the human body includes the eigenmatrix for the human body for being included, and the component feature matrix formed from the component feature that the component divides all parts for the human body that network exports from the target body image that human body detection network exports.The method provided by the above embodiment of the application can sufficiently obtain feature corresponding with all parts, avoid the minutia for ignoring widget, to improve the recall rate and accuracy rate of human testing.
Description
Technical field
The invention relates to field of computer technology, and in particular at Internet technical field more particularly to information
Manage method and apparatus.
Background technique
With the development of image detecting technique, Human Detection is more and more widely used.It is examined to human body
During survey, complete detection can be carried out to head, limbs.
In the related art, carry out detection image commonly using deep neural network.It is tied using deep neural network
When fruit is predicted, it can use and the modes such as subgraph, Lai Tigao detection accuracy are cut into image carry out level.
Summary of the invention
The embodiment of the present application proposes information processing method and device.
In a first aspect, the embodiment of the present application provides a kind of information processing method, comprising: target body image is obtained, it will
Human testing network and component in target body image input human testing model trained in advance divide network, wherein people
Body detection model is used to extract the characteristics of human body of the component feature including all parts；Obtain the mesh exported from human testing model
The characteristics of human body for the human body that mark human body image is included, wherein the characteristics of human body of human body includes exporting from human testing network
The eigenmatrix for the human body that target body image is included, and from component segmentation network export human body all parts portion
The component feature matrix of part feature composition.
In some embodiments, the human body for the human body for being included from the target body image that human testing model exports is obtained
Feature, comprising: obtain the mask matrix for dividing the human body that network exports from component, wherein mask matrix includes each of human body
The component feature of component.
In some embodiments, method further include: execute following characteristic processing step: component feature matrix is turned
It sets, obtains transposed matrix；Determine the result of the Matrix Multiplication between eigenmatrix and transposed matrix as the first result；To two spies
It levies matrix and carries out step-by-step multiplication, obtain step-by-step multiplied result, determine the result of the Matrix Multiplication of step-by-step multiplied result and transposed matrix
As the second result；Based on the first result and second as a result, identification human body.
In some embodiments, based on the first result and second as a result, identification human body, comprising: to the first result and second
As a result spliced, obtain splicing result；Similarity based on splicing result Yu the feature of the included human body of specified human body image,
Determine the human body that target body image is included and the human body that specified human body image is included, if the instruction same person.
In some embodiments, the training step of human testing model includes: to obtain human body image sample, is examined using human body
Survey grid network obtains the eigenmatrix of human body image sample；Divide network using component, obtains the component feature of human body image sample
Matrix；Eigenmatrix and component feature matrix to human body image sample execute characteristic processing step, obtain human body image sample
Corresponding first result and the second result；Based on corresponding first result of human body image sample and second as a result, the initial people of training
Body detection model.
In some embodiments, based on corresponding first result of human body image sample and second as a result, the initial human body of training
Detection model, including following prediction result generation step: by corresponding first result of the human body image sample comprising human sample
In, it indicates respectively the component feature of each component of human sample, inputs full articulamentum and the classification of initial human testing network
Layer is handled, and the first prediction result to human sample is obtained；By the human body image sample comprising human sample corresponding
In two results, the component feature of each component of human sample, the full articulamentum of input initial part segmentation network are indicated respectively
It is handled with classification layer, obtains the second prediction result to human sample.
In some embodiments, based on corresponding first result of human body image sample and second as a result, the initial human body of training
Detection model, further includes: network is divided to initial human testing network and initial part respectively, is performed the following operations: according to pre-
If loss function, and the prediction result to human sample, default mark, determine the penalty values of human sample, are based on penalty values
It is trained, wherein default loss function is associated with classification learning target；According to the prediction result, default to human sample
Mark and default metric function carry out metric learning, so that same human body determined by the human testing model that study obtains
Feature similarity be greater than different human body feature similarity.
Second aspect, the embodiment of the present application provide a kind of information processing unit, comprising: acquiring unit is configured to obtain
Target body image is taken, by the human testing network and component in target body image input human testing model trained in advance
Divide network, wherein human testing model is used to extract the characteristics of human body of the component feature including all parts；Output unit,
It is configured to obtain the characteristics of human body for the human body for being included from the target body image that human testing model exports, wherein human body
Characteristics of human body include the target body image human body that is included exported from human testing network eigenmatrix, and from portion
Part divides the component feature matrix of the component feature composition of all parts of the human body of network output.
In some embodiments, output unit is further configured to: obtaining the human body for dividing network output from component
Mask matrix, wherein mask matrix includes the component feature of all parts of human body.
In some embodiments, device further include: processing unit is configured to execute following characteristic processing step: to portion
Part eigenmatrix carries out transposition, obtains transposed matrix；Determine the result conduct of the Matrix Multiplication between eigenmatrix and transposed matrix
First result；Step-by-step multiplication is carried out to two eigenmatrixes, step-by-step multiplied result is obtained, determines step-by-step multiplied result and transposition square
The result of the Matrix Multiplication of battle array is as the second result；Recognition unit is configured to based on the first result and second as a result, identification people
Body.
In some embodiments, recognition unit, comprising: splicing module, be configured to the first result and the second result into
Row splicing, obtains splicing result；Determining module is configured to the spy based on splicing result Yu the included human body of specified human body image
The similarity of sign determines the human body that target body image is included and the human body that specified human body image is included, if instruction is same
One people.
In some embodiments, the training step of human testing model includes: to obtain human body image sample, is examined using human body
Survey grid network obtains the eigenmatrix of human body image sample；Divide network using component, obtains the component feature of human body image sample
Matrix；Eigenmatrix and component feature matrix to human body image sample execute characteristic processing step, obtain human body image sample
Corresponding first result and the second result；Based on corresponding first result of human body image sample and second as a result, the initial people of training
Body detection model.
In some embodiments, based on corresponding first result of human body image sample and second as a result, the initial human body of training
Detection model, including following prediction result generation step: by corresponding first result of the human body image sample comprising human sample
In, it indicates respectively the component feature of each component of human sample, inputs full articulamentum and the classification of initial human testing network
Layer is handled, and the first prediction result to human sample is obtained；By the human body image sample comprising human sample corresponding
In two results, the component feature of each component of human sample, the full articulamentum of input initial part segmentation network are indicated respectively
It is handled with classification layer, obtains the second prediction result to human sample.
In some embodiments, based on corresponding first result of human body image sample and second as a result, the initial human body of training
Detection model, further includes: network is divided to initial human testing network and initial part respectively, is performed the following operations: according to pre-
If loss function, and the prediction result to human sample, default mark, determine the penalty values of human sample, are based on penalty values
It is trained, wherein default loss function is associated with classification learning target；According to the prediction result, default to human sample
Mark and default metric function carry out metric learning, so that same human body determined by the human testing model that study obtains
Feature similarity be greater than different human body feature similarity.
The third aspect, the embodiment of the present application provide a kind of electronic equipment, comprising: one or more processors；Storage dress
It sets, for storing one or more programs, when one or more programs are executed by one or more processors, so that one or more
A processor realizes the method such as any embodiment in information processing method.
Fourth aspect, the embodiment of the present application provide a kind of computer readable storage medium, are stored thereon with computer journey
Sequence realizes the method such as any embodiment in information processing method when the program is executed by processor.
Information processing scheme provided by the embodiments of the present application, firstly, target body image is obtained, target body image is defeated
Enter the human testing network and component segmentation network in human testing model trained in advance, wherein human testing model is used for
Extract the characteristics of human body of the component feature including all parts.Later, the target body figure exported from human testing model is obtained
As the characteristics of human body for the human body for being included, wherein the characteristics of human body of human body includes the target body exported from human testing network
The eigenmatrix for the human body that image is included, and from component segmentation network export human body all parts component feature group
At component feature matrix.The scheme provided by the above embodiment of the application can sufficiently obtain spy corresponding with all parts
Sign, avoids the minutia for ignoring widget, to improve the recall rate and accuracy rate of human testing.Also, the application is real
Applying example can use human testing network and component segmentation network, determine the global feature of human body and the component of all parts respectively
Feature, so that the connection between all parts can be also got while being concerned about all parts independent minutia,
Make the feature of output very detailed, accurate.
Detailed description of the invention
By reading a detailed description of non-restrictive embodiments in the light of the attached drawings below, the application's is other
Feature, objects and advantages will become more apparent upon:
Fig. 1 is that this application can be applied to exemplary system architecture figures therein；
Fig. 2 is the flow chart according to one embodiment of the information processing method of the application；
Fig. 3 is the schematic diagram according to an application scenarios of the information processing method of the application；
Fig. 4 is the flow chart according to another embodiment of the information processing method of the application；
Fig. 5 is the structural schematic diagram according to one embodiment of the information processing unit of the application；
Fig. 6 is adapted for the structural schematic diagram for the computer system for realizing the electronic equipment of the embodiment of the present application.
Specific embodiment
The application is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining related invention, rather than the restriction to the invention.It also should be noted that in order to
Convenient for description, part relevant to related invention is illustrated only in attached drawing.
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.The application is described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
Fig. 1 is shown can be using the exemplary system of the embodiment of the information processing method or information processing unit of the application
System framework 100.
As shown in Figure 1, system architecture 100 may include terminal device 101,102,103, network 104 and server 105.
Network 104 between terminal device 101,102,103 and server 105 to provide the medium of communication link.Network 104 can be with
Including various connection types, such as wired, wireless communication link or fiber optic cables etc..
User can be used terminal device 101,102,103 and be interacted by network 104 with server 105, to receive or send out
Send message etc..Various telecommunication customer end applications can be installed on terminal device 101,102,103, such as information processing application,
Video class application, live streaming application, instant messaging tools, mailbox client, social platform software etc..
Here terminal device 101,102,103 can be hardware, be also possible to software.When terminal device 101,102,
103 be hardware when, can be the various electronic equipments with display screen, including but not limited to smart phone, tablet computer, electronics
Book reader, pocket computer on knee and desktop computer etc..It, can be with when terminal device 101,102,103 is software
It is mounted in above-mentioned cited electronic equipment.Multiple softwares or software module may be implemented into (such as providing distribution in it
The multiple softwares or software module of formula service), single software or software module also may be implemented into.It is not specifically limited herein.
Server 105 can be to provide the server of various services, such as provide support to terminal device 101,102,103
Background server.Background server can carry out the data such as the target body image received the processing such as analyzing, and will place
Reason result (such as characteristics of human body) feeds back to terminal device.
It should be noted that information processing method provided by the embodiment of the present application can be by server 105 or terminal
Equipment 101,102,103 executes, correspondingly, information processing unit can be set in server 105 or terminal device 101,
102, in 103.
It should be understood that the number of terminal device, network and server in Fig. 1 is only schematical.According to realization need
It wants, can have any number of terminal device, network and server.
With continued reference to Fig. 2, the process 200 of one embodiment of the information processing method according to the application is shown.The letter
Cease processing method, comprising the following steps:
Step 201, target body image is obtained, it will be in target body image input human testing model trained in advance
Human testing network and component divide network, wherein human testing model is used to extract the component feature including all parts
Characteristics of human body.
In the present embodiment, the executing subject (such as server shown in FIG. 1 or terminal device) of information processing method can
To obtain target body image from local or other electronic equipments, and the image is inputted into the people in human testing model respectively
Physical examination survey grid network and component divide network.Specifically, characteristics of human body includes each portion for the human body that target body image is included
The component feature of part.Here component can refer to the default part of human body, for example, arm, leg, face, head etc..Component feature
For show some component characteristic feature.In some cases, the characteristics of human body of output not only may include component feature,
It can also include the global feature etc. of human body.
Human testing model can be deep neural network, such as convolutional neural networks or depth residual error network etc., people
Body detection model may include human testing network and component segmentation network.Specifically, it can detecte using human testing model
The all parts of human body out, for example, human testing model can export all parts in the position of target body image.Here defeated
Characteristics of human body out can be the convolutional layer output of human testing model, and the full articulamentum for being also possible to human testing model is defeated
Out.
Human testing network can be used for extracting characteristics of image, may include convolutional layer etc..Here component divides network
It can be example partitioning algorithm (such as mask candidate region convolutional neural networks Mask-R-CNN), be also possible to semantic segmentation calculation
Method.Component segmentation network can extract the component feature of all parts of the included human body of image, may include convolutional layer etc..
Step 202, the characteristics of human body for the human body for being included from the target body image that human testing model exports is obtained,
In, the characteristics of human body of human body includes the eigenmatrix for the human body for being included from the target body image that human testing network exports,
And the component feature matrix that the component feature of all parts of the human body exported from component segmentation network forms.
In the present embodiment, the available characteristics of human body exported from human testing model of above-mentioned executing subject.It obtains
Characteristics of human body can show as the form of matrix.Specifically, eigenmatrix namely characteristic pattern can show as three-dimensional matrice (C, H
× W), the dimension that C therein is characterized, H, W are respectively the height and width of characteristic pattern.The component feature of above-mentioned each component can be with
Show as a line feature or a column feature.Correspondingly, the component feature of all parts can be with building block eigenmatrix.
In some optional implementations of the present embodiment, step 202 may include: obtain from component segmentation network it is defeated
The mask matrix of human body out, wherein mask matrix includes the component feature of all parts of human body.
In these optional implementations, the mask square of the above-mentioned available component segmentation network output of executing subject
Battle array.That is, component feature matrix can show as mask matrix.In mask matrix, the component feature of a component can be with table
It is now characterized the feature that the component is only presented in figure, the region where other features other than the component is blocked.Specifically, it covers
Modular matrix can be expressed as (N, H × W), and N therein is the number of human part, and H, W are respectively the height and width of mask matrix.This
In H, W can with human testing network export H, W of eigenmatrix it is identical.
These implementations can use the mask features composition mask matrix of all parts, thus expressing each component
Feature when, can maximumlly reduce the feature of other component for expressing the influence of the component feature, with more acurrate earth's surface
Reveal the feature of each component.
With continued reference to the schematic diagram that Fig. 3, Fig. 3 are according to the application scenarios of the information processing method of the present embodiment.In
In the application scenarios of Fig. 3, the available target body image 302 of executing subject 301 inputs target body image 302 in advance
Human testing network 3031 and component in trained human testing model 303 divide network 3032, obtain from human testing mould
The characteristics of human body 304 for the human body that is included of target body image 302 that type 303 exports, wherein the characteristics of human body of human body include from
The eigenmatrix for the human body that the target body image that human testing network 3031 exports is included, and divide network from component
The component feature matrix of the component feature composition of all parts of the human body of 3032 outputs, human testing model 303 is for extracting
The characteristics of human body of component feature including all parts.
The method provided by the above embodiment of the application can sufficiently obtain feature corresponding with all parts, avoid neglecting
The minutia of smaller component, to improve the recall rate and accuracy rate of human testing.Also, the present embodiment can use people
Physical examination survey grid network and component divide network, the global feature of human body and the component feature of all parts are determined respectively, thus closing
While infusing all parts independent minutia, the connection between all parts can be also got, allows the feature of output
It is very detailed, accurate.
With further reference to Fig. 4, it illustrates the processes 400 of another embodiment of information processing method.The information processing
The process 400 of method, comprising the following steps:
Step 401, target body image is obtained, it will be in target body image input human testing model trained in advance
Human testing network and component divide network, wherein human testing model is used to extract the component feature including all parts
Characteristics of human body.
In the present embodiment, the executing subject (such as server shown in FIG. 1 or terminal device) of information processing method can
To obtain target body image from local or other electronic equipments, and the image is inputted into the people in human testing model respectively
Physical examination survey grid network and component divide network.Specifically, characteristics of human body includes each portion for the human body that target body image is included
The component feature of part.Here component can refer to the default part of human body, for example, arm, leg, face, head etc..Component feature
For show some component characteristic feature.In some cases, the characteristics of human body of output not only may include component feature,
It can also include the global feature etc. of human body.
Step 402, the characteristics of human body for the human body for being included from the target body image that human testing model exports is obtained,
In, the characteristics of human body of human body includes the eigenmatrix for the human body for being included from the target body image that human testing network exports,
And the component feature matrix that the component feature of all parts of the human body exported from component segmentation network forms.
In the present embodiment, the available characteristics of human body exported from human testing model of above-mentioned executing subject.It obtains
Characteristics of human body can show as the form of matrix.Specifically, eigenmatrix namely characteristic pattern can show as three-dimensional matrice (C, H
× W), the dimension that C therein is characterized, H, W are respectively the height and width of characteristic pattern.
Step 403, it executes following characteristic processing step: step 4031, transposition being carried out to component feature matrix, obtains transposition
Matrix；Step 4032, determine the result of the Matrix Multiplication between eigenmatrix and transposed matrix as the first result；Step 4033,
Step-by-step multiplication is carried out to two eigenmatrixes, step-by-step multiplied result is obtained, determines the matrix of step-by-step multiplied result and transposed matrix
The result multiplied is as the second result.
In the present embodiment, above-mentioned executing subject can carry out transposition to above-mentioned component feature matrix, obtain transposed matrix.
Determine the result of the Matrix Multiplication for the eigenmatrix and the transposed matrix that human testing network exports as the first result.Specifically,
Each column feature in first result can be the component feature of a component.
Above-mentioned executing subject can carry out step-by-step multiplications to two features described above matrixes, the result being multiplied to step-by-step with it is above-mentioned
Transposed matrix carries out Matrix Multiplication, and using obtained result as the second result.Specifically, each column feature in the second result can
Think the component feature of a component.
In practice, the first result and the second result may appear as (C, N), and C therein is characterized dimension, and N is portion
The number of part.
Step 404, based on the first result and second as a result, identification human body.
In the present embodiment, above-mentioned executing subject can be based on the first result using various ways and the second result carries out people
Body identification.For example, above-mentioned executing subject can determine that the first result is similar to the feature of the included human body of specified human body image
Degree, and determine the similarity of the feature of the second result and above-mentioned specified the included human body of human body image.If the two similarities
It is all larger than similarity threshold, then can determine the human body that target body image is included and the human body that specified human body image is included
Indicate the same person.
In some optional implementations of the present embodiment, step 404 may include: to the first result and the second result
Spliced, obtains splicing result；Similarity based on splicing result Yu the feature of the included human body of specified human body image determines
The human body that the human body and specified human body image that target body image is included are included, if the instruction same person.
In these optional implementations, the first result and the second result are all matrix, therefore above-mentioned executing subject can
To splice the first result and the second result, to obtain splicing result.Later using splicing result as target body
The feature of image is compared, to be identified with the feature of the included human body of specified human body image.Specifically, above-mentioned execution
Main body can be in the similarity of the feature of the feature and the included human body of specified human body image of the included human body of target body image
In the case where greater than similarity threshold, determine that the included human body of target body image and the included human body of specified human body image indicate
The same person.In addition, above-mentioned executing subject can also be in the spy of the included human body of target body image in human body image set
The similarity with the feature of the included human body of specified human body image is levied, is included greater than other human body images in human body image set
In the case where the similarity of the feature of human body and the feature of the included human body of specified human body image, determine that target body image is wrapped
The same person is indicated containing human body and the included human body of specified human body image.
These implementations can allow the feature of the first result and the second result sufficiently to be merged by splicing, to obtain
Accurate characteristics of image, to improve the accuracy of human bioequivalence.
In some optional implementations of the present embodiment, the training step of human testing model includes: acquisition human body
Image pattern obtains the eigenmatrix of human body image sample using human testing network；Divide network using component, obtains people
The component feature matrix of body image pattern；Eigenmatrix and component feature matrix to human body image sample execute features described above
Processing step obtains corresponding first result of human body image sample and the second result；Based on human body image sample corresponding first
As a result with second as a result, the initial human testing model of training.
In these optional implementations, the human figure that the above-mentioned available training sample of executing subject is concentrated is decent
This, and human testing network and component the segmentation network being utilized respectively in human testing model, obtain the spy of human body image sample
Levy matrix and component feature matrix.Later, above-mentioned executing subject can execute features described above processing step, to obtain the first result
With second as a result, and the initial human testing model of training.Here initial human testing model refers to up for trained human body inspection
Survey model.
Above-mentioned executing subject can use various ways based on the first result of human body image sample and second as a result, training
Initial human testing model.For example, above-mentioned executing subject can splice the result of the first result and the second result, and will
Splicing result input convolutional layer further extracts feature.Above-mentioned executing subject can input the feature further extracted complete later
Articulamentum etc., to obtain the prediction result of human testing model.It is determined in this way, above-mentioned executing subject can use prediction result
Penalty values, and utilize the initial human testing model of penalty values training.
These implementations can use the first result and the second result is trained, thus the global feature for passing through human body
It is trained with the sufficiently fused detailed features of component feature, improves the recall rate and standard of the human testing model that training obtains
True rate.
It is above-mentioned based on human body image sample corresponding the in some optional application scenarios of these optional implementations
One result and second is as a result, the initial human testing model of training, may include following prediction result generation step: will include human body
In corresponding first result of the human body image sample of sample, the component feature of each component of human sample is indicated respectively, input
The full articulamentum and classification layer of initial human testing network are handled, and the first prediction result to human sample is obtained；It will packet
In corresponding second result of human body image sample containing human sample, indicate respectively that the component of each component of human sample is special
Sign, the full articulamentum and classification layer of input initial part segmentation network are handled, and the second prediction knot to human sample is obtained
Fruit.
In these optional application scenarios, above-mentioned executing subject can be by the component feature of all parts in the first result
It inputs full articulamentum and classification layer is handled, and the component feature of all parts in the second result is inputted into full articulamentum and is divided
Class layer (for example, classification layer may include softmax function) is handled, and carries out independent training to each component to realize.
In this way, the available processing result corresponding with each component respectively of full articulamentum.Above-mentioned executing subject can use a variety of
Processing result of the mode based on full articulamentum and classification layer, generates the prediction result of all parts.Here prediction result is specific
It can be the identity of the personnel identified.Later, above-mentioned executing subject or other electronic equipments can be utilized respectively
First prediction result is trained, and is trained using the second prediction result.For example, above-mentioned executing subject or other electronics
Equipment can use the initial human testing network of the first prediction result training, and utilize the second prediction result training initial part point
Network is cut, and/or, divide network using the first prediction result training initial part, and initial using the training of the second prediction result
Human testing network.Initial human testing network and initial part segmentation network are respectively up for trained human testing network
Divide network with component.Under normal conditions, the first prediction result and the second prediction result are respectively used to train, in some cases
Under, the first prediction result and the second prediction result, which can also merge, to be trained.
Human testing model in these application scenarios can use the component feature of each component, generate accurate
Prediction result, so that human testing model can accurately predict the component feature of all parts.
Optionally, based on corresponding first result of human body image sample and second as a result, the initial human testing model of training,
It can also include: that network is divided to initial human testing network and initial part respectively, perform the following operations: according to default loss
Function, and the prediction result to human sample, default mark, are determined the penalty values of human sample, are instructed based on penalty values
Practice, wherein default loss function is associated with classification learning target；According to the prediction result to human sample, default mark, with
And default metric function, metric learning is carried out, so that the feature of same human body determined by the human testing model that study obtains
Similarity be greater than different human body feature similarity.
Specifically, above-mentioned executing subject can use the prediction result of all parts, and marks and obtain in advance to human body
Default mark, determine the penalty values of human sample.And be trained in initial human testing model using the penalty values, than
Such as backpropagation, network is divided with the human testing network and component that are respectively trained in initial human testing model.In addition, above-mentioned
Executing subject be divided network by can also be allowed human testing network and component using default metric function and carry out metric learning respectively, with
The similarity of the feature for the same human body that human testing network and component segmentation network after realizing metric learning extract is greater than not
With the similarity of the feature of human body.
These optional modes can will be combined using the training and metric learning of penalty values, with improve training effectiveness and
Training accuracy.
The present embodiment can be by calculating the first result and second as a result, the global feature of human body is mutually melted with component feature
It closes, to obtain more accurate characteristics of human body, increases the accuracy of human bioequivalence.
With further reference to Fig. 5, as the realization to method shown in above-mentioned each figure, this application provides a kind of information processing apparatus
The one embodiment set, the Installation practice is corresponding with embodiment of the method shown in Fig. 2, which specifically can be applied to respectively
In kind electronic equipment.
As shown in figure 5, the information processing unit 500 of the present embodiment includes: acquiring unit 501 and output unit 502.Its
In, acquiring unit 501 is configured to obtain target body image, by target body image input human testing trained in advance
Human testing network and component in model divide network, wherein human testing model is for extracting the portion including all parts
The characteristics of human body of part feature.Output unit 502 is configured to obtain and be wrapped from the target body image that human testing model exports
The characteristics of human body of the human body contained, wherein the characteristics of human body of human body includes the target body image institute exported from human testing network
The eigenmatrix for the human body for including, and from component segmentation network export human body all parts component feature form portion
Part eigenmatrix.
In some embodiments, the acquiring unit 501 of information processing unit 500 can be from local or other electronic equipments
Target body image is obtained, and the human testing network and component that the image is inputted respectively in human testing model are divided into net
Network.Specifically, characteristics of human body includes the component feature of all parts for the human body that target body image is included.
In some embodiments, the available characteristics of human body exported from human testing model of output unit 502.It obtains
Characteristics of human body can show as the form of matrix.Specifically, eigenmatrix namely characteristic pattern can show as three-dimensional matrice (C, H
× W), the dimension that C therein is characterized, H, W are respectively the height and width of characteristic pattern.The component feature of above-mentioned each component can be with
Show as a line feature or a column feature.Correspondingly, the component feature of all parts can be with building block eigenmatrix.
In some optional implementations of the present embodiment, output unit is further configured to: being obtained from component point
Cut the mask matrix of the human body of network output, wherein mask matrix includes the component feature of all parts of human body.
In some optional implementations of the present embodiment, device further include: processing unit is configured to execute as follows
Characteristic processing step: transposition is carried out to component feature matrix, obtains transposed matrix；It determines between eigenmatrix and transposed matrix
The result of Matrix Multiplication is as the first result；Step-by-step multiplication is carried out to two eigenmatrixes, step-by-step multiplied result is obtained, determines step-by-step
The result of the Matrix Multiplication of multiplied result and transposed matrix is as the second result；Recognition unit, be configured to based on the first result and
Second as a result, identification human body.
In some optional implementations of the present embodiment, recognition unit, comprising: splicing module is configured to
One result and the second result are spliced, and splicing result is obtained；Determining module is configured to based on splicing result and specified human body
The similarity of the feature of the included human body of image determines that the human body that target body image is included is included with specified human body image
Human body, if instruction the same person.
In some optional implementations of the present embodiment, the training step of human testing model includes: acquisition human body
Image pattern obtains the eigenmatrix of human body image sample using human testing network；Divide network using component, obtains people
The component feature matrix of body image pattern；Eigenmatrix and component feature matrix to human body image sample execute characteristic processing
Step obtains corresponding first result of human body image sample and the second result；Based on corresponding first result of human body image sample
With second as a result, the initial human testing model of training.
In some optional implementations of the present embodiment, it is based on corresponding first result of human body image sample and second
As a result, the initial human testing model of training, including following prediction result generation step: the human figure comprising human sample is decent
In this corresponding first result, indicates respectively the component feature of each component of human sample, input initial human testing network
Full articulamentum and classification layer handled, obtain the first prediction result to human sample；By the human body comprising human sample
In corresponding second result of image pattern, the component feature of each component of human sample, input initial part point are indicated respectively
The full articulamentum and classification layer for cutting network are handled, and the second prediction result to human sample is obtained.
In some optional implementations of the present embodiment, it is based on corresponding first result of human body image sample and second
As a result, the initial human testing model of training, further includes: divide network to initial human testing network and initial part respectively, hold
The following operation of row: according to default loss function, and the prediction result to human sample, default mark, determine human sample
Penalty values are trained based on penalty values, wherein default loss function is associated with classification learning target；According to human body sample
This prediction result, default mark and default metric function, carry out metric learning, so that the human testing mould that study obtains
The similarity of the feature of same human body determined by type is greater than the similarity of the feature of different human body.
As shown in fig. 6, electronic equipment 600 may include processing unit (such as central processing unit, graphics processor etc.)
601, random access can be loaded into according to the program being stored in read-only memory (ROM) 602 or from storage device 608
Program in memory (RAM) 603 and execute various movements appropriate and processing.In RAM 603, it is also stored with electronic equipment
Various programs and data needed for 600 operations.Processing unit 601, ROM 602 and RAM 603 pass through the phase each other of bus 604
Even.Input/output (I/O) interface 605 is also connected to bus 604.
In general, following device can connect to I/O interface 605: including such as touch screen, touch tablet, keyboard, mouse, taking the photograph
As the input unit 606 of head, microphone, accelerometer, gyroscope etc.；Including such as liquid crystal display (LCD), loudspeaker, vibration
The output device 607 of dynamic device etc.；Storage device 608 including such as tape, hard disk etc.；And communication device 609.Communication device
609, which can permit electronic equipment 600, is wirelessly or non-wirelessly communicated with other equipment to exchange data.Although Fig. 6 shows tool
There is the electronic equipment 600 of various devices, it should be understood that being not required for implementing or having all devices shown.It can be with
Alternatively implement or have more or fewer devices.Each box shown in Fig. 6 can represent a device, can also root
According to needing to represent multiple devices.
Particularly, in accordance with an embodiment of the present disclosure, it may be implemented as computer above with reference to the process of flow chart description
Software program.For example, embodiment of the disclosure includes a kind of computer program product comprising be carried on computer-readable medium
On computer program, which includes the program code for method shown in execution flow chart.In such reality
It applies in example, which can be downloaded and installed from network by communication device 609, or from storage device 608
It is mounted, or is mounted from ROM 602.When the computer program is executed by processing unit 601, the implementation of the disclosure is executed
The above-mentioned function of being limited in the method for example.It should be noted that the computer-readable medium of embodiment of the disclosure can be meter
Calculation machine readable signal medium or computer readable storage medium either the two any combination.Computer-readable storage
Medium for example may be-but not limited to-system, device or the device of electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor,
Or any above combination.The more specific example of computer readable storage medium can include but is not limited to: have one
Or the electrical connections of multiple conducting wires, portable computer diskette, hard disk, random access storage device (RAM), read-only memory (ROM),
Erasable programmable read only memory (EPROM or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), light
Memory device, magnetic memory device or above-mentioned any appropriate combination.In embodiment of the disclosure, computer-readable to deposit
Storage media can be any tangible medium for including or store program, which can be commanded execution system, device or device
Part use or in connection.And in embodiment of the disclosure, computer-readable signal media may include in base band
In or as carrier wave a part propagate data-signal, wherein carrying computer-readable program code.This propagation
Data-signal can take various forms, including but not limited to electromagnetic signal, optical signal or above-mentioned any appropriate combination.Meter
Calculation machine readable signal medium can also be any computer-readable medium other than computer readable storage medium, which can
Read signal medium can be sent, propagated or be transmitted for being used by instruction execution system, device or device or being tied with it
Close the program used.The program code for including on computer-readable medium can transmit with any suitable medium, including but not
It is limited to: electric wire, optical cable, RF (radio frequency) etc. or above-mentioned any appropriate combination.
Flow chart and block diagram in attached drawing are illustrated according to the system of the various embodiments of the application, method and computer journey
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part of one module, program segment or code of table, a part of the module, program segment or code include one or more use
The executable instruction of the logic function as defined in realizing.It should also be noted that in some implementations as replacements, being marked in box
The function of note can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are actually
It can be basically executed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.Also it to infuse
Meaning, the combination of each box in block diagram and or flow chart and the box in block diagram and or flow chart can be with holding
The dedicated hardware based system of functions or operations as defined in row is realized, or can use specialized hardware and computer instruction
Combination realize.
Being described in unit involved in the embodiment of the present application can be realized by way of software, can also be by hard
The mode of part is realized.Described unit also can be set in the processor, for example, can be described as: a kind of processor packet
Include acquiring unit and output unit.Wherein, the title of these units does not constitute the limit to the unit itself under certain conditions
It is fixed, for example, acquiring unit is also described as " target body image being obtained, by target body image input training in advance
The unit of human testing network and component segmentation network in human testing model ".
As on the other hand, present invention also provides a kind of computer-readable medium, which be can be
Included in device described in above-described embodiment；It is also possible to individualism, and without in the supplying device.Above-mentioned calculating
Machine readable medium carries one or more program, when said one or multiple programs are executed by the device, so that should
Device: obtaining target body image, by the human testing net in target body image input human testing model trained in advance
Network and component divide network, wherein human testing model is used to extract the characteristics of human body of the component feature including all parts；
The characteristics of human body for the human body for being included to the target body image exported from human testing model, wherein the characteristics of human body of human body
Eigenmatrix including the human body that the target body image exported from human testing network is included, and divide network from component
The component feature matrix of the component feature composition of all parts of the human body of output.
Above description is only the preferred embodiment of the application and the explanation to institute's application technology principle.Those skilled in the art
Member is it should be appreciated that invention scope involved in the application, however it is not limited to technology made of the specific combination of above-mentioned technical characteristic
Scheme, while should also cover in the case where not departing from foregoing invention design, it is carried out by above-mentioned technical characteristic or its equivalent feature
Any combination and the other technical solutions formed.Such as features described above has similar function with (but being not limited to) disclosed herein
Can technical characteristic replaced mutually and the technical solution that is formed.
Claims (16)
1. a kind of information processing method, comprising:
Target body image is obtained, by the human testing in target body image input human testing model trained in advance
Network and component divide network, wherein the human testing model is used to extract the human body of the component feature including all parts
Feature；
Obtain the characteristics of human body for the human body for being included from the target body image that the human testing model exports, wherein
The characteristics of human body of the human body includes the human body for being included from the target body image that the human testing network exports
Eigenmatrix, and the component formed from the component feature that the component divides all parts for the human body that network exports are special
Levy matrix.
2. according to the method described in claim 1, wherein, the target person for obtaining exporting from the human testing model
The characteristics of human body for the human body that body image is included, comprising:
Obtain the mask matrix for dividing the human body that network exports from the component, wherein the mask matrix includes described
The component feature of all parts of human body.
3. according to the method described in claim 1, wherein, the method also includes:
It executes following characteristic processing step: transposition being carried out to the component feature matrix, obtains transposed matrix；Determine the feature
The result of Matrix Multiplication between matrix and the transposed matrix is as the first result；Step-by-step phase is carried out to two eigenmatrixes
Multiply, obtain step-by-step multiplied result, determines the result of the Matrix Multiplication of the step-by-step multiplied result and the transposed matrix as second
As a result；
Based on first result and described second as a result, identifying the human body.
4. according to the method described in claim 3, wherein, first result and described second that is based on is as a result, identification institute
State human body, comprising:
First result and second result are spliced, splicing result is obtained；
Similarity based on the splicing result Yu the feature of the included human body of specified human body image, determines the target body figure
The human body for being included as the human body for being included and the specified human body image, if the instruction same person.
5. according to the method described in claim 3, wherein, the training step of the human testing model includes:
It obtains human body image sample and obtains the eigenmatrix of the human body image sample using the human testing network；
Divide network using the component, obtains the component feature matrix of the human body image sample；
To the eigenmatrix and component feature matrix of the human body image sample, the characteristic processing step is executed, is obtained described
Corresponding first result of human body image sample and the second result；
Based on corresponding first result of the human body image sample and second as a result, the initial human testing model of training.
It is described based on corresponding first result of the human body image sample and the 6. according to the method described in claim 5, wherein
Two as a result, train initial human testing model, including following prediction result generation step:
By in corresponding first result of the human body image sample comprising human sample, the every of the human sample is indicated respectively
The component feature of a component, the full articulamentum and classification layer for inputting initial human testing network are handled, are obtained to the people
First prediction result of body sample；
By in corresponding second result of the human body image sample comprising the human sample, the human sample is indicated respectively
Each component component feature, input initial part segmentation network full articulamentum and classification layer handled, obtain to institute
State the second prediction result of human sample.
It is described based on corresponding first result of the human body image sample and the 7. according to the method described in claim 6, wherein
Two as a result, the initial human testing model of training, further includes:
Network is divided to the initial human testing network and the initial part respectively, is performed the following operations:
According to default loss function, and the prediction result to the human sample, default mark, determine the human sample
Penalty values are trained based on the penalty values, wherein the default loss function is associated with classification learning target；According to
To the prediction result of the human sample, default mark and default metric function, metric learning is carried out, so that study obtains
Human testing model determined by same human body feature similarity be greater than different human body feature similarity.
8. a kind of information processing unit, comprising:
Acquiring unit is configured to obtain target body image, and target body image input human body trained in advance is examined
The human testing network and component surveyed in model divide network, wherein the human testing model includes each portion for extracting
The characteristics of human body of the component feature of part；
Output unit is configured to obtain the human body for being included from the target body image that the human testing model exports
Characteristics of human body, wherein the characteristics of human body of the human body include from the human testing network export the target body figure
As the eigenmatrix of the human body that is included, and the component of all parts of the human body exported from component segmentation network
The component feature matrix of feature composition.
9. device according to claim 8, wherein the output unit is further configured to:
Obtain the mask matrix for dividing the human body that network exports from the component, wherein the mask matrix includes described
The component feature of all parts of human body.
10. device according to claim 8, wherein described device further include:
Processing unit is configured to execute following characteristic processing step: carrying out transposition to the component feature matrix, obtains transposition
Matrix；Determine the result of the Matrix Multiplication between the eigenmatrix and the transposed matrix as the first result；Described in two
Eigenmatrix carries out step-by-step multiplication, obtains step-by-step multiplied result, determines the square of the step-by-step multiplied result and the transposed matrix
The result that battle array multiplies is as the second result；
Recognition unit is configured to based on first result and described second as a result, identifying the human body.
11. device according to claim 10, wherein the recognition unit, comprising:
Splicing module is configured to splice first result and second result, obtains splicing result；
Determining module is configured to the similarity of the feature based on the splicing result Yu the included human body of specified human body image,
Determine the human body that the target body image is included and the human body that the specified human body image is included, if instruction is same
People.
12. device according to claim 10, wherein the training step of the human testing model includes:
It obtains human body image sample and obtains the eigenmatrix of the human body image sample using the human testing network；
Divide network using the component, obtains the component feature matrix of the human body image sample；
To the eigenmatrix and component feature matrix of the human body image sample, the characteristic processing step is executed, is obtained described
Corresponding first result of human body image sample and the second result；
Based on corresponding first result of the human body image sample and second as a result, the initial human testing model of training.
13. device according to claim 12, wherein it is described based on corresponding first result of the human body image sample and
Second as a result, train initial human testing model, including following prediction result generation step:
By in corresponding first result of the human body image sample comprising human sample, the every of the human sample is indicated respectively
The component feature of a component, the full articulamentum and classification layer for inputting initial human testing network are handled, are obtained to the people
First prediction result of body sample；
By in corresponding second result of the human body image sample comprising the human sample, the human sample is indicated respectively
Each component component feature, input initial part segmentation network full articulamentum and classification layer handled, obtain to institute
State the second prediction result of human sample.
14. device according to claim 13, wherein it is described based on corresponding first result of the human body image sample and
Second as a result, the initial human testing model of training, further includes:
Network is divided to the initial human testing network and the initial part respectively, is performed the following operations:
According to default loss function, and the prediction result to the human sample, default mark, determine the human sample
Penalty values are trained based on the penalty values, wherein the default loss function is associated with classification learning target；According to
To the prediction result of the human sample, default mark and default metric function, metric learning is carried out, so that study obtains
Human testing model determined by same human body feature similarity be greater than different human body feature similarity.
15. a kind of electronic equipment, comprising:
One or more processors；
Storage device, for storing one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processors are real
The now method as described in any in claim 1-7.
16. a kind of computer readable storage medium, is stored thereon with computer program, wherein when the program is executed by processor
Realize the method as described in any in claim 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910799755.5A CN110516603B (en) | 2019-08-28 | 2019-08-28 | Information processing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910799755.5A CN110516603B (en) | 2019-08-28 | 2019-08-28 | Information processing method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110516603A true CN110516603A (en) | 2019-11-29 |
| CN110516603B CN110516603B (en) | 2022-03-18 |
Family
ID=68627492
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910799755.5A Active CN110516603B (en) | 2019-08-28 | 2019-08-28 | Information processing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110516603B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107644209A (en) * | 2017-09-21 | 2018-01-30 | 百度在线网络技术（北京）有限公司 | Method for detecting human face and device |
| CN110046577A (en) * | 2019-04-17 | 2019-07-23 | 北京迈格威科技有限公司 | Pedestrian's attribute forecast method, apparatus, computer equipment and storage medium |
| CN110070073A (en) * | 2019-05-07 | 2019-07-30 | 国家广播电视总局广播电视科学研究院 | Pedestrian's recognition methods again of global characteristics and local feature based on attention mechanism |
| CN110110689A (en) * | 2019-05-15 | 2019-08-09 | 东北大学 | A kind of pedestrian's recognition methods again |
| CN110175595A (en) * | 2019-05-31 | 2019-08-27 | 北京金山云网络技术有限公司 | Human body attribute recognition approach, identification model training method and device |
| CN110174892A (en) * | 2019-04-08 | 2019-08-27 | 北京百度网讯科技有限公司 | Processing method, device, equipment and the computer readable storage medium of vehicle direction |
-
2019
- 2019-08-28 CN CN201910799755.5A patent/CN110516603B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107644209A (en) * | 2017-09-21 | 2018-01-30 | 百度在线网络技术（北京）有限公司 | Method for detecting human face and device |
| CN110174892A (en) * | 2019-04-08 | 2019-08-27 | 北京百度网讯科技有限公司 | Processing method, device, equipment and the computer readable storage medium of vehicle direction |
| CN110046577A (en) * | 2019-04-17 | 2019-07-23 | 北京迈格威科技有限公司 | Pedestrian's attribute forecast method, apparatus, computer equipment and storage medium |
| CN110070073A (en) * | 2019-05-07 | 2019-07-30 | 国家广播电视总局广播电视科学研究院 | Pedestrian's recognition methods again of global characteristics and local feature based on attention mechanism |
| CN110110689A (en) * | 2019-05-15 | 2019-08-09 | 东北大学 | A kind of pedestrian's recognition methods again |
| CN110175595A (en) * | 2019-05-31 | 2019-08-27 | 北京金山云网络技术有限公司 | Human body attribute recognition approach, identification model training method and device |
Non-Patent Citations (2)
| Title |
|---|
| WENQING HUANG ET AL.: "The Combination of Features Extracted from Different Parts for Person Re-Identification", 《2018 2ND INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION SCIENCES (ICRAS)》 * |
| *** 等: "区域块分割与融合的行人再识别", 《中国图象图形学报》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110516603B (en) | 2022-03-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109508681A (en) | The method and apparatus for generating human body critical point detection model | |
| CN107644209A (en) | Method for detecting human face and device | |
| CN109002842A (en) | Image-recognizing method and device | |
| CN108830235A (en) | Method and apparatus for generating information | |
| CN108898185A (en) | Method and apparatus for generating image recognition model | |
| CN109858445A (en) | Method and apparatus for generating model | |
| CN108985208A (en) | The method and apparatus for generating image detection model | |
| CN109410253B (en) | For generating method, apparatus, electronic equipment and the computer-readable medium of information | |
| CN109981787B (en) | Method and device for displaying information | |
| CN109086780A (en) | Method and apparatus for detecting electrode piece burr | |
| CN108121800A (en) | Information generating method and device based on artificial intelligence | |
| CN108989882A (en) | Method and apparatus for exporting the snatch of music in video | |
| CN109829432A (en) | Method and apparatus for generating information | |
| CN107958247A (en) | Method and apparatus for facial image identification | |
| CN109255767A (en) | Image processing method and device | |
| CN110516678A (en) | Image processing method and device | |
| CN108182472A (en) | For generating the method and apparatus of information | |
| CN108427941A (en) | Method, method for detecting human face and device for generating Face datection model | |
| CN110427915A (en) | Method and apparatus for output information | |
| CN110070076A (en) | Method and apparatus for choosing trained sample | |
| CN108171208A (en) | Information acquisition method and device | |
| CN109325996A (en) | Method and apparatus for generating information | |
| CN108509904A (en) | Method and apparatus for generating information | |
| CN108446659A (en) | Method and apparatus for detecting facial image | |
| CN110516099A (en) | Image processing method and device |
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 |