CN109840528A - The method and apparatus for extracting the characteristic pattern of image - Google Patents
The method and apparatus for extracting the characteristic pattern of image Download PDFInfo
- Publication number
- CN109840528A CN109840528A CN201910098620.6A CN201910098620A CN109840528A CN 109840528 A CN109840528 A CN 109840528A CN 201910098620 A CN201910098620 A CN 201910098620A CN 109840528 A CN109840528 A CN 109840528A
- Authority
- CN
- China
- Prior art keywords
- image
- characteristic pattern
- detected
- feature
- characteristic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
Embodiment of the disclosure discloses the method and apparatus for extracting the characteristic pattern of image.One specific embodiment of this method includes: acquisition image to be detected；Image to be detected input feature vector is extracted into network, feature extraction network includes at least three feature extraction layers；Three characteristic patterns are chosen from the characteristic pattern that at least three feature extraction layers export, and are based on three characteristic patterns, are obtained the corresponding characteristic pattern of image to be detected.The characteristic information that the embodiment realizes high, medium and low resolution ratio sufficiently merges.
Description
Technical field
Embodiment of the disclosure is related to field of computer technology, and in particular to extracts the method and dress of the characteristic pattern of image
It sets.
Background technique
When handling image, image input feature vector first can be extracted into network.Then, according to obtained characteristic pattern
A variety of predictions are carried out for image.For example, attitude detection, image segmentation etc..In the process, feature extraction network exports
Characteristic pattern can retain more contextual information.But with the intensification of the network number of plies, the resolution ratio of characteristic pattern is lower, often damages
Many detailed information are lost.
Summary of the invention
Embodiment of the disclosure proposes the characteristic pattern method and apparatus for extracting image.
In a first aspect, embodiment of the disclosure provides a kind of method of characteristic pattern for extracting image, comprising: obtain to be checked
Altimetric image；Image to be detected input feature vector is extracted into network, feature extraction network includes at least three feature extraction layers；From at least
Three characteristic patterns are chosen in the characteristic pattern of three feature extraction layers output, are based on three characteristic patterns, it is corresponding to obtain image to be detected
Characteristic pattern.
In some embodiments, the attitude prediction network that the input of the characteristic pattern of image to be detected is trained in advance, obtain to
The posture information of the human body shown in detection image, wherein attitude prediction network is used to characterize in the characteristic pattern and image of image
The corresponding relationship of the posture information of the human body of display.
In some embodiments, the image segmentation network that the input of the characteristic pattern of image to be detected is trained in advance, obtains pair
The segmentation result information of at least one object shown in image to be detected, wherein image segmentation network is for characterizing image
Characteristic pattern with to the corresponding relationship of the segmentation result information of at least one object shown in image.
In some embodiments, three characteristic patterns are chosen from the characteristic pattern that at least three feature extraction layers export, and are based on
Three characteristic patterns obtain the corresponding characteristic pattern of image to be detected, comprising: from the characteristic pattern that at least three feature extraction layers export
Three characteristic patterns that resolution ratio meets preset condition are chosen, three characteristic patterns are merged, it is corresponding to obtain image to be detected
Characteristic pattern.
In some embodiments, resolution ratio is chosen from the characteristic pattern that at least three feature extraction layers export meets default item
Three characteristic patterns of part, comprising: for the characteristic pattern of at least three feature extraction layers output, according to ascending suitable of resolution ratio
Sequence chooses two characteristic patterns as fisrt feature figure and second feature figure；Choose the resolution of resolution ratio and described image to be detected
The ratio of rate is equal to the characteristic pattern of preset threshold as third feature figure.
In some embodiments, resolution ratio is chosen from the characteristic pattern that at least three feature extraction layers export meets default item
Three characteristic patterns of part, comprising: for the characteristic pattern of at least three feature extraction layers output, choose resolution bits and differentiated in first
The characteristic pattern in rate section is as fisrt feature figure；Resolution bits are chosen in the characteristic pattern in second resolution section as second feature
Figure；Resolution bits are chosen in the characteristic pattern in third resolution ratio section as third feature figure.
Second aspect, embodiment of the disclosure provide a kind of device of characteristic pattern for extracting image, comprising: obtain single
Member is configured to obtain image to be detected；Feature extraction unit is configured to image to be detected input feature vector extracting network,
Feature extraction network includes at least three feature extraction layers；Fusion Features unit is configured to from least three feature extraction layers
Three characteristic patterns are chosen in the characteristic pattern of output, are based on three characteristic patterns, are obtained the corresponding characteristic pattern of image to be detected.
In some embodiments, device further include: attitude prediction unit is configured to the characteristic pattern of image to be detected
Input attitude prediction network trained in advance, obtains the posture information of human body shown in image to be detected, wherein attitude prediction
Network is used to characterize the corresponding relationship of the posture information of the human body shown in the characteristic pattern and image of image.
In some embodiments, device further include: cutting unit is configured to input the characteristic pattern of image to be detected
Trained image segmentation network in advance, obtains the segmentation result information of at least one object for showing in image to be detected,
Wherein, image segmentation network is used to characterize the characteristic pattern of image and the segmentation result at least one object shown in image is believed
The corresponding relationship of breath.
In some embodiments, Fusion Features unit is further configured to: being exported from least three feature extraction layers
Three characteristic patterns that resolution ratio meets preset condition are chosen in characteristic pattern, and three characteristic patterns are merged, mapping to be checked is obtained
As corresponding characteristic pattern.
In some embodiments, Fusion Features unit is further configured to: at least three feature extraction layers are exported
Characteristic pattern choose two characteristic patterns as fisrt feature figure and second feature figure according to the sequence that resolution ratio is ascending；Choosing
The ratio of the resolution ratio of resolution ratio and described image to be detected is taken to be equal to the characteristic pattern of preset threshold as third feature figure.
In some embodiments, Fusion Features unit is further configured to: at least three feature extraction layers are exported
Characteristic pattern, choose resolution bits in first resolution section characteristic pattern as fisrt feature figure；Resolution bits are chosen in the
The characteristic pattern in two resolution ratio sections is as second feature figure；Resolution bits are chosen in the characteristic pattern in third resolution ratio section as the
Three characteristic patterns.
The third aspect, embodiment of the disclosure provide a kind of server, which includes: one or more processing
Device；Storage device is stored thereon with one or more programs；When said one or multiple programs are by said one or multiple processing
Device executes, so that said one or multiple processors realize the method as described in implementation any in first aspect.
Fourth aspect, embodiment of the disclosure provide a kind of computer-readable medium, are stored thereon with computer program,
The method as described in implementation any in first aspect is realized when above procedure is executed by processor.
The method and apparatus that embodiment of the disclosure provides, can obtain image to be detected first.Later, by mapping to be checked
As input feature vector extraction network, feature extraction network includes at least three feature extraction layers.Finally, from least three feature extractions
Three characteristic patterns are chosen in the characteristic pattern of layer output, three characteristic patterns is based on, obtains the corresponding characteristic pattern of image to be detected.Its
In, the resolution ratio of the characteristic pattern of each feature extraction layer output is not identical.By choosing suitable characteristic pattern, can by it is high, in,
The characteristic information of low resolution sufficiently merges.To provide the foundation for processing such as subsequent attitude prediction or image segmentations.
Detailed description of the invention
By reading a detailed description of non-restrictive embodiments in the light of the attached drawings below, the disclosure is other
Feature, objects and advantages will become more apparent upon:
Fig. 1 is that one embodiment of the disclosure can be applied to exemplary system architecture figure therein；
Fig. 2 is the flow chart according to one embodiment of the method for the characteristic pattern of the extraction image of the disclosure；
Fig. 3 is the signal of an application scenarios of the method for the characteristic pattern according to an embodiment of the present disclosure for extracting image
Figure；
Fig. 4 is the flow chart according to another embodiment of the method for the characteristic pattern of the extraction image of the disclosure；
Fig. 5 is the structural schematic diagram according to one embodiment of the device of the characteristic pattern of the extraction image of the disclosure；
Fig. 6 is adapted for the structural schematic diagram for realizing the electronic equipment of embodiment of the disclosure.
Specific embodiment
The disclosure is described in further detail with reference to the accompanying drawings and examples.It is understood that this place
The specific embodiment of description is used only for explaining that correlation is open, rather than the restriction to the disclosure.It also should be noted that being
Convenient for description, is illustrated only in attached drawing and to related disclose relevant part.
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the disclosure can phase
Mutually combination.The disclosure 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 method for the characteristic pattern of the extraction image of embodiment of the disclosure or the spy of extraction image
Levy the exemplary system architecture 100 of the device of figure.
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 client applications, such as the application of picture processing class, figure can be installed on terminal device 101,102,103
Piece shoots class application etc..
Terminal device 101,102,103 can be hardware, be also possible to software.When terminal device 101,102,103 is hard
When part, it can be the various electronic equipments for supporting image storage or transmission.It, can when terminal device 101,102,103 is software
To be mounted in above-mentioned electronic equipment.Multiple softwares or software module may be implemented into (such as providing distributed clothes in it
Business), 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 to the figure that terminal device 101,102,103 uploads
Background server as carrying out the processing such as characteristic pattern extraction.Image to be detected input feature vector can be extracted net by background server
Network, and three characteristic patterns of output are merged, obtain the corresponding characteristic pattern of image to be detected.
It should be noted that extracting the method for the characteristic pattern of image provided by embodiment of the disclosure generally by server
105 execute, and correspondingly, the device for extracting the characteristic pattern of image is generally positioned in server 105.
It should be noted that server can be hardware, it is also possible to software.When server is hardware, may be implemented
At the distributed server cluster that multiple servers form, individual server also may be implemented into.It, can when server is software
To be implemented as multiple softwares or software module (such as providing Distributed Services), single software or software also may be implemented into
Module.It is not specifically limited herein.
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 stream of one embodiment of the method for the characteristic pattern of the extraction image according to the disclosure is shown
Journey 200.The method of the characteristic pattern of the extraction image, comprising the following steps:
Step 201, image to be detected is obtained.
In the present embodiment, the executing subject (such as server 105 shown in fig. 1) of the method for the characteristic pattern of image is extracted
Image to be detected can be obtained from local or communication connection electronic equipment.Image to be detected can be arbitrary image.It is to be checked
The determination of altimetric image can be specified by technical staff, can also be obtained according to certain conditional filtering.
Step 202, image to be detected input feature vector is extracted into network, feature extraction network includes at least three feature extractions
Layer.
In the present embodiment, image to be detected input feature vector can be extracted network by above-mentioned executing subject.Wherein, feature mentions
Taking network includes at least three feature extraction layers.As an example, feature extraction network can be existing various can be used for
The artificial neural network of image characteristics extraction.For example, it may be residual error network ResNet, convolutional neural networks CNN etc..Wherein,
Feature extraction layer can be the random layer in these networks.
Step 203, three characteristic patterns are chosen from the characteristic pattern that at least three feature extraction layers export, and are based on three features
Figure, obtains the corresponding characteristic pattern of image to be detected.
In the present embodiment, above-mentioned executing subject can adopt chooses the output of at least three feature extraction layers in various manners
Three characteristic patterns are chosen in characteristic pattern.Then, three characteristic patterns of selection are merged, obtains the corresponding spy of image to be detected
Sign figure.Specifically, the resolution ratio due to three characteristic patterns is all different with number of active lanes.Three characteristic patterns can be passed through respectively
At least one convolutional network, to keep the number of active lanes of three new characteristic patterns by convolution operation identical.On this basis,
Three new characteristic patterns can be subjected to resampling, to obtain the identical target signature of three resolution ratio.Finally, can be by three
A target signature is spliced (concat), to obtain the corresponding characteristic pattern of image to be detected.
As an example, can choose resolution ratio from the characteristic pattern that at least three feature extraction layers export meets preset condition
Three characteristic patterns, three characteristic patterns are merged, the corresponding characteristic pattern of image to be detected is obtained.For example, can be at low point
Resolution layer, intermediate-resolution layer, resolution layer choose a characteristic pattern respectively, to obtain three characteristic patterns.Wherein, feature mentions
Taking low-resolution layer, intermediate-resolution layer, the resolution layer of network can be preassigned by technical staff.
As an example, for the characteristic pattern of at least three feature extraction layers output, according to the sequence that resolution ratio is ascending,
Two characteristic patterns are chosen as fisrt feature figure and second feature figure；Choose the resolution ratio of resolution ratio and described image to be detected
Ratio is equal to the characteristic pattern of preset threshold as third feature figure.
In some optional implementations of the present embodiment, selected from the characteristic pattern that at least three feature extraction layers export
Resolution ratio is taken to meet three characteristic patterns of preset condition, comprising: for the characteristic pattern of at least three feature extraction layers output, to choose
Resolution bits in first resolution section characteristic pattern as fisrt feature figure；Resolution bits are chosen in second resolution section
Characteristic pattern is as second feature figure；Resolution bits are chosen in the characteristic pattern in third resolution ratio section as third feature figure.
With continued reference to one of application scenarios that Fig. 3, Fig. 3 are according to the method for the characteristic pattern of the extraction image of the present embodiment
Schematic diagram.In the application scenarios of Fig. 3, the executing subject for extracting the method for the characteristic pattern of image can be server.Should
With under scene, feature extraction network is residual error network ResNet, as shown in 302 in figure.In practice, many artificial neural networks exist
When design by the way of heap (block).In this way, typical image is inputted, can only considers five resolutions
The parameter of the block in rate stage (stage) and the quantity in channel.As shown, five resolution ratio stages are successively named as rank
Section 1- stage 5 (stage1-stage5).In this way, the mode of each block can be paid close attention to for network, simplify design and divided
Analysis process.On this basis, available image to be detected 301 of server.Later, 301 input feature vector of image to be detected is mentioned
Network 302 is taken, feature extraction network includes five feature extraction layers.What needs to be explained here is that in practice, each resolution ratio rank
Section may include multiple layers.Here, with reference to the mode of block, the specific structure in each resolution ratio stage is not analyzed,
And integrally it regard each resolution ratio stage as a layer.
On this basis, choose the stage 2, three characteristic patterns that stage 4 and stage 5 export, based on three characteristic patterns 303,
304,305, obtain the corresponding characteristic pattern 307 of image to be detected.Wherein, the more contextual information of the reservation of characteristic pattern 303.It is special
Sign Figure 30 4 remains more detailed information.Characteristic pattern 305 then remains more semantic information.The characteristic pattern obtained from
307 can merge the advantage of three, to improve the accuracy of extracted feature.
The method that embodiment of the disclosure provides, can obtain image to be detected first.Later, image to be detected is inputted
Feature extraction network, feature extraction network include at least three feature extraction layers.Finally, being exported from least three feature extraction layers
Characteristic pattern in choose three characteristic patterns, be based on three characteristic patterns, obtain the corresponding characteristic pattern of image to be detected.Wherein, each
The resolution ratio of the characteristic pattern of feature extraction layer output is not identical.It, can be by high, medium and low resolution by choosing suitable characteristic pattern
The characteristic information of rate sufficiently merges.To provide the foundation for processing such as subsequent attitude prediction or image segmentations.
With further reference to Fig. 4, it illustrates the processes 400 for extracting another embodiment of the method for characteristic pattern of image.It should
Extract the method 400 of the characteristic pattern of image, comprising the following steps:
Step 401, image to be detected is obtained.
Step 402, image to be detected input feature vector is extracted into network, feature extraction network includes at least three feature extractions
Layer.
Step 403, three characteristic patterns are chosen from the characteristic pattern that at least three feature extraction layers export, and are based on three features
Figure, obtains the corresponding characteristic pattern of image to be detected.
In the present embodiment, the specific implementation of step 401-403 and its brought technical effect can be corresponding with reference to Fig. 2
Embodiment in step 201-203, details are not described herein.
Step 404, the attitude prediction network that the characteristic pattern input of image to be detected is trained in advance, obtains image to be detected
The posture information of the human body of middle display.
In the present embodiment, attitude prediction network is used to characterize the posture of the human body shown in the characteristic pattern and image of image
The corresponding relationship of information.
In the present embodiment, as an example, feature extraction network in attitude prediction network and step 402 can by with
Under type training obtains:
The first step obtains training sample set.Wherein, training sample includes the people shown in sample image and sample image
The posture information of body.
Second step obtains initial predicted network.Here, it should be noted that initial predicted network may include initial spy
Sign extracts network and initial attitude predicts network.Wherein, the structure that initial characteristics extract network can be with reference to the spy in step 202
Sign extracts the structure of network.Initial attitude prediction network can be existing various Recurrent networks.For example, convolutional neural networks
CNN, Recognition with Recurrent Neural Network RNN etc..
Third step, for the training sample in training sample set, using the sample image of the training sample as initial pre-
The input of survey grid network, the posture information of the human body shown in the sample image of input is defeated as the expectation of initial predicted network
Out, using machine learning method, initial predicted network is trained.
Specifically, the posture information of output can be calculated first with preset loss function and training sample includes
The difference of posture information.It is then possible to adjust the network parameter of initial predicted network, and full based on resulting difference is calculated
In the case where the preset trained termination condition of foot, terminate training.The training termination condition here preset at can include but is not limited to
At least one of below: the training time is more than preset duration；Frequency of training is more than preset times；Resulting difference is calculated less than default
Discrepancy threshold.
The initial attitude prediction network that training obtains is determined as attitude prediction network by the 4th step.Training is obtained first
Beginning feature extraction network is determined as feature extraction network.
In some optional implementations of the present embodiment, this method further include: the characteristic pattern of image to be detected is defeated
Enter image segmentation network trained in advance, obtains the segmentation result letter at least one object shown in image to be detected
Breath, wherein the characteristic pattern and the segmentation knot at least one object shown in image that image segmentation network is used to characterize image
The corresponding relationship of fruit information.
, can be by training method similar to the above in these implementations, training obtains image segmentation network,
This is repeated no more.
Figure 4, it is seen that embodiment adds utilize image to be detected compared with the corresponding embodiment of Fig. 2
The step of carry out attitude prediction of characteristic pattern, so as to obtain the posture information of human body shown in image to be detected.Its
In, due to the characteristic pattern of image to be detected that three characteristic patterns of fusion obtain, keep feature corresponding to characteristic pattern more accurate, into
And the posture information of the human body made is more acurrate.Furthermore, it is possible to according to the actual situation, make some feature in three characteristic patterns
Figure retains more semantic information.To make the characteristic pattern of image to be detected that can also retain semantic information.Further increase appearance
State accuracy in detection.
With further reference to Fig. 5, as the realization to method shown in above-mentioned each figure, present disclose provides a kind of extraction images
The device of characteristic pattern, the device is corresponding with embodiment of the method shown in Fig. 2, which specifically can be applied to various electronics and set
In standby.
As shown in figure 5, the device 500 of the characteristic pattern of the image of the offer of the present embodiment includes: acquiring unit 501, feature
Extraction unit 502 and Fusion Features unit 503.Wherein, acquiring unit 501 is configured to obtain image to be detected.Feature extraction
Unit 502 is configured to extract image to be detected input feature vector network, and feature extraction network includes at least three feature extractions
Layer.Fusion Features unit 503 is configured to choose three characteristic patterns, base from the characteristic pattern that at least three feature extraction layers export
In three characteristic patterns, the corresponding characteristic pattern of image to be detected is obtained.
In the present embodiment, the acquiring unit 501 for including in the device 500 of the characteristic pattern of image, feature extraction list are extracted
Specific processing and brought technical effect of the member 502 with Fusion Features unit 503 can be with reference to the steps of the corresponding embodiment of Fig. 2
Rapid 201-203, details are not described herein.
In some optional implementations of the present embodiment, the device 500 further include: attitude prediction unit is (in figure not
Show), it is configured to the attitude prediction network that the characteristic pattern input of image to be detected is trained in advance, is obtained in image to be detected
The posture information of the human body of display, wherein attitude prediction network is used to characterize the human body shown in the characteristic pattern and image of image
Posture information corresponding relationship.
In some optional implementations of the present embodiment, the device 500 further include: cutting unit (does not show in figure
Out), it is configured to the image segmentation network that the characteristic pattern input of image to be detected is trained in advance, is obtained for image to be detected
The segmentation result information of at least one object of middle display, wherein image segmentation network be used to characterize the characteristic pattern of image with it is right
The corresponding relationship of the segmentation result information of at least one object shown in image.
In some optional implementations of the present embodiment, Fusion Features unit 503 is further configured to: from least
Choose resolution ratio in the characteristic pattern of three feature extraction layers output and meet three characteristic patterns of preset condition, by three characteristic patterns into
Row fusion, obtains the corresponding characteristic pattern of image to be detected.
In some optional implementations of the present embodiment, Fusion Features unit 503 is further configured to: for extremely
The characteristic pattern of few three feature extraction layers output chooses two characteristic patterns as first according to the sequence that resolution ratio is ascending
Characteristic pattern and second feature figure；The ratio for choosing the resolution ratio of resolution ratio and described image to be detected is equal to the feature of preset threshold
Figure is used as third feature figure.
In some optional implementations of the present embodiment, Fusion Features unit 503 is further configured to: for extremely
The characteristic pattern of few three feature extraction layers output, chooses resolution bits in the characteristic pattern in first resolution section as fisrt feature
Figure；Resolution bits are chosen in the characteristic pattern in second resolution section as second feature figure；Resolution bits are chosen to differentiate in third
The characteristic pattern in rate section is as third feature figure.
In the present embodiment, acquiring unit can obtain image to be detected first.Later, feature extraction unit will be to be detected
Image input feature vector extracts network, and feature extraction network includes at least three feature extraction layers.Finally, Fusion Features unit to
Three characteristic patterns are chosen in the characteristic pattern of few three feature extraction layers output, three characteristic patterns is based on, obtains image to be detected pair
The characteristic pattern answered.Wherein, the resolution ratio of the characteristic pattern of each feature extraction layer output is not identical.By choosing suitable feature
Figure, the characteristic information of high, medium and low resolution ratio can sufficiently be merged.To be the processing such as subsequent attitude prediction or image segmentation
It provides the foundation.
Below with reference to Fig. 6, it illustrates the electronic equipment that is suitable for being used to realize embodiment of the disclosure, (example is as shown in figure 1
Server) 600 structural schematic diagram.Electronic equipment shown in Fig. 6 is only an example, should not be to embodiment of the disclosure
Function and use scope bring any restrictions.
As shown in fig. 6, electronic equipment 600 includes processing unit (such as central processing unit, graphics processor etc.) 601,
Random access storage device can be loaded into according to the program being stored in read-only memory (ROM) 602 or from storage device 608
(RAM) program in 603 and execute various movements appropriate and processing.In RAM 603, it is also stored with the behaviour of electronic equipment 600
Various programs and data needed for making.Processing unit 601, ROM 602 and RAM 603 are connected with each other by bus 604.It is defeated
Enter/export (I/O) interface 605 and 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 middle processing unit 601, the reality of the disclosure is executed
Apply the above-mentioned function of limiting in the method for example.
It is situated between it should be noted that computer-readable medium described in embodiment of the disclosure can be computer-readable signal
Matter or computer readable storage medium either the two any combination.Computer readable storage medium for example can be with
System, device or the device of --- but being not limited to --- electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor, or it is any more than
Combination.The more specific example of computer readable storage medium can include but is not limited to: have one or more conducting wires
Electrical connection, portable computer diskette, hard disk, random access storage device (RAM), read-only memory (ROM), erasable type are programmable
Read-only memory (EPROM or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), light storage device, magnetic are deposited
Memory device or above-mentioned any appropriate combination.In embodiment of the disclosure, computer readable storage medium, which can be, appoints
What include or the tangible medium of storage program that the program can be commanded execution system, device or device use or and its
It is used in combination.And in embodiment of the disclosure, computer-readable signal media may include in a base band or as carrier wave
The data-signal that a part is propagated, wherein carrying computer-readable program code.The data-signal of this propagation can be adopted
With diversified forms, including but not limited to electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal is situated between
Matter can also be that any computer-readable medium other than computer readable storage medium, the computer-readable medium can be sent out
It send, propagate or transmits for by the use of instruction execution system, device or device or program in connection.It calculates
The program code for including on machine readable medium can transmit with any suitable medium, including but not limited to: electric wire, optical cable, RF
Etc. or above-mentioned any appropriate combination.
Above-mentioned computer-readable medium can be included in above-mentioned electronic equipment；It is also possible to individualism, and not
It is fitted into the electronic equipment.Above-mentioned computer-readable medium carries one or more program, when said one or more
When a program is executed by the electronic equipment, so that the electronic equipment: obtaining image to be detected；Image to be detected input feature vector is mentioned
Network is taken, feature extraction network includes at least three feature extraction layers；From the characteristic pattern that at least three feature extraction layers export
Three characteristic patterns are chosen, three characteristic patterns is based on, obtains the corresponding characteristic pattern of image to be detected.
The calculating of the operation for executing the disclosure can be write with one or more programming languages or combinations thereof
Machine program code, described program design language include object oriented program language-such as Java, Smalltalk, C+
+, it further include conventional procedural programming language-such as " C " language or similar programming language.Program code can
Fully to execute, partly execute on the user computer on the user computer, be executed as an independent software package,
Part executes on the remote computer or executes on a remote computer or server completely on the user computer for part.
In situations involving remote computers, remote computer can pass through the network of any kind --- including local area network (LAN)
Or wide area network (WAN)-is connected to subscriber computer, or, it may be connected to outer computer (such as utilize Internet service
Provider is connected by internet).
Flow chart and block diagram in attached drawing illustrate system, method and the computer of the various embodiments according to the disclosure
The architecture, function and operation in the cards of program product.In this regard, each box in flowchart or block diagram can be with
A part of a module, program segment or code is represented, a part of the module, program segment or code includes one or more
Executable instruction for implementing the specified logical function.It should also be noted that in some implementations as replacements, institute in box
The function of mark can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are practical
On 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 wants
It is noted that the combination of each box in block diagram and or flow chart and the box in block diagram and or flow chart, Ke Yiyong
The dedicated hardware based system of defined functions or operations is executed to realize, or can be referred to specialized hardware and computer
The combination of order is realized.
Being described in unit involved in embodiment of the disclosure can be realized by way of software, can also be passed through
The mode of hardware is realized.Described unit also can be set in the processor, for example, can be described as: a kind of processor
Including acquiring unit, feature extraction unit and Fusion Features unit.Wherein, the title of these units not structure under certain conditions
The restriction of the pairs of unit itself, for example, acquiring unit is also described as " obtaining the unit of image to be detected ".
Above description is only the preferred embodiment of the disclosure 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 disclosure, 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 in the disclosure
Can technical characteristic replaced mutually and the technical solution that is formed.
Claims (14)
1. a kind of method for the characteristic pattern for extracting image, comprising:
Obtain image to be detected；
Image to be detected input feature vector is extracted into network, the feature extraction network includes at least three feature extraction layers；
Three characteristic patterns are chosen from the characteristic pattern that at least three feature extraction layer exports, and are based on three characteristic patterns,
Obtain the corresponding characteristic pattern of described image to be detected.
2. according to the method described in claim 1, wherein, the method also includes:
By the characteristic pattern of described image to be detected input attitude prediction network trained in advance, obtain showing in image to be detected
The posture information of human body, wherein the attitude prediction network is used to characterize the human body shown in the characteristic pattern and image of image
The corresponding relationship of posture information.
3. according to the method described in claim 1, wherein, the method also includes:
The characteristic pattern input of described image to be detected image segmentation network trained in advance is obtained for showing in image to be detected
The segmentation result information of at least one object shown, wherein described image segmentation network be used to characterize the characteristic pattern of image with it is right
The corresponding relationship of the segmentation result information of at least one object shown in image.
4. method according to claim 1 to 3, wherein it is described from least three feature extraction layer export
Three characteristic patterns are chosen in characteristic pattern, three characteristic patterns is based on, obtains the corresponding characteristic pattern of described image to be detected, are wrapped
It includes:
Three characteristic patterns that resolution ratio meets preset condition are chosen from the characteristic pattern that at least three feature extraction layer exports,
Three characteristic patterns are merged, the corresponding characteristic pattern of described image to be detected is obtained.
5. described from the characteristic pattern that at least three feature extraction layer exports according to the method described in claim 4, wherein
Choose three characteristic patterns that resolution ratio meets preset condition, comprising:
Two are chosen according to the sequence that resolution ratio is ascending for the characteristic pattern of at least three feature extraction layer output
Characteristic pattern is as fisrt feature figure and second feature figure；The ratio for choosing resolution ratio and the resolution ratio of described image to be detected is equal to
The characteristic pattern of preset threshold is as third feature figure.
6. described from the characteristic pattern that at least three feature extraction layer exports according to the method described in claim 4, wherein
Choose three characteristic patterns that resolution ratio meets preset condition, comprising:
For the characteristic pattern of at least three feature extraction layer output, resolution bits are chosen in the feature in first resolution section
Figure is used as fisrt feature figure；Resolution bits are chosen in the characteristic pattern in second resolution section as second feature figure；It chooses and differentiates
Rate is located at the characteristic pattern in third resolution ratio section as third feature figure.
7. a kind of device for the characteristic pattern for extracting image, comprising:
Acquiring unit is configured to obtain image to be detected；
Feature extraction unit is configured to image to be detected input feature vector extracting network, the feature extraction network packet
Include at least three feature extraction layers；
Fusion Features unit is configured to choose three features from the characteristic pattern that at least three feature extraction layer exports
Figure is based on three characteristic patterns, obtains the corresponding characteristic pattern of described image to be detected.
8. device according to claim 7, wherein described device further include:
Attitude prediction unit is configured to the attitude prediction network that the characteristic pattern input of described image to be detected is trained in advance,
Obtain the posture information of human body shown in image to be detected, wherein the attitude prediction network is used to characterize the feature of image
The corresponding relationship of the posture information of the human body shown in figure and image.
9. device according to claim 7, wherein described device further include:
Cutting unit is configured to the image segmentation network that the characteristic pattern input of described image to be detected is trained in advance, obtains
Segmentation result information at least one object shown in image to be detected, wherein described image divides network and is used for table
Levy the characteristic pattern and the corresponding relationship of the segmentation result information at least one object shown in image of image.
10. according to the device any in claim 7-9, wherein the Fusion Features unit is further configured to:
Three characteristic patterns that resolution ratio meets preset condition are chosen from the characteristic pattern that at least three feature extraction layer exports,
Three characteristic patterns are merged, the corresponding characteristic pattern of described image to be detected is obtained.
11. device according to claim 10, wherein the Fusion Features unit is further configured to:
Two are chosen according to the sequence that resolution ratio is ascending for the characteristic pattern of at least three feature extraction layer output
Characteristic pattern is as fisrt feature figure and second feature figure；The ratio for choosing resolution ratio and the resolution ratio of described image to be detected is equal to
The characteristic pattern of preset threshold is as third feature figure.
12. device according to claim 11, wherein the Fusion Features unit is further configured to:
For the characteristic pattern of at least three feature extraction layer output, resolution bits are chosen in the feature in first resolution section
Figure is used as fisrt feature figure；Resolution bits are chosen in the characteristic pattern in second resolution section as second feature figure；It chooses and differentiates
Rate is located at the characteristic pattern in third resolution ratio section as third feature figure.
13. a kind of server, comprising:
One or more processors；
Storage device is stored thereon with 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
Now such as method as claimed in any one of claims 1 to 6.
14. a kind of computer-readable medium, is stored thereon with computer program, wherein real when described program is executed by processor
Now such as method as claimed in any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910098620.6A CN109840528A (en) | 2019-01-31 | 2019-01-31 | The method and apparatus for extracting the characteristic pattern of image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910098620.6A CN109840528A (en) | 2019-01-31 | 2019-01-31 | The method and apparatus for extracting the characteristic pattern of image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109840528A true CN109840528A (en) | 2019-06-04 |
Family
ID=66884486
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910098620.6A Pending CN109840528A (en) | 2019-01-31 | 2019-01-31 | The method and apparatus for extracting the characteristic pattern of image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109840528A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110222829A (en) * | 2019-06-12 | 2019-09-10 | 北京字节跳动网络技术有限公司 | Feature extracting method, device, equipment and medium based on convolutional neural networks |
| CN112990327A (en) * | 2021-03-25 | 2021-06-18 | 北京百度网讯科技有限公司 | Feature fusion method, device, apparatus, storage medium, and program product |
| CN115580721A (en) * | 2022-11-18 | 2023-01-06 | 南方电网数字电网研究院有限公司 | Quality detection method, device, equipment and medium for monitoring video of substation equipment |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170169313A1 (en) * | 2015-12-14 | 2017-06-15 | Samsung Electronics Co., Ltd. | Image processing apparatus and method based on deep learning and neural network learning |
| CN108062756A (en) * | 2018-01-29 | 2018-05-22 | 重庆理工大学 | Image, semantic dividing method based on the full convolutional network of depth and condition random field |
| CN108229523A (en) * | 2017-04-13 | 2018-06-29 | 深圳市商汤科技有限公司 | Image detection, neural network training method, device and electronic equipment |
| CN108710885A (en) * | 2018-03-29 | 2018-10-26 | 百度在线网络技术（北京）有限公司 | The detection method and device of target object |
| CN109003223A (en) * | 2018-07-13 | 2018-12-14 | 北京字节跳动网络技术有限公司 | Image processing method and device |
| CN109034198A (en) * | 2018-06-25 | 2018-12-18 | 中国科学院计算技术研究所 | The Scene Segmentation and system restored based on characteristic pattern |
| CN109284780A (en) * | 2018-09-10 | 2019-01-29 | 中山大学 | Ore mineral image automatic identification and classification method |
-
2019
- 2019-01-31 CN CN201910098620.6A patent/CN109840528A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170169313A1 (en) * | 2015-12-14 | 2017-06-15 | Samsung Electronics Co., Ltd. | Image processing apparatus and method based on deep learning and neural network learning |
| CN108229523A (en) * | 2017-04-13 | 2018-06-29 | 深圳市商汤科技有限公司 | Image detection, neural network training method, device and electronic equipment |
| CN108062756A (en) * | 2018-01-29 | 2018-05-22 | 重庆理工大学 | Image, semantic dividing method based on the full convolutional network of depth and condition random field |
| CN108710885A (en) * | 2018-03-29 | 2018-10-26 | 百度在线网络技术（北京）有限公司 | The detection method and device of target object |
| CN109034198A (en) * | 2018-06-25 | 2018-12-18 | 中国科学院计算技术研究所 | The Scene Segmentation and system restored based on characteristic pattern |
| CN109003223A (en) * | 2018-07-13 | 2018-12-14 | 北京字节跳动网络技术有限公司 | Image processing method and device |
| CN109284780A (en) * | 2018-09-10 | 2019-01-29 | 中山大学 | Ore mineral image automatic identification and classification method |
Non-Patent Citations (2)
| Title |
|---|
| 彭博: "《深度卷积网络原理与实践》", 31 May 2018, 机械工业出版社 * |
| 柳杨: "《数字图像物体识别理论详解与实战》", 31 January 2018 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110222829A (en) * | 2019-06-12 | 2019-09-10 | 北京字节跳动网络技术有限公司 | Feature extracting method, device, equipment and medium based on convolutional neural networks |
| CN112990327A (en) * | 2021-03-25 | 2021-06-18 | 北京百度网讯科技有限公司 | Feature fusion method, device, apparatus, storage medium, and program product |
| CN115580721A (en) * | 2022-11-18 | 2023-01-06 | 南方电网数字电网研究院有限公司 | Quality detection method, device, equipment and medium for monitoring video of substation equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109815964A (en) | The method and apparatus for extracting the characteristic pattern of image | |
| CN109902659B (en) | Method and apparatus for processing human body image | |
| CN108830235A (en) | Method and apparatus for generating information | |
| CN108985259A (en) | Human motion recognition method and device | |
| CN109002842A (en) | Image-recognizing method and device | |
| EP3889912B1 (en) | Method and apparatus for generating video | |
| CN110288547A (en) | Method and apparatus for generating image denoising model | |
| CN108989882A (en) | Method and apparatus for exporting the snatch of music in video | |
| CN109829432A (en) | Method and apparatus for generating information | |
| CN109740018A (en) | Method and apparatus for generating video tab model | |
| CN108345387A (en) | Method and apparatus for output information | |
| CN109308490A (en) | Method and apparatus for generating information | |
| CN109840528A (en) | The method and apparatus for extracting the characteristic pattern of image | |
| CN109981787B (en) | Method and device for displaying information | |
| CN109816037A (en) | The method and apparatus for extracting the characteristic pattern of image | |
| CN109815365A (en) | Method and apparatus for handling video | |
| CN109101919A (en) | Method and apparatus for generating information | |
| CN109993150A (en) | The method and apparatus at age for identification | |
| CN110009059A (en) | Method and apparatus for generating model | |
| CN109919244A (en) | Method and apparatus for generating scene Recognition model | |
| CN110213614A (en) | The method and apparatus of key frame are extracted from video file | |
| CN109389096A (en) | Detection method and device | |
| CN109977839A (en) | Information processing method and device | |
| CN108985228A (en) | Information generating method and device applied to terminal device | |
| CN109285181A (en) | The method and apparatus of image for identification |
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 |