CN108154169A - Image processing method and device - Google Patents
Image processing method and device Download PDFInfo
- Publication number
- CN108154169A CN108154169A CN201711311755.3A CN201711311755A CN108154169A CN 108154169 A CN108154169 A CN 108154169A CN 201711311755 A CN201711311755 A CN 201711311755A CN 108154169 A CN108154169 A CN 108154169A
- Authority
- CN
- China
- Prior art keywords
- model
- image
- classification results
- group
- position image
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Abstract
The disclosure is directed to image processing method and devices.This method includes obtaining one group of position image, and one group of position image includes n position images, a predetermined patterns of every position image shows object to be identified;The n is greater than 1 integer;According to one group of position image and default network, n classification results are determined;The default network includes feature extraction layer and identification layer;Last layer includes the energy function of attention mechanism in the identification layer, and the energy function is used for from a upper classification results, the corresponding output result of prediction current location image;The model of the object to be identified is determined according to the n classification results;Wherein, the object to be identified is one kind of the trained object.In the technical solution, the default neural network based on attention mechanism can be primarily focused on the crucial feature on the image of position, so as to ignore the same characteristic features with different model, can more accurately obtain the model of object to be identified.
Description
Technical field
This disclosure relates to image processing field more particularly to image processing method and device.
Background technology
At present, image identification is a key areas of artificial intelligence.If the result that image identifies is inaccurate, then
The information that processor processing inaccurate result obtains is problematic certainly, and therefore, the accuracy of image identification is to a certain degree
Limit the development of artificial intelligence.
Invention content
The embodiment of the present disclosure provides image processing method and device.The technical solution is as follows:
According to the embodiment of the present disclosure in a first aspect, provide a kind of image processing method, including:
One group of position image is obtained, one group of position image includes n position images, every position image shows
One predetermined patterns of object to be identified;The n is greater than 1 integer;
According to one group of position image and default network, n classification results are determined;Wherein, each classification results include
The model probability of one corresponding at least one model of position image and each model;The default network is by training object
The neural network that the model of the image at each position and the trained object trains;The default network include feature extraction layer and
Identification layer;Last layer includes the energy function of attention mechanism in the identification layer, and the energy function is used for from upper one
In classification results, the corresponding output result of prediction current location image;
The model of the object to be identified is determined according to the n classification results;
Wherein, the object to be identified is one kind of the trained object.
The technical scheme provided by this disclosed embodiment can include the following benefits:It is default based on attention mechanism
Neural network can be primarily focused on the crucial feature on the image of position, so as to ignore spy identical with different model
Sign can more accurately obtain the model of object to be identified.
In one embodiment, for i-th position image in one group of position image, the i is 1 whole to n
Number;It is described according to one group of position image and default network, determine that n classification results include:
The characteristic information of i-th position image is extracted by the feature extraction layer;
Classified by the identification layer for the characteristic information, it is i-th corresponding to obtain i-th position image
Classification results;
Wherein, in the identification layer in last layer, by (i-1)-th classification results, the identification layer it is second from the bottom
The output result of layer inputs the energy function, obtains i-th of classification results;I-th of classification results include described
The model probability of at least one model and each model corresponding to i-th of position image;0th classification results are default
's.
In one embodiment, it is described to determine that the model of the object to be identified includes according to the n classification results:
In the last one classification results, using the model of model maximum probability as the model of the object to be identified.
In one embodiment, it is described to determine that the model of the object to be identified includes according to the n classification results:
By the model probability in the n classification results according to type classification;
According to the default weighted value of the n position images, calculate the weighting of model probability for being classified as same model and put down
Mean value;The default weighted value is the weighted value of the model probability of each corresponding model of same position image;
In the weighted average of the model probability of each model, the model of weighted average maximum is waited to know as described in
The model of other object.
In one embodiment, the method further includes:
Obtain the p group images of the trained object;All include showing the trained object respectively for each group in the p groups image
The n of n predetermined patterns images;The p is positive integer;
Obtain the model of the trained object;
According to the model of the p groups image and the trained object, the default network is trained.
In one embodiment, one group of position image of the acquisition includes:
Obtain the displaying image for showing the object to be identified;
From the displaying image, n independent images are cut out;Each of the n independent images shows one
Different predetermined patterns;
By the n independent Image Adjustings to specified pixel and specified size;
Using the n after adjustment independent images as n position images of one group of position image.
According to the second aspect of the embodiment of the present disclosure, a kind of image processing apparatus is provided, including:
First acquisition module, for obtaining one group of position image, one group of position image includes n position images, institute
State a predetermined patterns of every position image shows object to be identified;The n is greater than 1 integer;
First determining module, for according to one group of position image and default network, determining n classification results;Wherein,
Each classification results include a corresponding at least one model of position image and the model probability of each model;The default net
Network is the neural network trained by the image at each position and the model of the trained object of training object;The default network
Including feature extraction layer and identification layer;Last layer includes the energy function of attention mechanism, the energy in the identification layer
Function is used for from a upper classification results, the corresponding output result of prediction current location image;
Second determining module, for determining the model of the object to be identified according to the n classification results;
Wherein, the object to be identified is one kind of the trained object.
In one embodiment, for i-th position image in one group of position image, the i is 1 whole to n
Number;First determining module includes:
First extracting sub-module, for passing through the characteristic information that the feature extraction layer extracts i-th position image;
First classification submodule, classifies for the characteristic information for passing through the identification layer, obtains described i-th
Corresponding i-th of the classification results of position image;
Wherein, in the identification layer in last layer, by (i-1)-th classification results, the identification layer it is second from the bottom
The output result of layer inputs the energy function, obtains i-th of classification results;I-th of classification results include described
The model probability of at least one model and each model corresponding to i-th of position image;0th classification results are default
's.
In one embodiment, second determining module includes:
First processing submodule, in the last one classification results, using the model of model maximum probability as described in
The model of object to be identified.
In one embodiment, second determining module includes:
Second classification submodule, for by the model probability in the n classification results according to type classification;
Computational submodule for the default weighted value according to the n position images, calculates the type for being classified as same model
The weighted average of number probability;The default weighted value is the weighting of the model probability of each corresponding model of same position image
Value;
Second processing submodule, in the weighted average of the model probability of each model, by weighted average most
Model of the big model as the object to be identified.
In one embodiment, described device further includes:
Second acquisition module, for obtaining the p group images of the trained object;All include dividing for each group in the p groups image
N images of n predetermined patterns of the trained object are not shown;The p is positive integer;
Third acquisition module, for obtaining the model of the trained object;
Training module for the model according to the p groups image and the trained object, trains the default network.
In one embodiment, first acquisition module includes:
Acquisition submodule, for obtaining the displaying image for showing the object to be identified;
Submodule is cut, for from the displaying image, cutting out n independent images;The n opens the every of independent image
One all shows a different predetermined patterns;
Submodule is adjusted, for the n to be opened independent Image Adjustings to specified pixel and specified size;
Third handles submodule, opens positions as the n of one group of position image for n independent images after adjusting
Image.
According to the third aspect of the embodiment of the present disclosure, a kind of image processing apparatus is provided, including:
Processor;
For storing the memory of processor-executable instruction;
Wherein, the processor is configured as:
One group of position image is obtained, one group of position image includes n position images, every position image exhibition
Show a predetermined patterns of object to be identified;The n is greater than 1 integer;
According to one group of position image and default network, n classification results are determined;Wherein, each classification results include
The model probability of one corresponding at least one model of position image and each model;The default network is by training object
The neural network that the model of the image at each position and the trained object trains;The default network include feature extraction layer and
Identification layer;Last layer includes the energy function of attention mechanism in the identification layer, and the energy function is used for from upper one
In classification results, the corresponding output result of prediction current location image;
The model of the object to be identified is determined according to the n classification results;
Wherein, the object to be identified is one kind of the trained object.
It should be understood that above general description and following detailed description are only exemplary and explanatory, not
The disclosure can be limited.
Description of the drawings
Attached drawing herein is incorporated into specification and forms the part of this specification, shows the implementation for meeting the disclosure
Example, and for explaining the principle of the disclosure together with specification.
Fig. 1 is the flow chart according to the image processing method shown in an exemplary embodiment.
Fig. 2 is the schematic diagram according to the default network shown in an exemplary embodiment.
Fig. 3 is the flow chart according to the image processing method shown in an exemplary embodiment.
Fig. 4 is the flow chart according to the image processing method shown in an exemplary embodiment.
Fig. 5 is the flow chart according to the image processing method shown in an exemplary embodiment.
Fig. 6 is the flow chart according to the image processing method shown in an exemplary embodiment.
Fig. 7 is the flow chart according to the image processing method shown in an exemplary embodiment.
Fig. 8 is according to the displaying image shown in an exemplary embodiment.
The independent image of vehicle headlights of the Fig. 9 according to an exemplary embodiment.
Figure 10 is the image of component according to the vehicle headlight shown in an exemplary implementation.
Figure 11 is the block diagram according to the image processing apparatus shown in an exemplary embodiment.
Figure 12 is the block diagram according to the image processing apparatus shown in an exemplary embodiment.
Figure 13 is the block diagram according to the image processing apparatus shown in an exemplary embodiment.
Figure 14 is the block diagram according to the image processing apparatus shown in an exemplary embodiment.
Figure 15 is the block diagram according to the image processing apparatus shown in an exemplary embodiment.
Figure 16 is the block diagram according to the image processing apparatus shown in an exemplary embodiment.
Figure 17 is the block diagram according to the image processing apparatus shown in an exemplary embodiment.
Specific embodiment
Here exemplary embodiment will be illustrated in detail, example is illustrated in the accompanying drawings.Following description is related to
During attached drawing, unless otherwise indicated, the same numbers in different attached drawings represent the same or similar element.Following exemplary embodiment
Described in embodiment do not represent all embodiments consistent with the disclosure.On the contrary, they be only with it is such as appended
The example of the consistent device and method of some aspects be described in detail in claims, the disclosure.
Fig. 1 is according to a kind of flow chart of image processing method shown in an exemplary embodiment, as shown in Figure 1, image
Processing method is used in image processing apparatus, which is applied in server, and this method includes the following steps 101-103:
In a step 101, one group of position image is obtained.
Here, one group of position image includes n position images.One default portion of every position image shows object to be identified
Position;N is greater than 1 integer.One group of position image needs to input default neural network, and therefore, image needs in position are sequentially input
Default neural network, therefore, the input sequence of position image can be the default of the predetermined patterns shown according to position image
What sequence was ranked up.
The object to be identified of the present embodiment can be vehicle to be identified, and corresponding predetermined patterns are the default automobile body parts of vehicle to be identified
Position, it is assumed that n is that 3,3 predetermined patterns are tail-light, vehicle headlight, logo, correspondingly, one group of position image can include displaying
The first position image of the tail-light of vehicle to be identified, the second position image of the vehicle headlight of displaying vehicle to be identified, displaying are to be identified
The third position image of the logo of vehicle.Assuming that the sequence of predetermined patterns is tail-light, logo, vehicle headlight, correspondingly, position image
Sequence be exactly first position figure, third station diagram, second position figure.Here, which can prompt user to need according to pre-
If the preset order at position is ranked up position image.
In a step 102, according to one group of position image and default network, n classification results are determined.
Wherein, each classification results include a corresponding at least one model of position image and the model of each model is general
Rate;Default network is the neural network trained by the image at each position of training object and the model of training object;Default net
Network includes feature extraction layer and identification layer;Last in identification layer layer includes the energy function of attention mechanism, and energy function is used
In from a upper classification results, the corresponding output result of prediction current location image.
The default neural network of the present embodiment can be designed based on Recognition with Recurrent Neural Network, only according to energy function
Last layer of identification layer is changed, other layers of structure does not change, by taking VGG16 as an example, as shown in Fig. 2, the network can wrap
Including has the third layer of energy function in feature extraction layer, first layer identification layer, second layer identification layer and the present embodiment of VGG16
Identification layer;Wherein, the input of third identification layer will also include last classification results.
In the present embodiment, by attention mechanism classify model when, each classification results all can be according to preceding state
Learn obtained classification results and the position image currently inputted, go the feature rather than position image of processing attention part
All features.Such benefit is exactly that less pixel needs to handle, and reduces the complexity of task.
In the present embodiment, energy function rt=tan (σ (I)+γ (rt-1)), wherein, σ and γ are attention mechanism
Practise parameter.
In step 103, the model of object to be identified is determined according to n classification results.
Wherein, object to be identified is one kind of trained object.
In one embodiment, as shown in figure 3, for i-th position image in one group of position image, i is 1 to n
Integer;Step 102 in Fig. 1 that is, according to one group of position image and default network, determines that n classification results include, can include:
In step 1021, the characteristic information of i-th position image is extracted by feature extraction layer.
In step 1022, information is characterized by identification layer and is classified, obtain i-th position image corresponding i-th
A classification results.
Wherein, in identification layer in last layer, by the output of (i-1)-th classification results, the layer second from the bottom of identification layer
As a result input energy function obtains i-th of classification results;I-th of classification results is included corresponding to i-th of position image at least
A kind of model probability of model and each model;0th classification results are preset.
What deserves to be explained is due to the 1st classification results be it is no it is previous output as a result, therefore, one can be preset
A 0th classification results, such as the 0th classification results include all models of training object, and the model probability phase of each model
Together, it is 1 that model probability, which adds up,.
In one embodiment, as shown in figure 4, step 103 in Fig. 1, i.e., determine object to be identified according to n classification results
Model can include:
In step 1031, in the last one classification results, using the model of model maximum probability as object to be identified
Model.
In the present embodiment, the last one classification results i.e. n-th of classification results.Here, one group of position image of input can
With without sequence, stochastic inputs.
In one embodiment, as shown in figure 5, step 103 in Fig. 1, i.e., determine object to be identified according to n classification results
Model can include:
In step 1032, by the model probability in n classification results according to type classification.
Current embodiment require that one group of image for input fixes input sequence, otherwise, point of the output of neural network is preset
Class result can not just determine the corresponding weighted value of model.
In step 1033, according to the default weighted value of n position images, the model probability for being classified as same model is calculated
Weighted average.
Default weighted value is the weighted value of the model probability of each corresponding model of same position image.
In step 1034, in the weighted average of the model probability of each model, by the type of weighted average maximum
Model number as object to be identified.
Assuming that predetermined patterns image has 3, the output result of first predetermined patterns image includes type-A and its model is general
Rate a1, B type and its model probability b1 and C type and its model probability c1;The output result packet of second predetermined patterns image
Include B types and its model probability b2, C type and its model probability c2 and D type and its model probability d2;Third predetermined patterns
The output result of image includes type-A and its model probability a3, C type and its model probability c3, E type and its model probability
E3, F type and its model probability f3.The default weighted value of first position image is w1, second predetermined patterns image it is pre-
If weighted value is w2, the default weighted value of third predetermined patterns is w3.
The model probability of A models is a1, a3;Model probability b1, b2 of Type B number;The model probability of c-type number be c1 and c2 and
c3;The model probability of D models is d2 and d3;The model probability of E models is e3;The model probability of F models is f3.
The weighted average of the model probability of A models is (a1*w1+a3*w3)/3;
The weighted average of the model probability of Type B number is (b1*w1+b2*w2)/3;
The weighted average of the model probability of c-type number is (c1*w1+c2*w2+c3*w3)/3;
The weighted average of the model probability of D models is (d2*w2+d3*w3)/3;
The weighted average of the model probability of E models is e3*w3/3;
The weighted average of the model probability of F models is f3*w3/3.
In one embodiment, as shown in fig. 6, method further includes:
At step 104, the p group images of training object are obtained.
N images of each group of n predetermined patterns for all including showing training object respectively in p group images.P is just whole
Number.
The P group pictures of the present embodiment seem training sample, for the parameter of the default neural network of training.
In step 105, the model of training object is obtained.
In step 106, according to the model of p groups image and training object, default network is trained.
In one embodiment, as shown in fig. 7, step 101 in Fig. 1, that is, obtain one group of position image, can include:
In step 1011, the displaying image for showing object to be identified is obtained.
In step 1012, from displaying image, n independent images are cut out.
Here, each of n independent images shows a different predetermined patterns.
In step 1013, by n independent Image Adjustings to specified pixel and specified size.
Specified pixel and specified size are all that default neural network requires in the present embodiment.
In step 1014, using the n after adjustment independent images as n position images of one group of position image.
Here, it seems to be ranked up according to the preset order of the predetermined patterns of its display to open station diagram.
It is exemplary, as shown in figure 8, Fig. 5 is a displaying figure image, which show a vehicle to be identified, in order to be treated to this
Identification vehicle is classified, and needs to cut the independent image of the predetermined patterns of under body, Fig. 9 is independent image, the figure for showing car light
10 be the independent image of exhibition vehicle target, and by independent Image Adjusting to specified size and specified pixel, the independent image after adjustment is
For position image.
Following is embodiment of the present disclosure, can be used for performing embodiments of the present disclosure.
Figure 11 is according to a kind of block diagram of image processing apparatus shown in an exemplary embodiment, which can be by soft
Part, hardware or both are implemented in combination with as some or all of of electronic equipment.As shown in figure 11, the image processing apparatus
Including:
First acquisition module 201, for obtaining one group of position image, one group of position image includes n position images,
One predetermined patterns of every position image shows object to be identified;The n is greater than 1 integer;
First determining module 202, for according to one group of position image and default network, determining n classification results;Its
In, each classification results include a corresponding at least one model of position image and the model probability of each model;It is described pre-
If network is the neural network trained by the image at each position and the model of the trained object of training object;It is described default
Network includes feature extraction layer and identification layer;Last layer includes the energy function of attention mechanism in the identification layer, described
Energy function is used for from a upper classification results, the corresponding output result of prediction current location image;
Second determining module 203, for determining the model of the object to be identified according to the n classification results;
Wherein, the object to be identified is one kind of the trained object.
In one embodiment, as shown in figure 12, for i-th group of position image, the i is 1 to the integer between n;Institute
The first determining module 202 is stated to include:
First extracting sub-module 2021, for extracting the feature of i-th position image by the feature extraction layer
Information;
First classification submodule 2022, classifies for passing through the identification layer for the characteristic information, obtains described
Corresponding i-th of the classification results of i-th position image;
Wherein, in the identification layer in last layer, by (i-1)-th classification results, the identification layer it is second from the bottom
The output result of layer inputs the energy function, obtains i-th of classification results;I-th of classification results include described
The model probability of at least one model and each model corresponding to i-th of position image;0th classification results are default
's.
In one embodiment, as shown in figure 13, second determining module 203 includes:
First processing submodule 2031, in the last one classification results, using the model of model maximum probability as
The model of the object to be identified.
In one embodiment, as shown in figure 14, second determining module 203 includes:
Second classification submodule 2032, for by the model probability in the n classification results according to type classification;
Computational submodule 2033, for the default weighted value according to the n position images, calculating is classified as same model
Model probability weighted average;The default weighted value is the model probability of each corresponding model of same position image
Weighted value;
Second processing submodule 2034, in the weighted average of the model probability of each model, by weighted average
It is worth model of the maximum model as the object to be identified.
In one embodiment, as shown in figure 15, described device further includes:
Second acquisition module 204, for obtaining the p group images of the trained object;All include for each group in the p groups image
N images of n predetermined patterns of the trained object are shown respectively;The p is positive integer;
Third acquisition module 205, for obtaining the model of the trained object;
Training module 206 for the model according to the p groups image and the trained object, trains the default network.
In one embodiment, first acquisition module 201 includes as shown in figure 16:
Acquisition submodule 2011, for obtaining the displaying image for showing the object to be identified;
Submodule 2012 is cut, for from the displaying image, cutting out n independent images;The n independent images
Each all show a different predetermined patterns;
Submodule 2013 is adjusted, for the n to be opened independent Image Adjustings to specified pixel and specified size;
Third handles submodule 2014, is opened for n independent images after adjusting as the n of one group of position image
Position image.
According to the third aspect of the embodiment of the present disclosure, a kind of image processing apparatus is provided, including:
Processor;
For storing the memory of processor-executable instruction;
Wherein, processor is configured as:
One group of position image is obtained, one group of position image includes n position images, every position image shows
One predetermined patterns of object to be identified;The n is greater than 1 integer;
According to one group of position image and default network, n classification results are determined;Wherein, each classification results include
The model probability of one corresponding at least one model of position image and each model;The default network is by training object
The neural network that the model of the image at each position and the trained object trains;The default network include feature extraction layer and
Identification layer;Last layer includes the energy function of attention mechanism in the identification layer, and the energy function is used for from upper one
In classification results, the corresponding output result of prediction current location image;
The model of the object to be identified is determined according to the n classification results;
Wherein, the object to be identified is one kind of the trained object.
Above-mentioned processor is also configured to:
For i-th position image in one group of position image, the i is 1 integer for arriving n;Described in the basis
One group of position image and default network determine that n classification results include:
The characteristic information of i-th position image is extracted by the feature extraction layer;
Classified by the identification layer for the characteristic information, it is i-th corresponding to obtain i-th position image
Classification results;
Wherein, in the identification layer in last layer, by (i-1)-th classification results, the identification layer it is second from the bottom
The output result of layer inputs the energy function, obtains i-th of classification results;I-th of classification results include described
The model probability of at least one model and each model corresponding to i-th of position image;0th classification results are default
's.
It is described to determine that the model of the object to be identified includes according to the n classification results:
In the last one classification results, using the model of model maximum probability as the model of the object to be identified.
It is described to determine that the model of the object to be identified includes according to the n classification results:
By the model probability in the n classification results according to type classification;
According to the default weighted value of the n position images, calculate the weighting of model probability for being classified as same model and put down
Mean value;The default weighted value is the weighted value of the model probability of each corresponding model of same position image;
In the weighted average of the model probability of each model, the model of weighted average maximum is waited to know as described in
The model of other object.
The method further includes:
Obtain the p group images of the trained object;All include showing the trained object respectively for each group in the p groups image
The n of n predetermined patterns images;The p is positive integer;
Obtain the model of the trained object;
According to the model of the p groups image and the trained object, the default network is trained.
One group of position image of the acquisition includes:
Obtain the displaying image for showing the object to be identified;
From the displaying image, n independent images are cut out;Each of the n independent images shows one
Different predetermined patterns;
By the n independent Image Adjustings to specified pixel and specified size;
Using the n after adjustment independent images as n position images of one group of position image.About above-described embodiment
In device, wherein modules perform operation concrete mode carried out retouching in detail in the embodiment in relation to this method
It states, explanation will be not set forth in detail herein.
Figure 17 is according to a kind of block diagram for image processing apparatus shown in an exemplary embodiment.For example, device
1900 may be provided as a server.Device 1900 includes processing component 1922, further comprises one or more processing
Device and memory resource represented by a memory 1932, can be by the instruction of the execution of processing component 1922, example for storing
Such as application program.The application program stored in memory 1932 can include it is one or more each correspond to one group
The module of instruction.In addition, processing component 1922 is configured as execute instruction, to perform the above method.
Device 1900 can also include a power supply module 1926 and be configured as the power management of executive device 1900, one
Wired or wireless network interface 1950 is configured as device 1900 being connected to network and input and output (I/O) interface
1958.Device 1900 can be operated based on the operating system for being stored in memory 1932, such as Windows ServerTM, Mac
OS XTM, UnixTM, LinuxTM, FreeBSDTM or similar.
A kind of non-transitorycomputer readable storage medium, when the instruction in the storage medium is by the processing of device 1900
When device performs so that device 1900 is able to carry out above-mentioned image processing method, the method includes:
One group of position image is obtained, one group of position image includes n position images, every position image shows
One predetermined patterns of object to be identified;The n is greater than 1 integer;
According to one group of position image and default network, n classification results are determined;Wherein, each classification results include
The model probability of one corresponding at least one model of position image and each model;The default network is by training object
The neural network that the model of the image at each position and the trained object trains;The default network include feature extraction layer and
Identification layer;Last layer includes the energy function of attention mechanism in the identification layer, and the energy function is used for from upper one
In classification results, the corresponding output result of prediction current location image;
The model of the object to be identified is determined according to the n classification results;
Wherein, the object to be identified is one kind of the trained object.
For i-th position image in one group of position image, the i is 1 integer for arriving n;Described in the basis
One group of position image and default network determine that n classification results include:
The characteristic information of i-th position image is extracted by the feature extraction layer;
Classified by the identification layer for the characteristic information, it is i-th corresponding to obtain i-th position image
Classification results;
Wherein, in the identification layer in last layer, by (i-1)-th classification results, the identification layer it is second from the bottom
The output result of layer inputs the energy function, obtains i-th of classification results;I-th of classification results include described
The model probability of at least one model and each model corresponding to i-th of position image;0th classification results are default
's.
It is described to determine that the model of the object to be identified includes according to the n classification results:
In the last one classification results, using the model of model maximum probability as the model of the object to be identified.
It is described to determine that the model of the object to be identified includes according to the n classification results:
By the model probability in the n classification results according to type classification;
According to the default weighted value of the n position images, calculate the weighting of model probability for being classified as same model and put down
Mean value;The default weighted value is the weighted value of the model probability of each corresponding model of same position image;
In the weighted average of the model probability of each model, the model of weighted average maximum is waited to know as described in
The model of other object.
The method further includes:
Obtain the p group images of the trained object;All include showing the trained object respectively for each group in the p groups image
The n of n predetermined patterns images;The p is positive integer;
Obtain the model of the trained object;
According to the model of the p groups image and the trained object, the default network is trained.
One group of position image of the acquisition includes:
Obtain the displaying image for showing the object to be identified;
From the displaying image, n independent images are cut out;Each of the n independent images shows one
Different predetermined patterns;
By the n independent Image Adjustings to specified pixel and specified size;
Using the n after adjustment independent images as n position images of one group of position image.
Those skilled in the art will readily occur to the disclosure its after considering specification and putting into practice disclosure disclosed herein
Its embodiment.This application is intended to cover any variations, uses, or adaptations of the disclosure, these modifications, purposes or
Person's adaptive change follows the general principle of the disclosure and including the undocumented common knowledge in the art of the disclosure
Or conventional techniques.Description and embodiments are considered only as illustratively, and the true scope and spirit of the disclosure are by following
Claim is pointed out.
It should be understood that the present disclosure is not limited to the precise structures that have been described above and shown in the drawings, and
And various modifications and changes may be made without departing from the scope thereof.The scope of the present disclosure is only limited by appended claim.
Claims (14)
1. a kind of image processing method, which is characterized in that including:
One group of position image is obtained, one group of position image includes n position images, and every position image shows are waited to know
One predetermined patterns of other object;The n is greater than 1 integer;
According to one group of position image and default network, n classification results are determined;Wherein, each classification results include one
The model probability of the corresponding at least one model of position image and each model;The default network is by each of training object
The neural network that the model of the image at position and the trained object trains;The default network includes feature extraction layer and identification
Layer;Last layer includes the energy function of attention mechanism in the identification layer, and the energy function is used for from upper one classification
As a result in, the corresponding output result of prediction current location image;
The model of the object to be identified is determined according to the n classification results;
Wherein, the object to be identified is one kind of the trained object.
2. according to the method described in claim 1, it is characterized in that, for i-th station diagram in one group of position image
Picture, the i are 1 integers for arriving n;It is described according to one group of position image and default network, determine that n classification results include:
The characteristic information of i-th position image is extracted by the feature extraction layer;
Classified by the identification layer for the characteristic information, obtain corresponding i-th of the classification of i-th position image
As a result;
Wherein, in the identification layer in last layer, by (i-1)-th classification results, the layer second from the bottom of the identification layer
It exports result and inputs the energy function, obtain i-th of classification results;I-th of classification results include described i-th
The model probability of at least one model and each model corresponding to the image of position;0th classification results are preset.
3. according to the method described in claim 1, it is characterized in that, described wait to know according to determining the n classification results
The model of other object includes:
In the last one classification results, using the model of model maximum probability as the model of the object to be identified.
4. according to the method described in claim 1, it is characterized in that, described wait to know according to determining the n classification results
The model of other object includes:
By the model probability in the n classification results according to type classification;
According to the default weighted value of the n position images, the weighted average for the model probability for being classified as same model is calculated;
The default weighted value is the weighted value of the model probability of each corresponding model of same position image;
In the weighted average of the model probability of each model, using the model of weighted average maximum as the object to be identified
Model.
5. according to the method described in claim 1, it is characterized in that, the method further includes:
Obtain the p group images of the trained object;Each group of all include showing the trained object respectively n in the p groups image
The n of predetermined patterns images;The p is positive integer;
Obtain the model of the trained object;
According to the model of the p groups image and the trained object, the default network is trained.
6. according to the method described in claim 2, it is characterized in that, one group of position image of the acquisition includes:
Obtain the displaying image for showing the object to be identified;
From the displaying image, n independent images are cut out;Each of the n independent images shows a difference
Predetermined patterns;
By the n independent Image Adjustings to specified pixel and specified size;
Using the n after adjustment independent images as n position images of one group of position image.
7. a kind of image processing apparatus, which is characterized in that including:
First acquisition module, for obtaining one group of position image, one group of position image includes n position images, described every
Open a predetermined patterns of position image shows object to be identified;The n is greater than 1 integer;
First determining module, for according to one group of position image and default network, determining n classification results;Wherein, each
Classification results include a corresponding at least one model of position image and the model probability of each model;The default network is
The neural network trained by the image at each position and the model of the trained object of training object;The default network includes
Feature extraction layer and identification layer;Last layer includes the energy function of attention mechanism, the energy function in the identification layer
For from a upper classification results, predicting the corresponding output result of current location image;
Second determining module, for determining the model of the object to be identified according to the n classification results;
Wherein, the object to be identified is one kind of the trained object.
8. device according to claim 7, which is characterized in that for i-th station diagram in one group of position image
Picture, the i are 1 integers for arriving n;First determining module includes:
First extracting sub-module, for passing through the characteristic information that the feature extraction layer extracts i-th position image;
First classification submodule, classifies for the characteristic information for passing through the identification layer, obtains i-th position
Corresponding i-th of the classification results of image;
Wherein, in the identification layer in last layer, by (i-1)-th classification results, the layer second from the bottom of the identification layer
It exports result and inputs the energy function, obtain i-th of classification results;I-th of classification results include described i-th
The model probability of at least one model and each model corresponding to the image of position;0th classification results are preset.
9. device according to claim 7, which is characterized in that second determining module includes:
First processing submodule, in the last one classification results, the model of model maximum probability being waited to know as described in
The model of other object.
10. device according to claim 7, which is characterized in that second determining module includes:
Second classification submodule, for by the model probability in the n classification results according to type classification;
Computational submodule, for the default weighted value according to the n position images, the model that calculating is classified as same model is general
The weighted average of rate;The default weighted value is the weighted value of the model probability of each corresponding model of same position image;
Second processing submodule, in the weighted average of the model probability of each model, weighted average is maximum
Model of the model as the object to be identified.
11. device according to claim 7, which is characterized in that described device further includes:
Second acquisition module, for obtaining the p group images of the trained object;All include opening up respectively for each group in the p groups image
Show n images of n predetermined patterns of the trained object;The p is positive integer;
Third acquisition module, for obtaining the model of the trained object;
Training module for the model according to the p groups image and the trained object, trains the default network.
12. device according to claim 8, which is characterized in that first acquisition module includes:
Acquisition submodule, for obtaining the displaying image for showing the object to be identified;
Submodule is cut, for from the displaying image, cutting out n independent images;Each of the n independent images
All show a different predetermined patterns;
Submodule is adjusted, for the n to be opened independent Image Adjustings to specified pixel and specified size;
Third handles submodule, opens station diagrams as the n of one group of position image for n independent images after adjusting
Picture.
13. a kind of image processing apparatus, which is characterized in that including:
Processor;
For storing the memory of processor-executable instruction;
Wherein, the processor is configured as:
One group of position image is obtained, one group of position image includes n position images, and every position image shows are waited to know
One predetermined patterns of other object;The n is greater than 1 integer;
According to one group of position image and default network, n classification results are determined;Wherein, each classification results include one
The model probability of the corresponding at least one model of position image and each model;The default network is by each of training object
The neural network that the model of the image at position and the trained object trains;The default network includes feature extraction layer and identification
Layer;Last layer includes the energy function of attention mechanism in the identification layer, and the energy function is used for from upper one classification
As a result in, the corresponding output result of prediction current location image;
The model of the object to be identified is determined according to the n classification results;
Wherein, the object to be identified is one kind of the trained object.
14. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is by processor
The step of any one of claim 1-6 the methods are realized during execution.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711311755.3A CN108154169A (en) | 2017-12-11 | 2017-12-11 | Image processing method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711311755.3A CN108154169A (en) | 2017-12-11 | 2017-12-11 | Image processing method and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108154169A true CN108154169A (en) | 2018-06-12 |
Family
ID=62466910
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711311755.3A Pending CN108154169A (en) | 2017-12-11 | 2017-12-11 | Image processing method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108154169A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109583318A (en) * | 2018-11-06 | 2019-04-05 | 平安科技(深圳)有限公司 | Medicinal plant recognition methods, device and computer equipment |
CN110136185A (en) * | 2019-05-23 | 2019-08-16 | 中国科学技术大学 | A kind of monocular depth estimation method and system |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102312616A (en) * | 2010-06-29 | 2012-01-11 | 马瑞利汽车电子(广州)有限公司 | The method and system that is used for intelligence entering and intelligent starting |
CN102968643A (en) * | 2012-11-16 | 2013-03-13 | 华中科技大学 | Multi-mode emotion recognition method based on Lie group theory |
CN104063677A (en) * | 2013-03-19 | 2014-09-24 | 北京三星通信技术研究有限公司 | Equipment used for estimating human body posture and method thereof |
CN105574820A (en) * | 2015-12-04 | 2016-05-11 | 南京云石医疗科技有限公司 | Deep learning-based adaptive ultrasound image enhancement method |
CN105574550A (en) * | 2016-02-02 | 2016-05-11 | 北京格灵深瞳信息技术有限公司 | Vehicle identification method and device |
CN106127747A (en) * | 2016-06-17 | 2016-11-16 | 史方 | Car surface damage classifying method and device based on degree of depth study |
CN106250819A (en) * | 2016-07-20 | 2016-12-21 | 上海交通大学 | Based on face's real-time monitor and detection facial symmetry and abnormal method |
CN106326857A (en) * | 2016-08-19 | 2017-01-11 | 乐视控股(北京)有限公司 | Gender identification method and gender identification device based on face image |
CN106340309A (en) * | 2016-08-23 | 2017-01-18 | 南京大空翼信息技术有限公司 | Dog bark emotion recognition method and device based on deep learning |
CN106600581A (en) * | 2016-12-02 | 2017-04-26 | 北京航空航天大学 | Train operation fault automatic detection system and method based on binocular stereoscopic vision |
-
2017
- 2017-12-11 CN CN201711311755.3A patent/CN108154169A/en active Pending
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102312616A (en) * | 2010-06-29 | 2012-01-11 | 马瑞利汽车电子(广州)有限公司 | The method and system that is used for intelligence entering and intelligent starting |
CN102968643A (en) * | 2012-11-16 | 2013-03-13 | 华中科技大学 | Multi-mode emotion recognition method based on Lie group theory |
CN104063677A (en) * | 2013-03-19 | 2014-09-24 | 北京三星通信技术研究有限公司 | Equipment used for estimating human body posture and method thereof |
CN105574820A (en) * | 2015-12-04 | 2016-05-11 | 南京云石医疗科技有限公司 | Deep learning-based adaptive ultrasound image enhancement method |
CN105574550A (en) * | 2016-02-02 | 2016-05-11 | 北京格灵深瞳信息技术有限公司 | Vehicle identification method and device |
CN106127747A (en) * | 2016-06-17 | 2016-11-16 | 史方 | Car surface damage classifying method and device based on degree of depth study |
CN106250819A (en) * | 2016-07-20 | 2016-12-21 | 上海交通大学 | Based on face's real-time monitor and detection facial symmetry and abnormal method |
CN106326857A (en) * | 2016-08-19 | 2017-01-11 | 乐视控股(北京)有限公司 | Gender identification method and gender identification device based on face image |
CN106340309A (en) * | 2016-08-23 | 2017-01-18 | 南京大空翼信息技术有限公司 | Dog bark emotion recognition method and device based on deep learning |
CN106600581A (en) * | 2016-12-02 | 2017-04-26 | 北京航空航天大学 | Train operation fault automatic detection system and method based on binocular stereoscopic vision |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109583318A (en) * | 2018-11-06 | 2019-04-05 | 平安科技(深圳)有限公司 | Medicinal plant recognition methods, device and computer equipment |
CN110136185A (en) * | 2019-05-23 | 2019-08-16 | 中国科学技术大学 | A kind of monocular depth estimation method and system |
CN110136185B (en) * | 2019-05-23 | 2022-09-06 | 中国科学技术大学 | Monocular depth estimation method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108229267B (en) | Object attribute detection, neural network training and region detection method and device | |
CN110858286B (en) | Image processing method and device for target recognition | |
CN108156519B (en) | Image classification method, television device and computer-readable storage medium | |
WO2018109505A1 (en) | Transforming source domain images into target domain images | |
CN112508975A (en) | Image identification method, device, equipment and storage medium | |
EP3989104A1 (en) | Facial feature extraction model training method and apparatus, facial feature extraction method and apparatus, device, and storage medium | |
CN110930296B (en) | Image processing method, device, equipment and storage medium | |
CN113688723A (en) | Infrared image pedestrian target detection method based on improved YOLOv5 | |
CN111192277A (en) | Instance partitioning method and device | |
CN106778820A (en) | Identification model determines method and device | |
CN113128478B (en) | Model training method, pedestrian analysis method, device, equipment and storage medium | |
WO2019214240A1 (en) | Generation of dynamic picture | |
CN111047088A (en) | Prediction image acquisition method and device, computer equipment and storage medium | |
CN110781976A (en) | Extension method of training image, training method and related device | |
CN111461213A (en) | Training method of target detection model and target rapid detection method | |
CN112465709B (en) | Image enhancement method, device, storage medium and equipment | |
CN110895802B (en) | Image processing method and device | |
CN112489143A (en) | Color identification method, device, equipment and storage medium | |
CN108121951A (en) | Characteristic point positioning method and device | |
CN110147833A (en) | Facial image processing method, apparatus, system and readable storage medium storing program for executing | |
CN108154169A (en) | Image processing method and device | |
CN108122001A (en) | Image-recognizing method and device | |
CN112712068B (en) | Key point detection method and device, electronic equipment and storage medium | |
CN107066980A (en) | A kind of anamorphose detection method and device | |
CN117253110A (en) | Diffusion model-based target detection model generalization capability improving method |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180612 |