CN106250921A - Image processing method and device - Google Patents
Image processing method and device Download PDFInfo
- Publication number
- CN106250921A CN106250921A CN201610597748.3A CN201610597748A CN106250921A CN 106250921 A CN106250921 A CN 106250921A CN 201610597748 A CN201610597748 A CN 201610597748A CN 106250921 A CN106250921 A CN 106250921A
- Authority
- CN
- China
- Prior art keywords
- samples pictures
- goods categories
- picture
- training pattern
- target training
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The disclosure is directed to a kind of image processing method and device.Wherein, method includes: obtain at least one samples pictures collection that at least one goods categories is corresponding, wherein, each samples pictures collection at least one goods categories corresponding, each samples pictures that each samples pictures is concentrated all includes the article under this goods categories corresponding to samples pictures collection, every kind of corresponding label of goods categories；Samples pictures, default convolutional neural networks agreement and the support utilizing at least one samples pictures described to concentrate exports the default convolutional neural networks model of multiple labels and is trained, obtaining the target training pattern that at least one goods categories described is corresponding, described target training pattern is for determining the target item classification corresponding to picture to be sorted of at least one goods categories corresponding.By this technical scheme, the target training pattern trained can be made more to meet practice, reduce the error rate of picture recognition.
Description
Technical field
It relates to picture Processing Technique field, particularly relate to image processing method and device.
Background technology
The concept of degree of depth study is proposed in 2006 by Hinton et al..Propose non-supervisory greedy based on deep Belief Network (DBN)
The heart successively training algorithm, brings hope for the optimization difficult problem solving deep structure relevant, proposes multilamellar autocoder subsequently deep
Rotating fields.In addition the convolutional neural networks that Lecun et al. proposes is first real multiple structure learning algorithm, and it utilizes space
Relativeness reduces number of parameters to improve training performance.
Convolutional neural networks can be used for the identification to picture.But its output of the convolutional neural networks in correlation technique is only propped up
Hold single label.Such as, an existing house pet, there is again the photo of people, if according to single label, it being set as house pet, but
Being other training sets inside training set, namely " other " set in addition to pets training collection, the most also has
Comprise the photo of people.Assume training set three classes: cat, Canis familiaris L., people, other.Also bag is had inside the training set of result cat or Canis familiaris L.
Photo containing people, " other " set inside also have the photo comprising people, this can allow model training when become the most puzzled, and
The model that so training obtains is when being identified, and recall rate and error rate all exist higher deviation.
Summary of the invention
Disclosure embodiment provides a kind of image processing method and device, including following technical scheme:
First aspect according to disclosure embodiment, it is provided that a kind of image processing method, including:
Obtaining at least one samples pictures collection that at least one goods categories is corresponding, wherein, each samples pictures collection is corresponding
At least one goods categories, all includes this thing corresponding to samples pictures collection in each samples pictures that each samples pictures is concentrated
Category not under article, the corresponding label of every kind of goods categories；
Utilize samples pictures, default convolutional neural networks agreement and support output that at least one samples pictures described is concentrated
The default convolutional neural networks model of multiple labels is trained, and obtains the target training that at least one goods categories described is corresponding
Model, described target training pattern is for determining the target item corresponding to picture to be sorted of at least one goods categories corresponding
Classification.
In one embodiment, the samples pictures that at least one samples pictures described in described utilization is concentrated, default convolution god
It is trained through procotol and default convolutional neural networks model, obtains the target instruction that at least one goods categories described is corresponding
Before practicing model, described method also includes:
The size of the samples pictures at least one samples pictures described concentrated is processed as pre-set dimension.
In one embodiment, described target training pattern determines the picture institute to be sorted of at least one goods categories corresponding
The step of corresponding target item classification includes:
The size of described pending picture is processed as pre-set dimension；
According to described target training pattern extraction feature vector from described pending picture；
According to described characteristic vector and described target training pattern, calculate described pending picture and at least one thing described
Category not in there is between each goods categories the probit of corresponding relation；
Probit is defined as described target item classification more than at least one goods categories of predetermined probabilities value.
In one embodiment, described method also includes:
Described pending picture is added in the pictures that extremely described target item classification is corresponding.
In one embodiment, described goods categories includes cat, Canis familiaris L., people.
Second aspect according to disclosure embodiment, it is provided that a kind of picture processing device, including:
Acquisition module, for obtaining at least one samples pictures collection that at least one goods categories is corresponding, wherein, each sample
These pictures at least one goods categories corresponding, all includes this samples pictures in each samples pictures that each samples pictures is concentrated
The article under goods categories corresponding to collection, every kind of corresponding label of goods categories；
Training module, for the samples pictures utilizing at least one samples pictures described to concentrate, presets convolutional neural networks
Agreement and support export the default convolutional neural networks model of multiple labels and are trained, and obtain at least one goods categories described
Corresponding target training pattern, described target training pattern is for determining the picture institute to be sorted of at least one goods categories corresponding
Corresponding target item classification.
In one embodiment, described device also includes:
First processing module, the size of the samples pictures at least one samples pictures described being concentrated is processed as presetting
Size.
In one embodiment, the picture to be sorted of at least one goods categories corresponding is determined in described target training pattern
During corresponding target item classification, described device also includes:
Second processing module, is processed as pre-set dimension by the size of described pending picture；
Abstraction module, for according to described target training pattern extraction feature vector from described pending picture；
Computing module, for according to described characteristic vector and described target training pattern, calculate described pending picture with
At least one goods categories described has between each goods categories the probit of corresponding relation；
Determine module, for probit is defined as described object more than at least one goods categories of predetermined probabilities value
Category is other.
In one embodiment, described device also includes:
Add module, for being added in the pictures corresponding to described target item classification by described pending picture.
In one embodiment, described goods categories includes cat, Canis familiaris L., people.
The third aspect according to disclosure embodiment, it is provided that a kind of picture processing device, including:
Processor；
For storing the memorizer of processor executable；
Wherein, described processor is configured to:
Obtaining at least one samples pictures collection that at least one goods categories is corresponding, wherein, each samples pictures collection is corresponding
At least one goods categories, all includes this thing corresponding to samples pictures collection in each samples pictures that each samples pictures is concentrated
Category not under article, the corresponding label of every kind of goods categories；
Utilize samples pictures, default convolutional neural networks agreement and support output that at least one samples pictures described is concentrated
The default convolutional neural networks model of multiple labels is trained, and obtains the target training that at least one goods categories described is corresponding
Model, described target training pattern is for determining the target item corresponding to picture to be sorted of at least one goods categories corresponding
Classification.
Embodiment of the disclosure that the technical scheme of offer can include following beneficial effect:
Technique scheme, utilizes at least one samples pictures collection corresponding at least one goods categories, presets convolution god
The default convolutional neural networks model exporting multiple labels through procotol and support is trained, thus obtains at least one thing
The target training pattern that category is not corresponding, in this manner it is possible to carry out the picture to be sorted of at least one goods categories corresponding point
Class, so that the target training pattern that training is out more meets practice, reduces the error rate of picture recognition.Such as, one
Picture to be sorted had not only comprised people and but also comprise house pet, then, by this technical scheme, house pet classification can be identified as
And/or the classification of people, and correlation technique can only treat compared with category images carries out Individual Items classification identification, recognition result
More comprehensive, reduce the error rate of picture recognition.
It should be appreciated that it is only exemplary and explanatory, not that above general description and details hereinafter describe
The disclosure can be limited.
Accompanying drawing explanation
Accompanying drawing herein is merged in description and constitutes the part of this specification, it is shown that meet the enforcement of the disclosure
Example, and for explaining the principle of the disclosure together with description.
Fig. 1 is the flow chart according to the image processing method shown in an exemplary embodiment.
Fig. 2 is the flow chart according to the another kind of image processing method shown in an exemplary embodiment.
Fig. 3 is the flow chart according to another image processing method shown in an exemplary embodiment.
Fig. 4 is the flow chart according to another image processing method shown in an exemplary embodiment.
Fig. 5 is the flow chart according to another image processing method shown in an exemplary embodiment.
Fig. 6 is the block diagram according to the picture processing device shown in an exemplary embodiment.
Fig. 7 is the block diagram according to the another kind of picture processing device shown in an exemplary embodiment.
Fig. 8 is the block diagram according to another the picture processing device shown in an exemplary embodiment.
Fig. 9 is the block diagram according to another the picture processing device shown in an exemplary embodiment.
Figure 10 is according to the block diagram being applicable to picture processing device shown in an exemplary embodiment.
Detailed description of the invention
Here will illustrate exemplary embodiment in detail, its example represents in the accompanying drawings.Explained below relates to
During accompanying drawing, unless otherwise indicated, the same numbers in different accompanying drawings represents same or analogous key element.Following exemplary embodiment
Described in embodiment do not represent all embodiments consistent with the disclosure.On the contrary, they are only with the most appended
The example of the apparatus and method that some aspects that described in detail in claims, the disclosure are consistent.
Disclosure embodiment provides a kind of image processing method, and the method can be used for the equipment needing to carry out picture processing
In, as it is shown in figure 1, the method comprising the steps of S101-S102:
In step S101, obtain at least one samples pictures collection that at least one goods categories is corresponding, wherein, each sample
These pictures at least one goods categories corresponding, all includes this samples pictures in each samples pictures that each samples pictures is concentrated
The article under goods categories corresponding to collection, every kind of corresponding label of goods categories；In one embodiment, goods categories bag
Include cat, Canis familiaris L., people.
In step s 102, utilize samples pictures that at least one samples pictures concentrates, preset convolutional neural networks agreement
It is trained with the default convolutional neural networks model supporting the multiple labels of output, obtains the mesh that at least one goods categories is corresponding
Mark training pattern, target training pattern is for determining the object corresponding to picture to be sorted of at least one goods categories corresponding
Category is other.
In this embodiment, utilize at least one samples pictures collection corresponding at least one goods categories, preset convolution god
The default convolutional neural networks model exporting multiple labels through procotol and support is trained, thus obtains at least one thing
The target training pattern that category is not corresponding, in this manner it is possible to carry out the picture to be sorted of at least one goods categories corresponding point
Class, so that the target training pattern that training is out more meets practice, reduces the error rate of picture recognition.Such as, one
Picture to be sorted had not only comprised people and but also comprise house pet, then, by this technical scheme, house pet classification can be identified as
And/or the classification of people, and correlation technique can only treat compared with category images carries out Individual Items classification identification, recognition result
More comprehensive, reduce the error rate of picture recognition.
As in figure 2 it is shown, in one embodiment, before above-mentioned steps S102, method also includes step S201:
In step s 201, the size of the samples pictures at least one samples pictures concentrated is processed as pre-set dimension.
As it is shown on figure 3, in one embodiment, target training pattern determines the to be sorted of at least one goods categories corresponding
The step of the target item classification corresponding to picture includes step S301-S304:
In step S301, the size of pending picture is processed as pre-set dimension；
In step s 302, according to target training pattern extraction feature vector from pending picture；
In step S303, according to characteristic vector and target training pattern, calculate pending picture and at least one article
Classification has between each goods categories the probit of corresponding relation；
In step s 304, probit is defined as object category more than at least one goods categories of predetermined probabilities value
Not.
In this embodiment, for the picture to be sorted of at least one goods categories of correspondence, can first it be carried out
Normalized, is processed as pre-set dimension, then extracts the characteristic vector of this picture, according to its characteristic vector and target training pattern
Calculate its probability having corresponding relation with each goods categories (such as cat, Canis familiaris L., people), and by probit more than predetermined probabilities value
Goods categories is defined as target item classification.Such as, for a pictures, calculating its probability having corresponding relation with cat is
0.5, the probability having corresponding relation with people is 0.5, it is determined that it belongs to the classification of cat and people.
As shown in Figure 4, in one embodiment, method also includes step S401:
In step S401, by pictures corresponding for pending picture interpolation to target item classification.
In this embodiment, it is also possible to by pictures corresponding for picture to be sorted interpolation to target item classification, as incited somebody to action
Target item classification is that the picture of cat and Canis familiaris L. adds to house pet pictures, thus realizes treating the classification of category images, is not required to
Want user manually to classify, promote the experience of user.
The technical scheme of the disclosure is described below in detail with a specific embodiment.Assume to need the photograph album to user terminal
In picture classify.
As it is shown in figure 5, include according to the image processing method of disclosure embodiment:
In step S501, obtain at least one samples pictures collection that at least one goods categories is corresponding, wherein, each sample
These pictures at least one goods categories corresponding, all includes this samples pictures in each samples pictures that each samples pictures is concentrated
The article under goods categories corresponding to collection, every kind of corresponding label of goods categories.As included: 50000, the picture of cat, Canis familiaris L.
50000, picture, 50000, the picture of people, other various foreign material picture 10w open, people photo 1w together with cat opens, people and
Canis familiaris L. photo 1w together opens.Training objective trains 4 class graders exactly.Training set 1 be numbered 1 training set 2
(1,3) training that is numbered being numbered 4 training set 5 being numbered 3 training set 4 being numbered 2 training set 3 gathers 6
It is numbered (2,3).
In step S502, all of samples pictures carries out size normalized, such as, process a size of 224*
224。
In step S503, amendment convolutional neural networks MODEL C NN and agreement, it is trained, allows the CNN network mould obtained
Type can support that multiaspect exports.
In step S504, training terminates, and preserves training protocol and CNN network model.
In step S505, a pictures of newly coming in, it is zoomed to 224*224, then utilizes amended support many
The CNN network model of label directly carries out output and judges, obtains the label that this picture is corresponding.
In step S506, all photos in photograph album are all carried out the operation of step S505, be judged to the most at last cat or
The photo of person Canis familiaris L., or the photo containing cat or Canis familiaris L. inside multiaspect is all classified as house pet photograph album.So, by this technical side
Case, can be identified as the classification of house pet classification and/or people, carries out single with the category images of can only treating in correlation technique
Goods categories identification is compared, and recognition result is more comprehensive, reduces the error rate of picture recognition.
Following for disclosure device embodiment, may be used for performing method of disclosure embodiment.
Fig. 6 is the block diagram according to a kind of picture processing device shown in an exemplary embodiment, and this device can be by soft
Part, hardware or both be implemented in combination with become the some or all of of electronic equipment.As shown in Figure 6, this picture processing device
Including:
Acquisition module 61, for obtaining at least one samples pictures collection that at least one goods categories is corresponding, wherein, each
Samples pictures collection at least one goods categories corresponding, all includes this sample graph in each samples pictures that each samples pictures is concentrated
The article under goods categories corresponding to sheet collection, every kind of corresponding label of goods categories；
Training module 62, for the samples pictures utilizing at least one samples pictures to concentrate, presets convolutional neural networks association
The default convolutional neural networks model supporting to export multiple labels of negotiating peace is trained, and obtains at least one goods categories corresponding
Target training pattern, target training pattern is for determining the target corresponding to picture to be sorted of at least one goods categories corresponding
Goods categories.
As it is shown in fig. 7, in one embodiment, device also includes:
First processing module 71, the size of the samples pictures at least one samples pictures being concentrated is processed as presetting chi
Very little.
As shown in Figure 8, in one embodiment, treating point of at least one goods categories corresponding is determined in target training pattern
During target item classification corresponding to class picture, device also includes:
Second processing module 81, is processed as pre-set dimension by the size of pending picture；
Abstraction module 82, for according to target training pattern extraction feature vector from pending picture；
Computing module 83, for according to characteristic vector and target training pattern, calculates pending picture and at least one thing
Category not in there is between each goods categories the probit of corresponding relation；
Determine module 84, for probit is defined as target item more than at least one goods categories of predetermined probabilities value
Classification.
As it is shown in figure 9, in one embodiment, device also includes:
Add module 91, for being added in the pictures corresponding to target item classification by pending picture.
In one embodiment, goods categories includes cat, Canis familiaris L., people.
The third aspect according to disclosure embodiment, it is provided that a kind of picture processing device, including:
Processor；
For storing the memorizer of processor executable；
Wherein, processor is configured to:
Obtaining at least one samples pictures collection that at least one goods categories is corresponding, wherein, each samples pictures collection is corresponding
At least one goods categories, all includes this thing corresponding to samples pictures collection in each samples pictures that each samples pictures is concentrated
Category not under article, the corresponding label of every kind of goods categories；
Utilize samples pictures, default convolutional neural networks agreement and support output that at least one samples pictures described is concentrated
The default convolutional neural networks model of multiple labels is trained, and obtains the target training that at least one goods categories described is corresponding
Model, described target training pattern is for determining the target item corresponding to picture to be sorted of at least one goods categories corresponding
Classification.
Above-mentioned processor is also configured to:
Described in described utilization at least one samples pictures concentrate samples pictures, preset convolutional neural networks agreement and preset
Convolutional neural networks model is trained, before obtaining the target training pattern that at least one goods categories described is corresponding, described
Method also includes:
The size of the samples pictures at least one samples pictures described concentrated is processed as pre-set dimension.
Above-mentioned processor is also configured to:
Described target training pattern determines the target item corresponding to picture to be sorted of at least one goods categories corresponding
The step of classification includes:
The size of described pending picture is processed as pre-set dimension；
According to described target training pattern extraction feature vector from described pending picture；
According to described characteristic vector and described target training pattern, calculate described pending picture and at least one thing described
Category not in there is between each goods categories the probit of corresponding relation；
Probit is defined as described target item classification more than at least one goods categories of predetermined probabilities value.
Above-mentioned processor is also configured to:
Described method also includes:
Described pending picture is added in the pictures that extremely described target item classification is corresponding.
Above-mentioned processor is also configured to:
Described goods categories includes cat, Canis familiaris L., people.
About the device in above-described embodiment, wherein modules performs the concrete mode of operation in relevant the method
Embodiment in be described in detail, explanation will be not set forth in detail herein.
Figure 10 is according to a kind of block diagram for picture processing device shown in an exemplary embodiment, and this device is applicable to
Terminal unit.Such as, device 1000 can be mobile phone, computer, digital broadcast terminal, messaging devices, game control
Platform processed, tablet device, armarium, body-building equipment, personal digital assistant etc..
Device 1000 can include following one or more assembly: processes assembly 1002, memorizer 1004, power supply module
1006, multimedia groupware 1008, audio-frequency assembly 1010, the interface 1011 of input/output (I/O), sensor cluster 1014, and
Communications component 1016.
Process assembly 1002 and generally control the integrated operation of device 1000, such as with display, call, data communication,
The operation that camera operation and record operation are associated.Process assembly 1002 and can include that one or more processor 1020 performs
Instruction, to complete all or part of step of above-mentioned method.Additionally, process assembly 1002 can include one or more mould
Block, it is simple to process between assembly 1002 and other assemblies is mutual.Such as, process assembly 1002 and can include multi-media module,
With facilitate multimedia groupware 1008 and process between assembly 1002 mutual.
Memorizer 1004 is configured to store various types of data to support the operation at device 1000.These data
Example include on device 1000 operation any application program or the instruction of method, contact data, telephone book data,
Message, picture, video etc..Memorizer 1004 can by any kind of volatibility or non-volatile memory device or they
Combination realizes, such as static RAM (SRAM), Electrically Erasable Read Only Memory (EEPROM), erasable can
Program read-only memory (EPROM), programmable read only memory (PROM), read only memory (ROM), magnetic memory, flash memory
Reservoir, disk or CD.
The various assemblies that power supply module 1006 is device 1000 provide electric power.Power supply module 1006 can include power management
System, one or more power supplys, and other generate, manage and distribute, with for device 1000, the assembly that electric power is associated.
The screen of one output interface of offer that multimedia groupware 1008 is included between described device 1000 and user.?
In some embodiments, screen can include liquid crystal display (LCD) and touch panel (TP).If screen includes touch panel,
Screen may be implemented as touch screen, to receive the input signal from user.Touch panel includes that one or more touch passes
Sensor is with the gesture on sensing touch, slip and touch panel.Described touch sensor can not only sense touch or slide dynamic
The border made, but also detect the persistent period relevant to described touch or slide and pressure.In certain embodiments, many
Media component 1008 includes a front-facing camera and/or post-positioned pick-up head.When device 1000 is in operator scheme, such as shooting mould
When formula or video mode, front-facing camera and/or post-positioned pick-up head can receive the multi-medium data of outside.Each preposition shooting
Head and post-positioned pick-up head can be a fixing optical lens system or have focal length and optical zoom ability.
Audio-frequency assembly 1010 is configured to output and/or input audio signal.Such as, audio-frequency assembly 1010 includes a wheat
Gram wind (MIC), when device 1000 is in operator scheme, during such as call model, logging mode and speech recognition mode, mike quilt
It is configured to receive external audio signal.The audio signal received can be further stored at memorizer 1004 or via communication
Assembly 1016 sends.In certain embodiments, audio-frequency assembly 1010 also includes a speaker, is used for exporting audio signal.
I/O interface 1011 provides interface, above-mentioned peripheral interface module for processing between assembly 1002 and peripheral interface module
Can be keyboard, put striking wheel, button etc..These buttons may include but be not limited to: home button, volume button, start button and
Locking press button.
Sensor cluster 1014 includes one or more sensor, for providing the state of various aspects to comment for device 1000
Estimate.Such as, what sensor cluster 1014 can detect device 1000 opens/closed mode, the relative localization of assembly, such as institute
Stating display and keypad that assembly is device 1000, sensor cluster 1014 can also detect device 1000 or device 1,000 1
The position change of individual assembly, the presence or absence that user contacts with device 1000, device 1000 orientation or acceleration/deceleration and dress
Put the variations in temperature of 1000.Sensor cluster 1014 can include proximity transducer, is configured to do not having any physics
The existence of object near detection during contact.Sensor cluster 1014 can also include optical sensor, as CMOS or ccd image sense
Device, for using in imaging applications.In certain embodiments, this sensor cluster 1014 can also include acceleration sensing
Device, gyro sensor, Magnetic Sensor, pressure transducer or temperature sensor.
Communications component 1016 is configured to facilitate the communication of wired or wireless mode between device 1000 and other equipment.Dress
Put 1000 and can access wireless network based on communication standard, such as WiFi, 2G or 3G, or combinations thereof.Exemplary at one
In embodiment, broadcast singal or broadcast that communications component 1016 receives from external broadcasting management system via broadcast channel are relevant
Information.In one exemplary embodiment, described communications component 1016 also includes near-field communication (NFC) module, to promote short distance
Communication.Such as, can be based on RF identification (RFID) technology in NFC module, Infrared Data Association (IrDA) technology, ultra broadband
(UWB) technology, bluetooth (BT) technology and other technologies realize.
In the exemplary embodiment, device 1000 can be by one or more application specific integrated circuits (ASIC), numeral
Signal processor (DSP), digital signal processing appts (DSPD), PLD (PLD), field programmable gate array
(FPGA), controller, microcontroller, microprocessor or other electronic building bricks realize, be used for performing said method.
In the exemplary embodiment, a kind of non-transitory computer-readable recording medium including instruction, example are additionally provided
As included the memorizer 1004 of instruction, above-mentioned instruction can have been performed said method by the processor 1020 of device 1000.Example
If, described non-transitory computer-readable recording medium can be ROM, random access memory (RAM), CD-ROM, tape, soft
Dish and optical data storage devices etc..
A kind of non-transitory computer-readable recording medium, when the instruction in described storage medium is by the process of device 1000
Device perform time so that device 1000 is able to carry out above-mentioned .... method, described method includes:
Obtaining at least one samples pictures collection that at least one goods categories is corresponding, wherein, each samples pictures collection is corresponding
At least one goods categories, all includes this thing corresponding to samples pictures collection in each samples pictures that each samples pictures is concentrated
Category not under article, the corresponding label of every kind of goods categories；
Utilize samples pictures, default convolutional neural networks agreement and support output that at least one samples pictures described is concentrated
The default convolutional neural networks model of multiple labels is trained, and obtains the target training that at least one goods categories described is corresponding
Model, described target training pattern is for determining the target item corresponding to picture to be sorted of at least one goods categories corresponding
Classification.
In one embodiment, the samples pictures that at least one samples pictures described in described utilization is concentrated, default convolution god
It is trained through procotol and default convolutional neural networks model, obtains the target instruction that at least one goods categories described is corresponding
Before practicing model, described method also includes:
The size of the samples pictures at least one samples pictures described concentrated is processed as pre-set dimension.
In one embodiment, described target training pattern determines the picture institute to be sorted of at least one goods categories corresponding
The step of corresponding target item classification includes:
The size of described pending picture is processed as pre-set dimension；
According to described target training pattern extraction feature vector from described pending picture；
According to described characteristic vector and described target training pattern, calculate described pending picture and at least one thing described
Category not in there is between each goods categories the probit of corresponding relation；
Probit is defined as described target item classification more than at least one goods categories of predetermined probabilities value.
In one embodiment, described method also includes:
Described pending picture is added in the pictures that extremely described target item classification is corresponding.
In one embodiment, described goods categories includes cat, Canis familiaris L., people.
Those skilled in the art, after considering description and putting into practice disclosure disclosed herein, will readily occur to its of the disclosure
Its embodiment.The application is intended to any modification, purposes or the adaptations of the disclosure, these modification, purposes or
Person's adaptations is followed the general principle of the disclosure and includes the undocumented common knowledge in the art of the disclosure
Or conventional techniques means.Description and embodiments is considered only as exemplary, and the true scope of the disclosure and spirit are by following
Claim is pointed out.
It should be appreciated that the disclosure is not limited to precision architecture described above and illustrated in the accompanying drawings, and
And various modifications and changes can carried out without departing from the scope.The scope of the present disclosure is only limited by appended claim.
Claims (11)
1. an image processing method, it is characterised in that including:
Obtaining at least one samples pictures collection that at least one goods categories is corresponding, wherein, each samples pictures collection correspondence is at least
A kind of goods categories, all includes this article class corresponding to samples pictures collection in each samples pictures that each samples pictures is concentrated
Article under not, every kind of corresponding label of goods categories；
The samples pictures, default convolutional neural networks agreement and the support output that utilize at least one samples pictures described to concentrate are multiple
The default convolutional neural networks model of label is trained, and obtains the target training mould that at least one goods categories described is corresponding
Type, described target training pattern is for determining the object category corresponding to picture to be sorted of at least one goods categories corresponding
Not.
Method the most according to claim 1, it is characterised in that the sample that at least one samples pictures described in described utilization is concentrated
This picture, default convolutional neural networks agreement and default convolutional neural networks model are trained, and obtain at least one thing described
Before the target training pattern that category is not corresponding, described method also includes:
The size of the samples pictures at least one samples pictures described concentrated is processed as pre-set dimension.
Method the most according to claim 1 and 2, it is characterised in that described target training pattern determine corresponding at least one
The step of the target item classification corresponding to picture to be sorted of goods categories includes:
The size of described pending picture is processed as pre-set dimension；
According to described target training pattern extraction feature vector from described pending picture；
According to described characteristic vector and described target training pattern, calculate described pending picture and at least one article class described
There is between each goods categories in not the probit of corresponding relation；
Probit is defined as described target item classification more than at least one goods categories of predetermined probabilities value.
The most according to the method in any one of claims 1 to 3, it is characterised in that described method also includes:
Described pending picture is added in the pictures that extremely described target item classification is corresponding.
The most according to the method in any one of claims 1 to 3, it is characterised in that described goods categories includes cat, Canis familiaris L., people.
6. a picture processing device, it is characterised in that including:
Acquisition module, for obtaining at least one samples pictures collection that at least one goods categories is corresponding, wherein, each sample graph
Sheet collection at least one goods categories corresponding, all includes this samples pictures collection institute in each samples pictures that each samples pictures is concentrated
The corresponding article under goods categories, every kind of corresponding label of goods categories；
Training module, for the samples pictures utilizing at least one samples pictures described to concentrate, presets convolutional neural networks agreement
It is trained with the default convolutional neural networks model supporting the multiple labels of output, obtains at least one goods categories described corresponding
Target training pattern, described target training pattern is for determining corresponding to the picture to be sorted of at least one goods categories corresponding
Target item classification.
Device the most according to claim 6, it is characterised in that the sample that at least one samples pictures described in described utilization is concentrated
This picture, default convolutional neural networks agreement and default convolutional neural networks model are trained, and obtain at least one thing described
Before the target training pattern that category is not corresponding, described method also includes:
First processing module, the size of the samples pictures at least one samples pictures described being concentrated is processed as presetting chi
Very little.
8. according to the device described in claim 6 or 7, it is characterised in that determine correspondence at least in described target training pattern
During the target item classification corresponding to picture to be sorted of individual goods categories, described device also includes:
Second processing module, is processed as pre-set dimension by the size of described pending picture；
Abstraction module, for according to described target training pattern extraction feature vector from described pending picture；
Computing module, for according to described characteristic vector and described target training pattern, calculates described pending picture with described
At least one goods categories has between each goods categories the probit of corresponding relation；
Determine module, for probit is defined as described object category more than at least one goods categories of predetermined probabilities value
Not.
9. according to the device according to any one of claim 6 to 8, it is characterised in that described device also includes:
Add module, for being added in the pictures corresponding to described target item classification by described pending picture.
10. according to the method according to any one of claim 6 to 8, it is characterised in that described goods categories include cat, Canis familiaris L.,
People.
11. 1 kinds of picture processing devices, it is characterised in that including:
Processor；
For storing the memorizer of processor executable；
Wherein, described processor is configured to:
Obtaining at least one samples pictures collection that at least one goods categories is corresponding, wherein, each samples pictures collection correspondence is at least
A kind of goods categories, all includes this article class corresponding to samples pictures collection in each samples pictures that each samples pictures is concentrated
Article under not, every kind of corresponding label of goods categories；
The samples pictures, default convolutional neural networks agreement and the support output that utilize at least one samples pictures described to concentrate are multiple
The default convolutional neural networks model of label is trained, and obtains the target training mould that at least one goods categories described is corresponding
Type, described target training pattern is for determining the object category corresponding to picture to be sorted of at least one goods categories corresponding
Not.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610597748.3A CN106250921A (en) | 2016-07-26 | 2016-07-26 | Image processing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610597748.3A CN106250921A (en) | 2016-07-26 | 2016-07-26 | Image processing method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106250921A true CN106250921A (en) | 2016-12-21 |
Family
ID=57604726
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610597748.3A Pending CN106250921A (en) | 2016-07-26 | 2016-07-26 | Image processing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106250921A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107145904A (en) * | 2017-04-28 | 2017-09-08 | 北京小米移动软件有限公司 | Determination method, device and the storage medium of image category |
| CN107480773A (en) * | 2017-08-09 | 2017-12-15 | 北京小米移动软件有限公司 | The method, apparatus and storage medium of training convolutional neural networks model |
| CN108009280A (en) * | 2017-12-21 | 2018-05-08 | 广东欧珀移动通信有限公司 | Image processing method, device, terminal and storage medium |
| CN108052860A (en) * | 2017-11-06 | 2018-05-18 | 珠海格力电器股份有限公司 | Article retrieval method and device |
| WO2018121018A1 (en) * | 2016-12-30 | 2018-07-05 | 腾讯科技（深圳）有限公司 | Picture identification method and device, server and storage medium |
| CN108491825A (en) * | 2018-03-30 | 2018-09-04 | 百度在线网络技术（北京）有限公司 | information generating method and device |
| CN108507270A (en) * | 2017-02-24 | 2018-09-07 | 九阳股份有限公司 | A kind of food materials of refrigerator determine method and food materials determining device |
| CN108921792A (en) * | 2018-07-03 | 2018-11-30 | 北京字节跳动网络技术有限公司 | Method and apparatus for handling picture |
| CN109107161A (en) * | 2018-08-17 | 2019-01-01 | 深圳市腾讯网络信息技术有限公司 | A kind of control method of game object, device, medium and equipment |
| CN110414541A (en) * | 2018-04-26 | 2019-11-05 | 京东方科技集团股份有限公司 | The method, equipment and computer readable storage medium of object for identification |
| CN112613950A (en) * | 2020-12-28 | 2021-04-06 | 上海寻梦信息技术有限公司 | Article type recognition model training method, article type recognition method and related equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105139033A (en) * | 2015-08-24 | 2015-12-09 | 小米科技有限责任公司 | Classifier construction method and device and image processing method and device |
| CN105184313A (en) * | 2015-08-24 | 2015-12-23 | 小米科技有限责任公司 | Classification model construction method and device |
-
2016
- 2016-07-26 CN CN201610597748.3A patent/CN106250921A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105139033A (en) * | 2015-08-24 | 2015-12-09 | 小米科技有限责任公司 | Classifier construction method and device and image processing method and device |
| CN105184313A (en) * | 2015-08-24 | 2015-12-23 | 小米科技有限责任公司 | Classification model construction method and device |
Non-Patent Citations (2)
| Title |
|---|
| YUNCHAO WEI 等: "CNN: Single-label to Multi-label", 《JOURNAL OF LATEX CLASS FILES》 * |
| 黎键成 等: "基于卷积神经网络的多标签图像自动标注", 《计算机科学》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10803359B2 (en) | 2016-12-30 | 2020-10-13 | Tencent Technology (Shenzhen) Company Limited | Image recognition method, apparatus, server, and storage medium |
| WO2018121018A1 (en) * | 2016-12-30 | 2018-07-05 | 腾讯科技（深圳）有限公司 | Picture identification method and device, server and storage medium |
| CN108507270A (en) * | 2017-02-24 | 2018-09-07 | 九阳股份有限公司 | A kind of food materials of refrigerator determine method and food materials determining device |
| CN108507270B (en) * | 2017-02-24 | 2020-07-10 | 九阳股份有限公司 | Food material determining method and food material determining device for refrigerator |
| CN107145904A (en) * | 2017-04-28 | 2017-09-08 | 北京小米移动软件有限公司 | Determination method, device and the storage medium of image category |
| CN107480773B (en) * | 2017-08-09 | 2020-11-13 | 北京小米移动软件有限公司 | Method and device for training convolutional neural network model and storage medium |
| CN107480773A (en) * | 2017-08-09 | 2017-12-15 | 北京小米移动软件有限公司 | The method, apparatus and storage medium of training convolutional neural networks model |
| CN108052860A (en) * | 2017-11-06 | 2018-05-18 | 珠海格力电器股份有限公司 | Article retrieval method and device |
| CN108009280A (en) * | 2017-12-21 | 2018-05-08 | 广东欧珀移动通信有限公司 | Image processing method, device, terminal and storage medium |
| CN108491825A (en) * | 2018-03-30 | 2018-09-04 | 百度在线网络技术（北京）有限公司 | information generating method and device |
| CN110414541A (en) * | 2018-04-26 | 2019-11-05 | 京东方科技集团股份有限公司 | The method, equipment and computer readable storage medium of object for identification |
| US11093800B2 (en) | 2018-04-26 | 2021-08-17 | Boe Technology Group Co., Ltd. | Method and device for identifying object and computer readable storage medium |
| CN110414541B (en) * | 2018-04-26 | 2022-09-09 | 京东方科技集团股份有限公司 | Method, apparatus, and computer-readable storage medium for identifying an object |
| CN108921792A (en) * | 2018-07-03 | 2018-11-30 | 北京字节跳动网络技术有限公司 | Method and apparatus for handling picture |
| CN108921792B (en) * | 2018-07-03 | 2023-06-27 | 北京字节跳动网络技术有限公司 | Method and device for processing pictures |
| CN109107161A (en) * | 2018-08-17 | 2019-01-01 | 深圳市腾讯网络信息技术有限公司 | A kind of control method of game object, device, medium and equipment |
| CN112613950A (en) * | 2020-12-28 | 2021-04-06 | 上海寻梦信息技术有限公司 | Article type recognition model training method, article type recognition method and related equipment |
| CN112613950B (en) * | 2020-12-28 | 2024-03-29 | 上海寻梦信息技术有限公司 | Class identification model training method, class identification method and related equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106250921A (en) | Image processing method and device | |
| CN105184313B (en) | Disaggregated model construction method and device | |
| CN106548145A (en) | Image-recognizing method and device | |
| CN106503617A (en) | Model training method and device | |
| CN106528709A (en) | Social information recommendation method and apparatus | |
| CN105631403A (en) | Method and device for human face recognition | |
| CN105139033A (en) | Classifier construction method and device and image processing method and device | |
| CN105426515A (en) | Video classification method and apparatus | |
| CN104717366A (en) | Method and device for recommending contact photos | |
| CN106652113A (en) | Access control method and device | |
| CN106778531A (en) | Face detection method and device | |
| CN107766820A (en) | Image classification method and device | |
| CN106485567A (en) | Item recommendation method and device | |
| CN104933419A (en) | Method and device for obtaining iris images and iris identification equipment | |
| CN108108671A (en) | Description of product information acquisition method and device | |
| CN105426878A (en) | Method and device for face clustering | |
| CN104111813B (en) | Method and device for picture synchronous display, mobile terminal and terminal device | |
| CN105203456A (en) | Plant species identification method and apparatus thereof | |
| CN106130873A (en) | Information processing method and device | |
| CN106547850A (en) | Expression annotation method and device | |
| CN106534658A (en) | Method and apparatus for controlling shooting of camera and mobile terminal | |
| CN104536753B (en) | Backlog labeling method and device | |
| CN104240274B (en) | Face image processing process and device | |
| CN105551047A (en) | Picture content detecting method and device | |
| CN104539497B (en) | Method for connecting network and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161221 |
|
| RJ01 | Rejection of invention patent application after publication |