CN110046586A - A kind of data processing method, equipment and storage medium - Google Patents
A kind of data processing method, equipment and storage medium Download PDFInfo
- Publication number
- CN110046586A CN110046586A CN201910319953.7A CN201910319953A CN110046586A CN 110046586 A CN110046586 A CN 110046586A CN 201910319953 A CN201910319953 A CN 201910319953A CN 110046586 A CN110046586 A CN 110046586A
- Authority
- CN
- China
- Prior art keywords
- cluster
- objects
- object data
- data
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 22
- 238000003860 storage Methods 0.000 title claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 65
- 238000000034 method Methods 0.000 claims abstract description 62
- 241001269238 Data Species 0.000 claims abstract description 32
- 238000004140 cleaning Methods 0.000 claims abstract description 20
- 239000000284 extract Substances 0.000 claims abstract description 13
- 238000001514 detection method Methods 0.000 claims description 146
- 238000012545 processing Methods 0.000 claims description 71
- 230000009466 transformation Effects 0.000 claims description 28
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims description 26
- 239000000463 material Substances 0.000 claims description 24
- 238000012360 testing method Methods 0.000 claims description 22
- 230000004044 response Effects 0.000 claims description 20
- 238000012217 deletion Methods 0.000 claims description 6
- 230000037430 deletion Effects 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 5
- 230000001815 facial effect Effects 0.000 description 16
- 238000010586 diagram Methods 0.000 description 15
- 238000007689 inspection Methods 0.000 description 11
- 238000000605 extraction Methods 0.000 description 8
- 238000004891 communication Methods 0.000 description 5
- 239000013598 vector Substances 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 241001465754 Metazoa Species 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000012372 quality testing Methods 0.000 description 2
- 238000004064 recycling Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000001960 triggered effect Effects 0.000 description 2
- 238000005267 amalgamation Methods 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Probability & Statistics with Applications (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the present invention discloses a kind of data processing method, equipment and storage medium, and wherein method includes: to obtain sample object data respectively from multiple multi-medium datas, extracts the characteristics of objects information of sample object data;Multiple sample object data are clustered according to characteristics of objects information, obtain multiple the first characteristics of objects clusters with different cluster labels;Foreign peoples's object data in each first characteristics of objects cluster is cleared up, the first characteristics of objects cluster after cleaning is determined as the second characteristics of objects cluster;Similar cluster merging is carried out between multiple second characteristics of objects clusters, generates multiple third characteristics of objects clusters with different cluster labels, and the object tag of sample object data is updated to the cluster label of affiliated third characteristics of objects cluster;Object tag and characteristics of objects information based on the sample object data in each third characteristics of objects cluster, training Object identifying model.Using the present invention, the accuracy rate of recognition of face can be improved.
Description
Technical field
The present invention relates to electronic technology field more particularly to a kind of data processing methods, equipment and storage medium.
Background technique
With the continuous popularization of face recognition technology in practical applications, recognition of face there are the problem of and difficult point also gradually
It highlights, in insecure acquisition environment, the face performance of user and ambient condition are ever-changing, and complexity is remote
Far beyond the picture obtained in standard evaluation and test, changes for example, blocking and will lead to face characteristic, will increase recognition of face
Difficulty increases reject rate;Illumination, the variation of posture and expression, the face characteristic also resulted under the varying environment of a people are poor
It is different very big, reduce discrimination, it is also possible to will increase wrong identification.Currently, in order to promote face recognition technology in practical application
Performance in scene, method the most effective is the face training data for largely having identity label by increasing, reality
A large amount of human face data present in scene (such as monitoring, market, community etc.), but the overwhelming majority is the people of not identity label
Face data can not largely obtain the face training data of identity label, then can not promote people by increasing face training data
Face recognition performance.
Summary of the invention
The embodiment of the present invention provides a kind of data processing method, equipment and storage medium, and the standard of recognition of face can be improved
True rate.
On the one hand the embodiment of the present invention provides a kind of data processing method, it may include:
It obtains sample object data respectively from multiple multi-medium datas, extracts the characteristics of objects of the sample object data
Information;
Multiple sample object data are clustered according to the characteristics of objects information, are obtained multiple with different cluster labels
The first characteristics of objects cluster;Cluster label phase of the object tag of all sample object data with the first affiliated characteristics of objects cluster
Together;
Foreign peoples's object data in each first characteristics of objects cluster is cleared up respectively, the first characteristics of objects cluster after cleaning is true
It is set to the second characteristics of objects cluster;
Similar cluster merging is carried out between multiple second characteristics of objects clusters, generates multiple thirds pair with different cluster labels
As feature cluster, the object tag of the sample object data is updated to the cluster label of the affiliated third characteristics of objects cluster;
Object tag and the object based on the sample object data in each third characteristics of objects cluster are special
Reference breath, training Object identifying model;The Object identifying model object data to be predicted and target object data for identification
Between similarity.
Wherein, described to obtain sample object data respectively from multiple multi-medium datas, extract the sample object data
Corresponding characteristics of objects information, comprising:
The object location information in multiple multi-medium datas is obtained, according to the object location information from each multimedia number
According to the middle subject area for obtaining respectively and there is target size, the picture material in the subject area is determined as sample object number
According to the characteristics of objects information of acquisition each sample object data;
If the multi-medium data is image data, set at random for the sample object data in described image data
Set object tag;
If the multi-medium data be video data, detect the sample object data in the video data with
Relating attribute will be arranged with the sample object data of the identical pursuit path information in track trace information, and for described
Identical object tag is arranged in the sample object data of relating attribute.
Wherein, the picture material by the subject area is determined as sample object data, comprising:
Picture material in the subject area is subjected to the affine transformation based on target bearing, by the figure after affine transformation
As content is determined as sample object data;Object in the sample object data is in the target bearing;
Then the method also includes:
Obtain multi-medium data to be predicted;
The target object area with target size is obtained according to the object location information in the multi-medium data to be predicted
Picture material to be predicted in the targeted object region is carried out the affine transformation based on the target bearing by domain, will be affine
Transformed picture material to be predicted is determined as object data to be predicted;
The similarity between object data to be predicted and target object data is obtained based on Object identifying model.
Wherein, described that multiple sample object data are clustered according to the characteristics of objects information, it obtains multiple having
First characteristics of objects cluster of different cluster labels, comprising:
First sample object data and the second sample object data are obtained from multiple sample object data, by described first
The object tag of sample object data is determined as the first object tag, and the object tag of the second sample object data is determined
For the second object tag;First object tag and second object tag be not identical;
The object of the characteristics of objects information and the second sample object data that obtain the first sample object data is special
The first image similarity between reference breath;
If the first image similarity is greater than first threshold, by least one sample with first object tag
This object data and at least one sample object data with the second object tag, are divided into identical first characteristics of objects
Cluster, and set the first object tag and the second object tag on the cluster label of the first divided characteristics of objects cluster;
When all sample object data are divided into the first affiliated characteristics of objects cluster, and each first object is special
Sample object data in sign cluster and the characteristics of objects cluster are stored.
Wherein, the foreign peoples's object data cleared up in each first characteristics of objects cluster respectively, by first pair after cleaning
As feature cluster is determined as the second characteristics of objects cluster, comprising:
Response is directed to the first marking operation of the first characteristics of objects cluster, by institute indicated by first marking operation
The sample object data in the first characteristics of objects cluster are stated, the corresponding Standard object data of the first characteristics of objects cluster is determined as;
Response is directed to the second marking operation of non-standard object data, will be non-standard indicated by second marking operation
Object data is determined as foreign peoples's object data;The non-standard object data is that the standard is removed in the first characteristics of objects cluster
Sample object data except object data;
Foreign peoples's object data is deleted from the first characteristics of objects cluster, the first of foreign peoples's object data will be deleted
Characteristics of objects cluster is determined as the second characteristics of objects cluster.
Wherein, described that similar cluster merging is carried out between multiple second characteristics of objects clusters, it generates multiple with different cluster marks
The object tag of the sample object data is updated to the affiliated third characteristics of objects cluster by the third characteristics of objects cluster of label
Cluster label, comprising:
The first detection feature cluster and the second detection feature cluster are obtained from multiple second characteristics of objects clusters;
Response is based on the merging for the merging request of the first detection feature cluster and the second detection feature cluster
In first quantity of the sample object data in the first detection feature cluster described in request and the second detection feature cluster
Second quantity of sample object data;
According to each sample in each sample object data in the first detection feature cluster and the second detection feature cluster
The image similarity of this object data, first quantity and second quantity obtain the first detection feature cluster and institute
State the classification similarity of the second detection feature cluster;
If the classification similarity is greater than second threshold, the first detection feature cluster and the second detection feature cluster are merged into
Third characteristics of objects cluster, and the object tag of the sample object data in the first detection feature cluster and the second detection feature cluster is set
The cluster label of the third characteristics of objects cluster belonging to being set to.
Wherein, described that similar cluster merging is carried out between multiple second characteristics of objects clusters, it generates multiple with different cluster marks
The object tag of the sample object data is updated to the affiliated third characteristics of objects cluster by the third characteristics of objects cluster of label
Cluster label after, further includes:
The third characteristics of objects cluster is determined as to recycle characteristics of objects cluster;
Foreign peoples's object data in each circulation characteristics of objects cluster is cleared up respectively;
Circulation characteristics of objects cluster after deletion foreign peoples's object data is subjected to similar cluster merging, it is special to obtain the third object
Levy cluster;
When meeting the condition of convergence or similar cluster merging number equal to frequency threshold value for the third characteristics of objects cluster,
Execute the object tag and characteristics of objects information based on the sample object data in each third characteristics of objects cluster, instruction
The step of practicing Object identifying model.
Wherein, further includes:
Between each third characteristics of objects cluster Plays object data and non-standard object data is obtained respectively
Two image similarities obtain the corresponding similarity of each third characteristics of objects cluster according to second image similarity
Mean value and similarity variance;
Similarity mean value is less than mean value threshold value and similarity variance is greater than the third characteristics of objects cluster of variance threshold values
It is determined as target object feature cluster;
Obtain the testing result for being directed to the target object feature cluster;
If the testing result is that foreign peoples's object data and similar cluster are not present in the target object feature cluster, it is determined that
The third characteristics of objects cluster meets the condition of convergence.
Wherein, the object tag and object based on the sample object data in each third characteristics of objects cluster are special
Reference breath, training Object identifying model, comprising:
The characteristics of objects information and initial object for obtaining the sample object data in the third characteristics of objects cluster identify mould
Sample image similarity between the characteristics of objects information of Standard object data in type;
According in the object tag of the sample object data in the Object identifying model, the third characteristics of objects cluster
The object tag of sample object data and the sample image similarity determine similarity error, and according to the similarity error
Backpropagation adjusts the model parameter of the initial object identification model;
When the adjustment number of the model parameter is equal to adjustment threshold value or the similarity error meets the condition of convergence,
Initial object identification model comprising model parameter adjusted is determined as the Object identifying model.
On the one hand the embodiment of the present invention provides a kind of data processing equipment, it may include:
Characteristic acquisition unit, for obtaining sample object data respectively from multiple multi-medium datas, described in extraction
The characteristics of objects information of sample object data;
Fisrt feature cluster acquiring unit, for being gathered according to the characteristics of objects information to multiple sample object data
Class obtains multiple the first characteristics of objects clusters with different cluster labels;The object tag of all sample object data with it is affiliated
The first characteristics of objects cluster cluster label it is identical;
Second feature cluster acquiring unit, foreign peoples's object data for being cleared up in each first characteristics of objects cluster respectively will
The first characteristics of objects cluster after cleaning is determined as the second characteristics of objects cluster;
Third feature cluster acquiring unit generates more for carrying out similar cluster merging between multiple second characteristics of objects clusters
A third characteristics of objects cluster with different cluster labels, by the object tag of the sample object data be updated to belonging to described in
The cluster label of third characteristics of objects cluster;
Model training unit, for the object based on the sample object data in each third characteristics of objects cluster
Label and the characteristics of objects information, training Object identifying model;Object identifying model number of objects to be predicted for identification
According to the similarity between target object data.
Wherein, the characteristic acquisition unit includes:
Location information obtains subelement, for obtaining the object location information in multiple multi-medium datas, according to described right
As location information obtains the subject area with target size respectively from each multi-medium data;
Object data obtains subelement, for the picture material in the subject area to be determined as sample object data,
Obtain the characteristics of objects information of each sample object data;
Subelement is arranged in object tag, if being image data for the multi-medium data, in described image data
The sample object data be randomly provided object tag;If the multi-medium data is video data, the sample is detected
Pursuit path information of the object data in the video data, by the sample object number with the identical pursuit path information
According to setting relating attribute, and for the sample object data with the relating attribute, identical object tag is set.
Wherein, the object data obtains subelement and is specifically used for:
Picture material in the subject area is subjected to the affine transformation based on target bearing, by the figure after affine transformation
As content is determined as sample object data, the characteristics of objects information of acquisition each sample object data;The sample object
Object in data is in the target bearing;
Then the method also includes:
Object data recognition unit, for obtaining multi-medium data to be predicted;According in the multi-medium data to be predicted
Object location information obtain have target size targeted object region, by the image to be predicted in the targeted object region
Content carries out the affine transformation based on the target bearing, and it is to be predicted right that the picture material to be predicted after affine transformation is determined as
Image data;The similarity between object data to be predicted and target object data is obtained based on Object identifying model.
Wherein, the fisrt feature cluster acquiring unit is specifically used for:
First sample object data and the second sample object data are obtained from multiple sample object data, by described first
The object tag of sample object data is determined as the first object tag, and the object tag of the second sample object data is determined
For the second object tag;First object tag and second object tag be not identical;
The object of the characteristics of objects information and the second sample object data that obtain the first sample object data is special
The first image similarity between reference breath;
If the first image similarity is greater than first threshold, by least one sample with first object tag
This object data and at least one sample object data with the second object tag, are divided into identical first characteristics of objects
Cluster, and set the first object tag and the second object tag on the cluster label of the first divided characteristics of objects cluster;
When all sample object data are divided into the first affiliated characteristics of objects cluster, and each first object is special
Sample object data in sign cluster and the characteristics of objects cluster are stored.
Wherein, the second feature cluster acquiring unit is specifically used for:
Response is directed to the first marking operation of the first characteristics of objects cluster, by institute indicated by first marking operation
The sample object data in the first characteristics of objects cluster are stated, the corresponding Standard object data of the first characteristics of objects cluster is determined as;
Response is directed to the second marking operation of non-standard object data, will be non-standard indicated by second marking operation
Object data is determined as foreign peoples's object data;The non-standard object data is that the standard is removed in the first characteristics of objects cluster
Sample object data except object data;
Foreign peoples's object data is deleted from the first characteristics of objects cluster, the first of foreign peoples's object data will be deleted
Characteristics of objects cluster is determined as the second characteristics of objects cluster.
Wherein, the third feature cluster acquiring unit is specifically used for:
The first detection feature cluster and the second detection feature cluster are obtained from multiple second characteristics of objects clusters;
Response is based on the merging for the merging request of the first detection feature cluster and the second detection feature cluster
In first quantity of the sample object data in the first detection feature cluster described in request and the second detection feature cluster
Second quantity of sample object data;
According to each sample in each sample object data in the first detection feature cluster and the second detection feature cluster
The image similarity of this object data, first quantity and second quantity obtain the first detection feature cluster and institute
State the classification similarity of the second detection feature cluster;
If the classification similarity is greater than second threshold, the first detection feature cluster and the second detection feature cluster are merged into
Third characteristics of objects cluster, and the object tag of the sample object data in the first detection feature cluster and the second detection feature cluster is set
The cluster label of the third characteristics of objects cluster belonging to being set to.
Wherein, further includes:
Feature cluster recycles acquiring unit, recycles characteristics of objects cluster for the third characteristics of objects cluster to be determined as;Respectively
Clear up foreign peoples's object data in each circulation characteristics of objects cluster;By delete foreign peoples's object data after circulation characteristics of objects cluster into
The similar cluster of row merges, and obtains the third characteristics of objects cluster;Meet the condition of convergence when being directed to the third characteristics of objects cluster, or same
When class cluster merges number equal to frequency threshold value, the model training unit is triggered.
Wherein, further includes:
Detection unit, for obtaining each third characteristics of objects cluster Plays object data and non-standard object respectively
The second image similarity between data obtains each third characteristics of objects cluster according to second image similarity and distinguishes
Corresponding similarity mean value and similarity variance;Similarity mean value is less than mean value threshold value and similarity variance is greater than variance threshold values
The third characteristics of objects cluster be determined as target object feature cluster;
Testing result acquiring unit, for obtaining the testing result for being directed to the target object feature cluster;If the detection
As a result for there is no foreign peoples's object data and similar clusters in the target object feature cluster, it is determined that the third characteristics of objects cluster
Meet the condition of convergence.
Wherein, the model training unit is specifically used for:
The characteristics of objects information and initial object for obtaining the sample object data in the third characteristics of objects cluster identify mould
Sample image similarity between the characteristics of objects information of Standard object data in type;
According in the object tag of the sample object data in the Object identifying model, the third characteristics of objects cluster
The object tag of sample object data and the sample image similarity determine similarity error, and according to the similarity error
Backpropagation adjusts the model parameter of the initial object identification model;
When the adjustment number of the model parameter is equal to adjustment threshold value or the similarity error meets the condition of convergence,
Initial object identification model comprising model parameter adjusted is determined as the Object identifying model.
On the one hand the embodiment of the present invention provides a kind of computer storage medium, the computer storage medium is stored with more
Item instruction, described instruction are suitable for being loaded by processor and executing above-mentioned method and step.
On the one hand the embodiment of the present invention provides a kind of data processing equipment, including processor and memory;Wherein, described
Memory is stored with computer program, and the computer program is suitable for being loaded by the processor and executing following steps:
It obtains sample object data respectively from multiple multi-medium datas, extracts the characteristics of objects of the sample object data
Information;
Multiple sample object data are clustered according to the characteristics of objects information, are obtained multiple with different cluster labels
The first characteristics of objects cluster;Cluster label phase of the object tag of all sample object data with the first affiliated characteristics of objects cluster
Together;
Foreign peoples's object data in each first characteristics of objects cluster is cleared up respectively, the first characteristics of objects cluster after cleaning is true
It is set to the second characteristics of objects cluster;
Similar cluster merging is carried out between multiple second characteristics of objects clusters, generates multiple thirds pair with different cluster labels
As feature cluster, the object tag of the sample object data is updated to the cluster label of the affiliated third characteristics of objects cluster;
Object tag and the object based on the sample object data in each third characteristics of objects cluster are special
Reference breath, training Object identifying model;The Object identifying model object data to be predicted and target object data for identification
Between similarity.
In embodiments of the present invention, by obtaining sample object data respectively from multiple multi-medium datas, described in extraction
The characteristics of objects information of sample object data;Multiple sample object data are clustered according to the characteristics of objects information, are obtained
To multiple the first characteristics of objects clusters with different cluster labels;The object tag of all sample object data is with affiliated first
The cluster label of characteristics of objects cluster is identical;Foreign peoples's object data in each first characteristics of objects cluster is cleared up respectively, after cleaning
First characteristics of objects cluster is determined as the second characteristics of objects cluster;Similar cluster merging is carried out between multiple second characteristics of objects clusters, it is raw
At multiple third characteristics of objects clusters with different cluster labels, by the object tag of the sample object data be updated to belonging to
The cluster label of the third characteristics of objects cluster;Pair based on the sample object data in each third characteristics of objects cluster
As label and the characteristics of objects information, training Object identifying model;Object identifying model object to be predicted for identification
Similarity between data and target object data.By being clustered to sample object data, foreign peoples's object data remove and
Similar cluster merges the sample object data for generating and having label, by largely the sample object data processing of label is not to have label
Sample object data, then using the sample object data training Object identifying model for largely having label, it can be to avoid Object identifying
The model lower problem of the very few discrimination for leading to Object identifying model of training data in the training process, it is a large amount of by obtaining
There are the sample object data of label, improves the discrimination to object data to be predicted.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of system architecture diagram of data processing method provided in an embodiment of the present invention;
Fig. 2 a is a kind of schematic diagram of a scenario of data processing method provided in an embodiment of the present invention;
Fig. 2 b is a kind of schematic diagram of a scenario of data processing method provided in an embodiment of the present invention;
Fig. 2 c is a kind of interface schematic diagram of data processing method provided in an embodiment of the present invention;
Fig. 3 is a kind of flow diagram of data processing method provided in an embodiment of the present invention;
Fig. 4 is the flow diagram of another data processing method provided in an embodiment of the present invention;
Fig. 5 is a kind of example schematic that label is arranged provided in an embodiment of the present invention;
Fig. 6 is the example schematic that a kind of foreign peoples's object data provided in an embodiment of the present invention is deleted;
Fig. 7 is the example schematic that a kind of feature cluster provided in an embodiment of the present invention merges;
Fig. 8 is a kind of example schematic of data processing method provided in an embodiment of the present invention;
Fig. 9 is a kind of structural schematic diagram of data processing equipment provided in an embodiment of the present invention;
Figure 10 is the structural schematic diagram of another data processing equipment provided in an embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
It referring to Figure 1, is a kind of system architecture diagram of data processing provided in an embodiment of the present invention.Server 10f passes through
Interchanger 10e and communication bus 10d and user terminal cluster establish connection, user terminal cluster can include: user terminal 10a,
User terminal 10b ..., user terminal 10c.The multi-medium data of multiple not labels is stored in database 10g.Server
10f extracts the multi-medium data of multiple not labels from database 10g, and sample object number is obtained from multi-medium data
According to each sample object data being divided to corresponding characteristics of objects cluster, and give characteristics of objects by the way of unsupervised learning
Corresponding label is arranged in cluster and sample object data, and label includes the object tag of sample object data and the cluster of characteristics of objects cluster
Label, the object tag of all sample object data are identical as the affiliated cluster label of characteristics of objects cluster.Server 10f is based on
Multiple sample object data for carrying label, by the way of supervised learning, training Object identifying model, and store the object
Identification model.Server 10f can be based on trained Object identifying model inspection object data to be predicted and target object number
Similarity between, when detecting that the similarity between object data to be predicted and target object data is less than threshold value, clothes
Device 10f be engaged in corresponding user terminal transmission standby signal, for prompting user's object data to be predicted and target object data
It is not belonging to the same object;When detecting that the similarity between object data to be predicted and target object data is more than or equal to
When threshold value, server 10f sends standby signal to corresponding user terminal, for prompting user's object data to be predicted and target
Object data belongs to the same object, for example, Object identifying model is human face recognition model, the face unlock for mobile terminal
Password then prompts user people to be predicted when the similarity between human face data to be predicted and target human face data is less than threshold value
Face data and target human face data are not the same face, mobile terminal unlock failure, when human face data to be predicted and target person
When similarity between face data is more than or equal to threshold value, then user's human face data to be predicted and target human face data is prompted to be
The same face, mobile terminal unlock successfully.Or when server 10f receives the inspection of the user terminal in terminal cluster
When surveying request, the corresponding object data to be predicted and target object data of above-mentioned detection request is identified based on Object identifying model
Between similarity whether be greater than threshold value, the result that server 10f will test is sent to corresponding user terminal.Certainly,
Trained Object identifying model can be sent to the user terminal each user terminal in cluster by server 10f, by user's end
Detect the similarity between the target object data of object data to be predicted and storage in the user terminal in end.
Data processing method provided in an embodiment of the present invention can be applied to do not have identity mark largely from reality scene
The sample object data for having label for Object identifying model training are excavated in the sample object data of label, the present invention is implemented
The data processing equipment that example is related to can be headend equipment, comprising: tablet computer, smart phone, PC (PC), notebook
The terminal devices such as computer, palm PC, headend equipment directly handle the sample object data obtained from multiple multi-medium datas,
The data processing equipment can also be server, and server process obtains sample object number from database or headend equipment
According to and handled, processing result is sent to database or headend equipment stores.
Since label is arranged in the sample object data for no label, and training Object identifying model needs are related to largely
Operation, therefore server 10f can use distributed structure/architecture, and data/model is carried out cutting, by multiple working nodes, divide
Cloth Optimized model training, finally model is polymerize, and above procedure can be linux system, Window system or
Application program or system program in person's Unix system etc. etc..It is following by taking the process in linux system as an example, be based on a large amount of nothings
The multi-medium data training Object identifying model of label.Wherein, user terminal may include mobile phone, tablet computer, notebook electricity
Brain, palm PC, intelligent sound, mobile internet device (MID, mobile internet device), POS (Point Of
Sales, point of sale) machine, wearable device (such as smartwatch, Intelligent bracelet etc.) etc..
Wherein, be able to achieve will be a large amount of for the server 10f in above-mentioned Fig. 1 corresponding embodiment or any one user terminal
The sample object data processing of no label is the sample object data with label, to realize based on the sample object with label
Data carry out Object identifying model training and are described by taking server 10f as an example below for ease of description.Please also refer to
Fig. 2 a- Fig. 2 b is a kind of schematic diagram of a scenario of data processing provided in an embodiment of the present invention.Server 10f obtains n without label
Multi-medium data, the object location information in multiple multi-medium datas is obtained, according to the object location information from each more
The subject area with target size is obtained in media data respectively, the picture material in the subject area is determined as sample
Object data, and obtain the characteristics of objects information of each sample object data.
So far, server 10f just obtains multiple sample object data and the corresponding object of each sample object data
Characteristic information.The sample object data are specifically as follows facial image or the image of other objects, next, service
Device 10f calculates the image similarity between multiple sample object data, and specific method is to calculate two sample object data pair
The distance between the characteristics of objects information answered information, i.e. image similarity between sample object data, if image similarity is big
In first threshold, then the two sample object data are divided into identical first characteristics of objects cluster, it should be noted that if it exists
The object tag of sample object data is identical as the object tag of sample object data for being divided to the first characteristics of objects cluster, then will
Sample object data with same object label are also divided to the first characteristics of objects cluster, for example, calculating first sample number of objects
When according to image similarity between the second sample object data, wherein first sample object data is provided with relating attribute, i.e.,
The first sample object data has pursuit path information with other at least one sample object data in video data, then
While first sample object data and the second sample object data are divided to the first characteristics of objects cluster, it can also will be provided with
Other sample object data of relating attribute are also divided to the first characteristics of objects cluster.It should be noted that multiple samples pair
When image data is clustered, since sample object data bulk is huge, cluster is takeed a long time, in order to improve the effect of cluster
Multiple sample object data can be divided into multiple sample object data acquisition systems by the way of Distributed Cluster here by rate,
Multiple sample object data sets are clustered respectively simultaneously, each sample object data set can be generated multiple first pairs
As feature cluster.
It should be noted that in addition to the clustering method of above-mentioned calculating image similarity, it can also be using the poly- of other modes
Class algorithm.
It is finally the third characteristics of objects cluster obtained according to multiple first characteristics of objects clusters for training Object identifying model,
First foreign peoples's object data in each first characteristics of objects cluster is deleted, to generate multiple first characteristics of objects clusters, then basis
Merge request to merge the merging cluster in the second characteristics of objects cluster, generate third characteristics of objects cluster, the merging cluster is to close
And request the indicated at least two second characteristics of objects clusters that can be merged.And then third characteristics of objects cluster can be used
In sample object data training Object identifying model 20a, it is subsequent directly to examine after Object identifying model 20a is trained
It surveys to the similarity between object data to be predicted and target object data.
As shown in Figure 2 b, trained Object identifying model 20a in advance is stored in server.User terminal will be to be predicted
Object data is sent to server for detecting the similarity between target object data.Server obtains number of objects to be predicted
According to characteristics of objects information, based on Object identifying model 20a identification object data to be predicted and target object data between
Similarity is less than threshold value, it is determined that does not pass through for detection, specifically, can be applied to field of face identification, if detecting to pre-
The similarity surveyed between human face data and target human face data is less than threshold value, and the result that server will test is sent to user's end
End, and prompting the user terminal human face data to be predicted is the human face data of stranger.
Fig. 2 c is referred to, is a kind of interface schematic diagram of data processing provided in an embodiment of the present invention.With recognition of face system
For system, monitor supervision platform training human face recognition model, with detect the face in video monitoring with server cluster (including service
Device 1, server 2 ..., the target face in server n) whether be the same person face.As shown in the 20c of interface, Yong Hudian
" performance upgrade " button in interface is hit, monitor supervision platform copies the human face data feature that there is label in part from server cluster
Cluster has the human face data feature cluster of label that can refer to the third pair in above-mentioned Fig. 2 a corresponding embodiment in the server cluster
As feature cluster, increase has the sample object data of label to be trained human face recognition model, human face recognition model can be improved
The accuracy rate of recognition of face.Monitor supervision platform is based on the above-mentioned human face data feature cluster for having label, training face detection model.Face
After detection model training, as shown in the 20d of interface, user clicks " starting to scan " button in interface, and monitor supervision platform can be with
Continuously and uninterruptedly, video data is obtained by camera, detects the target in the face and server cluster in video data
Whether face is the face of the same person, and will test result and shown.User can also be with certain interval of time again from clothes
New data is copied in business device cluster, for updating Face datection model.
Wherein, the first characteristics of objects cluster, the second characteristics of objects cluster and third characteristics of objects fasciation at detailed process can be with
Referring to embodiment corresponding to following figure 3 to Fig. 8.
Below in conjunction with attached drawing 3- attached drawing 8, describe in detail to data processing method provided in an embodiment of the present invention.
Fig. 3 is referred to, for the embodiment of the invention provides a kind of flow diagrams of data processing method.As shown in figure 3,
The embodiment of the present invention the method may include following steps S101- step S105.
S101 obtains sample object data respectively from multiple multi-medium datas, extracts pair of the sample object data
As characteristic information;
Specifically, data processing equipment can be any one in the user terminal cluster in above-mentioned Fig. 1 corresponding embodiment
A user terminal, is specifically as follows: tablet computer, smart phone, PC (PC), laptop, palm PC etc. are eventually
End equipment, alternatively, the data processing equipment may be the server 10f in above-mentioned Fig. 1 corresponding embodiment.Data processing is set
It is standby to obtain sample object data respectively from multiple multi-medium datas, the characteristics of objects information of the sample object data is extracted,
It is understood that the multi-medium data includes image data and video data, the video data includes an at least frame figure
As data, obtaining sample object data respectively from multiple multi-medium datas substantially is the acquisition sample pair from image data
Image data, for example, can obtain sample object data from several pictures either video file, the sample object data are
The image-region of target size in multi-medium data comprising sample object, sample object may include the face of face, animal
Either object may include at least one sample object data in each multi-medium data, can also not have sample object number
According to data processing equipment extracts characteristics of objects information from the sample object data, and the characteristics of objects information is target ruler
The information such as the textural characteristics, shape feature of sample object and spatial relation characteristics in very little image-region, usually to image ash
The characteristics of objects information of sample object data is extracted after degreeization processing using feature extracting method, feature extracting method includes scale
Invariant features convert (Scale-Invariant Feature Transform, SIFT), histograms of oriented gradients (Histogram
Of Oriented Gradient, HOG).
S102 clusters multiple sample object data according to the characteristics of objects information, obtains multiple with difference
First characteristics of objects cluster of cluster label;Cluster of the object tag of all sample object data with the first affiliated characteristics of objects cluster
Label is identical;
Specifically, data processing equipment clusters multiple sample object data according to the characteristics of objects information, obtain
To multiple the first characteristics of objects clusters with different cluster labels;The object tag of all sample object data is with affiliated first
The cluster label of characteristics of objects cluster is identical, it is to be understood that data processing equipment is using clustering algorithm to the multiple sample pair
Image data is clustered, and multiple the first characteristics of objects clusters with different cluster labels are obtained, and the first characteristics of objects cluster is tool
There is the set of the sample object data of identical clustering criteria, includes at least one sample object number in each first characteristics of objects cluster
According to image feature information corresponding with the sample object data, the cluster label is the identification information of the first characteristics of objects cluster,
The object tag is the identification information of sample object data, and the object tag of all sample object data is with affiliated first
The cluster label of characteristics of objects cluster is identical, and the object tag of the sample object data of each first characteristics of objects cluster is all the same, different
The first object data cluster label it is not identical, between the image feature information in this programme using sample object data
Image similarity judges whether two sample object data belong to the same first characteristics of objects cluster as clustering criteria, then counts
The image similarity between the corresponding image feature information of two sample object data is calculated, calculation formula is as follows:
Wherein, S indicates the image similarity between two sample object data, fA=[a0,a1,…,aN], fB=[b0,
b1,…,bN] indicating the feature vectors of two sample object data, N indicates characteristic dimension, and A and B indicate two sample object numbers
According to, if image similarity S >0, then sample object data A and B are divided to identical first characteristics of objects cluster, described first pair
As the cluster label of feature cluster can choose sample object data A and B one of them object tag as cluster label, can also set
The different cluster label of object tag with sample object data A and B is set, then by the object tag of sample object data A and B
It is set as the cluster label of the first characteristics of objects cluster, until all sample object data are divided to the first characteristics of objects cluster,
And set the object tag of all sample object data on the cluster label of the first affiliated characteristics of objects cluster.
S103 clears up foreign peoples's object data in each first characteristics of objects cluster respectively, and the first object after cleaning is special
Sign cluster is determined as the second characteristics of objects cluster;
Specifically, data processing equipment clears up foreign peoples's object data in each first characteristics of objects cluster respectively, will clear up
The first characteristics of objects cluster afterwards is determined as the second characteristics of objects cluster, it is to be understood that foreign peoples's object data is first pair
As the sample object data being labeled in feature cluster, user selects to be not belonging to the first object spy from the first characteristics of objects cluster
The sample object data of cluster are levied, and the sample object data of selection are marked, the labeled sample object data are
Foreign peoples's object data in each first characteristics of objects cluster is deleted respectively, will be deleted by foreign peoples's object data, data processing equipment
The first characteristics of objects cluster after foreign peoples's object data is determined as the second characteristics of objects cluster, and the second characteristics of objects cluster is with phase
With the set of the sample object data of clustering criteria, in each second characteristics of objects cluster include at least one sample object data and
The corresponding image feature information of the sample object data.
S104, carries out similar cluster merging between multiple second characteristics of objects clusters, generates multiple with different cluster labels
The object tag of the sample object data is updated to the cluster of the affiliated third characteristics of objects cluster by third characteristics of objects cluster
Label;
Specifically, data processing equipment carries out similar cluster merging between multiple second characteristics of objects clusters, multiple tools are generated
There is the third characteristics of objects cluster of different cluster labels, the object tag of the sample object data is updated to the affiliated third
The cluster label of characteristics of objects cluster, it is to be understood that multiple sample object data are being carried out by the way of Distributed Cluster
When cluster, since multiple sample object data are divided into multiple sample object data acquisition systems, due to different sample object data
Set is clustered respectively, there may be the first characteristics of objects cluster to be merged in different sample object data acquisition systems,
The the first characteristics of objects cluster for removing foreign peoples's object data is carried out and and can be improved the accuracy rate of cluster, the similar cluster is the
It is labeled one group of second characteristics of objects cluster in two characteristics of objects clusters, includes at least two second characteristics of objects in each group of similar cluster
Cluster, data processing equipment, which merges the second characteristics of objects cluster in each group, generates multiple third objects with different cluster labels
Feature cluster, there may be the similar cluster that multiple groups are labeled in the second characteristics of objects cluster, each group of similar cluster generates after merging
One third characteristics of objects cluster, there may also be the second characteristics of objects cluster is not labeled in the second characteristics of objects cluster, not by
Second characteristics of objects cluster of label merges without similar cluster, and the cluster label of the third characteristics of objects cluster can be combined into the
The cluster label of any one the second characteristics of objects cluster of three characteristics of objects clusters, can also regenerate new cluster label, and by
The object tag of sample object data is updated to the cluster label of the affiliated third characteristics of objects cluster, institute in three characteristics of objects clusters
Stating third characteristics of objects cluster is the set with the sample object data of identical clustering criteria, is wrapped in each third characteristics of objects cluster
Include at least one sample object data and the corresponding image feature information of the sample object data.
S105, object tag based on the sample object data in each third characteristics of objects cluster and described right
As characteristic information, training Object identifying model;The Object identifying model object data and target object to be predicted for identification
Similarity between data.
Specifically, pair of the data processing equipment based on the sample object data in each third characteristics of objects cluster
As label and the characteristics of objects information, training Object identifying model;Object identifying model object to be predicted for identification
Similarity between data and target object data, it is to be understood that data processing equipment is special by each third object
The object tag and the characteristics of objects information for levying the sample object data in cluster obtain Object identifying as input information
The model parameter of model, the Object identifying model are similar between object data to be predicted and target object data for identification
Degree, the sample object data are more, and the model parameter of Object identifying model is more accurate, calculated using the Object identifying model
Prediction object data and target object data between similarity accuracy it is higher.
In embodiments of the present invention, by obtaining sample object data respectively from multiple multi-medium datas, described in extraction
The characteristics of objects information of sample object data;Multiple sample object data are clustered according to the characteristics of objects information, are obtained
To multiple the first characteristics of objects clusters with different cluster labels;The object tag of all sample object data is with affiliated first
The cluster label of characteristics of objects cluster is identical;Foreign peoples's object data in each first characteristics of objects cluster is cleared up respectively, after cleaning
First characteristics of objects cluster is determined as the second characteristics of objects cluster;Similar cluster merging is carried out between multiple second characteristics of objects clusters, it is raw
At multiple third characteristics of objects clusters with different cluster labels, by the object tag of the sample object data be updated to belonging to
The cluster label of the third characteristics of objects cluster;Pair based on the sample object data in each third characteristics of objects cluster
As label and the characteristics of objects information, training Object identifying model;Object identifying model object to be predicted for identification
Similarity between data and target object data.By being clustered to sample object data, foreign peoples's object data remove and
Similar cluster merges the sample object data for generating and having label, by largely the sample object data of label do not generate the sample for having label
This object data, then using the sample object data training Object identifying model for largely having label, it can be to avoid Object identifying mould
The type lower problem of the very few discrimination for leading to Object identifying model of training data in the training process largely has by obtaining
The sample object data of label, improve the discrimination to object data to be predicted.
Fig. 4 is referred to, for the embodiment of the invention provides a kind of flow diagrams of data processing method.As shown in figure 4,
The embodiment of the present invention the method may include following steps S201- step S210.
S201 obtains the object location information in multiple multi-medium datas, according to the object location information from each more
The subject area with target size is obtained in media data respectively;
Specifically, data processing equipment obtains the object location information in multiple multi-medium datas, according to the object position
Confidence breath obtains the subject area with target size respectively from each multi-medium data, it is to be understood that more matchmakers
Volume data includes image data and video data, and the object location information is image district where sample object in multi-medium data
The location information in domain, sample object may include face, animal face either object, according to the object location information from
The subject area with target size is obtained in each multi-medium data respectively, the target size is preset fixed dimension.
Picture material in the subject area is determined as sample object data, obtains each sample pair by S202
The characteristics of objects information of image data;
Specifically, the picture material in the subject area is determined as sample object data by data processing equipment, obtain
The characteristics of objects information of each sample object data, it is to be understood that data processing equipment will be in the subject area
Picture material carry out the affine transformation based on target bearing, the picture material after affine transformation is determined as sample object number
According to the characteristics of objects information of acquisition each sample object data;Object in the sample object data is in the mesh
Orientation is marked, the affine transformation is that a vector space is carried out to once linear transformation and a translation transformation, is transformed to another
A vector space can be adjusted the sample object in sample object data to target bearing, the target by affine transformation
Orientation is preset azimuth information, and the target bearing can be adjusted by the way that the transformation parameter of affine transformation is arranged, below
The detailed process of the face characteristic information of acquisition facial image is described by image data: detecting own in image data
Face, with rectangle surround frame indicate face location;Face location rectangle surround frame in, detect human face five-sense-organ (eyes,
Nose, mouth etc.) location information, and to rectangle surround frame in image carry out affine transformation, by image face adjust
Face is usually adjusted to the direction to face in order to facilitate the extraction and comparison of face characteristic information to target bearing;It presses
According to preset target size, fixed-size face image data is obtained, in order to improve the quality of face image data, meeting
According to the indexs such as human face posture, image fog-level, ambient lighting and coverage extent, the relatively high face of filtering mass
Image data, and feature extracting method is utilized, the face characteristic information of the face image data after obtaining screening.
S203, if the multi-medium data is image data, for the sample object data in described image data
It is randomly provided object tag;If the multi-medium data is video data, the sample object data are detected in the video
Relating attribute will be arranged with the sample object data of the identical pursuit path information in pursuit path information in data, and
For the sample object data with the relating attribute, identical object tag is set;
Specifically, data processing equipment is the institute in described image data if the multi-medium data is image data
It states sample object data and is randomly provided object tag;If the multi-medium data is video data, the sample object is detected
Pursuit path information of the data in the video data sets the sample object data with the identical pursuit path information
Relating attribute is set, and identical object tag is set for the sample object data with the relating attribute, it is to be understood that
Corresponding object tag can be arranged according to the type of multi-medium data in data processing equipment, if the multi-medium data is image
Data, then be randomly provided object tag to the sample object data, and different sample object data correspond to different object marks
Label detect tracking rail of the sample object data in the video data if the multi-medium data is video data
Mark information, the pursuit path information are the change informations of sample object data position in the different frame of video data,
When the variation of sample object data position in the different frame of video data is in threshold range, it is determined that be in threshold value
The object data pursuit path information having the same changed in range, by the sample pair with the identical pursuit path information
Relating attribute is arranged in image data, and the relating attribute is a kind of mark information, having the same for characterizing sample object information
Pursuit path information, and identical object tag is set for the sample object data with the relating attribute.
S204 clusters multiple sample object data according to the characteristics of objects information, obtains multiple with difference
First characteristics of objects cluster of cluster label;Cluster of the object tag of all sample object data with the first affiliated characteristics of objects cluster
Label is identical;
The S204 of the embodiment of the present invention may comprise steps of S2041-S2044:
S2041 obtains first sample object data and the second sample object data, by institute from multiple sample object data
The object tag for stating first sample object data is determined as the first object tag, by the object mark of the second sample object data
Label are determined as the second object tag;First object tag and second object tag be not identical;
Specifically, data processing equipment obtains first sample object data and the second sample from multiple sample object data
The object tag of the first sample object data is determined as the first object tag by object data, by second sample pair
The object tag of image data is determined as the second object tag, it is to be understood that the first sample object data is multiple samples
A sample object data in this object data, the second sample object data be in multiple sample object data with it is described
The different sample object data of the object tag of first sample object data, by the object mark of the first sample object data
Label are determined as the first object tag, and the object tag of the second sample object data is determined as the second object tag.
S2042 obtains the characteristics of objects information and the second sample object data of the first sample object data
The first image similarity between characteristics of objects information;
Specifically, data processing equipment obtain the first sample object data characteristics of objects information and second sample
The first image similarity between the characteristics of objects information of this object data, it is to be understood that data processing equipment obtains institute
The characteristics of objects information of first sample object data and the characteristics of objects information of the second sample object data are stated, calculates first
The first image similarity between sample object data and the second sample object data, calculation formula are as follows:
Wherein, S indicates the first image similarity between two sample object data, fA=[a0,a1,…,aN], fB=
[b0,b1,…,bN] indicating the feature vectors of two sample object data, N indicates characteristic dimension, and A and B indicate two sample objects
Data.
S2043, if the first image similarity be greater than first threshold, by with first object tag at least
One sample object data and at least one sample object data with the second object tag, are divided into identical first object
Feature cluster, and set the first object tag and the second object tag on the cluster label of the first divided characteristics of objects cluster;
Specifically, data processing equipment will have described first if the first image similarity is greater than first threshold
At least one sample object data of object tag and at least one sample object data with the second object tag, are divided into
Identical first characteristics of objects cluster, and the first divided characteristics of objects is set by the first object tag and the second object tag
The cluster label of cluster, it is to be understood that if the first image similarity is greater than first threshold, the first threshold is to set in advance
It sets, data processing equipment then by least one sample object data with first object tag and has the second object
At least one sample object data of label, be divided into identical first characteristics of objects cluster, described to have the first object tag
At least one sample object data can be at least one sample in a sample object data or a characteristics of objects cluster
It is at least one sample object data with identical pursuit path information in object data or video data, it is described to have
At least one sample object data of second object tag can be a sample object data or a characteristics of objects cluster
In be at least one sample object with identical pursuit path information at least one sample object data or video data
Data, and set the first object tag and the second object tag on the cluster label of the first divided characteristics of objects cluster, it is described
The cluster label of first characteristics of objects cluster can be one in the first object tag and the second object tag, or set at random
It sets, please also refer to Fig. 5, for the embodiment of the invention provides a kind of example schematics that label is arranged.As shown in figure 5, sample
Object data 1 and the first image similarity of sample object data 2 are greater than first threshold, then sample object data 1 and sample pair
The cluster of image data 2 generates the first characteristics of objects cluster, and the object tag of the sample object data 1 is A, the sample object data
2 object tag is B, if being arranged by the cluster label that sample object data 1 and sample object data 2 generate the first characteristics of objects cluster
For the object tag A of sample object data 1, then the object tag of sample object data 1 and sample object data 2 is disposed as A,
If being set as sample object data 2 by the cluster label that sample object data 1 and sample object data 2 generate the first characteristics of objects cluster
Object tag B, then the object tag of sample object data 1 and sample object data 2 is disposed as B, if by sample object number
It is randomly set to C according to the cluster label that 1 and sample object data 2 generate the first characteristics of objects cluster, then sample object data 1 and sample
The object tag of object data 2 is disposed as C.
S2044, when all sample object data are divided into the first affiliated characteristics of objects cluster, and by each first
Sample object data in characteristics of objects cluster and the characteristics of objects cluster are stored;
Specifically, data processing equipment is divided into the first affiliated characteristics of objects cluster when all sample object data
When, and the sample object data in each first characteristics of objects cluster and the characteristics of objects cluster are stored, it is possible to understand that
, when all sample object data are divided into the first affiliated characteristics of objects cluster, can determine all samples pair
Image data has carried out cluster operation, then by the sample object number in each first characteristics of objects cluster and the characteristics of objects cluster
According to being stored, the first characteristics of objects cluster can store into the memory space of data processing equipment or store to database
In.
S205 clears up foreign peoples's object data in each first characteristics of objects cluster respectively, and the first object after cleaning is special
Sign cluster is determined as the second characteristics of objects cluster;
The S205 of the embodiment of the present invention may comprise steps of S2051-S2053:
S2051, response are directed to the first marking operation of the first characteristics of objects cluster, and first marking operation is signified
The sample object data in the first characteristics of objects cluster shown, are determined as the corresponding standard object of the first characteristics of objects cluster
Data;
Specifically, data processing equipment response is directed to the first marking operation of the first characteristics of objects cluster, by described the
Sample object data in the first characteristics of objects cluster indicated by one marking operation are determined as the first characteristics of objects cluster
Corresponding Standard object data, it is to be understood that data processing equipment is by preset quantity in the first characteristics of objects cluster
Sample object data are shown on a display screen, obtain the first label behaviour of the sample object data for display on a display screen
Make, first marking operation is the click commands or touching instruction of operator or user on a display screen, will be described
Sample object data indicated by first marking operation are determined as the corresponding Standard object data of the first characteristics of objects cluster,
Labeled sample object data are a sample object number at least one the sample object data of display on a display screen
According to, it should be noted that response is directed to the first marking operation of the first characteristics of objects cluster, and can be terminal device will be described
Sample object data in the first characteristics of objects cluster indicated by first marking operation are determined as first characteristics of objects
The corresponding Standard object data of cluster is also possible to terminal device for the first marking operation and is transmitted to server, and server is directed to institute
The first marking operation is stated, by the sample object data in the first characteristics of objects cluster indicated by first marking operation,
It is determined as the corresponding Standard object data of the first characteristics of objects cluster.
S2052, response are directed to the second marking operation of non-standard object data, will be indicated by second marking operation
Non-standard object data is determined as foreign peoples's object data;The non-standard object data in the first characteristics of objects cluster remove institute
State the sample object data except Standard object data;
Specifically, data processing equipment response is directed to the second marking operation of non-standard object data, described second is marked
The indicated non-standard object data of note operation is determined as foreign peoples's object data;The non-standard object data is described first pair
As the sample object data in feature cluster in addition to the Standard object data, it is to be understood that response is for non-standard right
The data processing equipment of second marking operation of image data can be terminal device, be also possible to server, when data processing is set
When for being terminal device, the Standard object data of the first characteristics of objects cluster is shown and is shown the first of display screen by terminal device
Show region, shows that the non-standard object data of preset quantity, data processing equipment obtain needle in the second display area of display screen
To the second marking operation for showing the non-standard object data of the second display area on a display screen.When data processing equipment is clothes
Be engaged in device when, the sample object data of the first characteristics of objects cluster are sent to terminal device by server, and terminal device is by standard
Object data is shown in the first display area of display screen, and the non-standard object data of preset quantity is shown the of display screen
Two display areas, server obtain the non-standard object data for the second display area on a display screen that terminal device is sent
The second marking operation, second marking operation be the click commands or touch of operator or user on a display screen
Instruction, is determined as foreign peoples's object data, labeled sample for non-standard object data indicated by second marking operation
Object data is at least one the non-standard object data of display on a display screen, is the embodiment of the present invention please also refer to Fig. 6
Provide a kind of example schematic that foreign peoples's object data is deleted.As shown in fig. 6, the display screen is for showing that the first object is special
The sample object data in cluster are levied, the Standard object data of the first characteristics of objects cluster is shown in the first viewing area of display screen
Domain 100, the non-standard object data of the first characteristics of objects cluster are shown in the second display area 200 of display screen, and described
Two display areas 200 can show the non-standard object data of at least one the first characteristics of objects cluster, obtain aobvious for being shown in
Second marking operation of the non-standard object data of the second display area in display screen, if operator or user think described
One in two display areas 200 or the Standard object data in multiple non-standard object datas and the first display area 100
It is not belonging to the same first characteristics of objects cluster, then obtains operator or user and is directed to the second display area on a display screen
Second marking operation of one in 200 or multiple non-standard object datas, and will be indicated by second marking operation
Non-standard object data is determined as foreign peoples's object data of the first characteristics of objects cluster.
S2053 deletes foreign peoples's object data from the first characteristics of objects cluster, will delete foreign peoples's object data
The first characteristics of objects cluster be determined as the second characteristics of objects cluster;
Specifically, data processing equipment deletes foreign peoples's object data from the first characteristics of objects cluster, will delete
Except the first characteristics of objects cluster of foreign peoples's object data is determined as the second characteristics of objects cluster, it is to be understood that data processing equipment
Foreign peoples's object data in first characteristics of objects cluster is deleted, the first characteristics of objects cluster after deletion foreign peoples's object data is determined
For the second characteristics of objects cluster, the sample object data in the second characteristics of objects cluster are less than or equal to the first characteristics of objects cluster
In sample object data, include at least one sample object data and the sample object number in each second characteristics of objects cluster
According to corresponding image feature information.
S206, carries out similar cluster merging between multiple second characteristics of objects clusters, generates multiple with different cluster labels
The object tag of the sample object data is updated to the cluster of the affiliated third characteristics of objects cluster by third characteristics of objects cluster
Label;
The S206 of the embodiment of the present invention may comprise steps of S2061-S2064:
S2061 obtains the first detection feature cluster and the second detection feature cluster from multiple second characteristics of objects clusters;
Specifically, data processing equipment obtains the first detection feature cluster and the second detection from multiple second characteristics of objects clusters
Feature cluster, it is to be understood that the first detection feature cluster is a characteristics of objects cluster in multiple second characteristics of objects clusters,
The second detection feature cluster is to detect the different characteristics of objects of feature cluster with first in multiple second characteristics of objects clusters
Cluster.
S2062, response are based on institute for the merging request of the first detection feature cluster and the second detection feature cluster
State the first quantity for merging the sample object data in the first detection feature cluster described in request and the second detection feature
Second quantity of the sample object data in cluster;
Specifically, conjunction of the data processing equipment response for the first detection feature cluster and the second detection feature cluster
And request, based on the first quantity of sample object data in the first detection feature cluster described in the merging request and described
Second quantity of the sample object data in the second detection feature cluster, it is to be understood that data processing equipment is by described first
The Standard object data of detection feature cluster is shown in the first display area of display screen, by the standard of the second detection feature cluster
Object data is shown in the second display area of display screen, obtains for the standard for showing the first detection feature cluster on a display screen
The merging request of the Standard object data of object data and the second detection feature cluster, the merging request is operator or use
The click commands or touching instruction of family on a display screen, based on the sample merged in the first detection feature cluster described in request
Second quantity of the first quantity of object data and the sample object data in the second detection feature cluster, needs to illustrate
It is that response can be terminal device and obtain for the merging request of the first detection feature cluster and the second detection feature cluster
It takes merging to request, and executes for the merging request based on the sample pair merged in the first detection feature cluster described in request
The step of second quantity of the first quantity of image data and the sample object data in the second detection feature cluster, response merges
Request, is also possible to the merging request that server obtains terminal device transmission, and server is executed for merging request and is based on
The first quantity and described second for merging the sample object data in the first detection feature cluster described in request detects feature cluster
In sample object data the second quantity the step of, please also refer to Fig. 7, for the embodiment of the invention provides a kind of feature clusters
Combined example schematic.As shown in fig. 7, the display screen is used to show the Standard object data in detection feature cluster, it is described
The Standard object data of first detection feature cluster is shown in the first display area 300 of display screen, the second detection feature cluster
Standard object data be shown in the second display area 400 of display screen, if operator or user are special according to the first detection
The Standard object data of the Standard object data and the second detection feature cluster of levying cluster thinks the first detection feature cluster and the first detection
Feature cluster can be merged with similar cluster, then the click for obtaining the merging request of the triggering of operator or user on a display screen refers to
Order or touching instruction, and obtain the first quantity and second inspection of the sample object data in the first detection feature cluster
Survey the second quantity of the sample object data in feature cluster.
S2063 is detected in feature cluster according to each sample object data and described second in the first detection feature cluster
The image similarity of each sample object data, first quantity and second quantity obtain the first detection feature
The classification similarity of cluster and the second detection feature cluster;
Specifically, data processing equipment is according to each sample object data and described second in the first detection feature cluster
Image similarity, first quantity and second quantity for detecting each sample object data in feature cluster, described in acquisition
The classification similarity of first detection feature cluster and the second detection feature cluster, it is to be understood that data processing equipment calculates
Each sample object data and described second detect each sample object data in feature cluster in the first detection feature cluster
Image similarity, according to image similarity, first quantity and second quantity, obtain the first detection feature cluster and
The classification similarity of the second detection feature cluster, the classification calculating formula of similarity are as follows:
M=P12/(N1*N2)
Wherein, M is the classification similarity of the first detection feature cluster and the second detection feature cluster, N1,N2Respectively indicate the first inspection
Survey the quantity of feature cluster and the corresponding sample object data of the second detection feature cluster, P12Indicate the first detection feature cluster and the second inspection
Survey the quantity that image similarity in feature cluster is greater than the sample object data pair of similarity threshold.
S2064, if the classification similarity is greater than second threshold, by the first detection feature cluster and the second detection feature cluster
Merge into third characteristics of objects cluster, and by the object of the sample object data in the first detection feature cluster and the second detection feature cluster
Label is set as the cluster label of the affiliated third characteristics of objects cluster;
Specifically, if the classification similarity be greater than second threshold, data processing equipment by first detection feature cluster and
Second detection feature cluster merges into third characteristics of objects cluster, and the first detection feature cluster and second are detected to the sample in feature cluster
The object tag of object data is set as the cluster label of the affiliated third characteristics of objects cluster, it is to be understood that if described
Classification similarity is greater than second threshold, and the first detection feature cluster and the second detection feature cluster are then merged into the by data processing equipment
Three characteristics of objects clusters, the sample object data in the third characteristics of objects cluster include the sample object in the first detection feature cluster
Data and second detection feature cluster in sample object data, and by first detection feature cluster and second detection feature cluster in sample
The object tag of this object data is set as the cluster label of the affiliated third characteristics of objects cluster, the third characteristics of objects cluster
Cluster label can be the first detection feature cluster or second detection feature cluster cluster label, the cluster label of third characteristics of objects cluster
Can arbitrarily it be arranged.
The third characteristics of objects cluster is determined as recycling characteristics of objects cluster by S207;It is special that each circulation object is cleared up respectively
Levy foreign peoples's object data in cluster;Circulation characteristics of objects cluster after deletion foreign peoples's object data is subjected to similar cluster merging, is obtained
The third characteristics of objects cluster;
S208, whether the third characteristics of objects cluster meets the condition of convergence or similar cluster merges number and meets time
Number threshold value;
When the third characteristics of objects cluster meets the condition of convergence or similar cluster merges number equal to frequency threshold value, execute
Step S209;When the third characteristics of objects cluster is unsatisfactory for the condition of convergence and similar cluster merges number less than frequency threshold value, weight
Step S207 is executed again.
Specifically, the third characteristics of objects cluster is determined as recycling characteristics of objects cluster by data processing equipment;It clears up respectively
Foreign peoples's object data in each circulation characteristics of objects cluster;The method and the method phase in step S205 for clearing up foreign peoples's object data
Together, then by the circulation characteristics of objects cluster after deletion foreign peoples's object data similar cluster merging is carried out, obtains the third characteristics of objects
Cluster, the method for carrying out similar cluster merging to circulation characteristics of objects cluster are identical as the method in step S206;It should be noted that clear
Reason recycles foreign peoples's object data in characteristics of objects cluster and can execute to the similar cluster merging of circulation characteristics of objects cluster progress more
Secondary, during carrying out repeatedly similar cluster merging, similar cluster merging needs to judge classification similarity every time, works as progress
When the classification similarity of the first detection feature cluster and the second detection feature cluster that merge is greater than second threshold, can just it carry out similar
Cluster merges, and during multiple similar cluster merges, the second threshold in each merging process can be identical, can also be according to same
The number of iterations that class cluster merges judges, for example, the second threshold that similar cluster merges according to the corresponding threshold value of the number of iterations is
[T0,T1,T2,…,TN], wherein N indicates the number of iterations that similar cluster merges, specifically, merging when carrying out second of similar cluster
When, second threshold T2, when the classification similarity of the first detection feature cluster merged and the second detection feature cluster is greater than T2
When, the first detection feature cluster and the second detection feature cluster are merged into third characteristics of objects cluster.When special for the third object
When sign cluster meets the condition of convergence or similar cluster and merges number and meet frequency threshold value condition, cleaning circulation characteristics of objects is no longer executed
Foreign peoples's object data in cluster and similar cluster merging is carried out to circulation characteristics of objects cluster, executes step S209, the convergence item
Part is that foreign peoples's object data and similar cluster are not present in the third characteristics of objects cluster.
Specifically, during determining whether the third characteristics of objects cluster meets the condition of convergence, data processing equipment
The second figure between each third characteristics of objects cluster Plays object data and non-standard object data can be obtained respectively
As similarity, the corresponding similarity mean value of each third characteristics of objects cluster is obtained according to second image similarity
With similarity variance;Similarity mean value is less than mean value threshold value and similarity variance is special greater than the third object of variance threshold values
Sign cluster is determined as target object feature cluster, it is to be understood that it is special that data processing equipment calculates separately each third object
The second image similarity between cluster Plays object data and non-standard object data is levied, calculation formula is as follows:
Wherein, S indicates the image similarity between two sample object data, fA=[a0,a1,…,aN], fB=[b0,
b1,…,bN] indicating the feature vectors of two sample object data, N indicates characteristic dimension, and A and B indicate two sample object numbers
According to, and obtain the similarity mean value and similarity variance of the second image similarity in each third characteristics of objects cluster;By similarity
Mean value, which is less than mean value threshold value and similarity variance and is greater than the third characteristics of objects clusters of variance threshold values, is determined as target object spy
Cluster is levied, the mean value threshold value and variance threshold values are to preset, and the target object feature cluster is in third characteristics of objects cluster
The relatively large characteristics of objects cluster of difference between sample object data.
Obtain the testing result for being directed to the target object feature cluster;If the testing result is the target object feature
Foreign peoples's object data and similar cluster are not present in cluster, it is determined that the third characteristics of objects cluster meets the condition of convergence;
Specifically, data processing equipment obtains the testing result for being directed to the target object feature cluster;If the detection knot
Fruit is that foreign peoples's object data and similar cluster are not present in the target object feature cluster, it is determined that the third characteristics of objects cluster is full
The sufficient condition of convergence, it is to be understood that data processing equipment is by the sample object number of at least one target object feature cluster
On a display screen according to display, it obtains for the testing result for showing target object feature cluster on a display screen, the testing result
For the click commands or touching instruction of operator or user on a display screen, if the testing result is the target pair
As foreign peoples's object data and similar cluster are not present in feature cluster, it is determined that the third characteristics of objects cluster meets the condition of convergence, if
The testing result is that there are foreign peoples's object data or similar clusters in the target object feature cluster, it is determined that the third object
Feature cluster is unsatisfactory for the condition of convergence.
S209, object tag based on the sample object data in each third characteristics of objects cluster and described right
As characteristic information, training Object identifying model;The Object identifying model object data and target object to be predicted for identification
Similarity between data.
Specifically, data processing equipment obtains the characteristics of objects letter of the sample object data in the third characteristics of objects cluster
Sample image similarity between breath and the characteristics of objects information of the Standard object data in initial object identification model, according to institute
State the object tag of the sample object data in Object identifying model, sample object data in the third characteristics of objects cluster
Object tag and the sample image similarity determine similarity error, and adjust institute according to the similarity error back propagation
The model parameter for stating initial object identification model, when the adjustment number of the model parameter is equal to adjustment threshold value or described similar
When degree error meets the condition of convergence, the initial object identification model comprising model parameter adjusted is determined as the object and is known
Other model, the condition of convergence are that the similarity error is less than similarity error threshold.
S210 obtains multi-medium data to be predicted;It is obtained according to the object location information in the multi-medium data to be predicted
The targeted object region with target size is taken, the picture material to be predicted in the targeted object region is carried out based on described
Picture material to be predicted after affine transformation is determined as object data to be predicted by the affine transformation of target bearing;Based on object
Identification model obtains the similarity between object data to be predicted and target object data.
Specifically, data processing equipment obtains multi-medium data to be predicted;According in the multi-medium data to be predicted
Object location information obtains the targeted object region with target size, will be in the image to be predicted in the targeted object region
Appearance carries out the affine transformation based on the target bearing, and the picture material to be predicted after affine transformation is determined as object to be predicted
Data;The similarity between object data to be predicted and target object data is obtained based on Object identifying model, it is possible to understand that
It is that the multi-medium data to be predicted is the image data or video data of not label, the object location information is more
The location information of image-region where sample object in media data, sample object may include face, animal face or
It is object, obtains the subject area with target size from multi-medium data to be predicted according to the object location information, it will
Picture material in the subject area carries out the affine transformation based on target bearing, will be in the image to be predicted after affine transformation
Appearance is determined as object data to be predicted, the characteristic information to be predicted of the object data to be predicted is obtained, using pair after training
As identification model obtains the similarity between object data to be predicted and target object data, the object data to be predicted is judged
It whether is the same sample object with the target object data, the target object data is stored in advance in memory space
Sample object data.
In embodiments of the present invention, by obtaining sample object data respectively from multiple multi-medium datas, described in extraction
The characteristics of objects information of sample object data;Multiple sample object data are clustered according to the characteristics of objects information, are obtained
To multiple the first characteristics of objects clusters with different cluster labels;The object tag of all sample object data is with affiliated first
The cluster label of characteristics of objects cluster is identical;Foreign peoples's object data in each first characteristics of objects cluster is cleared up respectively, after cleaning
First characteristics of objects cluster is determined as the second characteristics of objects cluster;Similar cluster merging is carried out between multiple second characteristics of objects clusters, it is raw
At multiple third characteristics of objects clusters with different cluster labels, by the object tag of the sample object data be updated to belonging to
The cluster label of the third characteristics of objects cluster;Pair based on the sample object data in each third characteristics of objects cluster
As label and the characteristics of objects information, training Object identifying model;Object identifying model object to be predicted for identification
Similarity between data and target object data.By being clustered to sample object data, foreign peoples's object data remove and
Similar cluster merges the sample object data for generating and having label, by largely the sample object data of label do not generate the sample for having label
This object data, then using the sample object data training Object identifying model for largely having label, it can be to avoid Object identifying mould
The type lower problem of the very few discrimination for leading to Object identifying model of training data in the training process, by sample object number
According to multiple foreign peoples's object data remove and similar cluster merge, and sampling check, improve the standard of sample object data clusters
True rate further improves Object identifying model to the discrimination of object data to be predicted.
A kind of data processing method provided in an embodiment of the present invention is illustrated below with reference to specific implement scene.Such as
Shown in Fig. 8, data processing method provided in an embodiment of the present invention can be applied to do not have identity mark largely from reality scene
The sample object data for having label for Object identifying model training are excavated in the sample object data of label, below not have
The human face data generation of label has the human face data of label to a kind of data processing method progress provided in an embodiment of the present invention
Illustrate, the invention mainly comprises data prediction, non-same people to clear up, with people's merging and four processes of quality testing, specifically,
Data processing equipment obtains human face data respectively from multiple multi-medium datas, and multi-medium data includes image data and video counts
According to extracting face characteristic information from human face data, and cluster to multiple human face datas according to face characteristic information, obtain
Multiple the first face characteristic clusters with different cluster labels, clear up the heterogeneous data in each first face characteristic cluster respectively, will
Cleaning heterogeneous data after the first face characteristic cluster be determined as the second face characteristic cluster, between multiple second face characteristic clusters into
The similar cluster of row merges, and generates multiple third face characteristic clusters with different cluster labels, and more by the face label of human face data
It is newly the cluster label of the affiliated third face characteristic cluster, using the face of the human face data in third face characteristic cluster
Label and the face characteristic information, training human face recognition model, human face recognition model can be used for identifying face number to be predicted
According to the similarity between target human face data.
Wherein, data prediction process are as follows:
By the local server for being stored in space abundance without label human face data in multi-medium data, then carry out original
Video cuts the pretreatment such as frame, image format conversion, detects all faces in multi-medium data, indicates face position with rectangle frame
It sets, obtains the subject area with target size, the target respectively from each multi-medium data according to the face location
Size is preset fixed dimension, for example, can be cut out 1 times and 2 times of sizes according to pre-specified training image size
Image corresponding face label is arranged according to the type of multi-medium data, if the multi-medium data be image data,
Face label is randomly provided to the human face data, different sample object data correspond to different face labels, if described more
Media data is video data, then detects pursuit path information of the sample object data in the video data, will have
There is the sample object data setting relating attribute of the identical pursuit path information, the relating attribute is a kind of mark information,
And for the human face data with the relating attribute, identical face label is set, and according to human face posture, image fog-level,
The indexs such as ambient lighting and coverage extent, the relatively high facial image of filtering mass, by filtered facial image and
Corresponding face characteristic information is uploaded to distal memory cell, using human face recognition model, extracts the face characteristic letter of facial image
Breath, clusters multiple face image datas according to face characteristic information, obtains multiple the first with different cluster labels
Face feature cluster.
Wherein, non-same people's clean-up process are as follows:
Remove the heterogeneous data in the first face characteristic cluster, heterogeneous data be not belonging in the first face characteristic cluster it is the first
The facial image of face feature cluster, specific method be the identical facial image of cluster labels is shown on mobile terminal display screen, and
Finally it is confirmed whether it is the same person with professional mark personnel.Data processing equipment responds professional mark personnel and is directed to first
Facial image indicated by first marking operation is determined as described the first by the first marking operation of face characteristic cluster
The corresponding standard faces image of face feature cluster, response are directed to the second marking operation of non-standard facial image, described second are marked
The indicated non-standard object data of note operation is determined as heterogeneous data, and foreign peoples's object data is special from first face
It is deleted in sign cluster, the first face characteristic cluster for deleting heterogeneous data is determined as the second face characteristic cluster.
If the removing heterogeneous data initiated for the first time, server is directly by cluster result information update to database;If
It is this process of iteration, server is inquired first with the presence or absence of new cluster result information (according to time inquiring), if it does, clothes
Being engaged in device will be according to new clustering information more new database corresponding field state, and completion database positioning pushes image to movement after updating
The display screen of terminal.
Wherein, merge process with people are as follows:
Merge the similar cluster in the second face characteristic cluster, specific method is at display screen display two of mobile terminal
The standard faces image of two face characteristic clusters, and finally it is confirmed whether it is the same person with professional mark personnel, at data
Reason equipment obtains the first detection feature cluster and the second detection feature cluster from multiple second face characteristic clusters, responds profession mark people
Member is based on merging request institute for the merging request of the first detection feature cluster and the second detection feature cluster
State the facial image in the first detection feature cluster the first quantity and it is described second detection feature cluster in facial image second
Quantity, according to each facial image in each facial image in the first detection feature cluster and the second detection feature cluster
Image similarity, the first quantity and the second quantity obtain the class of the first detection feature cluster and the second detection feature cluster
Other similarity merges the first detection feature cluster and the second detection feature cluster if the classification similarity is greater than second threshold
For third face characteristic cluster.
If first initiate to merge similar cluster, server is directly by cluster result information update to database;If it is repeatedly
For this process, server-side is inquired first with the presence or absence of new cluster result information, if it does, server-side will be according to new cluster letter
The more new database corresponding field state of breath completes the display screen that image to mobile terminal is pushed after database positioning updates.
After the same people for completing all categories pair merges, amalgamation result is updated to database by server, and activates quality
Testing process.
Wherein, quality inspection process are as follows:
Quality inspection process includes non-same people and examines two options with hostage, and non-same hostage's inspection is detection third face characteristic cluster
In whether there is heterogeneous data, be in the multiple third face characteristic clusters of detection with the presence or absence of annexable similar cluster with hostage's inspection,
Data processing equipment is obtained respectively between each third face characteristic cluster Plays facial image and non-standard facial image
Image similarity, the corresponding similarity mean value of each third face characteristic cluster and phase are obtained according to image similarity
Like degree variance, similarity mean value is less than mean value threshold value and similarity variance is greater than the third face characteristic cluster of variance threshold values
It is determined as needing to carry out the target face characteristic cluster of quality inspection, the screening technique of objectives face characteristic cluster is according to mean value to
Three face characteristic clusters carry out ascending order arrangement, or carry out descending arrangement to third face characteristic cluster according to variance, before sequence
10% third face characteristic cluster is determined as target face characteristic cluster, then obtains the detection for being directed to the target face characteristic cluster
As a result, if the testing result is that foreign peoples's object data and similar cluster are not present in the target object feature cluster, based on every
The face label and face characteristic information of facial image in a third face characteristic cluster, training human face recognition model, if described
Testing result is that there are foreign peoples's object data or similar clusters in the target object feature cluster, then carry out again non-same people cleaning,
Merge with people.
In embodiments of the present invention, by obtaining sample object data respectively from multiple multi-medium datas, described in extraction
The characteristics of objects information of sample object data;Multiple sample object data are clustered according to the characteristics of objects information, are obtained
To multiple the first characteristics of objects clusters with different cluster labels;The object tag of all sample object data is with affiliated first
The cluster label of characteristics of objects cluster is identical;Foreign peoples's object data in each first characteristics of objects cluster is cleared up respectively, after cleaning
First characteristics of objects cluster is determined as the second characteristics of objects cluster;Similar cluster merging is carried out between multiple second characteristics of objects clusters, it is raw
At multiple third characteristics of objects clusters with different cluster labels, by the object tag of the sample object data be updated to belonging to
The cluster label of the third characteristics of objects cluster;Pair based on the sample object data in each third characteristics of objects cluster
As label and the characteristics of objects information, training Object identifying model;Object identifying model object to be predicted for identification
Similarity between data and target object data.By being clustered to sample object data, foreign peoples's object data remove and
Similar cluster merges the sample object data for generating and having label, by largely the sample object data of label do not generate the sample for having label
This object data, then using the sample object data training Object identifying model for largely having label, it can be to avoid Object identifying mould
The type lower problem of the very few discrimination for leading to Object identifying model of training data in the training process largely has by obtaining
The sample object data of label, improve the discrimination to object data to be predicted.
Below in conjunction with attached drawing 9- attached drawing 10, describe in detail to data processing equipment provided in an embodiment of the present invention.
It should be noted that the attached equipment shown in Fig. 10 of attached drawing 9-, the method for executing Fig. 1-embodiment illustrated in fig. 8 of the present invention, are
Convenient for explanation, only parts related to embodiments of the present invention are shown, disclosed by specific technical details, please refers to the present invention
1- embodiment shown in Fig. 8.
Fig. 9 is referred to, for the embodiment of the invention provides a kind of structural schematic diagrams of data processing equipment.As shown in figure 9,
The data processing equipment 1 of the embodiment of the present invention may include: characteristic acquisition unit 11, fisrt feature cluster acquiring unit
12, second feature cluster acquiring unit 13, third feature cluster acquiring unit 14, model training unit 15, object data recognition unit
16, feature cluster recycles acquiring unit 17, detection unit 18, testing result acquiring unit 19.
Characteristic acquisition unit 11 extracts institute for obtaining sample object data respectively from multiple multi-medium datas
State the characteristics of objects information of sample object data;
Fisrt feature cluster acquiring unit 12, for being gathered according to the characteristics of objects information to multiple sample object data
Class obtains multiple the first characteristics of objects clusters with different cluster labels;The object tag of all sample object data with it is affiliated
The first characteristics of objects cluster cluster label it is identical;
Second feature cluster acquiring unit 13, foreign peoples's object data for being cleared up in each first characteristics of objects cluster respectively,
The first characteristics of objects cluster after cleaning is determined as the second characteristics of objects cluster;
Third feature cluster acquiring unit 14 is generated for carrying out similar cluster merging between multiple second characteristics of objects clusters
The object tag of the sample object data is updated to affiliated institute by multiple third characteristics of objects clusters with different cluster labels
State the cluster label of third characteristics of objects cluster;
Model training unit 15, for pair based on the sample object data in each third characteristics of objects cluster
As label and the characteristics of objects information, training Object identifying model;Object identifying model object to be predicted for identification
Similarity between data and target object data.
Fig. 9 is referred to, the characteristic acquisition unit 11 of the embodiment of the present invention may include: that location information obtains son
Unit 111, object data obtain subelement 112, subelement 113 is arranged in object tag.
Location information obtains subelement 111, for obtaining the object location information in multiple multi-medium datas, according to described
Object location information obtains the subject area with target size respectively from each multi-medium data;
Object data obtains subelement 112, for the picture material in the subject area to be determined as sample object number
According to the characteristics of objects information of acquisition each sample object data;
Subelement 113 is arranged in object tag, if being image data for the multi-medium data, for described image data
In the sample object data be randomly provided object tag;If the multi-medium data is video data, the sample is detected
Pursuit path information of this object data in the video data, by the sample object with the identical pursuit path information
Relating attribute is arranged in data, and identical object tag is arranged for the sample object data with the relating attribute;
The fisrt feature cluster acquiring unit 12 of the embodiment of the present invention is specifically used for:
It, will be described for obtaining first sample object data and the second sample object data from multiple sample object data
The object tag of first sample object data is determined as the first object tag, by the object tag of the second sample object data
It is determined as the second object tag;First object tag and second object tag be not identical;
For obtaining the characteristics of objects information of the first sample object data and pair of the second sample object data
As the first image similarity between characteristic information;
If being greater than first threshold for the first image similarity, by least one with first object tag
It is special to be divided into identical first object for a sample object data and at least one sample object data with the second object tag
Cluster is levied, and sets the first object tag and the second object tag on the cluster label of the first divided characteristics of objects cluster;
For when all sample object data are divided into the first affiliated characteristics of objects cluster, and by each first pair
As the sample object data in feature cluster and the characteristics of objects cluster are stored;
The second feature cluster acquiring unit 13 of the embodiment of the present invention is specifically used for:
It, will be indicated by first marking operation for responding the first marking operation for being directed to the first characteristics of objects cluster
The first characteristics of objects cluster in sample object data, be determined as the corresponding standard object number of the first characteristics of objects cluster
According to;
It, will be non-indicated by second marking operation for responding the second marking operation for being directed to non-standard object data
Standard object data is determined as foreign peoples's object data;The non-standard object data is in the first characteristics of objects cluster except described
Sample object data except Standard object data;
For deleting foreign peoples's object data from the first characteristics of objects cluster, foreign peoples's object data will be deleted
First characteristics of objects cluster is determined as the second characteristics of objects cluster;
The third feature cluster acquiring unit 14 of the embodiment of the present invention is specifically used for:
For obtaining the first detection feature cluster and the second detection feature cluster from multiple second characteristics of objects clusters;
For responding the merging request for being directed to the first detection feature cluster and the second detection feature cluster, based on described
The first quantity and described second for merging the sample object data in the first detection feature cluster described in request detects feature cluster
In sample object data the second quantity;
For according to every in each sample object data in the first detection feature cluster and the second detection feature cluster
The image similarity of a sample object data, first quantity and second quantity obtain the first detection feature cluster
With the classification similarity of the second detection feature cluster;
If being greater than second threshold for the classification similarity, the first detection feature cluster and the second detection feature cluster are closed
And be third characteristics of objects cluster, and by first detection feature cluster and second detection feature cluster in sample object data object mark
Label are set as the cluster label of the affiliated third characteristics of objects cluster;
Feature cluster recycles acquiring unit 17, recycles characteristics of objects cluster for the third characteristics of objects cluster to be determined as;Point
Foreign peoples's object data in each circulation characteristics of objects cluster is not cleared up;By the circulation characteristics of objects cluster after deletion foreign peoples's object data
Similar cluster merging is carried out, the third characteristics of objects cluster is obtained;Meet the condition of convergence when being directed to the third characteristics of objects cluster, or
When similar cluster merges number equal to frequency threshold value, object data recognition unit 16 is triggered;
Detection unit 18, for obtain respectively each third characteristics of objects cluster Plays object data with it is non-standard right
The second image similarity between image data obtains each third characteristics of objects cluster point according to second image similarity
Not corresponding similarity mean value and similarity variance;Similarity mean value is less than mean value threshold value and similarity variance is greater than variance threshold
The third characteristics of objects cluster of value is determined as target object feature cluster;
Testing result acquiring unit 19, for obtaining the testing result for being directed to the target object feature cluster;If the inspection
Surveying result is that foreign peoples's object data and similar cluster are not present in the target object feature cluster, it is determined that the third characteristics of objects
Cluster meets the condition of convergence;
Model training unit 15, for pair based on the sample object data in each third characteristics of objects cluster
As label and the characteristics of objects information, training Object identifying model;Object identifying model object to be predicted for identification
Similarity between data and target object data.
Object data recognition unit 16, for obtaining multi-medium data to be predicted;According to the multi-medium data to be predicted
In object location information obtain have target size targeted object region, by the figure to be predicted in the targeted object region
As content carries out the affine transformation based on the target bearing, the picture material to be predicted after affine transformation is determined as to be predicted
Object data;The similarity between object data to be predicted and target object data is obtained based on Object identifying model.
In embodiments of the present invention, by obtaining sample object data respectively from multiple multi-medium datas, described in extraction
The characteristics of objects information of sample object data;Multiple sample object data are clustered according to the characteristics of objects information, are obtained
To multiple the first characteristics of objects clusters with different cluster labels;The object tag of all sample object data is with affiliated first
The cluster label of characteristics of objects cluster is identical;Foreign peoples's object data in each first characteristics of objects cluster is cleared up respectively, after cleaning
First characteristics of objects cluster is determined as the second characteristics of objects cluster;Similar cluster merging is carried out between multiple second characteristics of objects clusters, it is raw
At multiple third characteristics of objects clusters with different cluster labels, by the object tag of the sample object data be updated to belonging to
The cluster label of the third characteristics of objects cluster;Pair based on the sample object data in each third characteristics of objects cluster
As label and the characteristics of objects information, training Object identifying model;Object identifying model object to be predicted for identification
Similarity between data and target object data.By being clustered to sample object data, foreign peoples's object data remove and
Similar cluster merges the sample object data for generating and having label, by largely the sample object data of label do not generate the sample for having label
This object data, then using the sample object data training Object identifying model for largely having label, it can be to avoid Object identifying mould
The type lower problem of the very few discrimination for leading to Object identifying model of training data in the training process, by sample object number
According to multiple foreign peoples's object data remove and similar cluster merge, and sampling check, improve the standard of sample object data clusters
True rate further improves Object identifying model to the discrimination of object data to be predicted.
The embodiment of the invention also provides a kind of computer storage medium, the computer storage medium can store more
Item instruction, described instruction are suitable for being loaded by processor and being executed the method and step such as above-mentioned Fig. 1-embodiment illustrated in fig. 8, specifically hold
Row process may refer to Fig. 1-embodiment illustrated in fig. 8 and illustrate, herein without repeating.
Referring to Figure 10, for the embodiment of the invention provides a kind of structural schematic diagrams of data processing equipment.Such as Figure 10 institute
Show, the data processing equipment 1000 may include: at least one processor 1001, such as CPU, at least one network interface
1004, user interface 1003, memory 1005, at least one communication bus 1002.Wherein, communication bus 1002 is for realizing this
Connection communication between a little components.Wherein, user interface 1003 may include display screen (Display), optional user interface
1003 can also include standard wireline interface and wireless interface.Network interface 1004 optionally may include that the wired of standard connects
Mouth, wireless interface (such as WI-FI interface).Memory 1005 can be high speed RAM memory, be also possible to non-labile storage
Device (non-volatile memory), for example, at least a magnetic disk storage.Memory 1005 optionally can also be at least one
A storage device for being located remotely from aforementioned processor 1001.As shown in Figure 10, the memory as a kind of computer storage medium
It may include operating system, network communication module, Subscriber Interface Module SIM and data process application in 1005.
In equipment 1000 shown in Fig. 10, user interface 1003 is mainly used for providing the interface of input for user, obtains
The data of user's input;And processor 1001 can be used for calling the data process application stored in memory 1005, and
It is specific to execute following operation:
It obtains sample object data respectively from multiple multi-medium datas, extracts the characteristics of objects of the sample object data
Information;
Multiple sample object data are clustered according to the characteristics of objects information, are obtained multiple with different cluster labels
The first characteristics of objects cluster;Cluster label phase of the object tag of all sample object data with the first affiliated characteristics of objects cluster
Together;
Foreign peoples's object data in each first characteristics of objects cluster is cleared up respectively, the first characteristics of objects cluster after cleaning is true
It is set to the second characteristics of objects cluster;
Similar cluster merging is carried out between multiple second characteristics of objects clusters, generates multiple thirds pair with different cluster labels
As feature cluster, the object tag of the sample object data is updated to the cluster label of the affiliated third characteristics of objects cluster;
Object tag and the object based on the sample object data in each third characteristics of objects cluster are special
Reference breath, training Object identifying model;The Object identifying model object data to be predicted and target object data for identification
Between similarity.
In embodiments of the present invention, by obtaining sample object data respectively from multiple multi-medium datas, described in extraction
The characteristics of objects information of sample object data;Multiple sample object data are clustered according to the characteristics of objects information, are obtained
To multiple the first characteristics of objects clusters with different cluster labels;The object tag of all sample object data is with affiliated first
The cluster label of characteristics of objects cluster is identical;Foreign peoples's object data in each first characteristics of objects cluster is cleared up respectively, after cleaning
First characteristics of objects cluster is determined as the second characteristics of objects cluster;Similar cluster merging is carried out between multiple second characteristics of objects clusters, it is raw
At multiple third characteristics of objects clusters with different cluster labels, by the object tag of the sample object data be updated to belonging to
The cluster label of the third characteristics of objects cluster;Pair based on the sample object data in each third characteristics of objects cluster
As label and the characteristics of objects information, training Object identifying model;Object identifying model object to be predicted for identification
Similarity between data and target object data.By being clustered to sample object data, foreign peoples's object data remove and
Similar cluster merges the sample object data for generating and having label, by largely the sample object data of label do not generate the sample for having label
This object data, then using the sample object data training Object identifying model for largely having label, it can be to avoid Object identifying mould
The type lower problem of the very few discrimination for leading to Object identifying model of training data in the training process largely has by obtaining
The sample object data of label, improve the discrimination to object data to be predicted.
It should be appreciated that embodiment corresponding to executable Fig. 3 to the Fig. 9 above of equipment 1000 described in the embodiment of the present invention
In description to the data processing method, also can be performed in embodiment corresponding to Figure 10 above to the data processing equipment 1
Description, details are not described herein.In addition, being described to using the beneficial effect of same procedure, also no longer repeated.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a computer-readable storage medium
In, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the storage medium can be magnetic
Dish, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random Access
Memory, RAM) etc..
The above disclosure is only the preferred embodiments of the present invention, cannot limit the right model of the present invention with this certainly
It encloses, therefore equivalent changes made in accordance with the claims of the present invention, is still within the scope of the present invention.
Claims (12)
1. a kind of data processing method characterized by comprising
It obtains sample object data respectively from multiple multi-medium datas, extracts the characteristics of objects letter of the sample object data
Breath;
Multiple sample object data are clustered according to the characteristics of objects information, obtain multiple the with different cluster labels
An object feature cluster;The object tag of all sample object data is identical as the affiliated cluster label of the first characteristics of objects cluster;
Foreign peoples's object data in each first characteristics of objects cluster is cleared up respectively, and the first characteristics of objects cluster after cleaning is determined as
Second characteristics of objects cluster;
Similar cluster merging is carried out between multiple second characteristics of objects clusters, and it is special to generate multiple third objects with different cluster labels
Cluster is levied, the object tag of the sample object data is updated to the cluster label of the affiliated third characteristics of objects cluster;
Object tag and characteristics of objects letter based on the sample object data in each third characteristics of objects cluster
Breath, training Object identifying model;The Object identifying model is for identification between object data to be predicted and target object data
Similarity.
2. the method according to claim 1, wherein described obtain sample pair from multiple multi-medium datas respectively
Image data extracts the corresponding characteristics of objects information of the sample object data, comprising:
The object location information in multiple multi-medium datas is obtained, according to the object location information from each multi-medium data
The subject area with target size is obtained respectively, and the picture material in the subject area is determined as sample object data,
Obtain the characteristics of objects information of each sample object data;
If the multi-medium data is image data, it is randomly provided pair for the sample object data in described image data
As label;
If the multi-medium data is video data, tracking rail of the sample object data in the video data is detected
Relating attribute will be arranged with the sample object data of the identical pursuit path information in mark information, and for the association
Identical object tag is arranged in the sample object data of attribute.
3. according to the method described in claim 2, it is characterized in that, the picture material by the subject area is determined as
Sample object data, comprising:
Picture material in the subject area is subjected to the affine transformation based on target bearing, it will be in the image after affine transformation
Appearance is determined as sample object data;Object in the sample object data is in the target bearing;
Then the method also includes:
Obtain multi-medium data to be predicted;
The targeted object region with target size is obtained according to the object location information in the multi-medium data to be predicted, it will
Picture material to be predicted in the targeted object region carries out the affine transformation based on the target bearing, after affine transformation
Picture material to be predicted be determined as object data to be predicted;
The similarity between object data to be predicted and target object data is obtained based on Object identifying model.
4. the method according to claim 1, wherein it is described according to the characteristics of objects information to multiple samples pair
Image data is clustered, and multiple the first characteristics of objects clusters with different cluster labels are obtained, comprising:
First sample object data and the second sample object data are obtained from multiple sample object data, by the first sample
The object tag of object data is determined as the first object tag, and the object tag of the second sample object data is determined as
Two object tags;First object tag and second object tag be not identical;
Obtain the characteristics of objects information of the first sample object data and the characteristics of objects letter of the second sample object data
The first image similarity between breath;
If the first image similarity is greater than first threshold, by least one sample pair with first object tag
Image data and at least one sample object data with the second object tag, are divided into identical first characteristics of objects cluster, and
Set the first object tag and the second object tag on the cluster label of the first divided characteristics of objects cluster;
When all sample object data are divided into the first affiliated characteristics of objects cluster, and by each first characteristics of objects cluster
And the sample object data in the characteristics of objects cluster are stored.
5. the method according to claim 1, wherein it is described clear up respectively it is different in each first characteristics of objects cluster
The first characteristics of objects cluster after cleaning is determined as the second characteristics of objects cluster by class object data, comprising:
Response is directed to the first marking operation of the first characteristics of objects cluster, by indicated by first marking operation described the
Sample object data in an object feature cluster are determined as the corresponding Standard object data of the first characteristics of objects cluster;
Response is directed to the second marking operation of non-standard object data, by non-standard object indicated by second marking operation
Data are determined as foreign peoples's object data;The non-standard object data is that the standard object is removed in the first characteristics of objects cluster
Sample object data except data;
Foreign peoples's object data is deleted from the first characteristics of objects cluster, the first object of foreign peoples's object data will be deleted
Feature cluster is determined as the second characteristics of objects cluster.
6. the method according to claim 1, wherein it is described carried out between multiple second characteristics of objects clusters it is similar
Cluster merges, and generates multiple third characteristics of objects clusters with different cluster labels, more by the object tag of the sample object data
It is newly the cluster label of the affiliated third characteristics of objects cluster, comprising:
The first detection feature cluster and the second detection feature cluster are obtained from multiple second characteristics of objects clusters;
Response is requested for the merging request of the first detection feature cluster and the second detection feature cluster based on the merging
Obtain the sample in the first quantity and the second detection feature cluster of the sample object data in the first detection feature cluster
Second quantity of object data;
According to each sample pair in each sample object data in the first detection feature cluster and the second detection feature cluster
The image similarity of image data, first quantity and second quantity obtain the first detection feature cluster and described the
The classification similarity of two detection feature clusters;
If the classification similarity is greater than second threshold, the first detection feature cluster and the second detection feature cluster are merged into third
Characteristics of objects cluster, and set the object tag of the sample object data in the first detection feature cluster and the second detection feature cluster to
The cluster label of the affiliated third characteristics of objects cluster.
7. the method according to claim 1, wherein it is described carried out between multiple second characteristics of objects clusters it is similar
Cluster merges, and generates multiple third characteristics of objects clusters with different cluster labels, more by the object tag of the sample object data
Newly for the affiliated third characteristics of objects cluster cluster label after, further includes:
The third characteristics of objects cluster is determined as to recycle characteristics of objects cluster;
Foreign peoples's object data in each circulation characteristics of objects cluster is cleared up respectively;
Circulation characteristics of objects cluster after deletion foreign peoples's object data is subjected to similar cluster merging, obtains the third characteristics of objects
Cluster;
When meeting the condition of convergence or similar cluster merging number equal to frequency threshold value for the third characteristics of objects cluster, execute
The object tag and characteristics of objects information based on the sample object data in each third characteristics of objects cluster, training pair
As the step of identification model.
8. the method according to the description of claim 7 is characterized in that further include:
The second figure between each third characteristics of objects cluster Plays object data and non-standard object data is obtained respectively
As similarity, the corresponding similarity mean value of each third characteristics of objects cluster is obtained according to second image similarity
With similarity variance;
Similarity mean value is less than mean value threshold value and similarity variance is determining greater than the third characteristics of objects cluster of variance threshold values
For target object feature cluster;
Obtain the testing result for being directed to the target object feature cluster;
If the testing result is that foreign peoples's object data and similar cluster are not present in the target object feature cluster, it is determined that described
Third characteristics of objects cluster meets the condition of convergence.
9. the method according to claim 1, wherein the sample based in each third characteristics of objects cluster
The object tag and characteristics of objects information of this object data, training Object identifying model, comprising:
It obtains in the characteristics of objects information and initial object identification model of the sample object data in the third characteristics of objects cluster
Standard object data characteristics of objects information between sample image similarity;
According to the sample in the object tag of the sample object data in the Object identifying model, the third characteristics of objects cluster
The object tag of object data and the sample image similarity determine similarity error, and reversed according to the similarity error
Propagate the model parameter for adjusting the initial object identification model;
When the adjustment number of the model parameter is equal to adjustment threshold value or the similarity error meets the condition of convergence, will wrap
Initial object identification model containing model parameter adjusted is determined as the Object identifying model.
10. a kind of data processing equipment characterized by comprising
Characteristic acquisition unit extracts the sample for obtaining sample object data respectively from multiple multi-medium datas
The characteristics of objects information of object data;
Fisrt feature cluster acquiring unit is obtained for being clustered according to the characteristics of objects information to multiple sample object data
To multiple the first characteristics of objects clusters with different cluster labels;The object tag of all sample object data is with affiliated first
The cluster label of characteristics of objects cluster is identical;
Second feature cluster acquiring unit, foreign peoples's object data for being cleared up in each first characteristics of objects cluster respectively, will clear up
The first characteristics of objects cluster afterwards is determined as the second characteristics of objects cluster;
Third feature cluster acquiring unit generates multiple tools for carrying out similar cluster merging between multiple second characteristics of objects clusters
There is the third characteristics of objects cluster of different cluster labels, the object tag of the sample object data is updated to the affiliated third
The cluster label of characteristics of objects cluster;
Model training unit, for the object tag based on the sample object data in each third characteristics of objects cluster
With the characteristics of objects information, training Object identifying model;The Object identifying model for identification object data to be predicted with
Similarity between target object data.
11. a kind of data processing equipment, which is characterized in that it is characterised by comprising: processor and memory;Wherein, described to deposit
Reservoir is stored with computer program, and the computer program is suitable for being loaded by the processor and being executed claim 1-9 such as and appoints
Method described in one.
12. a kind of computer storage medium, which is characterized in that the computer storage medium is stored with a plurality of instruction, the finger
It enables and is suitable for being loaded by processor and being executed the method and step such as claim 1-9 any one.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910319953.7A CN110046586A (en) | 2019-04-19 | 2019-04-19 | A kind of data processing method, equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910319953.7A CN110046586A (en) | 2019-04-19 | 2019-04-19 | A kind of data processing method, equipment and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110046586A true CN110046586A (en) | 2019-07-23 |
Family
ID=67278189
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910319953.7A Pending CN110046586A (en) | 2019-04-19 | 2019-04-19 | A kind of data processing method, equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110046586A (en) |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110427862A (en) * | 2019-07-26 | 2019-11-08 | 北京奇艺世纪科技有限公司 | A kind of face picture collection method, device and electronic equipment |
CN110443190A (en) * | 2019-07-31 | 2019-11-12 | 腾讯科技(深圳)有限公司 | A kind of object identifying method and device |
CN110502651A (en) * | 2019-08-15 | 2019-11-26 | 深圳市商汤科技有限公司 | Image processing method and device, electronic equipment and storage medium |
CN110909824A (en) * | 2019-12-09 | 2020-03-24 | 天津开心生活科技有限公司 | Test data checking method and device, storage medium and electronic equipment |
CN111062407A (en) * | 2019-10-15 | 2020-04-24 | 深圳市商汤科技有限公司 | Image processing method and device, electronic equipment and storage medium |
CN111340084A (en) * | 2020-02-20 | 2020-06-26 | 北京市商汤科技开发有限公司 | Data processing method and device, processor, electronic equipment and storage medium |
CN111709473A (en) * | 2020-06-16 | 2020-09-25 | 腾讯科技(深圳)有限公司 | Object feature clustering method and device |
CN111783734A (en) * | 2020-07-21 | 2020-10-16 | 腾讯科技(深圳)有限公司 | Original edition video identification method and device |
CN111783517A (en) * | 2020-05-13 | 2020-10-16 | 北京达佳互联信息技术有限公司 | Image recognition method and device, electronic equipment and storage medium |
CN112132220A (en) * | 2020-09-24 | 2020-12-25 | 杭州海康威视数字技术股份有限公司 | Self-training method, system, device, electronic equipment and storage medium |
CN113127668A (en) * | 2019-12-31 | 2021-07-16 | 深圳云天励飞技术有限公司 | Data annotation method and related product |
CN113239202A (en) * | 2021-05-25 | 2021-08-10 | 北京达佳互联信息技术有限公司 | Data processing method, device, server and storage medium |
CN114707617A (en) * | 2022-05-31 | 2022-07-05 | 每日互动股份有限公司 | Data processing system for acquiring pkg cluster |
CN115795318A (en) * | 2022-11-17 | 2023-03-14 | 北京白龙马云行科技有限公司 | Classification method of use case object, model training method, equipment and storage medium |
WO2023151529A1 (en) * | 2022-02-11 | 2023-08-17 | 华为技术有限公司 | Facial image processing method and related device |
CN116841716A (en) * | 2023-08-29 | 2023-10-03 | 四川三思德科技有限公司 | Multi-point cooperation method for basic-level society management |
CN117332303A (en) * | 2023-12-01 | 2024-01-02 | 太极计算机股份有限公司 | Label correction method for clusters |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101151851B1 (en) * | 2011-11-08 | 2012-06-01 | (주)올라웍스 | Method, apparatus and computer-readable recording medium for image tagging by using image clustering |
CN104391879A (en) * | 2014-10-31 | 2015-03-04 | 小米科技有限责任公司 | Method and device for hierarchical clustering |
CN106228188A (en) * | 2016-07-22 | 2016-12-14 | 北京市商汤科技开发有限公司 | Clustering method, device and electronic equipment |
US10025950B1 (en) * | 2017-09-17 | 2018-07-17 | Everalbum, Inc | Systems and methods for image recognition |
CN109583332A (en) * | 2018-11-15 | 2019-04-05 | 北京三快在线科技有限公司 | Face identification method, face identification system, medium and electronic equipment |
-
2019
- 2019-04-19 CN CN201910319953.7A patent/CN110046586A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101151851B1 (en) * | 2011-11-08 | 2012-06-01 | (주)올라웍스 | Method, apparatus and computer-readable recording medium for image tagging by using image clustering |
CN104391879A (en) * | 2014-10-31 | 2015-03-04 | 小米科技有限责任公司 | Method and device for hierarchical clustering |
CN106228188A (en) * | 2016-07-22 | 2016-12-14 | 北京市商汤科技开发有限公司 | Clustering method, device and electronic equipment |
US10025950B1 (en) * | 2017-09-17 | 2018-07-17 | Everalbum, Inc | Systems and methods for image recognition |
CN109583332A (en) * | 2018-11-15 | 2019-04-05 | 北京三快在线科技有限公司 | Face identification method, face identification system, medium and electronic equipment |
Cited By (31)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110427862B (en) * | 2019-07-26 | 2021-08-27 | 北京奇艺世纪科技有限公司 | Face picture collecting method and device and electronic equipment |
CN110427862A (en) * | 2019-07-26 | 2019-11-08 | 北京奇艺世纪科技有限公司 | A kind of face picture collection method, device and electronic equipment |
CN110443190A (en) * | 2019-07-31 | 2019-11-12 | 腾讯科技(深圳)有限公司 | A kind of object identifying method and device |
CN110443190B (en) * | 2019-07-31 | 2024-02-02 | 腾讯科技(成都)有限公司 | Object recognition method and device |
WO2021027344A1 (en) * | 2019-08-15 | 2021-02-18 | 深圳市商汤科技有限公司 | Image processing method and device, electronic apparatus, and storage medium |
CN110502651A (en) * | 2019-08-15 | 2019-11-26 | 深圳市商汤科技有限公司 | Image processing method and device, electronic equipment and storage medium |
TWI761851B (en) * | 2019-08-15 | 2022-04-21 | 大陸商深圳市商湯科技有限公司 | Image processing method, image processing apparatus, electronic device, and computer-readable storage medium |
CN110502651B (en) * | 2019-08-15 | 2022-08-02 | 深圳市商汤科技有限公司 | Image processing method and device, electronic equipment and storage medium |
CN111062407A (en) * | 2019-10-15 | 2020-04-24 | 深圳市商汤科技有限公司 | Image processing method and device, electronic equipment and storage medium |
CN111062407B (en) * | 2019-10-15 | 2023-12-19 | 深圳市商汤科技有限公司 | Image processing method and device, electronic equipment and storage medium |
CN110909824A (en) * | 2019-12-09 | 2020-03-24 | 天津开心生活科技有限公司 | Test data checking method and device, storage medium and electronic equipment |
CN113127668A (en) * | 2019-12-31 | 2021-07-16 | 深圳云天励飞技术有限公司 | Data annotation method and related product |
CN111340084B (en) * | 2020-02-20 | 2024-05-17 | 北京市商汤科技开发有限公司 | Data processing method and device, processor, electronic equipment and storage medium |
CN111340084A (en) * | 2020-02-20 | 2020-06-26 | 北京市商汤科技开发有限公司 | Data processing method and device, processor, electronic equipment and storage medium |
CN111783517A (en) * | 2020-05-13 | 2020-10-16 | 北京达佳互联信息技术有限公司 | Image recognition method and device, electronic equipment and storage medium |
CN111783517B (en) * | 2020-05-13 | 2024-05-07 | 北京达佳互联信息技术有限公司 | Image recognition method, device, electronic equipment and storage medium |
CN111709473B (en) * | 2020-06-16 | 2023-09-19 | 腾讯科技(深圳)有限公司 | Clustering method and device for object features |
CN111709473A (en) * | 2020-06-16 | 2020-09-25 | 腾讯科技(深圳)有限公司 | Object feature clustering method and device |
CN111783734A (en) * | 2020-07-21 | 2020-10-16 | 腾讯科技(深圳)有限公司 | Original edition video identification method and device |
CN111783734B (en) * | 2020-07-21 | 2024-02-27 | 腾讯科技(深圳)有限公司 | Original edition video recognition method and device |
CN112132220A (en) * | 2020-09-24 | 2020-12-25 | 杭州海康威视数字技术股份有限公司 | Self-training method, system, device, electronic equipment and storage medium |
CN113239202B (en) * | 2021-05-25 | 2024-03-05 | 北京达佳互联信息技术有限公司 | Data processing method, device, server and storage medium |
CN113239202A (en) * | 2021-05-25 | 2021-08-10 | 北京达佳互联信息技术有限公司 | Data processing method, device, server and storage medium |
WO2023151529A1 (en) * | 2022-02-11 | 2023-08-17 | 华为技术有限公司 | Facial image processing method and related device |
CN114707617B (en) * | 2022-05-31 | 2022-08-26 | 每日互动股份有限公司 | Data processing system for acquiring pkg cluster |
CN114707617A (en) * | 2022-05-31 | 2022-07-05 | 每日互动股份有限公司 | Data processing system for acquiring pkg cluster |
CN115795318A (en) * | 2022-11-17 | 2023-03-14 | 北京白龙马云行科技有限公司 | Classification method of use case object, model training method, equipment and storage medium |
CN116841716A (en) * | 2023-08-29 | 2023-10-03 | 四川三思德科技有限公司 | Multi-point cooperation method for basic-level society management |
CN116841716B (en) * | 2023-08-29 | 2023-11-28 | 四川三思德科技有限公司 | Multi-point cooperation method for basic-level society management |
CN117332303A (en) * | 2023-12-01 | 2024-01-02 | 太极计算机股份有限公司 | Label correction method for clusters |
CN117332303B (en) * | 2023-12-01 | 2024-03-26 | 太极计算机股份有限公司 | Label correction method for clusters |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110046586A (en) | A kind of data processing method, equipment and storage medium | |
US11526549B2 (en) | Method and system for interfacing with a user to facilitate an image search for an object-of-interest | |
CN109947975B (en) | Image search device, image search method, and setting screen used therein | |
US11086924B2 (en) | Image search device and image search method | |
TWI742300B (en) | Method and system for interfacing with a user to facilitate an image search for a person-of-interest | |
CN110175549A (en) | Face image processing process, device, equipment and storage medium | |
IL263532A (en) | Method, system and computer program product for interactively identifying same individuals or objects present in video recordings | |
CN110263613A (en) | Monitor video processing method and processing device | |
CN109644255A (en) | Mark includes the method and apparatus of the video flowing of a framing | |
JPWO2012073421A1 (en) | Image classification device, image classification method, program, recording medium, integrated circuit, model creation device | |
JP6787831B2 (en) | Target detection device, detection model generation device, program and method that can be learned by search results | |
CN108229347A (en) | For the method and apparatus of the deep layer displacement of the plan gibbs structure sampling of people's identification | |
US11429985B2 (en) | Information processing device calculating statistical information | |
CN112001932A (en) | Face recognition method and device, computer equipment and storage medium | |
Geng et al. | An improved helmet detection method for YOLOv3 on an unbalanced dataset | |
JPWO2020026643A1 (en) | Information processing equipment, information processing methods and information processing programs | |
CN110232331A (en) | A kind of method and system of online face cluster | |
CN110569918A (en) | sample classification method and related device | |
WO2020171066A1 (en) | Image search device and training data extraction method | |
CN112084812A (en) | Image processing method, image processing device, computer equipment and storage medium | |
CN111428087B (en) | Video interception method and device, computer equipment and storage medium | |
CN113705650B (en) | Face picture set processing method, device, medium and computing equipment | |
CN111767880B (en) | Living body identity recognition method and device based on facial features and storage medium | |
CN115953650B (en) | Training method and device for feature fusion model | |
KR20190101692A (en) | Video watch method based on transfer of learning |
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 |