CN108491543A - Image search method, image storage method and image indexing system - Google Patents
Image search method, image storage method and image indexing system Download PDFInfo
- Publication number
- CN108491543A CN108491543A CN201810296979.XA CN201810296979A CN108491543A CN 108491543 A CN108491543 A CN 108491543A CN 201810296979 A CN201810296979 A CN 201810296979A CN 108491543 A CN108491543 A CN 108491543A
- Authority
- CN
- China
- Prior art keywords
- image
- node
- characteristic
- feature data
- classification code
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Processing Or Creating Images (AREA)
Abstract
The invention discloses a kind of image search method, image storage method and image indexing systems, image feature data is stored in each node of retrieval service cluster, the characteristic of input picture is distributed to each node, it is independently matched from the image feature data being locally stored by each node, compared with characteristic is obtained to traversal in database in the prior art and carries out matched method, reduce from database traversal and obtain taking for characteristic, compared to retrieval rate can be improved.
Description
Technical field
The present invention relates to image retrieval technologies fields, more particularly to a kind of image search method and image storage side
Method.The invention further relates to a kind of image indexing systems.
Background technology
With the rapid development and progress of computer technology, internetworking technologies and multimedia technology, on network
Various information resource becomes increasingly abundant, and the multimedia messages of especially visual pattern receive the favor of people.Number in global range
Word image increases with surprising rapidity, how in immense image library it is convenient, fast, accurately inquire and retrieve use
Needed for family or interested picture, becomes the hot spot of multimedia information retrieval area research.
Currently, image retrieval can be divided into two classes in the way of describing picture material, one kind is text based image
It retrieves (Text Based Image Retrieval, TBIR), another kind of is content-based image retrieval (Content
Based Image Retrieval, CBIR).
Wherein, text based image retrieval is that the content to image in the way of text marking is described, to
The keyword for describing this picture material is formed for each image, for example object, scene etc. in image, this mode can be people
Work notation methods can also carry out semi-automatic mark by image recognition technology.When being retrieved, user can be according to oneself
Demand key word of the inquiry is provided, searching system finds out according to the key word of the inquiry that user provides and is labeled with the key word of the inquiry
Picture, the result inquired is finally returned into user.This image search method precision ratio based on text description is opposite
Height, but the defect present in it is also that obviously, it needs manpower intervention annotation process so that and it is only applicable on a small scale
Image data, this process is completed in large-scale image data and needs to expend a large amount of manpower and financial resources, it is very unreal
With.
And content-based image retrieval is analyzed image using computer, is established characteristics of image vector and is stored in
Database extracts feature to query image using identical feature extracting method, obtains query vector, so in query image
Query vector is calculated under certain similarity measurement criterion afterwards and stores the similitude size of characteristic vector in database, is finally pressed
It is ranked up according to similitude size and Sequential output corresponds to picture.Content-based image retrieval technology is by the expression of picture material
Computer is given with similarity measurement to be automatically processed, and overcomes and the defect that image retrieval is faced is carried out using text, and
And the advantage that computer is longer than calculating has been given full play to, retrieval effect is substantially increased compared with text based image search method
Rate.
Shown in please referring to Fig.1, content-based image retrieval includes mainly following three processes：
First part is the extraction of characteristics of image.Inspection to the image zooming-out characteristic relation to be stored to image indexing system
Suo Jingdu needs to pre-process picture, obtains the characteristic of image for the picture that mass users generate in real time.
Second part is the storage of characteristics of image.It is later storing process to extract characteristics of image, including storage image sheet
The characteristic of body and storage image.
Part III is the retrieval matching process of image.After system receives Client-initiated retrieval request, press first
According to identical image characteristic extracting method, the feature of input picture is extracted, then according in the characteristics of image to database of extraction
Matching finally exports the corresponding image of the characteristic for meeting condition retrieved.
In the prior art, original image and the characteristics of image of extraction are stored together in storing process, rank is matched in retrieval
Section traversal from database obtains stored characteristics of image and is matched, and in entire retrieving, is traversed from database
It is relatively time consuming compared with the matching primitives between characteristics of image to obtain image feature data, therefore reduces image retrieval speed.
Invention content
The object of the present invention is to provide a kind of image search method and image storage methods, improve compared with prior art
The retrieval rate of image.The present invention also provides a kind of image indexing systems.
To achieve the above object, the present invention provides the following technical solutions：
A kind of image search method, including：
Characteristic is extracted to the target image of input, and the characteristic extracted is distributed to retrieval service cluster
Each node；
Each node matches the characteristic received with the image feature data being locally stored, and will inspection
The image feature data that the similarity that rope goes out meets condition returns；
According to the image feature data that each node returns, corresponding original image is obtained from database and is exported.
Optionally, further include：The target image is learnt using trained identification model, obtains the target
The classification code is distributed together each section of the retrieval service cluster by the classification code of image with the characteristic
Point；
The characteristic received and the image feature data being locally stored are carried out matching packet by each node
It includes：
Each node is according to the classification code of the target image, from local file corresponding with the classification code
Middle reading image feature data matches the characteristic of the target image with the image feature data of reading.
Optionally, the image feature data returned according to each node, obtains corresponding original from database
Image and export include：
The image feature data that each node is returned, sorts successively from high to low according to similarity, and from database
It is middle to obtain original image corresponding with the image feature data retrieved, and export.
A kind of image storage method, including：
It will be in the image storage to database of input；
The node of retrieval service cluster obtains the image of input from the database, extracts the characteristic of described image；
The node stores the characteristic extracted to local.
Optionally, further include：The node learns described image using trained identification model, obtains described
The classification code of image；
The node stores the characteristic extracted to local：
The node is according to the obtained classification code, by the characteristic extracted storage to the local and classification
In the corresponding file of code.
A kind of image indexing system, including preprocessing module, acquisition module and the retrieval service cluster including several nodes；
The preprocessing module is used to extract characteristic to the target image of input, and the characteristic extracted is divided
It is dealt into each node of retrieval service cluster；
The node include retrieval submodule, the retrieval submodule for by the characteristic received be locally stored
Image feature data matched, and by the similarity retrieved meet condition image feature data return；
The image feature data that the acquisition module is used to be returned according to each node, obtains corresponding from database
Original image simultaneously exports.
Optionally, the preprocessing module is additionally operable to using trained identification model to the target image
It practises, obtains the classification code of the target image, the classification code is distributed together the retrieval with the characteristic
Each node of service cluster；
The retrieval submodule is specifically used for the classification code according to the target image, from the local and classification code
Image feature data is read in corresponding file, the characteristic of the target image and the image feature data of reading are carried out
Matching.
Optionally, the acquisition module is specifically used for the image feature data for returning to each node, according to similarity
It sorts successively from high to low, and obtains original image corresponding with the image feature data retrieved from database, and export.
Optionally, further include database, the image storage for that will input；
The node further includes：
Submodule is pre-processed, the image for obtaining input from the database extracts the characteristic of described image；
Sub-module stored, for storing the characteristic extracted to local.
Optionally, the pretreatment submodule is additionally operable to learn described image using trained identification model,
Obtain the classification code of described image；
The sub-module stored is additionally operable to according to the obtained classification code, by the characteristic extracted storage to originally
In ground file corresponding with the classification code.
As shown from the above technical solution, image search method provided by the present invention, in each section of retrieval service cluster
Point storage image feature data, each node is distributed to by the characteristic of input picture, by each node independently from this
It is matched in the image feature data of ground storage, is matched with characteristic is obtained to traversal in database in the prior art
Method compare, reduce from database traversal and obtain taking for characteristic, compared to retrieval rate can be improved.
Image storage method provided by the present invention, the node of retrieval service cluster, will to the image zooming-out feature of input
The characteristic of image and image is stored separately, and in raw image storage to database, characteristic is stored in each node,
It is and existing in this way so that each node can be matched independently from the image feature data being locally stored in image retrieval
Have in technology and compared to the matched method of traversal acquisition characteristic progress in database, reduces and traverse acquisition from database
Characteristic takes, compared to can improve retrieval rate.
Image indexing system provided by the present invention can reach above-mentioned advantageous effect.
Description of the drawings
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 technology 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
Obtain other attached drawings according to these attached drawings.
Fig. 1 is the process schematic of content-based image retrieval method；
Fig. 2 is a kind of flow chart of image search method provided in an embodiment of the present invention；
Fig. 3 is a kind of flow chart for image search method that further embodiment of this invention provides；
Fig. 4 is a kind of flow chart of image storage method provided in an embodiment of the present invention；
Fig. 5 is a kind of flow chart for image storage method that further embodiment of this invention provides；
Fig. 6 is a kind of schematic diagram of image indexing system provided in an embodiment of the present invention；
Fig. 7 is a kind of schematic diagram for image indexing system that further embodiment of this invention provides；
Fig. 8 is the schematic diagram of image indexing system interior joint shown in Fig. 7.
Specific implementation mode
In order to make those skilled in the art more fully understand the technical solution in the present invention, below in conjunction with of the invention real
The attached drawing in example is applied, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described implementation
Example is only a part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, this field is common
The every other embodiment that technical staff is obtained without making creative work, should all belong to protection of the present invention
Range.
Referring to FIG. 2, a kind of image search method provided in an embodiment of the present invention, includes the following steps：
S10：Characteristic is extracted to the target image of input, and the characteristic extracted is distributed to retrieval service collection
Each node of group.
The present embodiment image search method is applied to scheme to search figure, i.e., according to known picture input by user, from picture number
According to searching similar pictures in library.
In this step, target image input by user is handled, extracts the characteristic of target image.To target figure
As the method for extraction feature uses and the identical image feature extracting method that uses when storage image.
Illustratively, CEDD (Color and Edge Directivity can be calculated to image in the specific implementation
Descriptor) feature, CEDD features are a kind of features indicating color of image and texture, and extraction CEDD features, which have, extracts spy
The advantage that sign speed is fast, feature descriptor occupies little space.It is not limited to it is understood that extracting characteristics of image in this step
Using the above method, the method that other extraction characteristics of image can also be used in other embodiment of the invention, also in this hair
In bright protection domain.
After extracting the characteristic of target image, characteristic is distributed to each node of retrieval service cluster.
The retrieval service cluster includes several nodes, and image feature data is stored on each node.
S11：Each node matches the characteristic received with the image feature data being locally stored, and
The image feature data that the similarity retrieved is met to condition returns.
Each node is according to the characteristic of the target image received, from middle reading image feature data is locally stored,
The characteristic of target image is matched with the image feature data of reading, retrieves the image spy that similarity meets condition
Levy data,
In the specific implementation, respective file can be looked for from being locally stored according to the relevant parameter of target image, then from
Image feature data is read in respective file.Illustratively, the relevant parameter of image can be image date, according to target image
Date search corresponding with date file, image feature data is read from respective file, helps to improve recall precision.
In this method, the image feature data for meeting certain condition can be buffered in the memory of node server, be saved
Point can preferentially read the image feature data in memory according to the characteristic of target image and be matched, in node server
In the case that memory is sufficiently large, cacheable characteristic amount is huge in memory, can further increase retrieval rate
And recall precision.
S12：According to the image feature data that each node returns, corresponding original image and defeated is obtained from database
Go out.
In the specific implementation, corresponding with image feature data to store original image in being locally stored of each node
Storage address.In this step, according to the image feature data that each Nodes Retrieval goes out, according to corresponding raw image storage
Location obtains original image from database.
Specifically, the image feature data for returning to each node in this step, sorts successively from high to low according to similarity,
And original image corresponding with the image feature data retrieved is obtained from database, and export.
The present embodiment image search method stores image feature data in each node of retrieval service cluster, will input
The characteristic of image is distributed to each node, is independently carried out from the image feature data being locally stored by each node
Matching reduces compared with obtaining characteristic to traversal in database in the prior art and carrying out matched method from database
Middle traversal obtains taking for characteristic, compared to can improve retrieval rate.
Referring to FIG. 3, further embodiment of this invention provides a kind of image search method, include the following steps：
S20：Characteristic is extracted to the target image of input, and using trained identification model to the target image
Learnt, obtain the classification code of the target image, the characteristic extracted and the classification code are distributed to inspection
Each node of rope service cluster.
To target image input by user, characteristic is extracted to target image.Characteristics of image is extracted to target image
Method can refer to step S10 in an embodiment and specifically describe.
In this step, also image is learnt using trained identification model, the object information that identification image includes,
Corresponding classification code is generated, as label to indicate object information that image includes.
Training obtains the identification model in advance, and the image object recognition methods based on deep learning specifically may be used,
First with the image data set for including a large amount of tape label images, after the training of certain time, an identification model is obtained.
The object identification demarcated in input picture can be gone out with obtained identification model, identifies the kind of object in practical applications
Class can be manually set, and different schemes can be arranged according to different application scenarios, as long as the content in update training library and label are
It can.
S21：Each node is according to the classification code of the target image, from local corresponding with the classification code
Image feature data is read in file, and the characteristic of the target image is matched with the image feature data of reading,
And the image feature data that the similarity retrieved is met to condition returns.
In each node, text corresponding with the classification code is found from local according to the classification code of target image
Part reads image feature data from respective file, and the characteristic of target image and the image feature data of reading are carried out
Matching, retrieves the image feature data that similarity meets condition.
In the specific implementation, it can be looked for from being locally stored according to the relevant parameter of classification code and combining target image pair
File is answered, image feature data is then read from respective file.
S22：According to the image feature data that each node returns, corresponding original image and defeated is obtained from database
Go out.
In the specific implementation, corresponding with image feature data to store original image in being locally stored of each node
Storage address.In this step, according to the image feature data that each Nodes Retrieval goes out, according to corresponding raw image storage
Location obtains original image from database.
Specifically, the image feature data for returning to each node in this step, sorts successively from high to low according to similarity,
And original image corresponding with the image feature data retrieved is obtained from database, and export.
The present embodiment image search method, extraction target image characteristic simultaneously, target image is learnt
The classification code of instruction target image classification is obtained, each node only reads image feature data from the file of local corresponding classification
It is matched, All Files need not be searched, can not only improve image retrieval precision, image inspection can also be further increased
Suo Sudu.
Accordingly referring to FIG. 4, the embodiment of the present invention also provides a kind of image storage method, include the following steps：
S30：It will be in the image storage to database of input.
For the image that user's application generates in real time, in storage to database.In practical applications to the picture number of input
According to can first carry out validity checking, image is stored into database if image is met the requirements.
Illustratively, it can be used distributed PostgreSQL database (HBase) towards row, HBase databases have can be with
The advantages of realizing high performance concurrent reading and writing operation, while can also transparent cutting be carried out to data so that image storage has can
Retractility and autgmentability.
S31：The node of retrieval service cluster obtains the image of input from the database, extracts the characteristic of described image
According to.
The retrieval service cluster includes several nodes, and each node fifty-fifty obtains the image of input from database respectively.
Node is handled image after database gets image, extracts the characteristic of image.Illustratively,
CEDD (Color and Edge Directivity Descriptor) feature, CEDD can be calculated to image in the specific implementation
It is characterized in a kind of feature indicating color of image and texture, extraction CEDD features fast, feature descriptor with extraction feature speed
The advantage to occupy little space.It is not limited to use the above method it is understood that extracting characteristics of image in this step, in this hair
The method that other extraction characteristics of image can also be used in bright other embodiment, also all falls in the scope of protection of the present invention.
S32：The node stores the characteristic extracted to local.
In the specific implementation, node the storage address of the characteristic of extraction and original image is formatted and two into
The storage address of characteristic and original image is written in local file by inhibition and generation.
Preferably, respective file can be found according to the relevant parameter of image, by depositing for characteristic and original image
Storage address is written in respective file.Illustratively, the relevant parameter of image can be image date, will according to the date of image
Data are written in date corresponding file, help to improve retrieval rate in the image retrieval stage in this way.
It is very small to node local server disk occupied space since the characteristic capacity of extraction is very small, it is depositing
A large amount of characteristic can be stored when storage in single file so that being read in image retrieval once can be a large amount of number
According to reading in server memory.
Preferably, more image feature datas is written as possible in single file, once being read in retrieval in this way can
A large amount of image feature data is obtained, the IO performances of server can be given full play to.
Each node of retrieval service cluster equably obtains image from database, and characteristics of image is stored by each node
The image feature data of data, storage is unique.
The characteristic of image and image is stored separately by the present embodiment image storage method, raw image storage to number
According in library, and characteristic is stored in each node, so that each node can be independently from this in image retrieval
It is matched in the image feature data of ground storage, is matched with characteristic is obtained to traversal in database in the prior art
Method compare, reduce from database traversal and obtain taking for characteristic, compared to retrieval rate can be improved.
Referring to FIG. 5, further embodiment of this invention provides a kind of image storage method, include the following steps：
S40：It will be in the image storage to database of input.
For the image that user's application generates in real time, in storage to database.In practical applications to the picture number of input
According to can first carry out validity checking, image is stored into database if image is met the requirements.The specific embodiment party of this step
Formula can refer to step S30 in an embodiment and specifically describe.
S41：The node of retrieval service cluster obtains the image of input from the database, extracts the characteristic of described image
According to, and described image is learnt using trained identification model, obtain the classification code of described image.
The retrieval service cluster includes several nodes, and each node fifty-fifty obtains the image of input from database respectively.
Node is handled image after database gets image, extracts the characteristic of image.To image zooming-out
The method of characteristic can refer to the specific descriptions of step S31 in an embodiment.
Meanwhile node learns described image using trained identification model, the object letter that identification image includes
Breath, generates corresponding classification code, as label to indicate object information that image includes.
The identification model is obtained by training in advance, and the image object identification side based on deep learning specifically may be used
Method after the training of certain time, obtains an identification mould first with the image data set for including a large amount of tape label images
Type.The object identification demarcated in input picture can be gone out with obtained identification model, identifies object in practical applications
Type can be manually set, can according to different application scenarios be arranged different schemes, as long as update training library content and mark
Label.
S42：The node according to the obtained classification code, by the characteristic extracted storage to it is local with it is described
In the corresponding file of classification code.
In the specific implementation, node formats the characteristic of extraction and the storage address of original image, further
File corresponding with the classification code is found from local according to classification code, by the storage address of characteristic and original image
It is written in respective file.
In the specific implementation, according to classification code and the relevant parameter of image can be combined to look for corresponding text from being locally stored
The storage address of the characteristic of extraction and original image is then written in respective file by part.Illustratively, the phase of image
Related parameter can be image date, be write data into corresponding file according to the classification code of image and date, in this way
Retrieval rate is helped to improve in the image retrieval stage.
Preferably, more image feature datas is written as possible in single file, once being read in retrieval in this way can
A large amount of image feature data is obtained, the IO performances of server can be given full play to.
The present embodiment image storage method, the node of retrieval service cluster is to the image zooming-out feature of input, while to figure
As being learnt to obtain the classification code of instruction image classification, the characteristic classification of image is deposited according to the classification code of image
It stores up in respective file, so that when retrieving image, node only reads characteristics of image number from the file of local corresponding classification
According to being matched, All Files need not be searched, can not only improve image retrieval precision, image can also be further increased
Retrieval rate.
Correspondingly, referring to FIG. 6, the embodiment of the present invention also provides a kind of image indexing system, including preprocessing module 50,
Acquisition module 52 and retrieval service cluster including several nodes 51；
The preprocessing module 50 is used to extract characteristic, and the characteristic that will be extracted to the target image of input
It is distributed to each node 51 of retrieval service cluster；
The node 51 includes retrieval submodule 510, the retrieval submodule 510 for will the characteristic that receive with
The image feature data being locally stored is matched, and the similarity retrieved is met to the image feature data return of condition；
The image feature data that the acquisition module 52 is used to be returned according to each node 51, the acquisition pair from database
The original image answered and output.
The present embodiment image indexing system is can be seen that, characteristics of image number is stored in each node of retrieval service cluster
According to the characteristic of input picture being distributed to each node, by each node independently from the characteristics of image being locally stored
It is matched in data, compared with obtaining characteristic to traversal in database in the prior art and carrying out matched method, is reduced
Traversal obtains taking for characteristic from database, compared to can improve retrieval rate.
Further, in the present embodiment image indexing system, the preprocessing module 50 is additionally operable to utilize trained knowledge
Other model learns the target image, obtains the classification code of the target image, by the classification code with described
Characteristic is distributed together each node 51 of the retrieval service cluster.
The retrieval submodule 510 is specifically used for the classification code according to the target image, from the local and classification
Image feature data is read in the corresponding file of code, by the image feature data of the characteristic of the target image and reading
It is matched.
The present embodiment image indexing system, extraction target image characteristic simultaneously, target image is learnt
The classification code of instruction target image classification is obtained, each node only reads image feature data from the file of local corresponding classification
It is matched, All Files need not be searched, can not only improve image retrieval precision, image inspection can also be further increased
Suo Sudu.
Specifically, in the present embodiment image indexing system, the acquisition module 52 is specifically used for returning each node 51
The image feature data returned, sorts successively from high to low according to similarity, and the image for obtaining and retrieving from database is special
The corresponding original image of data is levied, and is exported.
Further to please refer to Fig. 7 and Fig. 8, further embodiment of this invention provides a kind of image indexing system, including locates in advance
Reason module 60, acquisition module 62 include the retrieval service cluster and database 63 of several nodes 61.
The preprocessing module 60 is used to extract characteristic, and the characteristic that will be extracted to the target image of input
It is distributed to each node 61 of retrieval service cluster.
The node 61 includes retrieval submodule 610, the retrieval submodule 610 for will the characteristic that receive with
The image feature data being locally stored is matched, and the similarity retrieved is met to the image feature data return of condition.
The image feature data that the acquisition module 62 is used to be returned according to each node 61, obtains from database 63
Corresponding original image simultaneously exports.
The image storage that database 63 is used to input.
The node 61 further includes：
Submodule 611 is pre-processed, the image for obtaining input from the database extracts the characteristic of described image
According to.
Sub-module stored 612, for storing the characteristic extracted to local.
The characteristic of image and image is stored separately by the present embodiment image indexing system, raw image storage to number
According in library, and characteristic is stored in each node, each node can be independently from being locally stored in image retrieval
It is matched in image feature data, matched method phase is carried out with characteristic is obtained to traversal in database in the prior art
Than reducing from database traversal and obtaining taking for characteristic, compared to retrieval rate can be improved.
Further specifically, the pretreatment submodule 610 is additionally operable to using trained identification model to described image
Learnt, obtains the classification code of described image；
The sub-module stored 612 is additionally operable to, according to the obtained classification code, the characteristic extracted be stored
Into local file corresponding with the classification code.
Therefore the present embodiment image indexing system, the node of retrieval service cluster is to the image zooming-out feature of input, simultaneously
Image is learnt to obtain the classification code of instruction image classification, is divided the characteristic of image according to the classification code of image
Class storage in respective file, when retrieve image node only from the file of local corresponding classification reading image feature data into
Row matching, need not search All Files, can not only improve image retrieval precision, can also further increase image retrieval
Speed.
The present embodiment image indexing system uses Distributed Design, mutual indepedent work between each node of retrieval service cluster
Make, the characteristic of image by rationally, be evenly distributed in cluster, in practical applications, can be according to the big of system load
It is small come flexible configuration, have good retractility.
Image search method provided by the present invention, image storage method and image indexing system have been carried out in detail above
It introduces.Principle and implementation of the present invention are described for specific case used herein, the explanation of above example
It is merely used to help understand the method and its core concept of the present invention.It should be pointed out that for the ordinary skill people of the art
Member for, without departing from the principle of the present invention, can with several improvements and modifications are made to the present invention, these improve and
Modification is also fallen within the protection scope of the claims of the present invention.
Claims (10)
1. a kind of image search method, which is characterized in that including：
Characteristic is extracted to the target image of input, and the characteristic extracted is distributed to each of retrieval service cluster
Node；
Each node matches the characteristic received with the image feature data being locally stored, and will retrieve
Similarity meet condition image feature data return；
According to the image feature data that each node returns, corresponding original image is obtained from database and is exported.
2. image search method according to claim 1, which is characterized in that further include：Utilize trained identification model
The target image is learnt, the classification code of the target image is obtained, by the classification code with the characteristic
According to each node for being distributed together the retrieval service cluster；
Each node the characteristic received is carried out match with the image feature data being locally stored including：
Each node is read according to the classification code of the target image from local file corresponding with the classification code
Image feature data is taken, the characteristic of the target image is matched with the image feature data of reading.
3. image search method according to claim 1 or 2, which is characterized in that described to be returned according to each node
Image feature data, obtained from database corresponding original image and export include：
The image feature data that each node is returned, sorts, and obtained from database successively from high to low according to similarity
Original image corresponding with the image feature data retrieved is taken, and is exported.
4. a kind of image storage method, which is characterized in that including：
It will be in the image storage to database of input；
The node of retrieval service cluster obtains the image of input from the database, extracts the characteristic of described image；
The node stores the characteristic extracted to local.
5. image storage method according to claim 4, which is characterized in that further include：The node utilizes trained
Identification model learns described image, obtains the classification code of described image；
The node stores the characteristic extracted to local：
The node is according to the obtained classification code, by the characteristic extracted storage to the local and classification code
In corresponding file.
6. a kind of image indexing system, which is characterized in that including preprocessing module, acquisition module and the retrieval for including several nodes
Service cluster；
The preprocessing module is used to extract characteristic to the target image of input, and the characteristic extracted is distributed to
Each node of retrieval service cluster；
The node includes retrieval submodule, and the retrieval submodule is used for the characteristic received and the figure being locally stored
As characteristic is matched, and the image feature data that the similarity retrieved is met to condition returns；
The image feature data that the acquisition module is used to be returned according to each node, obtains corresponding original from database
Image simultaneously exports.
7. image indexing system according to claim 6, which is characterized in that the preprocessing module is additionally operable to utilize training
Good identification model learns the target image, the classification code of the target image is obtained, by the classification code
Each node of the retrieval service cluster is distributed together with the characteristic；
The retrieval submodule is specifically used for according to the classification code of the target image, from local corresponding with the classification code
File in read image feature data, the characteristic of the target image and the image feature data of reading are carried out
Match.
8. image indexing system according to claim 6, which is characterized in that the acquisition module is specifically used for will be each described
The image feature data that node returns, sorts successively from high to low according to similarity, and obtain and retrieve from database
The corresponding original image of image feature data, and export.
9. according to claim 6-8 any one of them image indexing systems, which is characterized in that further include database, being used for will
The image of input stores；
The node further includes：
Submodule is pre-processed, the image for obtaining input from the database extracts the characteristic of described image；
Sub-module stored, for storing the characteristic extracted to local.
10. image indexing system according to claim 9, which is characterized in that the pretreatment submodule is additionally operable to utilize
Trained identification model learns described image, obtains the classification code of described image；
The sub-module stored is additionally operable to according to the obtained classification code, by the characteristic extracted storage to it is local with
In the corresponding file of the classification code.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810296979.XA CN108491543A (en) | 2018-03-30 | 2018-03-30 | Image search method, image storage method and image indexing system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810296979.XA CN108491543A (en) | 2018-03-30 | 2018-03-30 | Image search method, image storage method and image indexing system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108491543A true CN108491543A (en) | 2018-09-04 |
Family
ID=63314509
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810296979.XA Pending CN108491543A (en) | 2018-03-30 | 2018-03-30 | Image search method, image storage method and image indexing system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108491543A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109740019A (en) * | 2018-12-14 | 2019-05-10 | 上海众源网络有限公司 | A kind of method, apparatus to label to short-sighted frequency and electronic equipment |
| CN110334763A (en) * | 2019-07-04 | 2019-10-15 | 北京字节跳动网络技术有限公司 | Model data file generation, image-recognizing method, device, equipment and medium |
| WO2020125544A1 (en) * | 2018-12-18 | 2020-06-25 | 华为技术有限公司 | Distributed image analysis method and system, and storage medium |
| CN113678120A (en) * | 2018-11-20 | 2021-11-19 | 多伦多道明银行 | Automated image retrieval using images |
| CN115525181A (en) * | 2022-11-28 | 2022-12-27 | 深圳飞蝶虚拟现实科技有限公司 | Method and device for manufacturing 3D content, electronic device and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101196905A (en) * | 2007-12-05 | 2008-06-11 | 覃征 | Intelligent pattern searching method |
| CN103310460A (en) * | 2013-06-24 | 2013-09-18 | 安科智慧城市技术（中国）有限公司 | Image characteristic extraction method and system |
| CN103927387A (en) * | 2014-04-30 | 2014-07-16 | 成都理想境界科技有限公司 | Image retrieval system, method and device |
| CN106354735A (en) * | 2015-07-22 | 2017-01-25 | 杭州海康威视数字技术股份有限公司 | Image target searching method and device |
| CN107193965A (en) * | 2017-05-24 | 2017-09-22 | 哈尔滨工业大学 | A kind of quick indoor orientation method based on BoVW algorithms |
-
2018
- 2018-03-30 CN CN201810296979.XA patent/CN108491543A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101196905A (en) * | 2007-12-05 | 2008-06-11 | 覃征 | Intelligent pattern searching method |
| CN103310460A (en) * | 2013-06-24 | 2013-09-18 | 安科智慧城市技术（中国）有限公司 | Image characteristic extraction method and system |
| CN103927387A (en) * | 2014-04-30 | 2014-07-16 | 成都理想境界科技有限公司 | Image retrieval system, method and device |
| CN106354735A (en) * | 2015-07-22 | 2017-01-25 | 杭州海康威视数字技术股份有限公司 | Image target searching method and device |
| CN107193965A (en) * | 2017-05-24 | 2017-09-22 | 哈尔滨工业大学 | A kind of quick indoor orientation method based on BoVW algorithms |
Non-Patent Citations (1)
| Title |
|---|
| 李洁: "基于 Hadoop 的海量视频的分布式存储与检索研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113678120A (en) * | 2018-11-20 | 2021-11-19 | 多伦多道明银行 | Automated image retrieval using images |
| CN109740019A (en) * | 2018-12-14 | 2019-05-10 | 上海众源网络有限公司 | A kind of method, apparatus to label to short-sighted frequency and electronic equipment |
| WO2020125544A1 (en) * | 2018-12-18 | 2020-06-25 | 华为技术有限公司 | Distributed image analysis method and system, and storage medium |
| CN110334763A (en) * | 2019-07-04 | 2019-10-15 | 北京字节跳动网络技术有限公司 | Model data file generation, image-recognizing method, device, equipment and medium |
| CN110334763B (en) * | 2019-07-04 | 2021-07-23 | 北京字节跳动网络技术有限公司 | Model data file generation method, model data file generation device, model data file identification device, model data file generation apparatus, model data file identification apparatus, and model data file identification medium |
| CN115525181A (en) * | 2022-11-28 | 2022-12-27 | 深圳飞蝶虚拟现实科技有限公司 | Method and device for manufacturing 3D content, electronic device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zheng et al. | SIFT meets CNN: A decade survey of instance retrieval | |
| US11048966B2 (en) | Method and device for comparing similarities of high dimensional features of images | |
| CN108491543A (en) | Image search method, image storage method and image indexing system | |
| CN110390054B (en) | Interest point recall method, device, server and storage medium | |
| US8392430B2 (en) | Concept-structured image search | |
| US8577882B2 (en) | Method and system for searching multilingual documents | |
| Afifi et al. | Image retrieval based on content using color feature | |
| CN112347284B (en) | Combined trademark image retrieval method | |
| CN107329987A (en) | A kind of search system based on mongo databases | |
| CN102855298A (en) | Image retrieval method and system | |
| CN110990597A (en) | Cross-modal data retrieval system based on text semantic mapping and retrieval method thereof | |
| Truong et al. | Video search based on semantic extraction and locally regional object proposal | |
| WO2010071617A1 (en) | Method and apparatus for performing image processing | |
| Rossetto et al. | Query by semantic sketch | |
| Tian et al. | Automatic image annotation with real-world community contributed data set | |
| CN114077682B (en) | Intelligent recognition matching processing method and system for image retrieval and storage medium | |
| CN115687663A (en) | Video retrieval and marking method, system and storage medium based on full text search | |
| CN115455249A (en) | Double-engine driven multi-modal data retrieval method, equipment and system | |
| CN111506754B (en) | Picture retrieval method, device, storage medium and processor | |
| CN115203474A (en) | Automatic database classification and extraction technology | |
| Khawandi et al. | A survey on image indexing and retrieval based on content based image | |
| CN113505257A (en) | Image search method, trademark search method, electronic device, and storage medium | |
| Sebastine et al. | Semantic web for content based video retrieval | |
| TW202004519A (en) | Method for automatically classifying images | |
| US11893012B1 (en) | Content extraction using related entity group metadata from reference objects |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180904 |
|
| RJ01 | Rejection of invention patent application after publication |