CN110502650A - A kind of image indexing system and method based on natural language description - Google Patents
A kind of image indexing system and method based on natural language description Download PDFInfo
- Publication number
- CN110502650A CN110502650A CN201910738598.7A CN201910738598A CN110502650A CN 110502650 A CN110502650 A CN 110502650A CN 201910738598 A CN201910738598 A CN 201910738598A CN 110502650 A CN110502650 A CN 110502650A
- Authority
- CN
- China
- Prior art keywords
- image
- mark
- user
- natural language
- language description
- 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
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000002372 labelling Methods 0.000 claims abstract description 16
- 230000004044 response Effects 0.000 claims abstract description 10
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000000605 extraction Methods 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 16
- 238000013528 artificial neural network Methods 0.000 claims description 15
- 239000000284 extract Substances 0.000 claims description 15
- 238000003062 neural network model Methods 0.000 claims description 11
- 230000007246 mechanism Effects 0.000 claims description 8
- 210000005036 nerve Anatomy 0.000 claims description 7
- 230000001537 neural effect Effects 0.000 claims description 5
- 230000000306 recurrent effect Effects 0.000 claims description 3
- 239000000463 material Substances 0.000 abstract description 3
- 230000001755 vocal effect Effects 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 11
- 238000012905 input function Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000012790 confirmation Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007717 exclusion Effects 0.000 description 2
- 230000009191 jumping Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 1
- 238000013549 information retrieval technique Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/53—Querying
- G06F16/532—Query formulation, e.g. graphical querying
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
Abstract
The present invention discloses a kind of image indexing system based on natural language description, including user interface and system two parts；The system includes preprocessing module, text matches module, automatic marking module；A kind of image search method based on natural language description, including pretreatment: before system the first secondary response user request, data needed for being packed into respond request in system running environment accelerate the speed of respond request；Image retrieval: system responds the retrieval request of user, and the image labeling having in the input of user and database i.e. natural language description is carried out text matches, obtains the high search result of matching degree.The present invention is labeled by the iamge description generation of automation, and rapid pin generates content-descriptive text to emerging picture；Solves the demand that image retrieval is carried out based on longer descriptive matter in which there, user only need to carry out verbal description to picture material, can be obtained the picture for relatively accurately meeting description.
Description
Technical field
The image indexing system method based on natural language description that the present invention relates to a kind of, belongs to computer picture vision skill
Art, natural language processing technique, information retrieval technique crossing domain.
Background technique
The rapid development of information technology and network technology provides greatly just for the generation and propagation of magnanimity the Internet images
Benefit.In this context, traditional image retrieval mode based on keyword, the retrieval precision provided are difficult to meet user's need
It asks.To scheme to search the retrieval mode of figure (image retrieval based on image), although being capable of providing higher retrieval precision, it is applicable in field
Scape is relatively limited.Therefore, research can be directly based upon descriptive sentence and carry out method and technology that image is precisely retrieved,
As the forward position focus problem of field of image search, it is with a wide range of applications.
The present invention is directed to the urgent need based on natural language description retrieval image, uses the automatic marking of image first
Technology automatically generates its natural language description for given image.Then, by the natural language description of the Internet images and the image
Be associated, construct the key-value pair of image retrieval, as based on natural language description carry out image retrieval basis, then pass through by
Key assignments carries out text justification marking and is retrieved, and has designed and Implemented and a set of has been integrated with the system of automatic marking and search function.
The system has to database uploading pictures and automatic marking, can substitute existing artificial mark, substantially increase image mark
The efficiency of note, to adapt to the rapidly growth of amount of images；Meanwhile the system can based on one section of longer descriptive sentence and
Non-keyword carries out accurate image retrieval, to overcome the limitation of existing mainstream search engine.
Summary of the invention
A kind of image indexing system and method based on natural language description of the present invention, technology solve the problems, such as: needle
To the automatic marking technology that large-scale image uploads, a kind of neural network model for automatically generating iamge description is provided, for mutual
The image newly uploaded in networking automatically generates the verbal description to picture material, as image using the neural network model
Mark；Further, a kind of algorithm is provided for based on natural language description progress image retrieval, it can be longer based on one section
Iamge description is matched and is retrieved to image.
A kind of image indexing system based on natural language description of the present invention, including user interface and system two parts；Its
In, the system includes preprocessing module, text matches module, automatic marking module；
The user interface proposes retrieval image for user or uploads the request of image, system is transferred to respond, specific to wrap
It includes:
Search interface: the input function area and submission request function area for needing to use when comprising user search.Input function
Area is the input of text provided by retrieving image for receiving user, can be completed by an input frame；Submitting request function area is
After user's confirmation input, the trigger means used when submitting and input to system can be designed as a button.
Upload interface: the upload file function area and submission request function area for needing to use when uploading image comprising user.
It uploads file function area and is used to receive the image that user wants upload, can be a towing file and put area, allow user will be literary
Part towing uploads after so far unclamping, and is also possible to a navigation button, browses file directory select file by clicking；Submission is asked
Asking functional areas is after user confirms image, and the trigger means used when submitting image to system can be designed as a button.
The preprocessing module, for being filled in system running environment before system the first secondary response user request
Data needed for entering respond request are specifically included with accelerating the speed of respond request:
Image access structure load units: the unit is used to arrive the path storage of the corresponding top catalogue of image file
In running environment, to obtain complete database images access structure.
The stem of mark extracts and load units: the unit is used to the content of image labeling file carrying out stem extraction,
Tense, part of speech and capital and small letter are excluded to storage after the influence of word into running environment, to obtain complete database mark
Set.
Neural network load units: the unit is used to read pre- for the neural net layer for needing to load weight in model
First trained model file, and the whole weight informations wherein having are extracted, and be loaded into corresponding layer, then by mould
Type is loaded into running environment.
The text matches module, when responding the retrieval request of user for system, by the input of user and database
In the image labeling that has i.e. natural language description carry out text matches, obtain the high search result of matching degree；It specifically includes:
Stem extraction unit: the sentence that the unit is used to input user carries out stem extraction, exclusion tense, part of speech and
Influence of the capital and small letter to word.
Matching primitives unit: the unit is used to traverse the mark extracted by stem having had in running environment, by it
The calculating that BLEU-4 score value is carried out with the sentence of user's input extracted by stem, gives a mark institute's score value as matching.
Return the result unit: the unit is used to mark according to matching score value and is ranked up with sequence from high to low, and
The corresponding picture of mark sequentially is returned as search result to user using this.
The automatic marking module: for system response user upload request when, by user upload picture according to
Neural network model, which generates, meets the content type description that the mankind state habit, by picture and description storage into database；Specifically
Include:
Mark unit: the unit is generated for that will upload image as input afferent nerve network model by the model
Natural language description is as to the mark for uploading image.
Save result unit: the unit is used to upload in image and the mark deposit database generated to this image,
And access path and the mark generated are updated to database images access structure and database mark set into running environment
In.
A kind of image search method based on natural language description, including pretreatment and image retrieval, detailed process is as follows:
S1 pretreatment: before system the first secondary response user request, respond request institute is packed into system running environment
The data needed, to accelerate the speed of respond request, concrete operations are as follows:
S1.1 image access structure loads: since having total data point top catalogue, subdirectory is gradually accessed, until
Access obtains image file, and the path of its opposed tips catalogue is stored into running environment.The process is to all image files
It carries out, obtains complete database images access structure after the completion.
S1.2 neural network loads: providing the model framework of neural network first.Neural network model is used with attention
The coder-decoder framework of power mechanism, encoder are a depth convolutional network ResNet101, and decoder is a length
When memory unit LSTM, attention mechanism is the soft attention mechanism realized by a feedforward neural network.Thereafter, for model
The middle neural net layer for needing to load weight reads trained model file in advance, and extracts the whole weights wherein having
Information, and be loaded into corresponding layer, then by model loading to running environment.
The stem of S1.3 mark is extracted and is loaded: since having total data point top catalogue, subdirectory is gradually accessed,
Until accessing the file for obtaining storing image labeling, (wherein, image labeling is produced when picture is uploaded by neural network model
It is raw), its content is subjected to stem extraction, excludes the influence of tense, part of speech and capital and small letter to word, then storage to operation ring
In border.The process carries out all image labeling files, obtains complete database mark set after the completion.
S2 image retrieval: text matches system responds the retrieval request of user, will have in the input of user and database
Image labeling, that is, natural language description carry out text matches, obtain the high search result of matching degree, the specific steps of which are as follows:
S2.1 stem extracts: carrying out stem extraction to user's read statement, excludes tense, part of speech and capital and small letter to word
Influence.For Indo-European family of languages language, this process, which can be, to be carried out small letter to each word, removes tense, extracts root etc.
Operation；To Han-Tibetan family language, this process can be the operation such as participle, synonymous conflation of words.
S2.2 matching primitives: the mark extracted by stem having had in traversal running environment, by itself and user's input
By stem extract sentence carry out BLEU-4 score value calculating, using institute's score value as matching marking.
S2.3 is returned the result: according to matching score value, mark is ranked up with sequence from high to low, with this sequentially to
Family returns to the corresponding picture of mark as search result.
A kind of image search method based on natural language description further comprises having automatic marking step, specific as follows:
S3 automatic marking: when system responds the upload request of user, the picture that user is uploaded is according to neural network model
It generates and meets the content type description that the mankind state habit, by picture and description storage into database, concrete operations are as follows:
S3.1 generates mark: will upload image as input afferent nerve network model, model is first by depth convolution net
Network extracts characteristics of image, characteristics of image is then screened incoming Recognition with Recurrent Neural Network by attention mechanism, sequence generates nature
Language description is as mark, this process can carry out several times, to generate multiple and different descriptions as mark.
S3.2 saves result: will upload in image and the mark deposit database generated to this image, as new images
Access path and the mark that generates update in database images access structure into running environment and database mark set
It goes, so that next new images can be retrieved.
The present invention is a kind of image indexing system and method based on natural language description, and advantage and effect are: solution
Newly there is the demand of picture in mark on a large scale under big data of having determined environment, is labeled by the iamge description generation of automation,
Rapid pin generates the content-descriptive text that the suitable mankind state habit to emerging picture；It solves based on longer description
Property text carry out image retrieval demand, user only need to picture material carry out verbal description, can be obtained and relatively accurately accord with
The picture for closing description, provides a great convenience for acquisition of information.
Detailed description of the invention
Fig. 1 is the composition block diagram of present system
Fig. 2 is the work flow diagram of neural network model in the present invention
Fig. 3 is one of user interface search operation chart of the present invention
Fig. 4 is the two of user interface search operation chart of the present invention
Fig. 5 is user interface upload operation schematic diagram of the present invention
Specific embodiment
With reference to the accompanying drawing, the following further describes the technical solution of the present invention.
As shown in Figure 1, a kind of image indexing system based on natural language description of the embodiment of the present invention, including user interface
With system two parts.Wherein, the system includes preprocessing module, text matches module, automatic marking module；
The user interface proposes retrieval image for user or uploads the request of image, system is transferred to respond, specific to wrap
It includes:
Search interface: the input function area and submission request function area for needing to use when comprising user search.Input function
Area is the input of text provided by retrieving image for receiving user, can be completed by an input frame；Submitting request function area is
After user's confirmation input, the trigger means used when submitting and input to system can be designed as a button.
Upload interface: the upload file function area and submission request function area for needing to use when uploading image comprising user.
It uploads file function area and is used to receive the image that user wants upload, can be a towing file and put area, allow user will be literary
Part towing uploads after so far unclamping, and is also possible to a navigation button, browses file directory select file by clicking；Submission is asked
Asking functional areas is after user confirms image, and the trigger means used when submitting image to system can be designed as a button.
The preprocessing module, for being filled in system running environment before system the first secondary response user request
Data needed for entering respond request are specifically included with accelerating the speed of respond request:
Image access structure load units: the unit is used to arrive the path storage of the corresponding top catalogue of image file
In running environment, to obtain complete database images access structure.
The stem of mark extracts and load units: the unit is used to the content of image labeling file carrying out stem extraction,
Tense, part of speech and capital and small letter are excluded to storage after the influence of word into running environment, to obtain complete database mark
Set.
Neural network load units: the unit is used to read pre- for the neural net layer for needing to load weight in model
First trained model file, and the whole weight informations wherein having are extracted, and be loaded into corresponding layer, then by mould
Type is loaded into running environment.
The text matches module, when responding the retrieval request of user for system, by the input of user and database
In the image labeling that has i.e. natural language description carry out text matches, obtain the high search result of matching degree；It specifically includes:
Stem extraction unit: the sentence that the unit is used to input user carries out stem extraction, exclusion tense, part of speech and
Influence of the capital and small letter to word.
Matching primitives unit: the unit is used to traverse the mark extracted by stem having had in running environment, by it
The calculating that BLEU-4 score value is carried out with the sentence of user's input extracted by stem, gives a mark institute's score value as matching.
Return the result unit: the unit is used to mark according to matching score value and is ranked up with sequence from high to low, and
The corresponding picture of mark sequentially is returned as search result to user using this.
The automatic marking module: for system response user upload request when, by user upload picture according to
Neural network model, which generates, meets the content type description that the mankind state habit, by picture and description storage into database；Specifically
Include:
Mark unit: the unit is generated for that will upload image as input afferent nerve network model by the model
Natural language description is as to the mark for uploading image.
Save result unit: the unit is used to upload in image and the mark deposit database generated to this image,
And access path and the mark generated are updated to database images access structure and database mark set into running environment
In.
The present invention is a kind of image indexing system based on natural language description, and the working condition of the system is as follows:
S1. data prediction and model loading state: system needs before the upload or retrieval request for beginning to respond to user
Database data is pre-processed to and loaded neural network model file, then enters wait state, so that waiting
Data need not be reprocessed when each secondary response user requests after state or load model.
S2. wait state: when system is waited for, will not actively carry out activity, but wait user's read statement
When carrying out search operaqtion, retrieval responsive state is entered back into；Or when uploading image progress upload operation, upload response shape is entered back into
State.
S3. retrieve responsive state: after system enters retrieval responsive state, the sentence inputted according to user is right in the database
Mark carries out marking matching, the corresponding image of the mark of high matching degree is returned to user, and return to wait state；
S4. upload responsive state: after system enters upload responsive state, the image that user is uploaded is based on neural network mould
Type generates one or more natural language descriptions, and by these marks of the description as image, then saves image and mark
In the database, while the database file access structure in running environment is updated, allows new uploading pictures in quilt later
Retrieval, then returns to wait state；
Wherein, data prediction and model loading work status process are specific as follows:
S1.1 access structure loads
Since having total data point top catalogue, subdirectory is gradually accessed, until access obtains image file, by it
The path of opposed tips catalogue is stored into running environment.The process carries out all image files, obtains after the completion complete
Database images access structure.
S1.2 neural network loads
The model framework of neural network is provided first, as shown in Figure 2.Thereafter, for the mind for needing to load weight in model
Through network layer, trained model file in advance is read, and extract the whole weight informations wherein having, and is loaded into corresponding
Layer in, then by model loading to running environment.
The stem of S1.3 mark is extracted and is loaded
Since having total data point top catalogue, subdirectory is gradually accessed, until access obtains storage image labeling
File, its content is subjected to stem extraction, excludes the influence to word of tense, part of speech and capital and small letter, then storage to fortune
In row environment.The process carries out all image labeling files, obtains complete database mark set after the completion.
Wherein, retrieval responsive state workflow is specific as follows:
S3.1 stem extracts
Stem extraction is carried out to user's read statement, excludes the influence of tense, part of speech and capital and small letter to word.
S3.2 matching primitives
Traversal running environment in have by stem extract mark, by itself and by stem extract user input into
The calculating of row BLEU-4 score value is given a mark institute's score value as matching.
S3.3 is returned the result
According to matching score value, mark is ranked up with sequence from high to low, mark institute is sequentially returned to user with this
Corresponding picture is as search result.
Wherein, it is specific as follows to upload responsive state workflow:
S4.1 generates mark
Image will be uploaded as input afferent nerve network model, it is special that model extracts image by depth convolutional network first
Characteristics of image is then screened incoming Recognition with Recurrent Neural Network by attention mechanism by sign, and sequence generates natural language description and makees, this
A process can carry out several times, to generate multiple and different descriptions as mark.
S4.2 saves result
It will upload in image and the description deposit database generated to this image, and by the access path and production of new images
Raw mark updates in image access structure and mark set into running environment, so that new images can be in next quilt
It retrieves.
Embodiment
As shown in figure 3, when being shown which directly to retrieve " riding a horse jumping over a fence "
Result.And in the image of leaping over obstacles as shown in figure 5, upload one description, one jockey rides, system, which automatically generates, retouches
State " a person riding a horse jumping over a hurdle ".After upload, then the result retrieved is such as
Shown in Fig. 4, it can be seen that the picture being just uploaded at this time can be retrieved, although the sentence of user's input and description are only
It is similar rather than completely the same.
Claims (9)
1. a kind of image indexing system based on natural language description, it is characterised in that: including user interface and system two parts；
Wherein, the system includes preprocessing module, text matches module, automatic marking module；
The user interface proposes retrieval image for user or uploads the request of image, system is transferred to respond, comprising: retrieval
Interface and upload interface；
The preprocessing module, for being packed into and ringing in system running environment before system the first secondary response user request
Required data should be requested, to accelerate the speed of respond request；
The text matches module will deposit when responding the retrieval request of user for system in the input of user and database
Some image labelings, that is, natural language description carries out text matches, obtains the high search result of matching degree；
The automatic marking module: when responding the upload request of user for system, the picture that user is uploaded is according to nerve
Network model, which generates, meets the content type description that the mankind state habit, by picture and description storage into database.
2. a kind of image indexing system based on natural language description according to claim 1, it is characterised in that: described
Preprocessing module specifically includes:
Image access structure load units: the unit is used for the path storage of the corresponding top catalogue of image file to operation
In environment, to obtain complete database images access structure；
The stem of mark extracts and load units: the unit is used to the content of image labeling file carrying out stem extraction, excludes
Tense, part of speech and capital and small letter into running environment, gather storage after the influence of word to obtain complete database mark；
Neural network load units: the unit is used to read instruction in advance for the neural net layer for needing to load weight in model
The model file perfected, and the whole weight informations wherein having are extracted, and be loaded into corresponding layer, then model is filled
It is downloaded in running environment.
3. a kind of image indexing system based on natural language description according to claim 1, it is characterised in that: described
Text matches module specifically includes:
Stem extraction unit: the sentence that the unit is used to input user carries out stem extraction, excludes tense, part of speech and size
Write the influence to word；
Matching primitives unit: the unit is used to traverse the mark extracted by stem having had in running environment, by it and uses
The sentence of family input extracted by stem carries out the calculating of BLEU-4 score value, gives a mark institute's score value as matching；
Return the result unit: the unit is used to mark according to matching score value and is ranked up with sequence from high to low, and with this
Sequence returns to the corresponding picture of mark as search result to user.
4. a kind of image indexing system based on natural language description according to claim 1, it is characterised in that: described
Automatic marking module specifically includes:
Mark unit: the unit generates nature by the model for that will upload image as input afferent nerve network model
Language description is as to the mark for uploading image；
Save result unit: the unit is used to upload in image and the mark deposit database for generating this image, and will
Access path and the mark generated update in database images access structure and database mark set into running environment.
5. a kind of image search method based on natural language description, it is characterised in that: including pretreatment and image retrieval, specifically
Process is as follows:
S1 pretreatment: it before system the first secondary response user request, is packed into needed for respond request in system running environment
Data, to accelerate the speed of respond request；
S2 image retrieval: system responds the retrieval request of user, i.e. by the image labeling having in the input of user and database
Natural language description carries out text matches, obtains the high search result of matching degree.
6. a kind of image search method based on natural language description according to claim 5, it is characterised in that: the step
Detailed process is as follows for rapid S1 pretreatment:
S1.1 image access structure loads: since having total data point top catalogue, subdirectory is gradually accessed, until access
Image file is obtained, the path of its opposed tips catalogue is stored into running environment；The process carries out all image files,
Complete database images access structure is obtained after the completion；
S1.2 neural network loads: providing the model framework of neural network first；Neural network model, which uses, has attention machine
The coder-decoder framework of system, encoder are a depth convolutional network ResNet101, and decoder is remembered in short-term to be one long
Recall unit LSTM, attention mechanism is the soft attention mechanism realized by a feedforward neural network；Thereafter, for being needed in model
The neural net layer of weight is loaded, reads trained model file in advance, and extract the whole weight informations wherein having,
And be loaded into corresponding layer, then by model loading to running environment；
The stem of S1.3 mark is extracted and is loaded: since having total data point top catalogue, subdirectory is gradually accessed, until
Access obtains the file of storage image labeling, its content is carried out stem extraction, excludes tense, part of speech and capital and small letter to word
Influence, then store in running environment；The process carries out all image labeling files, is completely counted after the completion
It marks and gathers according to library.
7. a kind of image search method based on natural language description according to claim 5, it is characterised in that: the step
Specific step is as follows by rapid S2:
S2.1 stem extracts: carrying out stem extraction to user's read statement, excludes tense, part of speech and capital and small letter to the shadow of word
It rings；
S2.2 matching primitives: the mark extracted by stem having had in traversal running environment, by the warp of itself and user's input
The sentence for crossing stem extraction carries out the calculating of BLEU-4 score value, gives a mark institute's score value as matching；
S2.3 is returned the result: according to matching score value, mark being ranked up with sequence from high to low, is sequentially returned to user with this
The corresponding picture of mark is returned as search result.
8. a kind of image search method based on natural language description according to claim 5, it is characterised in that: this method
It further comprise having automatic marking step, specific as follows:
S3 automatic marking: when system responds the upload request of user, the picture that user uploads is generated according to neural network model
Meet the content type description that the mankind state habit, by picture and description storage into database.
9. a kind of image search method based on natural language description according to claim 8, it is characterised in that: the step
Detailed process is as follows for rapid S3 automatic marking:
S3.1 generates mark: will upload image as input afferent nerve network model, model is mentioned by depth convolutional network first
Characteristics of image is taken, characteristics of image is then screened into incoming Recognition with Recurrent Neural Network by attention mechanism, sequence generates natural language
Description is as mark, this process can carry out several times, to generate multiple and different descriptions as mark；
S3.2 saves result: will upload in image and the mark deposit database generated to this image, the visit as new images
It asks the way diameter and the mark that generates updates in database images access structure into running environment and database mark set, with
Just next new images can be retrieved.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910738598.7A CN110502650A (en) | 2019-08-12 | 2019-08-12 | A kind of image indexing system and method based on natural language description |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910738598.7A CN110502650A (en) | 2019-08-12 | 2019-08-12 | A kind of image indexing system and method based on natural language description |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110502650A true CN110502650A (en) | 2019-11-26 |
Family
ID=68587194
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910738598.7A Pending CN110502650A (en) | 2019-08-12 | 2019-08-12 | A kind of image indexing system and method based on natural language description |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110502650A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111949815A (en) * | 2020-08-17 | 2020-11-17 | 西南石油大学 | Artificial intelligent corrosion material selection system based on image recognition technology and working method |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020081024A1 (en) * | 2000-12-26 | 2002-06-27 | Sung Hee Park | Apparatus and method for retrieving color and shape of image based on natural language |
| CN101110077A (en) * | 2007-08-24 | 2008-01-23 | 新诺亚舟科技（深圳）有限公司 | Method for implementing associated searching on handhold learning terminal |
| CN104966031A (en) * | 2015-07-01 | 2015-10-07 | 复旦大学 | Method for identifying permission-irrelevant private data in Android application program |
| CN106202422A (en) * | 2016-07-12 | 2016-12-07 | 腾讯科技（深圳）有限公司 | The treating method and apparatus of Web page icon |
| CN106708565A (en) * | 2016-11-29 | 2017-05-24 | 武汉斗鱼网络科技有限公司 | Method and device for removing useless picture resources in APK (Android Package) |
| CN107563409A (en) * | 2017-08-04 | 2018-01-09 | 汕头大学 | A kind of description method based on area image feature concern network with arest neighbors sequence |
| CN108681541A (en) * | 2018-01-17 | 2018-10-19 | 百度在线网络技术（北京）有限公司 | Image searching method, device and computer equipment |
| CN109190471A (en) * | 2018-07-27 | 2019-01-11 | 天津大学 | The attention model method of video monitoring pedestrian search based on natural language description |
| CN109685103A (en) * | 2018-11-13 | 2019-04-26 | 成都四方伟业软件股份有限公司 | A kind of text Multi-label learning method based on broad sense K mean algorithm |
| CN110083729A (en) * | 2019-04-26 | 2019-08-02 | 北京金山数字娱乐科技有限公司 | A kind of method and system of picture search |
-
2019
- 2019-08-12 CN CN201910738598.7A patent/CN110502650A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020081024A1 (en) * | 2000-12-26 | 2002-06-27 | Sung Hee Park | Apparatus and method for retrieving color and shape of image based on natural language |
| CN101110077A (en) * | 2007-08-24 | 2008-01-23 | 新诺亚舟科技（深圳）有限公司 | Method for implementing associated searching on handhold learning terminal |
| CN104966031A (en) * | 2015-07-01 | 2015-10-07 | 复旦大学 | Method for identifying permission-irrelevant private data in Android application program |
| CN106202422A (en) * | 2016-07-12 | 2016-12-07 | 腾讯科技（深圳）有限公司 | The treating method and apparatus of Web page icon |
| CN106708565A (en) * | 2016-11-29 | 2017-05-24 | 武汉斗鱼网络科技有限公司 | Method and device for removing useless picture resources in APK (Android Package) |
| CN107563409A (en) * | 2017-08-04 | 2018-01-09 | 汕头大学 | A kind of description method based on area image feature concern network with arest neighbors sequence |
| CN108681541A (en) * | 2018-01-17 | 2018-10-19 | 百度在线网络技术（北京）有限公司 | Image searching method, device and computer equipment |
| CN109190471A (en) * | 2018-07-27 | 2019-01-11 | 天津大学 | The attention model method of video monitoring pedestrian search based on natural language description |
| CN109685103A (en) * | 2018-11-13 | 2019-04-26 | 成都四方伟业软件股份有限公司 | A kind of text Multi-label learning method based on broad sense K mean algorithm |
| CN110083729A (en) * | 2019-04-26 | 2019-08-02 | 北京金山数字娱乐科技有限公司 | A kind of method and system of picture search |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111949815A (en) * | 2020-08-17 | 2020-11-17 | 西南石油大学 | Artificial intelligent corrosion material selection system based on image recognition technology and working method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110633409B (en) | Automobile news event extraction method integrating rules and deep learning | |
| CN106528845B (en) | Retrieval error correction method and device based on artificial intelligence | |
| JP6423845B2 (en) | Method and system for dynamically ranking images to be matched with content in response to a search query | |
| CN104216913B (en) | Question answering method, system and computer-readable medium | |
| CN109829104A (en) | Pseudo-linear filter model information search method and system based on semantic similarity | |
| CN107145496A (en) | The method for being matched image with content item based on keyword | |
| CN108228571B (en) | Method and device for generating couplet, storage medium and terminal equipment | |
| CN101251852B (en) | Integrating system and method of Web data facing to field | |
| WO2020010834A1 (en) | Faq question and answer library generalization method, apparatus, and device | |
| CN105335487A (en) | Agricultural specialist information retrieval system and method on basis of agricultural technology information ontology library | |
| CN110427483A (en) | Text snippet evaluating method, device, system and evaluation and test server | |
| CN110008309A (en) | A kind of short phrase picking method and device | |
| CN107463592A (en) | For by the method, equipment and data handling system of content item and images match | |
| CN101661469A (en) | System and method for indexing and retrieving keywords of academic documents | |
| CN103455896A (en) | Paperless assembling quality control method based on internet of things | |
| CN106802937A (en) | The conversion method and system of Word document | |
| CN110110047B (en) | Topic content aggregation analysis method based on TF-IDF and domain dictionary | |
| US20100174719A1 (en) | System, method, and program product for personalization of an open network search engine | |
| US11379527B2 (en) | Sibling search queries | |
| Darmawiguna et al. | The development of integrated Bali tourism information portal using web scrapping and clustering methods | |
| CN110502650A (en) | A kind of image indexing system and method based on natural language description | |
| CN110222252A (en) | Information retrieval method, device and equipment | |
| CN111859074B (en) | Network public opinion information source influence evaluation method and system based on deep learning | |
| CN112463896B (en) | Archive catalogue data processing method, archive catalogue data processing device, computing equipment and storage medium | |
| Álvarez et al. | A Task-specific Approach for Crawling the Deep Web. |
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: 20191126 |
|
| RJ01 | Rejection of invention patent application after publication |