CN110263199A - It is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method - Google Patents
It is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method Download PDFInfo
- Publication number
- CN110263199A CN110263199A CN201910542851.1A CN201910542851A CN110263199A CN 110263199 A CN110263199 A CN 110263199A CN 201910542851 A CN201910542851 A CN 201910542851A CN 110263199 A CN110263199 A CN 110263199A
- Authority
- CN
- China
- Prior art keywords
- image
- feature
- picture
- search
- scheme
- 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.)
- Withdrawn
Links
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
- G06F16/53—Querying
- G06F16/538—Presentation of query results
-
- 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
- G06F16/55—Clustering; Classification
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Library & Information Science (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of based on the cartographical sketching of deep learning to scheme to search drawing method, including the picture input Edge Gradient Feature device in image data base is obtained edge feature, the edge feature is inputted into depth characteristic extractor again, the linear feature for obtaining an image indicates vector, it is stored by characteristics of image persistence service, is cached to after image border depth characteristic vector to be searched is carried out binaryzation compression to scheme to search in figure engine;The cartographical sketching of user's input is sent into feature extractor, obtaining linear feature indicates vector, it is searched figure engine with this feature vector and required most like picture number Xiang Yitu and is made requests, binary conversion treatment is carried out to scheme to search after figure engine obtains feature vector, then the characteristics of image in this feature and caching is subjected to similarity calculation, it is made requests after the most like Image ID of respective request quantity is ranked up by similarity to image data base, by returning result to requestor after ID index picture.
Description
Technical field
The present invention relates to figure retrieval technique field, it is specially a kind of based on the cartographical sketching of deep learning to scheme to search figure side
Method.
Background technique
To scheme to search figure, it is a kind of technology for searching for similar picture by one picture of input, provides correlation for user
The search technique of graph image data-searching.Relate to database, data buffer storage, computer vision, image procossing, information retrieval
Equal subjects.Its focus technology is character representation and similarity measurement.With scheme to search figure service big data image retrieval, mutually
The multiple fields such as networking picture materials search, shopping search are all widely used.Currently, using point based on deep neural network
Class device can obtain very ideal effect on visual task.
Karen Simonyan and Andrew Zisserman propose a kind of deep neural network structure for classification, quilt
Referred to as VGG (Visual Geometry Group) network, the network have very strong feature extraction ability, classify and detect with
And it is showed on other visual tasks good;The hidden layer feature of VGG network contains certain visual signature and semantic information.
Perceptual hash (Perceptual Hash) is a kind of skill according to its eap-message digest of image content and feature extraction
Art.Different from cryptographic Hash, cryptographic Hash can all create a difference very big Hash for the extremely small variation of input
Value, and be if two width pictures are similar the characteristics of perceptual hash, their perceptual hash value is also close.Perceptual hash is being schemed
As being widely used in information retrieval technique, tradition often uses perceptual hash to carry out image characteristics extraction to scheme to search diagram technology.
For, to scheme to search nomography, being more abstracted by the formation of neural network ensemble low-level feature based on deep learning
High level indicates attribute classification or feature, to find that the distributed nature of data indicates.Its significant advantage be can take out it is advanced
Feature constructs complicated high performance model.Feature extraction effective solution feature pumping is carried out using the middle layer of VGG network
Take problem.
Saining Xie and Zhuowen Tu propose a kind of deep neural network HED for edge detection
(Holistically-nested Edge Detection), the network are mentioned by combining the feature of different scale to carry out edge
It takes, obtains good effect, result is more natural compared to traditional edge detection operator.
But traditional is to be scanned for by existing picture, however there is a kind of situation, i.e. conduct to scheme to search figure all
The picture of searching request is simultaneously not present, the cartographical sketching of only artificial conceptual representation, and it is desirable to pass through simple Freehandhand-drawing grass
Graph search goes out form or the conceptive picture to match.To solve the disadvantage that the prior art and deficiency, the present invention propose one kind
Based on the cartographical sketching search technique of depth migration study, the linear feature of picture is extracted with deep neural network, uses step-by-step
Hamming distance carries out similarity measurement, to realize high performance similarity calculation.
Summary of the invention
The purpose of the present invention is to provide a kind of based on the cartographical sketching of deep learning to scheme to search drawing method, above-mentioned to solve
The problem of being proposed in background technique.
To achieve the above object, the invention provides the following technical scheme: it is a kind of based on the cartographical sketching of deep learning to scheme
Search drawing method, including following operating procedure:
S1: a neural network is constructed for extracting image border using training Edge Gradient Feature device, passes through edge spy
Sign data are trained, and the neural network can be used as the information that image edge information extractor extracts the edge of image later;
S2: passing through the deep neural network trained or directly use pre-training using depth characteristic extractor is obtained, will
Wherein the feature of interbed is as depth characteristic;
S3: the picture input Edge Gradient Feature device in image data base is obtained into edge feature, then by the edge feature
Depth characteristic extractor is inputted, the linear feature for obtaining an image indicates vector, carries out by characteristics of image persistence service
Storage, is indexed with Image ID;
S4: by read characteristics of image persistence service data, by image border depth characteristic vector to be searched into
It is cached to after the compression of row binaryzation to scheme to search in figure engine;
S5: when scanning for request, the cartographical sketching of user's input is sent into feature extractor, obtains linear feature expression
Vector is searched figure engine and is made requests, drawn with scheming to search figure with this feature vector and required most like picture number Xiang Yitu
It holds up and carries out binary conversion treatment after obtaining feature vector, the characteristics of image in this feature and caching is then subjected to similarity meter
It calculates, makes requests, pass through to image data base after the most like Image ID of respective request quantity is ranked up by similarity
Requestor is returned result to after ID index picture.
Preferably, this is searched drawing method and searches drawing system based on one kind, and the drawing system of searching includes image data base, depth characteristic
Extractor, picture edge characteristic extractor, the service of characteristics of image persistent storage and characteristics of image search engine.
Preferably, the coding characteristic for obtaining depth characteristic extractor needs to have good content representation ability, energy
It is under European measurement apart from size by the similarity map between picture.
Preferably, the S1) middle utilization training Edge Gradient Feature device extraction image border, including following operating procedure:
S11: downloading edge detection data collection, according to HED network establishment multilayer depth convolutional network;Depth convolutional network packet
Containing a large amount of convolutional layer, the last layer exports the picture that single channel includes picture edge characteristic;
S12: cutting, scaling pictures are to meet the input requirements of network, requires input surplus than the fixation for 1 to 1 here
Size;
S13: using Adam optimizer, using squared error function as loss function training deep neural network;Square
Error function is expressed as:
Wherein, output represents the output of network, and true_edge represents the real image edges manually marked.
Preferably, the S2) in using VGG16 deep neural network intermediate features layer be used as depth characteristic extractor, wrap
Include following operating procedure:
S21: the VGG16 neural network of the pre-training on ImageNet data set is obtained;
S22: removing its Nonlinear Classifier, and directly wherein interbed coding characteristic, obtained depth characteristic are to belong to for output
R1Vector.
Preferably, the S3) in for depth characteristic vector each component, set 1 greater than 0, set 0 less than or equal to 0;
Fixed length bit array is constructed, 1 is set on corresponding bit or sets 0, the character representation conduct output compressed.
Preferably, the S5) in carry out image similarity calculating, including following operating procedure in a search engine:
S51: the picture feature after this feature progress binaryzation with caching is subjected to step-by-step XOR operation;Compare in statistical result
1 number on special position obtains the request picture and caches the Hamming distance of picture, Image ID is pressed distance-taxis;
S52: when request, required picture number n can be incoming as parameter, and search engine takes apart from the smallest preceding n
The ID of picture is exported;
S53: search engine inquires picture address to picture database according to the Image ID of output, and according to image category into
Row sorts out and polymerization, and picture address is finally returned to requestor;
Preferably, the Adam optimizer, iteration update rule are as follows:
mt=β1mt-1+(1-β1)gt
Wherein β1=0.9, β2=0.999, ∈=10-8
Compared with prior art, the beneficial effects of the present invention are: including the following aspects:
One, the present invention carries out edge extracting by deep learning, compared to the edge detection that tradition is carried out by operator, originally
The method that invention proposes can preferably take out the edge feature of picture, be more in line with understanding and perception of the mankind for sketch.
Two, the present invention takes full advantage of the feature extraction ability of depth convolutional network, avoids artificial design features, extracts
Characteristic properties it is preferable.
Three, the present invention carries out measuring similarity using step-by-step Hamming distance, can obtain very high search by optimization
Performance.
Four, the present invention is scanning for not needing to provide the photo of physical presence when request, it is only necessary to sketch is provided,
User is greatly facilitated to use.
Five, the present invention can classify search result, allow user it is more accurate check search result.
Detailed description of the invention
Fig. 1 is general frame figure of the 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.
In the description of the present invention, it should be noted that the orientation of the instructions such as term "vertical", "upper", "lower", "horizontal"
Or positional relationship is to be based on the orientation or positional relationship shown in the drawings, and is merely for convenience of description of the present invention and simplification of the description, and
It is not that the device of indication or suggestion meaning or element must have a particular orientation, be constructed and operated in a specific orientation, therefore
It is not considered as limiting the invention.
In the description of the present invention, it is also necessary to which explanation is unless specifically defined or limited otherwise, term " setting ",
" installation ", " connected ", " connection " shall be understood in a broad sense, for example, it may be fixedly connected, may be a detachable connection or one
Connect to body;It can be mechanical connection, be also possible to be electrically connected;It can be directly connected, it can also be indirect by intermediary
It is connected, can be the connection inside two elements.For the ordinary skill in the art, it can manage as the case may be
Solve the concrete meaning of above-mentioned term in the present invention.
Referring to Fig. 1, the present invention provides a kind of technical solution: it is a kind of based on the cartographical sketching of deep learning to scheme to search figure side
Method, including following operating procedure:
S1: a neural network is constructed for extracting image border using training Edge Gradient Feature device, passes through edge spy
Sign data are trained, and the neural network can be used as the information that image edge information extractor extracts the edge of image later;
S2: passing through the deep neural network trained or directly use pre-training using depth characteristic extractor is obtained, will
Wherein the feature of interbed is as depth characteristic;
S3: the picture input Edge Gradient Feature device in image data base is obtained into edge feature, then by the edge feature
Depth characteristic extractor is inputted, the linear feature for obtaining an image indicates vector, carries out by characteristics of image persistence service
Storage, is indexed with Image ID;
S4: by read characteristics of image persistence service data, by image border depth characteristic vector to be searched into
It is cached to after the compression of row binaryzation to scheme to search in figure engine;
S5: when scanning for request, the cartographical sketching of user's input is sent into feature extractor, obtains linear feature expression
Vector is searched figure engine and is made requests, drawn with scheming to search figure with this feature vector and required most like picture number Xiang Yitu
It holds up and carries out binary conversion treatment after obtaining feature vector, the characteristics of image in this feature and caching is then subjected to similarity meter
It calculates, makes requests, pass through to image data base after the most like Image ID of respective request quantity is ranked up by similarity
Requestor is returned result to after ID index picture.
Further, this is searched drawing method and searches drawing system based on one kind, and the drawing system of searching includes image data base, depth spy
Levy extractor, picture edge characteristic extractor, the service of characteristics of image persistent storage and characteristics of image search engine.
Further, the coding characteristic for obtaining depth characteristic extractor needs to have good content representation ability,
It can be under European measurement apart from size by the similarity map between picture.
Further, the S1) middle utilize trains Edge Gradient Feature device to extract image border, including following operation walks
It is rapid:
S11: downloading edge detection data collection, according to HED network establishment multilayer depth convolutional network;Depth convolutional network packet
Containing a large amount of convolutional layer, the last layer exports the picture that single channel includes picture edge characteristic;
S12: cutting, scaling pictures are to meet the input requirements of network, requires input surplus than the fixation for 1 to 1 here
Size;
S13: using Adam optimizer, using squared error function as loss function training deep neural network;Square
Error function is expressed as:
Wherein, output represents the output of network, and true_edge represents the real image edges manually marked.
Further, the S2) in using VGG16 deep neural network intermediate features layer as depth characteristic extractor,
Including following operating procedure:
S21: the VGG16 neural network of the pre-training on ImageNet data set is obtained;
S22: removing its Nonlinear Classifier, and directly wherein interbed coding characteristic, obtained depth characteristic are to belong to for output
R1Vector.
Further, the S3) in for depth characteristic vector each component, set 1 greater than 0, setting less than or equal to 0
0;Fixed length bit array is constructed, 1 is set on corresponding bit or sets 0, the character representation conduct output compressed.
Further, the S5) in carry out image similarity calculating, including following operating procedure in a search engine:
S51: the picture feature after this feature progress binaryzation with caching is subjected to step-by-step XOR operation;Compare in statistical result
1 number on special position obtains the request picture and caches the Hamming distance of picture, Image ID is pressed distance-taxis;
S52: when request, required picture number n can be incoming as parameter, and search engine takes apart from the smallest preceding n
The ID of picture is exported;
S53: search engine inquires picture address to picture database according to the Image ID of output, and according to image category into
Row sorts out and polymerization, and picture address is finally returned to requestor;
Further, the Adam optimizer, iteration update rule are as follows:
mt=β1mt-1+(1-β1)gt
Wherein β1=0.9, β2=0,999, ∈=10-8。
Working principle: a neural network is constructed for extracting image border using training Edge Gradient Feature device, is passed through
Edge feature data are trained, and the neural network can be used as the edge of image edge information extractor extraction image later
Information;The deep neural network for passing through training using depth characteristic extractor is obtained or directly using pre-training, will be in-between
The feature of layer obtains edge feature as depth characteristic, by the picture input Edge Gradient Feature device in image data base, then will
The edge feature inputs depth characteristic extractor, and the linear feature for obtaining an image indicates vector, lasting by characteristics of image
Change service to be stored, be indexed with Image ID;By reading the data of characteristics of image persistence service, by figure to be searched
It is cached to after carrying out binaryzation compression as edge depth characteristic vector to scheme to search in figure engine;
Scan for request when, by user input cartographical sketching be sent into feature extractor, obtain linear feature indicate to
Amount, is searched figure engine and is made requests, to scheme to search figure engine with this feature vector and required most like picture number Xiang Yitu
It obtains feature vector and carries out binary conversion treatment later, the characteristics of image in this feature and caching is then subjected to similarity calculation,
It is made requests after the most like Image ID of respective request quantity is ranked up by similarity to image data base, passes through ID rope
Draw picture and returns result to requestor later.
It although an embodiment of the present invention has been shown and described, for the ordinary skill in the art, can be with
A variety of variations, modification, replacement can be carried out to these embodiments without departing from the principles and spirit of the present invention by understanding
And modification, the scope of the present invention is defined by the appended.
Claims (8)
1. it is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method, it is characterised in that: including following operating procedure:
S1: a neural network is constructed for extracting image border using training Edge Gradient Feature device, passes through edge feature number
According to being trained, the neural network can be used as the information that image edge information extractor extracts the edge of image later;
S2: passing through the deep neural network trained or directly use pre-training using depth characteristic extractor is obtained, will wherein
The feature of interbed is as depth characteristic;
S3: the picture input Edge Gradient Feature device in image data base is obtained into edge feature, then the edge feature is inputted
Depth characteristic extractor obtains the linear feature expression vector an of image, is stored by characteristics of image persistence service,
It is indexed with Image ID;
S4: image border depth characteristic vector to be searched is carried out two by the data by reading characteristics of image persistence service
It is cached to after value compression to scheme to search in figure engine;
S5: scan for request when, by user input cartographical sketching be sent into feature extractor, obtain linear feature indicate to
Amount, is searched figure engine and is made requests, to scheme to search figure engine with this feature vector and required most like picture number Xiang Yitu
It obtains feature vector and carries out binary conversion treatment later, the characteristics of image in this feature and caching is then subjected to similarity calculation,
It is made requests after the most like Image ID of respective request quantity is ranked up by similarity to image data base, passes through ID rope
Draw picture and returns result to requestor later.
2. it is according to claim 1 it is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method, it is characterised in that: should
It searches drawing method and drawing system is searched based on one kind, the drawing system of searching includes image data base, depth characteristic extractor, image border spy
Levy extractor, the service of characteristics of image persistent storage and characteristics of image search engine.
3. it is according to claim 1 it is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method, it is characterised in that: institute
The coding characteristic for stating acquisition depth characteristic extractor needs to have good content representation ability, can be by the similitude between picture
Be mapped as under European measurement apart from size.
4. it is according to claim 1 it is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method, it is characterised in that: institute
State S1) middle utilization training Edge Gradient Feature device extraction image border, including following operating procedure:
S11: downloading edge detection data collection, according to HED network establishment multilayer depth convolutional network;Depth convolutional network includes big
The convolutional layer of amount, the last layer export the picture that single channel includes picture edge characteristic;
S12: cutting, scaling pictures are to meet the input requirements of network, requires input surplus than the fixed dimension for 1 to 1 here;
S13: using Adam optimizer, using squared error function as loss function training deep neural network;Square error
Function representation are as follows:
Wherein, output represents the output of network, and true_edge represents the real image edges manually marked.
5. it is according to claim 1 it is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method, it is characterised in that: institute
State S2) in using VGG16 deep neural network intermediate features layer as depth characteristic extractor, including following operating procedure:
S21: the VGG16 neural network of the pre-training on ImageNet data set is obtained;
S22: removing its Nonlinear Classifier, and directly wherein interbed coding characteristic, obtained depth characteristic are to belong to R for output1To
Amount.
6. it is according to claim 1 it is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method, it is characterised in that: institute
State S3) in for depth characteristic vector each component, set 1 greater than 0, set 0 less than or equal to 0;Fixed length bit array is constructed,
1 is set on corresponding bit or sets 0, the character representation conduct output compressed.
7. it is according to claim 1 it is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method, it is characterised in that: institute
State S5) in carry out image similarity calculating, including following operating procedure in a search engine:
S51: the picture feature after this feature progress binaryzation with caching is subjected to step-by-step XOR operation;Bit in statistical result
Upper 1 number obtains the request picture and caches the Hamming distance of picture, Image ID is pressed distance-taxis;
S52: when request, required picture number n can be incoming as parameter, and search engine takes apart from the smallest preceding n picture
ID output;
S53: search engine inquires picture address to picture database according to the Image ID of output, and is returned according to image category
Class and polymerization, finally return to requestor for picture address.
8. it is according to claim 4 it is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method, it is characterised in that: institute
Adam optimizer is stated, iteration updates rule are as follows:
mt=β1mt-1+(1-β1)gt
Wherein β1=0.9, β2=0.999, ∈=10-8。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910542851.1A CN110263199A (en) | 2019-06-21 | 2019-06-21 | It is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910542851.1A CN110263199A (en) | 2019-06-21 | 2019-06-21 | It is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110263199A true CN110263199A (en) | 2019-09-20 |
Family
ID=67920232
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910542851.1A Withdrawn CN110263199A (en) | 2019-06-21 | 2019-06-21 | It is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110263199A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110737792A (en) * | 2019-10-22 | 2020-01-31 | 广东小天才科技有限公司 | Exercise searching method, exercise searching device, exercise searching equipment and storage medium |
CN114168770A (en) * | 2022-02-14 | 2022-03-11 | 成都四方伟业软件股份有限公司 | Deep learning-based method and device for searching images by images |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106126581A (en) * | 2016-06-20 | 2016-11-16 | 复旦大学 | Cartographical sketching image search method based on degree of depth study |
EP3166021A1 (en) * | 2015-11-06 | 2017-05-10 | Thomson Licensing | Method and apparatus for image search using sparsifying analysis and synthesis operators |
CN108009286A (en) * | 2017-12-25 | 2018-05-08 | 合肥阿巴赛信息科技有限公司 | A kind of Sketch Searching method based on deep learning |
CN109299303A (en) * | 2018-10-19 | 2019-02-01 | 中国石油大学(华东) | Freehand sketch retrieval method based on deformable convolution Yu depth network |
CN109408655A (en) * | 2018-10-19 | 2019-03-01 | 中国石油大学(华东) | The freehand sketch retrieval method of incorporate voids convolution and multiple dimensioned sensing network |
-
2019
- 2019-06-21 CN CN201910542851.1A patent/CN110263199A/en not_active Withdrawn
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP3166021A1 (en) * | 2015-11-06 | 2017-05-10 | Thomson Licensing | Method and apparatus for image search using sparsifying analysis and synthesis operators |
CN106126581A (en) * | 2016-06-20 | 2016-11-16 | 复旦大学 | Cartographical sketching image search method based on degree of depth study |
CN108009286A (en) * | 2017-12-25 | 2018-05-08 | 合肥阿巴赛信息科技有限公司 | A kind of Sketch Searching method based on deep learning |
CN109299303A (en) * | 2018-10-19 | 2019-02-01 | 中国石油大学(华东) | Freehand sketch retrieval method based on deformable convolution Yu depth network |
CN109408655A (en) * | 2018-10-19 | 2019-03-01 | 中国石油大学(华东) | The freehand sketch retrieval method of incorporate voids convolution and multiple dimensioned sensing network |
Non-Patent Citations (1)
Title |
---|
刘玉杰等: "基于条件生成对抗网络的手绘图像检索", 《计算机辅助设计与图形学学报》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110737792A (en) * | 2019-10-22 | 2020-01-31 | 广东小天才科技有限公司 | Exercise searching method, exercise searching device, exercise searching equipment and storage medium |
CN114168770A (en) * | 2022-02-14 | 2022-03-11 | 成都四方伟业软件股份有限公司 | Deep learning-based method and device for searching images by images |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107256246B (en) | printed fabric image retrieval method based on convolutional neural network | |
KR101388638B1 (en) | Annotating images | |
CN105469096B (en) | A kind of characteristic bag image search method based on Hash binary-coding | |
CN103258000B (en) | Method and device for clustering high-frequency keywords in webpages | |
Xu et al. | Remote sensing image scene classification based on generative adversarial networks | |
US20090049062A1 (en) | Method for Organizing Structurally Similar Web Pages from a Web Site | |
US20060155684A1 (en) | Systems and methods to present web image search results for effective image browsing | |
WO2021093308A1 (en) | Method and apparatus for extracting poi name, device, and computer storage medium | |
CN110688474B (en) | Embedded representation obtaining and citation recommending method based on deep learning and link prediction | |
CN107748799A (en) | A kind of method of multi-data source movie data entity alignment | |
Xie et al. | Fast and accurate near-duplicate image search with affinity propagation on the ImageWeb | |
CN102112987A (en) | Statistical approach to large-scale image annotation | |
Weyand et al. | Visual landmark recognition from internet photo collections: A large-scale evaluation | |
Qian et al. | Landmark summarization with diverse viewpoints | |
CN104346438A (en) | Data management service system based on large data | |
CN107392238A (en) | Outdoor knowledge of plants based on moving-vision search expands learning system | |
Zhou et al. | Relevance feature mapping for content-based multimedia information retrieval | |
CN110580302A (en) | Sketch image retrieval method based on semi-heterogeneous joint embedded network | |
CN109308324A (en) | A kind of image search method and system based on hand drawing style recommendation | |
CN109284411A (en) | One kind being based on having supervision hypergraph discretized image binary-coding method | |
CN110263199A (en) | It is a kind of based on the cartographical sketching of deep learning to scheme to search drawing method | |
CN111680506A (en) | External key mapping method and device of database table, electronic equipment and storage medium | |
CN107016566A (en) | User model construction method based on body | |
CN105159898A (en) | Searching method and searching device | |
CN106993240A (en) | Many video summarization methods based on sparse coding |
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 | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190920 |
|
WW01 | Invention patent application withdrawn after publication |