CN110069643A - The method and system of similar building picture searching - Google Patents
The method and system of similar building picture searching Download PDFInfo
- Publication number
- CN110069643A CN110069643A CN201910175878.1A CN201910175878A CN110069643A CN 110069643 A CN110069643 A CN 110069643A CN 201910175878 A CN201910175878 A CN 201910175878A CN 110069643 A CN110069643 A CN 110069643A
- Authority
- CN
- China
- Prior art keywords
- feature
- picture
- building
- points
- building picture
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of method and systems of similar building picture searching, wherein method includes: to carry out feature extraction to building picture using deep neural network, obtains dense feature；The picture of building training sample and the data set of picture mark are obtained, landmarks categorizations device is obtained with training to the summation pondization operation that dense feature and data set are weighted；Dense feature is chosen according to terrestrial reference classifier and preset attention mechanism, obtains key feature points；Dimensionality reduction is carried out to key feature points using Principal Component Analysis, retains K Important Characteristic Points；The search for carrying out multidimensional space data in building picture library using KD tree according to K Important Characteristic Points, finds the most similar characteristic strong point of feature, and export corresponding similar building picture.The present invention can bring the characteristic matching and geometric verification of higher precision；And the high-efficient and accuracy of search is high.
Description
Technical field
The present invention relates to picture searching technical fields, more particularly to a kind of method and system of similar building picture searching.
Background technique
Internet presents explosive growth in recent years, and daily-life related that picture resource becomes more
Exacerbation is wanted, and building picture is due to shooting angle, shooting time, and the reasons such as environment when shooting, the picture of the same building is very
It might have numerous different forms of expression.Since building picture has a variety of forms of expression, so that building picture searching is more
Add difficulty.
Currently, due to building the particularity of picture, the overall situation usually using building picture is special in some existing researchs
The either whole local feature of sign is realized to indicate a picture using the either whole local feature of global characteristics
Similar pictures search.But the calculation amount that this method has whole characteristic similarities is very huge, characteristic matching precision is low, leads
The efficiency and accuracy for causing similar building picture searching reduce.
Summary of the invention
The method and system of similar building picture searching provided by the invention, main purpose are to overcome existing similar build
Build picture searching there are calculation amounts huge, characteristic matching precision is low, the problem of causing efficiency and accuracy to reduce.
In order to solve the above technical problems, the present invention adopts the following technical scheme:
A kind of method of similar building picture searching, includes the following steps；
Feature extraction is carried out to building picture using deep neural network, obtains dense feature；
The picture of building training sample and the data set of picture mark are obtained, the dense feature and data set are added
The summation pondization operation of power obtains landmarks categorizations device with training；According to the landmarks categorizations device and preset attention mechanism to close
Collection feature is chosen, and key feature points are obtained；
Dimensionality reduction is carried out to the key feature points using Principal Component Analysis, retains K Important Characteristic Points；
The search for being carried out multidimensional space data in building picture library using KD tree according to K Important Characteristic Points, is found special
Most similar characteristic strong point is levied, and exports corresponding similar building picture.
As an embodiment, described that feature extraction is carried out to building picture using deep neural network, it obtains close
Collect feature, includes the following steps；
Building picture is converted, RGB picture is obtained；
The RGB picture is input in ResNet50 network, carries out spy using the full articulamentum of ResNet50 network
Sign extracts, and obtains dense feature.
As an embodiment, described that dimensionality reduction is carried out to the key feature points using Principal Component Analysis, retain K
A Important Characteristic Points, include the following steps；
According to the key feature points construct original sample set, and be calculated covariance matrix, feature vector and
Characteristic value；Dimensionality reduction is carried out to the key feature points according to characteristic value, retains K Important Characteristic Points.
As an embodiment, described to construct original sample set according to the key feature points, and be calculated
Covariance matrix, feature vector and characteristic value, include the following steps；
Original sample set is constructed according to the key feature points, to original sample matrix normalization, and with dimension initial value
The average value for subtracting the dimension obtains new sample matrix, then multiplied by the transposition of sample matrix, then divided by (N-1) obtains association side
Poor matrix；And covariance matrix is carried out corresponding feature vector and characteristic value is calculated.
As an embodiment, all kinds of building pictures, the architectural drawing are previously stored in the building picture library
The characteristic strong point of piece is a kind of data structure divided in K dimensional feature space.
Correspondingly, the present invention also provides a kind of system of similar building picture searching, including abstraction module, choose module,
Dimensionality reduction module and index extraction module；
The abstraction module obtains dense feature for carrying out feature extraction to building picture using deep neural network；
The selection module, for obtaining the picture of building training sample and the data set of picture mark, to described intensive
The summation pondization operation that feature and data set are weighted obtains landmarks categorizations device with training；According to the landmarks categorizations device and in advance
If attention mechanism dense feature is chosen, obtain key feature points；
It is a important to retain K for carrying out dimensionality reduction to the key feature points using Principal Component Analysis for the dimensionality reduction module
Characteristic point；
The index extraction module, for carrying out multidimensional in building picture library using KD tree according to K Important Characteristic Points
The search of spatial data finds the most similar characteristic strong point of feature, and exports corresponding similar building picture.
As an embodiment, the selection module is also used to, and building picture is converted, RGB picture is obtained；
The RGB picture is input in ResNet50 network, carries out spy using the full articulamentum of ResNet50 network
Sign extracts, and obtains dense feature.
As an embodiment, the dimensionality reduction module is also used to；Original sample is constructed according to the key feature points
Set, and covariance matrix, feature vector and characteristic value is calculated；The key feature points are dropped according to characteristic value
Dimension retains K Important Characteristic Points.
As an embodiment, the dimensionality reduction module includes PCA unit；
The PCA unit, for constructing original sample set according to the key feature points, to original sample matrix normalizing
Change, and obtain new sample matrix with the average value that dimension initial value subtracts the dimension, then multiplied by the transposition of sample matrix, then removes
Covariance matrix is obtained with (N-1)；And covariance matrix is carried out corresponding feature vector and characteristic value is calculated.
As an embodiment, all kinds of building pictures, the architectural drawing are previously stored in the building picture library
The characteristic strong point of piece is a kind of data structure divided in K dimensional feature space.
Compared with prior art, the technical program has the advantage that
It is provided by the invention it is similar building picture searching method and system, using deep neural network to building picture into
Row feature extraction, obtains dense feature；It is selected on the basis of dense feature according to terrestrial reference classifier and preset attention mechanism
The key feature points obtained can bring the characteristic matching and geometric verification of higher precision；And utilize Principal Component Analysis pair
Key feature points carry out dimensionality reduction, retain K Important Characteristic Points, carry out hyperspace number in building picture library with Important Characteristic Points
According to search, and then realize it is similar building picture search；To improve the efficiency and accuracy of search.
Detailed description of the invention
Fig. 1 is the flow diagram of the method for the similar building picture searching that the embodiment of the present invention one provides；
Fig. 2 is the schematic illustration for building picture feature extraction；
Fig. 3 is the schematic illustration building the key feature points of picture and choosing；
Fig. 4 is the structural schematic diagram of the system of similar building picture searching provided by Embodiment 2 of the present invention.
In figure: 100, abstraction module；200, module is chosen；300, dimensionality reduction module；400, extraction module is indexed.
Specific embodiment
Below in conjunction with attached drawing, the technical characteristic and advantage above-mentioned and other to the present invention are clearly and completely described,
Obviously, described embodiment is only section Example of the invention, rather than whole embodiments.
Referring to Fig. 1, the method for the similar building picture searching that the embodiment of the present invention one provides, includes the following steps；
S100, feature extraction is carried out to building picture using deep neural network, obtains dense feature；
The data set of S200, the picture for obtaining building training sample and picture mark, carry out dense feature and data set
The summation pondization operation of weighting obtains landmarks categorizations device with training；According to terrestrial reference classifier and preset attention mechanism to intensive
Feature is chosen, and key feature points are obtained；
S300, dimensionality reduction is carried out to key feature points using Principal Component Analysis, retains K Important Characteristic Points；
S400, the search for being carried out multidimensional space data in building picture library using KD tree according to K Important Characteristic Points, are sought
The most similar characteristic strong point of feature is looked for, and exports corresponding similar building picture.
It should be noted that classifying for any picture recognition, it is required to first extract feature, that is, passes through depth nerve
Network extracts the abstract characteristic of picture profound level, and this characteristic, which is abstracted into, can distinguish two different pictures.For example, its
In include picture color gray scale, profile, lines bending degree etc., these features can distinguish two different pictures.Yu Ben
In embodiment, the extraction of feature is realized using convolutional neural networks (CNN) framework.Specific step is permissible are as follows: by architectural drawing
Piece is converted, and RGB picture is obtained；RGB picture is input in ResNet50 network, the full connection of ResNet50 network is utilized
Layer carries out feature extraction, obtains dense feature.It is input to it is, building picture processing is converted to RGB picture
In ResNet50 network, the advanced features of acquisition are connected entirely, the institute of one layer of each node of full articulamentum and front
There is node (feature) to be connected, advanced features is integrated, that is, the calculating process of a linear weighted sum, obtained close
It is as shown in Figure 2 to collect feature.In other embodiments, it can be realized using VGG, GoogleNet and Resnet etc..
In this present embodiment, the selection of key feature points is committed step, in order to ensure extraction feature to building picture
The validity of retrieval tasks chooses dense feature according to terrestrial reference classifier and preset attention mechanism, obtains key
Characteristic point.Landmarks categorizations device is a miniature neural network, is for measuring the correlation between local feature.Landmarks categorizations device
Training be the summation pond that is weighted of data set by picture and picture mark to dense feature and building training sample
Change what operation was obtained with training.Attention is focused directly in dense feature and is carried out.That is, directly on high-level semantic
Key feature points detection is carried out, rather than is detected in original image, characteristic matching and the geometric verification such as Fig. 3 institute of higher precision are brought
Show.It can be more accurate using the K Important Characteristic Points that Principal Component Analysis (PCA) obtains key feature points progress dimensionality reduction
Characterize a picture.
In building picture library, all kinds of building pictures and characteristic strong point corresponding with every building picture are store, are built
Build the characteristic strong point of picture is a kind of data structure divided in K dimensional feature space.It can be understood as every building picture
It can be the pre- process for first passing through step S100 to step S300.And in this present embodiment, using in KD tree (K-
Dimension tree) be indexed in building picture library, retrieve in k-d tree with query point apart from nearest data point and
Related corresponding building picture realizes the search of similar building picture.
It is provided by the invention it is similar building picture searching method and system, using deep neural network to building picture into
Row feature extraction, obtains dense feature；It is selected on the basis of dense feature according to terrestrial reference classifier and preset attention mechanism
The key feature points obtained can bring the characteristic matching and geometric verification of higher precision；And utilize Principal Component Analysis pair
Key feature points carry out dimensionality reduction, retain K Important Characteristic Points, carry out hyperspace number in building picture library with Important Characteristic Points
According to search, and then realize it is similar building picture search；To improve the efficiency and accuracy of search.
Further, dimensionality reduction is carried out to key feature points using Principal Component Analysis, retains K Important Characteristic Points, including
Following steps；
Original sample set is constructed according to key feature points, and covariance matrix, feature vector and feature is calculated
Value；Dimensionality reduction is carried out to key feature points according to characteristic value, retains K Important Characteristic Points.It is to be understood that find out first it is main at
The direction divided, the i.e. maximum direction of data variance；Similarly the direction of second principal component is successively found out (with first according to variance again
A principal component direction is orthogonal, just cries if it is two-dimensional space vertical), by the covariance matrix, the feature vector that calculate data set
And characteristic value, so that it may retain maximum K Important Characteristic Points.
And original sample set is constructed according to key feature points, and covariance matrix, feature vector and spy is calculated
Value indicative includes the following steps；According to key feature points construct original sample set, to original sample matrix normalization, and with tie up
The average value that degree initial value subtracts the dimension obtains new sample matrix, then multiplied by the transposition of sample matrix, then must divided by (N-1)
To covariance matrix；And covariance matrix is carried out corresponding feature vector and characteristic value is calculated.For example, key feature points
For 10 i.e. 10 samples, each there are 3 dimensions.It must be integer in order to facilitate the data in each dimension of computational rules,
Finally to construct an INTEGER MATRICES.
So INTEGER MATRICES are as follows:
Known by above-mentioned formula, about the row according to matrix either column count the problem of before highlight, covariance
Matrix, which calculates, to be focused on calculating between different dimensions.A line of MySample matrix represents a sample in initial data, a column definition
For one tie up, so will by dimension rather than sample calculate mean value.It is assigned to three dimensions respectively: 108.3222 74.5333-
10.0889；74.5333 260.6222 -106.4000；-10.0889 -106.4000 94.1778.
The calculating process of covariance matrix is to normalize to sample matrix, the average value of the dimension is subtracted with dimension initial value
New sample matrix is obtained, then multiplied by the transposition of sample matrix, then divided by (N-1) is exactly final covariance matrix.It calculates
Obtained covariance matrix is:
So feature vector and characteristic value of the covariance matrix are as follows:
To characteristic value and vector order, the smallest value of weight is deleted to reduce dimension, certain letters will necessarily be lost here
Breath, as long as but the information of the sufficiently small loss of specific gravity is worth to be ignored.Final result is to obtain less dimension, such as
Data of the fruit before selecting in the initial data of N-dimensional after K main component dimensionality reduction just only have K dimension.
Based on the same inventive concept, the embodiment of the present invention also provides a kind of system of similar building picture searching, the system
Implementation can refer to the above method process realize, repeat place it is no longer redundant later.
As shown in figure 4, being the structural schematic diagram of the system of similar building picture searching provided by Embodiment 2 of the present invention, packet
It includes abstraction module 100, choose module 200, dimensionality reduction module 300 and index extraction module 400；
Abstraction module 100 is used to carry out feature extraction to building picture using deep neural network, obtains dense feature；
Choose module 200 be used for obtains building training sample picture and picture mark data set, to dense feature with
The summation pondization operation that data set is weighted obtains landmarks categorizations device with training；According to terrestrial reference classifier and preset attention
Mechanism chooses dense feature, obtains key feature points；
Dimensionality reduction module 300 is used to carry out dimensionality reduction to key feature points using Principal Component Analysis, retains K important feature
Point；
Extraction module 400 is indexed to be used to carry out multidimensional sky in building picture library using KD tree according to K Important Characteristic Points
Between data search, find the most similar characteristic strong point of feature, and export corresponding similar building picture.
The present invention can bring the characteristic matching and geometric verification of higher precision；And utilize Principal Component Analysis to crucial special
Sign point carries out dimensionality reduction, retains K Important Characteristic Points, carries out searching for multidimensional space data in building picture library with Important Characteristic Points
Rope, and then realize the search of similar building picture；To improve the efficiency and accuracy of search.
Further, it chooses module 200 to be also used to, building picture is converted, RGB picture is obtained；RGB picture is defeated
Enter into ResNet50 network, carries out feature extraction using the full articulamentum of ResNet50 network, obtain dense feature.
In order to improve dimensionality reduction efficiency, dimensionality reduction module 300 is also used to；Original sample set is constructed according to key feature points, and
Covariance matrix, feature vector and characteristic value is calculated；Dimensionality reduction is carried out to key feature points according to characteristic value, retains K
Important Characteristic Points.
Further, dimensionality reduction module 300 includes PCA unit；PCA unit is used to construct original sample according to key feature points
Set, obtains new sample matrix to original sample matrix normalization, and with the average value that dimension initial value subtracts the dimension, then
Covariance matrix is obtained multiplied by the transposition of sample matrix, then divided by (N-1)；And covariance matrix be calculated corresponding
Feature vector and characteristic value.
Further, it builds in picture library and is previously stored with all kinds of building pictures, the characteristic strong point for building picture is in K
A kind of data structure that dimensional feature space divides.
Although the invention has been described by way of example and in terms of the preferred embodiments, but it is not for limiting the present invention, any this field
Technical staff without departing from the spirit and scope of the present invention, may be by the methods and technical content of the disclosure above to this hair
Bright technical solution makes possible variation and modification, therefore, anything that does not depart from the technical scheme of the invention, and according to the present invention
Technical spirit any simple modifications, equivalents, and modifications to the above embodiments, belong to technical solution of the present invention
Protection scope.
Claims (10)
1. a kind of method of similar building picture searching, which is characterized in that include the following steps；
Feature extraction is carried out to building picture using deep neural network, obtains dense feature；
The picture of building training sample and the data set of picture mark are obtained, the dense feature and data set are weighted
Summation pondization operation obtains landmarks categorizations device with training；According to the landmarks categorizations device and preset attention mechanism to intensive spy
Sign is chosen, and key feature points are obtained；
Dimensionality reduction is carried out to the key feature points using Principal Component Analysis, retains K Important Characteristic Points；
The search for carrying out multidimensional space data in building picture library using KD tree according to K Important Characteristic Points, finds feature most
Similar characteristic strong point, and export corresponding similar building picture.
2. the method for similar building picture searching as described in claim 1, which is characterized in that described to utilize deep neural network
Feature extraction is carried out to building picture, dense feature is obtained, includes the following steps；
Building picture is converted, RGB picture is obtained；
The RGB picture is input in ResNet50 network, carries out feature pumping using the full articulamentum of ResNet50 network
It takes, obtains dense feature.
3. the method for similar building picture searching as described in claim 1, which is characterized in that described to utilize Principal Component Analysis
Dimensionality reduction is carried out to the key feature points, retains K Important Characteristic Points, includes the following steps；
Original sample set is constructed according to the key feature points, and covariance matrix, feature vector and feature is calculated
Value；Dimensionality reduction is carried out to the key feature points according to characteristic value, retains K Important Characteristic Points.
4. the method for similar building picture searching as claimed in claim 3, which is characterized in that described according to the key feature
Point building original sample set, and covariance matrix, feature vector and characteristic value is calculated, include the following steps；
Original sample set is constructed according to the key feature points, is subtracted to original sample matrix normalization, and with dimension initial value
The average value of the dimension obtains new sample matrix, then multiplied by the transposition of sample matrix, then divided by (N-1) obtains covariance square
Battle array；And covariance matrix is carried out corresponding feature vector and characteristic value is calculated.
5. the method for similar building picture searching as described in claim 1, which is characterized in that in the building picture library in advance
All kinds of building pictures are stored with, the characteristic strong point of the building picture is a kind of data structure divided in K dimensional feature space.
6. a kind of system of similar building picture searching, which is characterized in that including abstraction module, choose module, dimensionality reduction module with
And index extraction module；
The abstraction module obtains dense feature for carrying out feature extraction to building picture using deep neural network；
The selection module, for obtaining the picture of building training sample and the data set of picture mark, to the dense feature
The summation pondization operation being weighted with data set obtains landmarks categorizations device with training；According to the landmarks categorizations device and preset
Attention mechanism chooses dense feature, obtains key feature points；
The dimensionality reduction module retains K important feature for carrying out dimensionality reduction to the key feature points using Principal Component Analysis
Point；
The index extraction module, for carrying out hyperspace in building picture library using KD tree according to K Important Characteristic Points
The search of data finds the most similar characteristic strong point of feature, and exports corresponding similar building picture.
7. the system of similar building picture searching as claimed in claim 6, which is characterized in that the selection module is also used to,
Building picture is converted, RGB picture is obtained；
The RGB picture is input in ResNet50 network, carries out feature pumping using the full articulamentum of ResNet50 network
It takes, obtains dense feature.
8. the system of similar building picture searching as claimed in claim 6, which is characterized in that the dimensionality reduction module is also used to；
Original sample set is constructed according to the key feature points, and covariance matrix, feature vector and characteristic value is calculated；Root
Dimensionality reduction is carried out to the key feature points according to characteristic value, retains K Important Characteristic Points.
9. the system of similar building picture searching as claimed in claim 8, which is characterized in that the dimensionality reduction module includes PCA
Unit；
The PCA unit, for constructing original sample set according to the key feature points, to original sample matrix normalization,
And new sample matrix is obtained with the average value that dimension initial value subtracts the dimension, then multiplied by the transposition of sample matrix, then divided by
(N-1) covariance matrix is obtained；And covariance matrix is carried out corresponding feature vector and characteristic value is calculated.
10. the system of similar building picture searching as claimed in claim 6, which is characterized in that pre- in the building picture library
All kinds of building pictures are first stored with, the characteristic strong point of the building picture is a kind of data knot divided in K dimensional feature space
Structure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910175878.1A CN110069643A (en) | 2019-03-08 | 2019-03-08 | The method and system of similar building picture searching |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910175878.1A CN110069643A (en) | 2019-03-08 | 2019-03-08 | The method and system of similar building picture searching |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110069643A true CN110069643A (en) | 2019-07-30 |
Family
ID=67366107
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910175878.1A Pending CN110069643A (en) | 2019-03-08 | 2019-03-08 | The method and system of similar building picture searching |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110069643A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112862762A (en) * | 2021-01-21 | 2021-05-28 | 博云视觉科技(青岛)有限公司 | Deep learning-based food material feature extraction and compression method |
| CN116704403A (en) * | 2023-05-11 | 2023-09-05 | 杭州晶彩数字科技有限公司 | Building image vision identification method and device, electronic equipment and medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108021693A (en) * | 2017-12-18 | 2018-05-11 | 北京奇艺世纪科技有限公司 | A kind of image search method and device |
| CN108614884A (en) * | 2018-05-03 | 2018-10-02 | 桂林电子科技大学 | A kind of image of clothing search method based on convolutional neural networks |
-
2019
- 2019-03-08 CN CN201910175878.1A patent/CN110069643A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108021693A (en) * | 2017-12-18 | 2018-05-11 | 北京奇艺世纪科技有限公司 | A kind of image search method and device |
| CN108614884A (en) * | 2018-05-03 | 2018-10-02 | 桂林电子科技大学 | A kind of image of clothing search method based on convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| HYEONWOO NOH ET AL.: "Large-Scale Image Retrieval with Attentive Deep Local Features", 《2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV)》 * |
| SHAWN0901: "大规模图像检索深度特征：Large-Scale Image Retrieval with Attentive Deep Local Features", 《HTTPS://BLOG.CSDN.NET/WANGXINSHENG0901/ARTICLE/DETAILS/81906696》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112862762A (en) * | 2021-01-21 | 2021-05-28 | 博云视觉科技(青岛)有限公司 | Deep learning-based food material feature extraction and compression method |
| CN116704403A (en) * | 2023-05-11 | 2023-09-05 | 杭州晶彩数字科技有限公司 | Building image vision identification method and device, electronic equipment and medium |
| CN116704403B (en) * | 2023-05-11 | 2024-05-24 | 杭州晶彩数字科技有限公司 | Building image vision identification method and device, electronic equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8232996B2 (en) | Image learning, automatic annotation, retrieval method, and device | |
| CN110188225B (en) | Image retrieval method based on sequencing learning and multivariate loss | |
| CN110472652B (en) | Small sample classification method based on semantic guidance | |
| CN109918537A (en) | A kind of method for quickly retrieving of the ship monitor video content based on HBase | |
| CN102324047A (en) | High spectrum image atural object recognition methods based on sparse nuclear coding SKR | |
| CN109871454B (en) | Robust discrete supervision cross-media hash retrieval method | |
| CN110458175B (en) | Unmanned aerial vehicle image matching pair selection method and system based on vocabulary tree retrieval | |
| CN105224961B (en) | A kind of infrared spectrum feature extracting and matching method of high resolution | |
| CN109241813A (en) | The sparse holding embedding grammar of differentiation for unconstrained recognition of face | |
| Wang et al. | Image retrieval based on exponent moments descriptor and localized angular phase histogram | |
| CN113889228A (en) | Semantic enhanced Hash medical image retrieval method based on mixed attention | |
| CN102663447A (en) | Cross-media searching method based on discrimination correlation analysis | |
| Gonzalez-Diaz et al. | Neighborhood matching for image retrieval | |
| CN109710804A (en) | A kind of instructional video image knowledge point Dimension Reduction Analysis method | |
| CN110069643A (en) | The method and system of similar building picture searching | |
| CN112036511B (en) | Image retrieval method based on attention mechanism graph convolution neural network | |
| Sundara Vadivel et al. | An efficient CBIR system based on color histogram, edge, and texture features | |
| Ma et al. | Spatial-content image search in complex scenes | |
| CN113536986B (en) | Dense target detection method in remote sensing image based on representative features | |
| CN115221281A (en) | Intellectual property retrieval system and retrieval method thereof | |
| CN109582743A (en) | A kind of data digging method for the attack of terrorism | |
| CN109409407A (en) | A kind of industry monitoring data clustering method based on LE algorithm | |
| CN111027609B (en) | Image data weighted classification method and system | |
| Huang et al. | A method for object-based color image retrieval | |
| CN112084353A (en) | Bag-of-words model method for rapid landmark-convolution feature matching |
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: 20190730 |
|
| RJ01 | Rejection of invention patent application after publication |