CN108898165A - A kind of recognition methods of billboard style - Google Patents
A kind of recognition methods of billboard style Download PDFInfo
- Publication number
- CN108898165A CN108898165A CN201810599682.0A CN201810599682A CN108898165A CN 108898165 A CN108898165 A CN 108898165A CN 201810599682 A CN201810599682 A CN 201810599682A CN 108898165 A CN108898165 A CN 108898165A
- Authority
- CN
- China
- Prior art keywords
- billboard
- style
- model
- picture
- identification
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of recognition methods of billboard style, include the following steps：According to the characteristic of billboard, by billboard picture segmentation at main body, text and background three parts, and extract the color histogram of every part, GIST descriptor, saliency map, binary element category feature and histograms of oriented gradients, as feature, model is trained and is assessed, the billboard style identification model that can accurately identify billboard style is obtained, then identifies billboard picture style using the billboard style identification model.This method can accurately identify billboard style, meet the requirement in advertisement the Automation Design for the identification of quick and precisely style.
Description
Technical field
The present invention relates to field of image processings, and in particular to a kind of recognition methods of billboard style.
Background technique
As the continuous enhancing and the continuous of internet of multimedia technology are popularized, so that digital advertisement picture gradually starts
The characteristics of demand is very big, quality requirement is low, easy consumption is presented.
Since advertisement plays the development of very important effect, especially e-commerce in business, lead to advertisement
Demand constantly increases, and businessman may require that designer designs difference according to the difference of theme to adapt to the development of trend
The advertisement of style.Such as summer conception design teacher tends to design clean and tidy, natural stype advertisement.
Ad style is the bridge that the spirit between advertisement designer and consumer is linked up.Therefore, in order to cope with people couple
In the personalized affection need of digital advertisement picture, the demand that enterprise automates advertisement design is combined.Enterprise can
With according to consumer for ad style demand carry out mass, automatically generate specific style advertising pictures.In life
During at specific style advertisement, largely automated system can be helped preferably to set the assessment of the style of advertisement
Count out the advertisement of specific style.Therefore, quickly and effectively style identification is carried out to advertisement to have important practical significance.
Existing picture style recognizer, the style study either based on picking feature is still based on deep learning
Style calculates, and is extracted to global feature, is then trained with the good data of mark and is shown that style identifies mould
Type.These methods, the feature extracted for all pictures is all identical, however in the picture of different types, feature
The influence degree of style is different.Billboard is computer synthesising picture, he is generally by text, main body, the groups such as background
At.The position of different elements, collocation etc. can all impact style.Therefore it studies a kind of for ad style knowledge method for distinguishing
It has important practical significance.
The patent application of Publication No. CN201510922662.9 discloses a kind of picture style recognition methods and device.It is public
The patent application that the number of opening is CN201510922684.5 discloses style recognition methods and the device of a kind of commodity.The two patents
Apply for that disclosed technology contents are：It first obtains samples pictures and forms training set, then to pre-set multiple target convolutional Neural
Network carries out parameter initialization, and the picture in training set is trained to obtain picture style identification model, finally utilizes mould
Type identifies picture to be identified.These methods are all based on depth convolutional neural networks, computationally intensive, iteration time
It is long.Need the tape label data of much larger number that could train to obtain accurate style identification model, however this can spend largely
Cost.
Billboard picture is by by text, main body, background composition, different from common two-dimension picture, so above-mentioned side
Method and algorithm are not suitable for identifying billboard style, that is, identify billboard style using the above method,
Accuracy rate can be very low, is unable to satisfy the requirement in advertisement the Automation Design for the identification of quick and precisely style.
Summary of the invention
The object of the present invention is to provide a kind of recognition methods of billboard style, this method can accurately identify flat
Face ad style meets the requirement in advertisement the Automation Design for the identification of quick and precisely style.
For achieving the above object, the present invention provides following technical scheme：
A kind of recognition methods of billboard style, includes the following steps：
(1) genre labels of billboard picture and billboard picture are obtained；
(2) main body, text and the background of billboard picture are extracted using deep learning method, and are led respectively
Body, the color histogram of text and background, GIST descriptor, saliency map, binary element category feature and direction gradient are straight
Fang Tu, by the color histogram of acquisition, GIST descriptor, saliency map, binary element category feature, histograms of oriented gradients and
The genre labels of billboard picture constitute sample set as a sample, and sample set is divided into training set and test set；
(3) decision-tree model, non-linear largest interval model and linear model is respectively trained using training set；
(4) trained decision-tree model, non-linear largest interval model and linear mould are tested respectively using test set
Type, and three models are assessed according to test result, the model optimal using assessment result is identified as billboard style
Model；
(5) it for billboard to be identified, is obtained using the method for step (2) straight using the color of billboard to be identified
Fang Tu, GIST descriptor, saliency map, binary element category feature and histograms of oriented gradients, and by color histogram, GIST
Descriptor, saliency map, binary element category feature and histograms of oriented gradients are input in billboard style identification model,
It is computed the style probability for obtaining billboard picture to be identified.
Wherein, the step (1) includes：
(1-1) obtains billboard picture, and rejects to the non-ad elements in the billboard picture；
(1-2) treated that billboard picture is labeled to rejecting, and obtains the genre labels of billboard picture.
More specifically, the step (1-2) includes：
(1-2-1) building style describes word set, and style adjective concentration includes several antagonism of symbols and statement wind
The adjective of lattice；
(1-2-2) describes word set according to the style, using the method compared in pairs to rejecting treated billboard figure
Piece is labeled, and obtains the genre labels of billboard picture.
Wherein, in the step (1-2-2)：Word set is described according to the style, is put down using Bradley-Terry model
The genre labels of face advertising pictures.
In step (2)：
Using the main body of FCIS model identification billboard picture；
Using the text of CTPN model identification billboard picture；
Main body, text in removing billboard picture, remaining is the background of billboard picture.
The step (4) includes：
(4-1) is directed to trained decision-tree model, and the billboard picture in test set is input to trained determine
In plan tree-model, the identification style probability of billboard picture is obtained, according to the identification style probability and mark of billboard picture
The ROC curve of style probability building decision-tree model is signed, and calculates the area below the ROC curve for obtaining decision-tree model
AUC1；
(4-2) obtains the area below the ROC curve of non-linear largest interval model using the method in step (4-1)
AUC2；
(4-3) obtains the area AUC below the ROC curve of linear model using the method in step (4-1)3；
(4-4) chooses AUC1、AUC2And AUC3In the corresponding model of maximum AUC as billboard style identify mould
Type.
The device have the advantages that being：
The present invention is according to the characteristic of billboard, by billboard picture segmentation at main body, text and background three parts,
And extract the color histogram of every part, GIST descriptor, saliency map, binary element category feature and direction gradient histogram
Figure, as feature, is trained model and assesses, and obtains the billboard wind that can accurately identify billboard style
Lattice identification model can accurately identify billboard picture style using the billboard style identification model.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to do simply to introduce, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art, can be with root under the premise of not making the creative labor
Other accompanying drawings are obtained according to these attached drawings.
Fig. 1 is the flow chart of the recognition methods of billboard style provided by the invention；
Fig. 2 is the procedure chart provided by the invention for obtaining billboard style identification model.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention more comprehensible, with reference to the accompanying drawings and embodiments to this
Invention is described in further detail.It should be appreciated that the specific embodiments described herein are only used to explain the present invention,
And the scope of protection of the present invention is not limited.
Fig. 1 is the flow chart of the recognition methods of billboard style provided by the invention.Fig. 2 is acquisition provided by the invention
The procedure chart of billboard style identification model.As shown in Figure 1 and Figure 2, the recognition methods of billboard style provided by the invention
Including：
S101 obtains the genre labels of billboard picture and billboard picture.
Billboard picture is various aspects (such as fashionable dress, makeups, food and drink, the amusement collected from major website (such as petal net)
Deng) advertising pictures.It can include some non-ad elements on those advertising pictures, those non-ad elements can be link letter
Breath, image credit information etc..To guarantee training sample reliability, need to carry out billboard picture to arrange and clear, specifically
The non-ad elements in the billboard picture of acquisition are rejected on ground.
After the non-ad elements for rejecting billboard picture, it is also necessary to mark the genre labels of billboard picture.
Detailed process is：
Firstly, existing document is studied in research, therefrom obtains several antagonism of symbols and state the adjective of style, composition
Style describes word set, style adjective concentrate comprising adjective romantic (Romantic), sweet (Pretty), clean and tidy (Clear),
Natural (Natural), arbitrarily (Casual), graceful (Elegant), leisure (Cool Casual), fashionable (Chic), luxury
(Gorgeous), dynamic (Dynamic), classic (Classic), magnificent (Dandy), modern (Modern) etc..
Then, word set is described according to style, using the method compared in pairs to reject treated billboard picture into
Rower note, obtains the genre labels of billboard picture.
Specifically, the genre labels of billboard picture can be obtained, using Bradley-Terry model to be had
The data set of scientific Dimension style.
S102 constructs training set and test set according to billboard picture.
It is made of due to billboard picture main body, text and background three parts, wherein main body is and advertiser
Inscribe identical content, advertisement is based on then the shoes in billboard picture are, in billboard picture about shoes
Text is text, and remaining content is background.
In order to obtain better billboard style identification model, billboard figure is extracted using deep learning method
Main body, text and the background of piece, specifically process be：
(a) using the main body of FCIS model identification billboard picture；
(b) using the text of CTPN model identification billboard picture；
(c) main body in removing billboard picture, text, remaining is the background of billboard picture, specifically,
Mask is established to main body and text, is rejected main body and text by Mask, remaining is background.
FCIS (Fully Convolutional Instance-aware Semantic Segmentation) model is
The Image Segmentation Model that model parameter has determined can rapidly and accurately split the main body in billboard picture.
CTPN (connectionist text proposal network) model is the text that model parameter has determined
Parted pattern can rapidly and accurately come out the text segmentation in billboard picture.
After obtaining main body, text and the background of billboard picture, proceed as follows：
For main body, extract the color histogram of main body, GIST descriptor, saliency map, binary element category feature and
Histograms of oriented gradients feature；
For text, extract the color histogram of text, GIST descriptor, saliency map, binary element category feature and
Histograms of oriented gradients feature；
For background, extract the color histogram of background, GIST descriptor, saliency map, binary element category feature and
Histograms of oriented gradients feature.
Color histogram is constructed based on CELAB color channel：The style of many images all show to color this
The strong depend-ence of dimension.For example the image of " romance " style likes partially red form and aspect mostly, the image of " clean and tidy " style is usual
The form and aspect of whole image are single and saturation degree is relatively low, and the image of " luxury " style is then with color ratio relatively exaggeration etc..Therefore present invention choosing
It selects from the channel LAB, by building color histogram as characteristic of division.This feature is 784 vectors tieed up by a length
Composition, is the pixels statistics histogram of each color channel of the image under CELAB color mode.Its specific method is：L is led to
Road is equally divided into 4 sections, and equally, A channel and channel B are respectively divided into 14 sections, then by picture under each color channel
The statistics of element value.It thus can achieve from color dimension and distinguish the purpose of different images style.
GIST descriptor：Classical GIST descriptor has obtained widely in scene classification and image similarity searching field
It uses, the feature in terms of available image construction, this feature are the vector that a length is 960 dimensions in a way.
Saliency map：The purpose of this feature is to identify image position that people are usually paid attention to from the angle of vision attention
It sets, it is important so as to soon predict position that people may notice that and its corresponding vision from an image
Degree.This has very big help for style follow-up study which type of feature affects image.This feature is by length
What the vector of 1024 dimensions was constituted.
Binary element category feature (Meta-class binary features)：The content of image usually influences whether image
Style.Such as have the people moved in an advertisement, then the style of this advertisement may be very much " innervation " wind greatly
Lattice, and when having the contents such as forest-tree in an advertisement, style may be very much " nature " style greatly.And binary element category feature
Be one 15000 dimension binary vector, this vector be from existing public data focusing study to classifier obtained by
, that is to say, that this feature has the ability that can identify picture material.The vector be a length be 15232 dimensions to
Amount composition, wherein being integrated with the style and features of a large amount of low layer.This feature is weight coefficient maximum one in numerous style and features
A feature, it is, this feature plays conclusive effect for distinguishing the genre category of image.
Histograms of oriented gradients (Histogram of Oriented Gradient, HOG)：This feature is widely used in image
In the task of identification, especially in pedestrian detection, it by calculate and statistical picture regional area gradient orientation histogram come
Constitutive characteristic can be used for extracting the correlated characteristics such as the texture of image.
Above-mentioned color histogram, GIST descriptor, saliency map, binary element category feature, histograms of oriented gradients covering
The color dimension of billboard picture, composition dimension, content dimension and texture dimension, can comprehensively be presented billboard
The feature of picture can obtain more accurately model using those feature training patterns.
The main body of billboard picture, three parts of text and background color histogram, GIST descriptor, significance
Figure, binary element category feature, histograms of oriented gradients and genre labels constitute a sample, and a large amount of sample forms sample
Collection, and most in sample set is as training set, it is remaining to be used as test set.
Decision-tree model, non-linear largest interval model and linear model is respectively trained using training set in S103.
There are many kinds of classifiers, more preferably billboard style identification model is obtained to train, to multiple points in this step
Class device is trained.Specifically, by the sample in training set be separately input to decision-tree model, non-linear largest interval model with
And in linear model, until model convergence, trained decision-tree model, non-linear largest interval model and linear mould are obtained
Type.
S104 tests trained decision-tree model, non-linear largest interval model and linear using test set respectively
Model, and three models are assessed according to test result, the model optimal using assessment result is known as billboard style
Other model.
For trained decision-tree model, non-linear largest interval model and linear model, assessed with
Select optimal model as billboard style identification model, specially：
Firstly, carrying out feature extraction to the test sample in test set using S102, the color histogram of test sample is obtained
Figure, GIST descriptor, saliency map, binary element category feature, histograms of oriented gradients.
Then, for trained decision-tree model, the billboard picture in test set is input to trained determine
In plan tree-model, the identification style probability of billboard picture is obtained, according to the identification style probability and mark of billboard picture
The ROC curve of style probability building decision-tree model is signed, and calculates the area below the ROC curve for obtaining decision-tree model
AUC1；
For non-linear largest interval model, the billboard picture in test set is input to training after feature extraction
In good decision-tree model, the identification style probability of billboard picture is obtained, it is general according to the identification style of billboard picture
Rate and label style probability construct the ROC curve of non-linear largest interval model, and calculate and obtain non-linear largest interval model
ROC curve below area AUC2；
Billboard picture in test set is input in trained decision-tree model by linear model, is obtained
The identification style probability of billboard picture constructs line according to the identification style probability of billboard picture and label style probability
Property model ROC curve, and calculate obtain linear model ROC curve below area AUC3。
Finally, choosing AUC1、AUC2And AUC3In the corresponding model of maximum AUC as billboard style identify mould
Type.
S105 identifies billboard picture to be identified using billboard style identification model, obtains to be identified
The identification style probability of billboard picture.
Specifically, it obtained using the method for S102 using the color histogram of billboard to be identified, GIST descriptor, shown
Work degree figure, binary element category feature and histograms of oriented gradients, and by color histogram, GIST descriptor, saliency map, two
System member category feature and histograms of oriented gradients are input in billboard style identification model, are computed and are obtained to be identified put down
The identification style probability of face advertising pictures.
Billboard picture style can be accurately identified using the above method, met in advertisement the Automation Design for fast
The requirement of fast accurate style identification.
Technical solution of the present invention and beneficial effect is described in detail in above-described specific embodiment, Ying Li
Solution is not intended to restrict the invention the foregoing is merely presently most preferred embodiment of the invention, all in principle model of the invention
Interior done any modification, supplementary, and equivalent replacement etc. are enclosed, should all be included in the protection scope of the present invention.
Claims (6)
1. a kind of recognition methods of billboard style, includes the following steps：
(1) genre labels of billboard picture and billboard picture are obtained；
(2) main body, text and the background of billboard picture are extracted using deep learning method, and obtain main body, text respectively
Color histogram, GIST descriptor, saliency map, binary element category feature and the histograms of oriented gradients of this and background,
By the color histogram of acquisition, GIST descriptor, saliency map, binary element category feature, histograms of oriented gradients and plane
The genre labels of advertising pictures constitute sample set as a sample, and sample set is divided into training set and test set；
(3) decision-tree model, non-linear largest interval model and linear model is respectively trained using training set；
(4) trained decision-tree model, non-linear largest interval model and linear model are tested respectively using test set, and
Three models are assessed according to test result, the model optimal using assessment result is as billboard style identification model；
(5) for billboard to be identified, the color histogram using billboard to be identified is obtained using the method for step (2)
Figure, GIST descriptor, saliency map, binary element category feature and histograms of oriented gradients, and color histogram, GIST are retouched
It states symbol, saliency map, binary element category feature and histograms of oriented gradients to be input in billboard style identification model, pass through
Calculate the style probability for obtaining billboard picture to be identified.
2. the recognition methods of billboard style as described in claim 1, which is characterized in that the step (1) includes：
(1-1) obtains billboard picture, and rejects to the non-ad elements in the billboard picture；
(1-2) treated that billboard picture is labeled to rejecting, and obtains the genre labels of billboard picture.
3. the recognition methods of billboard style as claimed in claim 2, which is characterized in that the step (1-2) includes：
(1-2-1) building style describes word set, and style adjective concentration includes several antagonism of symbols and statement style
Adjective；
(1-2-2) describes word set according to the style, using the method compared in pairs to reject treated billboard picture into
Rower note, obtains the genre labels of billboard picture.
4. the recognition methods of billboard style as claimed in claim 3, which is characterized in that in the step (1-2-2)：
Word set is described according to the style, and the genre labels of billboard picture are obtained using Bradley-Terry model.
5. the recognition methods of billboard style as described in claim 1, which is characterized in that in step (2)：
Using the main body of FCIS model identification billboard picture；
Using the text of CTPN model identification billboard picture；
Main body, text in removing billboard picture, remaining is the background of billboard picture.
6. the recognition methods of billboard style as described in claim 1, which is characterized in that the step (4) includes：
(4-1) is directed to trained decision-tree model, and the billboard picture in test set is input to trained decision tree
In model, the identification style probability of billboard picture is obtained, according to the identification style probability and label wind of billboard picture
Lattice probability constructs the ROC curve of decision-tree model, and calculates the area AUC below the ROC curve for obtaining decision-tree model1；
(4-2) obtains the area AUC below the ROC curve of non-linear largest interval model using the method in step (4-1)2；
(4-3) obtains the area AUC below the ROC curve of linear model using the method in step (4-1)3；
(4-4) chooses AUC1、AUC2And AUC3In the corresponding model of maximum AUC as billboard style identification model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810599682.0A CN108898165B (en) | 2018-06-12 | 2018-06-12 | Method for identifying plane advertisement style |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810599682.0A CN108898165B (en) | 2018-06-12 | 2018-06-12 | Method for identifying plane advertisement style |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108898165A true CN108898165A (en) | 2018-11-27 |
| CN108898165B CN108898165B (en) | 2021-11-30 |
Family
ID=64344592
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810599682.0A Active CN108898165B (en) | 2018-06-12 | 2018-06-12 | Method for identifying plane advertisement style |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108898165B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110458236A (en) * | 2019-08-14 | 2019-11-15 | 有米科技股份有限公司 | A kind of Advertising Copy style recognition methods and system |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102207966A (en) * | 2011-06-01 | 2011-10-05 | 华南理工大学 | Video content quick retrieving method based on object tag |
| CN102842046A (en) * | 2012-08-07 | 2012-12-26 | 天津大学 | Calligraphic style identification method based on overall feature extracting and training |
| CN103503463A (en) * | 2011-11-23 | 2014-01-08 | 华为技术有限公司 | Video advertisement broadcasting method, device and system |
| US8842965B1 (en) * | 2011-11-02 | 2014-09-23 | Google Inc. | Large scale video event classification |
| CN104657468A (en) * | 2015-02-12 | 2015-05-27 | 中国科学院自动化研究所 | Fast video classification method based on images and texts |
| CN105260398A (en) * | 2015-09-17 | 2016-01-20 | 中国科学院自动化研究所 | Quick sorting method for movie types based on poster and plot summary |
| CN105912935A (en) * | 2016-05-03 | 2016-08-31 | 腾讯科技（深圳）有限公司 | Advertisement detection method and advertisement detection apparatus |
| CN107590491A (en) * | 2016-07-07 | 2018-01-16 | 阿里巴巴集团控股有限公司 | A kind of image processing method and device |
-
2018
- 2018-06-12 CN CN201810599682.0A patent/CN108898165B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102207966A (en) * | 2011-06-01 | 2011-10-05 | 华南理工大学 | Video content quick retrieving method based on object tag |
| US8842965B1 (en) * | 2011-11-02 | 2014-09-23 | Google Inc. | Large scale video event classification |
| CN103503463A (en) * | 2011-11-23 | 2014-01-08 | 华为技术有限公司 | Video advertisement broadcasting method, device and system |
| CN102842046A (en) * | 2012-08-07 | 2012-12-26 | 天津大学 | Calligraphic style identification method based on overall feature extracting and training |
| CN104657468A (en) * | 2015-02-12 | 2015-05-27 | 中国科学院自动化研究所 | Fast video classification method based on images and texts |
| CN105260398A (en) * | 2015-09-17 | 2016-01-20 | 中国科学院自动化研究所 | Quick sorting method for movie types based on poster and plot summary |
| CN105912935A (en) * | 2016-05-03 | 2016-08-31 | 腾讯科技（深圳）有限公司 | Advertisement detection method and advertisement detection apparatus |
| CN107590491A (en) * | 2016-07-07 | 2018-01-16 | 阿里巴巴集团控股有限公司 | A kind of image processing method and device |
Non-Patent Citations (4)
| Title |
|---|
| JAN C.VAN GEMERT: "Exploiting Photographic Style for Category-Level Image Classification by Generalizing the Spatial Pyramid", 《ICMR"11》 * |
| JIANBIN JIAO 等: "A configurable method for multi-style license plate Recognition", 《PATTERN RECOGNITION》 * |
| 冀中: "基于多模态信息的新闻视频内容分析技术研究", 《万方数据知识服务平台》 * |
| 袁军辉: "基于多特征信息融合的WEB广告分类方法研究", 《中国优秀博硕士学位论文全文数据库（硕士）信息科技辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110458236A (en) * | 2019-08-14 | 2019-11-15 | 有米科技股份有限公司 | A kind of Advertising Copy style recognition methods and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108898165B (en) | 2021-11-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107346420B (en) | Character detection and positioning method in natural scene based on deep learning | |
| CN107391703B (en) | The method for building up and system of image library, image library and image classification method | |
| Lee et al. | Adaboost for text detection in natural scene | |
| CN106815604B (en) | Method for viewing points detecting based on fusion of multi-layer information | |
| Xie et al. | DeRPN: Taking a further step toward more general object detection | |
| CN102012939B (en) | Method for automatically tagging animation scenes for matching through comprehensively utilizing overall color feature and local invariant features | |
| CN105069481B (en) | Natural scene multiple labeling sorting technique based on spatial pyramid sparse coding | |
| Greenhalgh et al. | Traffic sign recognition using MSER and random forests | |
| CN107545271B (en) | Image recognition method, device and system | |
| CN103810274B (en) | Multi-characteristic image tag sorting method based on WordNet semantic similarities | |
| CN106126581A (en) | Cartographical sketching image search method based on degree of depth study | |
| CN109740686A (en) | A kind of deep learning image multiple labeling classification method based on pool area and Fusion Features | |
| CN102385592B (en) | Image concept detection method and device | |
| CN108109055B (en) | Cross-scene clothing retrieval method based on image rendering | |
| CN106897681A (en) | A kind of remote sensing images comparative analysis method and system | |
| CN108052952A (en) | A kind of the clothes similarity determination method and its system of feature based extraction | |
| CN110377727A (en) | A kind of multi-tag file classification method and device based on multi-task learning | |
| CN110119688A (en) | A kind of Image emotional semantic classification method using visual attention contract network | |
| Chen et al. | Recognizing the style of visual arts via adaptive cross-layer correlation | |
| CN107305691A (en) | Foreground segmentation method and device based on images match | |
| CN104268552B (en) | One kind is based on the polygonal fine classification sorting technique of part | |
| CN109993187A (en) | A kind of modeling method, robot and the storage device of object category for identification | |
| CN103839078A (en) | Hyperspectral image classifying method based on active learning | |
| CN113221770B (en) | Cross-domain pedestrian re-recognition method and system based on multi-feature hybrid learning | |
| CN106326451B (en) | A kind of webpage heat transfer agent block decision method of view-based access control model feature extraction |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |