CN113362096A - Frame advertisement image matching method based on deep learning - Google Patents
Frame advertisement image matching method based on deep learning Download PDFInfo
- Publication number
- CN113362096A CN113362096A CN202010149359.0A CN202010149359A CN113362096A CN 113362096 A CN113362096 A CN 113362096A CN 202010149359 A CN202010149359 A CN 202010149359A CN 113362096 A CN113362096 A CN 113362096A
- Authority
- CN
- China
- Prior art keywords
- advertisement
- deep
- calculating
- image matching
- method based
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 24
- 238000013135 deep learning Methods 0.000 title claims abstract description 13
- 239000013598 vector Substances 0.000 claims abstract description 27
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 4
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 238000012549 training Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 5
- 238000000926 separation method Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 239000000654 additive Substances 0.000 description 2
- 230000000996 additive effect Effects 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000012937 correction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Development Economics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Accounting & Taxation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Entrepreneurship & Innovation (AREA)
- Evolutionary Biology (AREA)
- Game Theory and Decision Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a frame advertisement image matching method based on deep learning, which comprises the following steps: s1, establishing a picture data set; s2, calculating an embedded vector of a low-dimensional space for each picture in the picture data set by using a deep convolutional neural network as a deep advertisement characteristic, wherein the deep advertisement characteristic comprises distinguishing information of advertisement content; s3, randomly selecting two advertisement pictures and calculating the depth advertisement characteristics, calculating the cosine similarity of the two depth advertisement characteristics, and judging whether the two advertisements are matched according to a predefined threshold; and S4, replacing two new advertisement pictures, and repeating the step S3. The technical scheme of the invention has the beneficial effects that: the accuracy of image matching identification is improved: from 98.2% to 99%; and (3) reducing the amount of manually recognized photos: the number of the sheets per year is reduced from 198w to 110w per year.
Description
Technical Field
The invention belongs to the field of application of an image recognition and matching technology to information propagation, and particularly relates to a frame advertisement image matching method based on deep learning.
Background
Due to the development of advertisement media services, a large number of advertisement schemes are put in the elevator of the cell every week, the putting period is shortened continuously, and the advertisement scheme for correctly identifying and classifying the slots faces a huge challenge. In the traditional method, the advertisement scheme identification is operated manually, and a large amount of manpower is consumed. Moreover, because the picture replacing period is short, a large number of picture recognition and classification tasks need to be completed in a short time, and the accuracy is often not guaranteed. On the basis, if a system capable of accurately and automatically identifying and matching photos taken by workers can be developed, the system is very meaningful and can greatly improve the working efficiency.
In recent years, computer vision technology has matured day by day, and great transformation is brought to the traditional identification method. Advanced computer vision technology can not only liberate manpower from fussy manual identification matching, but also greatly improve accuracy, and an advertisement identification system is developed under the background. The advertisement recognition system automatically performs image matching based on a deep learning method, so that not only are errors caused by manpower reduced, but also part of manpower can be liberated, and the value of the advertisement recognition system is more greatly exerted. The traditional automatic matching system is based on feature point (SIFT) matching, uses SIFT to extract features, matches according to the feature points, and if the number of matched feature point pairs meets a specific threshold value, the matching is considered to be successful; if not, the match is considered to fail (see FIG. 1). For similar reference, CN 106066887B discloses a method for fast searching and analyzing advertisement sequence images, which comprises: dividing an advertisement image database into LOGO and scene images by using an image complexity characteristic and decision classification tree method; secondly, extracting and storing HOG characteristics and SIFT characteristics for images in an advertisement database; extracting HOG features and SIFT features of an image to be matched, calculating Euclidean distances between the image to be matched and HOG feature vectors of all images of the same type in an advertisement database, sequencing the Euclidean distances from small to large, and screening first S candidate images; calculating the logarithm of SIFT feature matching points of the image to be matched and the image with the most logarithm as an image corresponding to the image to be matched; and aiming at the images to be matched of the advertisement sequence, calculating the time length of each advertisement to obtain advertisement playing information. The above method mainly has two disadvantages: 1. the accuracy of matching is low when the content of different advertising schemes is mostly the same. For example, in the advertisement of fig. 2, "i love my home", different schemes only have difference in the information of the salesman, and the number of matched feature points is higher than the threshold value in the conventional method, so that the case of successful matching is erroneously determined (actually, different schemes); 2. the inverted advertisement content cannot be accurately detected. In the case of upside down magazine, because it is the same graph, the conventional method can still match a large number of pairs of feature points, resulting in the number of matched pairs of feature points exceeding the threshold value and returning a false matching result (see fig. 3).
Disclosure of Invention
In view of the above, an object of the present invention is to provide a frame advertisement image matching method based on deep learning, which matches a shot picture with a corresponding advertisement scheme, so as to ensure that an advertisement is accurately delivered to a specified location, thereby solving the deficiencies in the prior art.
In order to achieve the purpose, the invention is realized by the following technical scheme:
the method for matching the frame advertisement image based on deep learning is provided, and comprises the following steps:
s1, establishing a picture data set;
s2, calculating an embedded vector of a low-dimensional space for each picture in the picture data set by using a deep convolutional neural network as a deep advertisement characteristic, wherein the deep advertisement characteristic comprises distinguishing information of advertisement content;
s3, randomly selecting two advertisement pictures and calculating the depth advertisement characteristics, calculating the cosine similarity of the two depth advertisement characteristics, and judging whether the two advertisements are matched according to a predefined threshold;
and S4, replacing two new advertisement pictures, and repeating the step S3.
According to the frame advertisement image matching method based on deep learning, Softmax is adopted as a Loss function to train a deep convolutional neural network to obtain the feature vector of the image, and meanwhile the Large-Margin Softmax Loss method is adopted to reduce the included angle between the weight vector and the feature vector.
The frame advertisement image matching method based on deep learning includes normalizing the feature vector x on the basis of w x of Softmax, and multiplying the normalized feature vector x by a scale factor to amplify:
wherein w is a weight vector and α is a scale factor.
z＝α·y
The technical scheme of the invention has the beneficial effects that:
-improving the accuracy of image matching recognition: the accuracy is improved by 0.8%, and the error rate is reduced by 44.4%;
-reducing the amount of manually recognized photos: the number of the sheets per year is reduced from 198w to 110w per year.
Drawings
FIG. 1 is a diagram illustrating a conventional feature point matching technique;
FIG. 2 is a schematic diagram of an advertisement to be identified with only local subtle changes;
FIG. 3 is a schematic diagram of feature point matching for an inverted picture;
FIG. 4 is a schematic flow chart of a matching method according to the present invention;
FIG. 5 is a graph of the angle distribution of matched and unmatched pairs of images according to the present invention;
fig. 6 is a schematic diagram of feature differentiation of 10 digital pictures on the MNIST;
FIG. 7 is a schematic diagram of the Large-Margin Softmax Loss algorithm flow.
Detailed Description
The invention is further described with reference to the following drawings and specific examples, which are not intended to be limiting.
Referring to fig. 4, the method for matching a frame advertisement image based on deep learning of the present invention includes the following steps:
and S1, establishing a picture data set. The small frame pictures and the periodical pictures of the past year can be collected, and the total number of the pictures is 9935, and 373471 photos. And randomly extracting 5000 groups of matching photos and 5000 groups of non-matching photos, taking the photos as a verification set after manual error correction, and taking the other photos as a training set.
S2, calculating an embedded vector (Embedding) of a low-dimensional space (512 dimensions) for each picture in the picture data set by using a deep Convolutional Neural Network (CNN) as a deep advertisement characteristic, wherein the deep advertisement characteristic comprises distinguishing information of advertisement contents.
S3, randomly selecting two advertisement pictures and calculating the depth advertisement characteristics, calculating the cosine similarity of the two depth advertisement characteristics, and judging whether the two advertisements are matched according to a predefined threshold value. And randomly selecting equivalent advertisement picture pairs which are matched with (1) and not matched with (0) from the advertisement picture verification set, calculating the characteristics of each pair of pictures, then calculating the vector inner product, and obtaining the included angle theta by using an inverse cosine function. The matched and unmatched picture diagonal angle distribution graph shown in fig. 5 can be obtained, a fitted normal distribution curve is presented, and the mean value mu and the standard deviation sigma corresponding to the algorithm are displayed at the same time. Specific values are as follows in table 1:
TABLE 1
| Threshold value (degree) | mu_0 | sigma_0 | mu_1 | sigma_1 |
| 23.0420 | 42.4932 | 6.7359 | 9.6546 | 3.5097 |
The core of influencing the matching readiness lies in whether the deep convolutional neural network can obtain the advertisement characteristics with high discrimination. The Convolutional Neural Network (CNN) is trained with a relatively difficult task of 'super-classification', and the network is forced to form a relatively compact and strong-discriminability deep advertisement feature at the first full connection layer (FC layer) for later advertisement identification.
And training the CNN by using Softmax as a loss function to obtain the feature vector of the picture. The 2-dimensional feature mapping visualization of the 10 classes on the MNIST as shown in FIG. 6, the different classes are obviously separated, but the condition does not meet the requirement of feature vector comparison in advertisement identification. Cosine distance (cosine distance) is commonly used for calculating the similarity of the feature vectors, and the larger the cosine is, the smaller the cosine distance (included angle) is, and the higher the similarity of the vectors is. For feature mapping (feature embedding) for advertisement identification, Softmax encourages separation of features of different classes, but does not encourage much separation of features.
The depth features of Softmax training will partition the entire hyperspace or hypersphere according to the number of classes, ensuring that the classes are separable, which is very suitable for multi-classification tasks such as MNIST and ImageNet, since the test classes must be in the training classes. However, Softmax does not require intra-class compactness and inter-class separation, which is not very suitable for the task of advertisement recognition because the number of 1W advertisements in the training set is very small compared with the number of advertisements in the test set, and it is impossible to obtain training samples of all advertisements. In particular, it is also generally required that the training set and the test set do not overlap. Therefore, Softmax needs to be modified, separability is guaranteed, and the feature vector classes are as compact as possible and are separated as far as possible.
Therefore, the method of Large Margin (Large Margin) is adopted, so that the included angle between the weight vector W and the feature vector f is smaller. Specifically, the following method is used:
on the basis of w x of Softmax, normalizing the feature vector x, and multiplying by a scale factor to amplify:
z＝α·y
the scale factor alpha uses a fixed value of 64. The weight and the characteristic normalization enable the CNN to be more concentrated on the optimized included angle, and the obtained deep advertisement characteristics are more separated.
After feature normalization, the feature vectors are fixedly mapped to the hypersphere with the radius of 1, so that understanding and optimization are facilitated. But this also compresses the space of feature expression; multiplying by the scale factor α is equivalent to enlarging the radius of the hypersphere to α, the hypersphere becomes larger and the space for feature expression is larger (the larger the radius the larger the surface area of the sphere). In addition, after the characteristics are normalized, the similarity of characteristic vectors is calculated by advertisement identification, the L2 distance and the cosine distance are equivalent in meaning, the calculated amount is the same, and the convenience is improved.
On this basis, an Additive Margin (Additive Margin) is used. For example, the included angle between the advertisement feature xi and the corresponding weight W is θ, and a fixed value m (0.5) is added on the basis of the included angle. The result of this is to "push" the features away from the weights, as shown in FIG. 7, thereby increasing the difficulty of the classification task and making the resulting features more cohesive.
And S4, replacing two new advertisement pictures, and repeating the step S3. Based on the feature angle θ and the above parameters, the matching probability of the pair of pictures can be calculated, and the specific code refers to the following:
def get_prob(theta):
prob_0＝norm.pdf(theta,mu_0,sigma_0)
prob_1＝norm.pdf(theta,mu_1,sigma_1)
total＝prob_0+prob_1
return prob_1/total
the technical scheme of the invention improves the accuracy of image matching identification: from 98.2% to 99% (accuracy improved by 0.8%, error rate reduced by 44.4%), while reducing the amount of manually recognized photos: the number of the sheets per year is reduced from 198w to 110w per year.
While the invention has been described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the invention.
Claims (3)
1. A frame advertisement image matching method based on deep learning is characterized by comprising the following steps:
s1, establishing a picture data set;
s2, calculating an embedded vector of a low-dimensional space for each picture in the picture data set by using a deep convolutional neural network as a deep advertisement characteristic, wherein the deep advertisement characteristic comprises distinguishing information of advertisement content;
s3, randomly selecting two advertisement pictures and calculating the depth advertisement characteristics, calculating the cosine similarity of the two depth advertisement characteristics, and judging whether the two advertisements are matched according to a predefined threshold;
and S4, replacing two new advertisement pictures, and repeating the step S3.
2. The frame advertisement image matching method based on deep learning as claimed in claim 1, wherein Softmax is used as a Loss function to train a deep convolutional neural network to obtain the feature vector of the image, and meanwhile, a Large-Margin Softmax Loss method is used to reduce the included angle between the weight vector and the feature vector.
3. The frame advertisement image matching method based on deep learning of claim 2, wherein the feature vector x is normalized on the basis of w x of Softmax and multiplied by a scale factor to be amplified:
z＝α·y
wherein w is a weight vector and α is a scale factor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010149359.0A CN113362096A (en) | 2020-03-04 | 2020-03-04 | Frame advertisement image matching method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010149359.0A CN113362096A (en) | 2020-03-04 | 2020-03-04 | Frame advertisement image matching method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113362096A true CN113362096A (en) | 2021-09-07 |
Family
ID=77523871
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010149359.0A Pending CN113362096A (en) | 2020-03-04 | 2020-03-04 | Frame advertisement image matching method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113362096A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114529713A (en) * | 2022-01-14 | 2022-05-24 | 电子科技大学 | Underwater image enhancement method based on deep learning |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529996A (en) * | 2016-10-24 | 2017-03-22 | 北京百度网讯科技有限公司 | Deep learning-based advertisement display method and device |
| CN108062575A (en) * | 2018-01-03 | 2018-05-22 | 广东电子工业研究院有限公司 | A kind of high similarity graph picture identification and sorting technique |
| CN108549883A (en) * | 2018-08-06 | 2018-09-18 | 国网浙江省电力有限公司 | A kind of face recognition methods again |
| CN108846694A (en) * | 2018-06-06 | 2018-11-20 | 厦门集微科技有限公司 | A kind of elevator card put-on method and device, computer readable storage medium |
| CN108960258A (en) * | 2018-07-06 | 2018-12-07 | 江苏迪伦智能科技有限公司 | A kind of template matching method based on self study depth characteristic |
| CN109800306A (en) * | 2019-01-10 | 2019-05-24 | 深圳Tcl新技术有限公司 | It is intended to analysis method, device, display terminal and computer readable storage medium |
| WO2019128367A1 (en) * | 2017-12-26 | 2019-07-04 | 广州广电运通金融电子股份有限公司 | Face verification method and apparatus based on triplet loss, and computer device and storage medium |
| WO2020015075A1 (en) * | 2018-07-18 | 2020-01-23 | 平安科技（深圳）有限公司 | Facial image comparison method and apparatus, computer device, and storage medium |
| CN110781917A (en) * | 2019-09-18 | 2020-02-11 | 北京三快在线科技有限公司 | Method and device for detecting repeated image, electronic equipment and readable storage medium |
-
2020
- 2020-03-04 CN CN202010149359.0A patent/CN113362096A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529996A (en) * | 2016-10-24 | 2017-03-22 | 北京百度网讯科技有限公司 | Deep learning-based advertisement display method and device |
| WO2019128367A1 (en) * | 2017-12-26 | 2019-07-04 | 广州广电运通金融电子股份有限公司 | Face verification method and apparatus based on triplet loss, and computer device and storage medium |
| CN108062575A (en) * | 2018-01-03 | 2018-05-22 | 广东电子工业研究院有限公司 | A kind of high similarity graph picture identification and sorting technique |
| CN108846694A (en) * | 2018-06-06 | 2018-11-20 | 厦门集微科技有限公司 | A kind of elevator card put-on method and device, computer readable storage medium |
| CN108960258A (en) * | 2018-07-06 | 2018-12-07 | 江苏迪伦智能科技有限公司 | A kind of template matching method based on self study depth characteristic |
| WO2020015075A1 (en) * | 2018-07-18 | 2020-01-23 | 平安科技（深圳）有限公司 | Facial image comparison method and apparatus, computer device, and storage medium |
| CN108549883A (en) * | 2018-08-06 | 2018-09-18 | 国网浙江省电力有限公司 | A kind of face recognition methods again |
| CN109800306A (en) * | 2019-01-10 | 2019-05-24 | 深圳Tcl新技术有限公司 | It is intended to analysis method, device, display terminal and computer readable storage medium |
| CN110781917A (en) * | 2019-09-18 | 2020-02-11 | 北京三快在线科技有限公司 | Method and device for detecting repeated image, electronic equipment and readable storage medium |
Non-Patent Citations (13)
| Title |
|---|
| "人脸识别的LOSS（下）", pages 2, Retrieved from the Internet <URL:https://zhuanlan.zhihu.com/p/34436551> * |
| PEANUT_范: "AM-Softmax Loss", pages 1, Retrieved from the Internet <URL:https://blog.csdn.net/u013841196/article/details/89920902> * |
| 刘施乐;: "基于深度学习的人脸识别技术研究", 电子制作, no. 24 * |
| 孙洁;丁笑君;杜磊;李秦曼;邹奉元;: "基于卷积神经网络的织物图像特征提取与检索研究进展", 纺织学报, no. 12 * |
| 张惠凡等: "基于卷积神经网络的鸟类视频图像检索研究", 科研信息化技术与应用, vol. 8, no. 5, pages 50 * |
| 彭晏飞;高艺;杜婷婷;桑雨;訾玲玲;: "生成对抗网络的单图像超分辨率重建方法", 计算机科学与探索, no. 09 * |
| 彭骋;: "基于深度学习的图像检索系统", 通讯世界, no. 06 * |
| 有三AI: "一文道尽softmax loss及其变种", pages 3, Retrieved from the Internet <URL:https://blog.csdn.net/u013841196/article/details/89920902> * |
| 李振东;钟勇;曹冬平;: "深度卷积特征向量用于快速人脸图像检索", 计算机辅助设计与图形学学报, no. 12 * |
| 李振东;钟勇;陈蔓;曹冬平;: "基于深度特征的快速人脸图像检索方法", 光学学报, no. 10 * |
| 邓建国;张素兰;张继福;荀亚玲;刘爱琴;: "监督学习中的损失函数及应用研究", 大数据, no. 01 * |
| 黄旭;凌志刚;李绣心;: "融合判别式深度特征学习的图像识别算法", 中国图象图形学报, no. 04 * |
| 龚锐;丁胜;章超华;苏浩;: "基于深度学习的轻量级和多姿态人脸识别方法", 计算机应用, no. 03 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114529713A (en) * | 2022-01-14 | 2022-05-24 | 电子科技大学 | Underwater image enhancement method based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102414680B (en) | Utilize the semantic event detection of cross-domain knowledge | |
| US8358856B2 (en) | Semantic event detection for digital content records | |
| CN102542058B (en) | Hierarchical landmark identification method integrating global visual characteristics and local visual characteristics | |
| CN108733778B (en) | Industry type identification method and device of object | |
| CN103699523B (en) | Product classification method and apparatus | |
| Tarawneh et al. | Invoice classification using deep features and machine learning techniques | |
| EP2808827A1 (en) | System and method for OCR output verification | |
| WO2017016240A1 (en) | Banknote serial number identification method | |
| CN104732413A (en) | Intelligent individuation video advertisement pushing method and system | |
| CN109376796A (en) | Image classification method based on active semi-supervised learning | |
| CN105389593A (en) | Image object recognition method based on SURF | |
| CN106845358A (en) | A kind of method and system of handwritten character characteristics of image identification | |
| CN111460961A (en) | CDVS-based similarity graph clustering static video summarization method | |
| Gordo et al. | Document classification and page stream segmentation for digital mailroom applications | |
| CN111340032A (en) | Character recognition method based on application scene in financial field | |
| CN110442736B (en) | Semantic enhancer spatial cross-media retrieval method based on secondary discriminant analysis | |
| CN115062186A (en) | Video content retrieval method, device, equipment and storage medium | |
| CN114357307A (en) | News recommendation method based on multi-dimensional features | |
| CN114495139A (en) | Operation duplicate checking system and method based on image | |
| CN110020638A (en) | Facial expression recognizing method, device, equipment and medium | |
| CN113362096A (en) | Frame advertisement image matching method based on deep learning | |
| CN113657377A (en) | Structured recognition method for airplane ticket printing data image | |
| CN109472307A (en) | A kind of method and apparatus of training image disaggregated model | |
| TWI761090B (en) | Dialogue data processing system and method thereof and computer readable medium | |
| CN104778478A (en) | Handwritten numeral identification method |
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: 20210907 |