CN110135248A - A kind of natural scene Method for text detection based on deep learning - Google Patents
A kind of natural scene Method for text detection based on deep learning Download PDFInfo
- Publication number
- CN110135248A CN110135248A CN201910270269.4A CN201910270269A CN110135248A CN 110135248 A CN110135248 A CN 110135248A CN 201910270269 A CN201910270269 A CN 201910270269A CN 110135248 A CN110135248 A CN 110135248A
- Authority
- CN
- China
- Prior art keywords
- text
- natural scene
- text detection
- detection
- deep learning
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 86
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000013135 deep learning Methods 0.000 title claims abstract description 19
- 230000001629 suppression Effects 0.000 claims abstract description 6
- 238000013528 artificial neural network Methods 0.000 claims description 17
- 238000012549 training Methods 0.000 claims description 13
- 230000000306 recurrent effect Effects 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 5
- 238000013100 final test Methods 0.000 claims description 4
- 210000000988 bone and bone Anatomy 0.000 claims description 3
- 230000006403 short-term memory Effects 0.000 claims description 3
- 238000007689 inspection Methods 0.000 claims 1
- 210000005036 nerve Anatomy 0.000 claims 1
- 239000000284 extract Substances 0.000 abstract description 2
- 230000008901 benefit Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 244000025254 Cannabis sativa Species 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 239000011449 brick Substances 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/40—Document-oriented image-based pattern recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/40—Document-oriented image-based pattern recognition
- G06V30/41—Analysis of document content
- G06V30/413—Classification of content, e.g. text, photographs or tables
Abstract
The invention discloses a kind of natural scene Method for text detection based on deep learning.This method extracts the Analysis On Multi-scale Features of text using CNN network, then encodes these features using RNN to make full use of the contextual properties of text；Then, characteristic pattern is inputted into the pond ROI layer and exports a series of text and proposed.After by non-maxima suppression, the text proposal of generation is connected finally by a text connector, to realize multiple dimensioned, multidirectional text detection flexibly and efficiently.The present invention improves multi-direction, natural scene text detection under the conditions of mutative scale accuracy rate and recall rate.
Description
Technical field
The invention belongs to technical field of image processing, and in particular to a kind of natural scene text detection based on deep learning
Method.
Background technique
Scene text detection is the important prerequisite of text identification, is often used in image retrieval, machine translation, automatic Pilot
Equal fields.But text detection complex background, multiple dimensioned, multilingual, uneven illumination is even, it is fuzzy when detection still
So there is many difficulties.
The diversity of natural scene text is with variability: compared with the text in document, the text of natural scene may be
Multiple dimensioned, multilingual, shape, direction, ratio, color are possibly different from, these variations are all brought to the detection of text
Lot of challenges.
Complex background: scene text may occur in arbitrary background, including signal signature, brick or thick grass, grid
Column, these backgrounds may have the feature closely similar with text, be likely to become the judgement of influence of noise text.Meanwhile also
The missing of text caused by the blocking of foreign matter, causes potentially to detect mistake.
Irregular image quality: due to uncontrollable collect means, the quality of imaging is not can guarantee.For detection
Image may cause distortion, empty burnt due to different shooting angle or shooting distance, or when due to shooting illumination difference
Form noise, shade.
For natural scene text detection problem, the method that can will test is divided into two classes, and one kind is traditional detection method, separately
One kind is the detection method based on deep learning.Traditional method has the method based on texture, such as uses local strength, filter
Response, wavelet coefficient etc.；Have the method based on region, as stroke width transformation (Stroke Width Transform, SWT),
Maximum extreme value stability region (Maximally Stable Extremal Regions, MSER), stroke feature convert (Stroke
Feature Transform, SFT) etc..In recent years, with the development of deep neural network, deep learning is led in computer vision
Domain shows increasing advantage.Currently, it is most popular or be based on convolutional neural networks (Convolutional Neural
Networks, CNN) deep learning method.After having used deep learning, the accuracy of text detection is substantially increased,
And people are freed from complicated characteristic Design work.The commonly natural scene text detection based on deep learning
Model is typically based on common target detection model, such as RCNN, YOLO, SSD.The basic structure of these models is usually to use number
A convolutional layer and pond layer extract feature, and the classification and recurrence of detection block are finally carried out using full articulamentum.
Summary of the invention
In order to more accurately and efficiently carry out text detection in natural scene, solve text in natural scene it is multi-direction,
The test problems of mutative scale, the invention proposes a kind of natural scene Method for text detection based on deep learning.
The purpose of the present invention is realized at least through one of following technical solution.
A kind of natural scene Method for text detection based on deep learning, includes the following steps:
(1) natural scene text detection model neural network based is constructed and is trained, includes following sub-step:
(1.1) building is based on the feature extraction of feature pyramid network (Feature Pyramid Networks, FPN)
Device；
(1.2) feature extractor is extracted using Recognition with Recurrent Neural Network (Recurrent Neural Network, RNN)
Feature encoded；
(1.3) precision of detection is further increased using the pond ROI layer；
(1.4) classification and recurrence that detection block is finally carried out using full articulamentum, form text detection model；
It (1.5) will be by the training figure input model of mark；Use the multitask comprising Classification Loss and recurrence loss
Loss function calculates penalty values with training pattern；
(2) using it is above-mentioned training complete natural scene text detection model to the natural scene text in given image into
Row detection, includes following sub-step:
(2.1) image to be detected is inputted, literary version is carried out to given image using model after above-mentioned training and is detected, a system is exported
The score and coordinate of column text proposal detection block.
(2.2) obtained text is proposed to carry out non-maxima suppression, to remove partial redundance detection block.
(2.3) proposal of a series of text is attached using text connector, generates final testing result.
Compared with prior art, the invention has the advantages that and technical effect:
(1) present invention has used feature pyramid network (Feature Pyramid for the text detection of mutative scale
Networks, FPN), it can be efficiently simultaneously using the information of each different size of convolutional layer, compared to using the last layer
The method of characteristic pattern, while the high resolution information of high-rise strong semantic information and bottom is utilized, to realize higher call together
Return rate and accuracy rate；Compared to the method based on image pyramid, then greatly reduce calculation amount.
(2) multidirectional text detection is connected by the way of exporting a series of texts and proposing finally by text
Device connects the proposal of these texts, compared to using arbitrary quadrilateral or rotating the method for rectangle, has used less
Parameter, thus more flexible to the detection of multi-direction text efficient.
Detailed description of the invention
Fig. 1 is natural scene text detection flow chart in embodiment.
Fig. 2 is natural scene text detection model support composition used in embodiment.
Fig. 3 is the actual result figure detected under different scenes in embodiment using Method for text detection of the invention.
Specific embodiment
In order to which technical solution of the present invention and advantage is more clearly understood, below in conjunction with attached drawing, carry out further detailed
It is without being limited thereto to describe bright but of the invention implementation and protection in detail.
Illustrate the term in the present invention first:
Feature pyramid network (Feature Pyramid Networks, FPN): FPN is directly in original back bone network
On make an amendment, the characteristic pattern of each resolution ratio, which introduces latter resolution ratio and scales twice of characteristic pattern, does that each element is corresponding to be added
Operation.Connection in this way, each layer of prediction characteristic pattern used have all merged the spy of different resolution, different semantic intensity
Sign, the characteristic pattern of the different resolution of fusion do the object detection of corresponding resolution sizes respectively.This ensure that each layer is all
There are suitable resolution ratio and strong semantic feature.
Residual error network (ResNet): it is the depth convolutional network model that He Kaiming was proposed in 2015, is adopted according to model
The difference of the number of plies is respectively designated as ResNet-34, ResNet-50, ResNet-101, ResNet-152 etc..
Non-maxima suppression (Non-Maximum Suppression, NMS): inhibition is not the element of maximum, Ke Yili
Solution is that local maxima is searched for.Each detection block of output has a score, these detection blocks there may be comprising with intersection
Situation, the detection block that the detection block of highest scoring in selection field is carried out using NMS, and inhibits those scores low.
As shown in Figure 1, the natural scene text detection model in the present invention based on deep learning, comprising the following steps:
(1) it constructs and trains natural scene text detection model neural network based, as shown in Fig. 2, including following son
Step:
(1.1) building is based on the feature extraction of feature pyramid network (Feature Pyramid Networks, FPN)
Device.Use ResNet-101 as back bone network, generates feature pyramid, use the feature of the level therein from P2 to P5.
(1.2) feature extracted is carried out using Recognition with Recurrent Neural Network (Recurrent Neural Network, RNN)
Coding.Use two-way long short-term memory Recognition with Recurrent Neural Network (the Bi-directional Long Short- of 512 hidden layers
Term Memory, Bi-LSTM) feature extracted is encoded as RNN.
(1.3) precision of detection is further increased using the pond ROI layer.The concrete operations in the pond ROI are as follows:
ROI is mapped to the corresponding position of characteristic pattern according to the image of input by (1.3.1)；
Region division after mapping is the part of same size by (1.3.2), and the quantity of division and the dimension of output are identical；
(1.3.3) carries out maximum pondization operation to each part.
(1.4) classification and recurrence of detection block are finally carried out using full articulamentum.Pass through respectively by the feature in the pond ROI
Two full articulamentums are classified and are returned.If the quantity of the detection block of output is k, wherein the dimension of classification layer output is 2k,
Correspond to text and background；The dimension for returning layer output is 4k, the upper left of corresponding detection block and the coordinate of bottom right 2.
(1.5) input is trained model by the training figure of mark.Wherein, quadrangle can be used in training image
Rectangle mark also can be used in mark.But before input model, it need to be divided by given width, if training image mark
Note is quadrangle, then the minimum circumscribed rectangle after taking it to divide；If being labeled as rectangle, directly divide.
Design is comprising Classification Loss and returns the multitask loss function lost.It is lost using the loss function of design
Calculating:
Wherein L, LclsAnd LregRespectively total losses, Classification Loss and recurrence loss, λ are balanced sort loss and recurrence damage
Weight coefficient between mistake.piIt is the classification of i-th of detection block prediction,It is the true classification of i-th of detection block.tiIt is i-th
The prediction coordinate of detection block,It is the true coordinate of i-th of detection block.
(2) using it is above-mentioned training complete natural scene text detection model to the natural scene text in given image into
Row detection, includes following sub-step:
(2.1) image to be detected is inputted, text detection is carried out to given image using model after above-mentioned training, exports a system
The score and coordinate of column text proposal detection block.
(2.2) obtained text is proposed to carry out non-maxima suppression, to remove partial redundance detection block.Concrete operations are such as
Under:
The list B and its corresponding score S for proposing detection block for text, using following calculation.Selection has
Largest score
Detection block M, by it from removing and be added in final testing result D in set B.Usually examined remaining in B
It surveys in frame
It is removed from B with the IoU of the M frame for being greater than threshold value.This process is repeated, until B is sky.
(2.3) proposal of a series of text is attached using text connector, generates final testing result.It uses
Such as
Lower step carries out the connection of text proposal:
If proposing PjWith proposal Pi(i, j herein indicates different proposals) meets following two conditions, will propose PjDefinition
To propose PiNeighbours:
(1) propose PjWith proposal PiIt is nearest and they the distance between be less than wj+wi
(2) propose PjWith proposal PiThere is the registration greater than 0.5 in vertical direction
Wherein wiAnd wjRespectively propose PiWith proposal PjWidth, if propose PiIt is to propose PjNeighbours and propose Pj
It is to propose PiNeighbours, the two proposals are connected as the same detection block by this.Above-mentioned steps are repeated, until all mention
View connection is completed, then detection block is final output result.From Fig. 2 and Fig. 3 it is found that detection effect of the present invention in natural scene
Fruit, it is seen that the present invention can detect well mutative scale, multidirectional text in natural scene.
Claims (8)
1. a kind of natural scene Method for text detection based on deep learning, it is characterised in that the following steps are included:
(1) it constructs and trains natural scene text detection model neural network based, comprising:
(1.1) building is based on the feature extractor of feature pyramid network (Feature Pyramid Networks, FPN)；
(1.2) spy that feature extractor is extracted using Recognition with Recurrent Neural Network (Recurrent Neural Network, RNN)
Sign is encoded；
(1.3) precision of detection is further increased using the pond ROI layer；
(1.4) classification and recurrence that detection block is finally carried out using full articulamentum, form text detection model；
It (1.5) will be by the training figure input model of mark；
(1.6) penalty values are calculated with training pattern using the multitask loss function comprising Classification Loss and recurrence loss；
(2) the natural scene text in given image is examined using the natural scene text detection model that training is completed
It surveys, includes following sub-step:
(2.1) image to be detected is inputted, text inspection is carried out to given image using the natural scene text detection model after training
It surveys, exports score and coordinate that a series of texts propose detection block；
(2.2) obtained text is proposed to carry out non-maxima suppression, to remove partial redundance detection block；
(2.3) proposal of a series of text is attached using text connector, generates final testing result.
2. the natural scene Method for text detection according to claim 1 based on deep learning, which is characterized in that constructing
In natural scene text detection model neural network based, feature pyramid network (Feature Pyramid Networks,
FPN the level from P2 to P5) has been only used.
3. the natural scene Method for text detection according to claim 1 based on deep learning, which is characterized in that constructing
In natural scene text detection model neural network based, feature pyramid network (Feature Pyramid Networks,
FPN) used ResNet-101 as back bone network.
4. the natural scene Method for text detection according to claim 1 based on deep learning, which is characterized in that constructing
In natural scene text detection model neural network based, nerve is recycled using the two-way long short-term memory of 512 hidden layers
Network (Bi-directional Long Short-Term Memory, Bi-LSTM) is used as Recognition with Recurrent Neural Network (Recurrent
Neural Network, RNN) feature extracted is encoded.
5. the natural scene Method for text detection according to claim 1 based on deep learning, which is characterized in that constructing
In natural scene text detection model neural network based, the calculating lost using following loss function:
Wherein L, LclsAnd LregRespectively total losses, Classification Loss and recurrence loss, λ are balanced sort loss and recurrence loss
Between weight coefficient,It is the true classification of i-th of detection block.
6. the natural scene Method for text detection according to claim 5 based on deep learning, which is characterized in that classification damage
Mistake is defined as follows:
Wherein, piIt is the prediction classification of i-th of detection block,It is the true classification of i-th of detection block.
7. the natural scene Method for text detection according to claim 5 based on deep learning, which is characterized in that return damage
Mistake is defined as follows:
Wherein, tiIt is the prediction coordinate of i-th of detection block,It is the true coordinate of i-th of detection block.
8. the natural scene Method for text detection according to claim 1 based on deep learning, which is characterized in that giving
Determine to use following steps to carry out the connection of text proposal during the natural scene text in image detected:
If proposing PjWith proposal PiMeet following two conditions, will propose PjIt is defined as proposing PiNeighbours:
(1) propose PjWith proposal PiIt is nearest and they the distance between be less than wj+wi；
(2) propose PjWith proposal PiThere is the registration greater than 0.5 in vertical direction
Wherein wiAnd wjRespectively propose PiWith proposal PjWidth, if propose PiIt is to propose PjNeighbours and propose PjIt is to mention
Discuss PiNeighbours, the two proposals are connected as the same detection block by this.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910270269.4A CN110135248A (en) | 2019-04-03 | 2019-04-03 | A kind of natural scene Method for text detection based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910270269.4A CN110135248A (en) | 2019-04-03 | 2019-04-03 | A kind of natural scene Method for text detection based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110135248A true CN110135248A (en) | 2019-08-16 |
Family
ID=67569376
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910270269.4A Pending CN110135248A (en) | 2019-04-03 | 2019-04-03 | A kind of natural scene Method for text detection based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110135248A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110766020A (en) * | 2019-10-30 | 2020-02-07 | 哈尔滨工业大学 | System and method for detecting and identifying multi-language natural scene text |
| CN110807422A (en) * | 2019-10-31 | 2020-02-18 | 华南理工大学 | Natural scene text detection method based on deep learning |
| CN111753714A (en) * | 2020-06-23 | 2020-10-09 | 中南大学 | Multidirectional natural scene text detection method based on character segmentation |
| WO2020221298A1 (en) * | 2019-04-30 | 2020-11-05 | 北京金山云网络技术有限公司 | Text detection model training method and apparatus, text region determination method and apparatus, and text content determination method and apparatus |
| CN113591829A (en) * | 2021-05-25 | 2021-11-02 | 上海一谈网络科技有限公司 | Character recognition method, device, equipment and storage medium |
Citations (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101436299A (en) * | 2008-11-19 | 2009-05-20 | 哈尔滨工业大学 | Method for detecting natural scene image words |
| CN103942550A (en) * | 2014-05-04 | 2014-07-23 | 厦门大学 | Scene text recognition method based on sparse coding characteristics |
| CN104537362A (en) * | 2015-01-16 | 2015-04-22 | 中国科学院自动化研究所 | Domain-based self-adaptive English scene character recognition method |
| CN105740909A (en) * | 2016-02-02 | 2016-07-06 | 华中科技大学 | Text recognition method under natural scene on the basis of spatial transformation |
| CN106650725A (en) * | 2016-11-29 | 2017-05-10 | 华南理工大学 | Full convolutional neural network-based candidate text box generation and text detection method |
| CN107122342A (en) * | 2017-04-21 | 2017-09-01 | 东莞中国科学院云计算产业技术创新与育成中心 | Text code recognition methods and device |
| CN107203606A (en) * | 2017-05-17 | 2017-09-26 | 西北工业大学 | Text detection and recognition methods under natural scene based on convolutional neural networks |
| CN107402947A (en) * | 2017-03-29 | 2017-11-28 | 北京粉笔未来科技有限公司 | Picture retrieval method for establishing model and device, picture retrieval method and device |
| CN107622267A (en) * | 2017-10-16 | 2018-01-23 | 天津师范大学 | A kind of scene character recognition method based on embedded bilateral convolution activation |
| CN108288088A (en) * | 2018-01-17 | 2018-07-17 | 浙江大学 | A kind of scene text detection method based on end-to-end full convolutional neural networks |
| CN108304835A (en) * | 2018-01-30 | 2018-07-20 | 百度在线网络技术（北京）有限公司 | character detecting method and device |
| CN108399419A (en) * | 2018-01-25 | 2018-08-14 | 华南理工大学 | Chinese text recognition methods in natural scene image based on two-dimentional Recursive Networks |
| CN108427665A (en) * | 2018-03-15 | 2018-08-21 | 广州大学 | A kind of text automatic generation method based on LSTM type RNN models |
| CN108549893A (en) * | 2018-04-04 | 2018-09-18 | 华中科技大学 | A kind of end-to-end recognition methods of the scene text of arbitrary shape |
| CN108573257A (en) * | 2017-03-14 | 2018-09-25 | 奥多比公司 | Image is divided automatically based on natural language phrase |
| CN108764133A (en) * | 2018-05-25 | 2018-11-06 | 北京旷视科技有限公司 | Image-recognizing method, apparatus and system |
| US20180365560A1 (en) * | 2017-06-19 | 2018-12-20 | International Business Machines Corporation | Context aware sensitive information detection |
| CN109299274A (en) * | 2018-11-07 | 2019-02-01 | 南京大学 | A kind of natural scene Method for text detection based on full convolutional neural networks |
| CN109344824A (en) * | 2018-09-21 | 2019-02-15 | 泰康保险集团股份有限公司 | A kind of line of text method for detecting area, device, medium and electronic equipment |
| CN109492638A (en) * | 2018-11-07 | 2019-03-19 | 北京旷视科技有限公司 | Method for text detection, device and electronic equipment |
-
2019
- 2019-04-03 CN CN201910270269.4A patent/CN110135248A/en active Pending
Patent Citations (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101436299A (en) * | 2008-11-19 | 2009-05-20 | 哈尔滨工业大学 | Method for detecting natural scene image words |
| CN103942550A (en) * | 2014-05-04 | 2014-07-23 | 厦门大学 | Scene text recognition method based on sparse coding characteristics |
| CN104537362A (en) * | 2015-01-16 | 2015-04-22 | 中国科学院自动化研究所 | Domain-based self-adaptive English scene character recognition method |
| CN105740909A (en) * | 2016-02-02 | 2016-07-06 | 华中科技大学 | Text recognition method under natural scene on the basis of spatial transformation |
| CN106650725A (en) * | 2016-11-29 | 2017-05-10 | 华南理工大学 | Full convolutional neural network-based candidate text box generation and text detection method |
| CN108573257A (en) * | 2017-03-14 | 2018-09-25 | 奥多比公司 | Image is divided automatically based on natural language phrase |
| CN107402947A (en) * | 2017-03-29 | 2017-11-28 | 北京粉笔未来科技有限公司 | Picture retrieval method for establishing model and device, picture retrieval method and device |
| CN107122342A (en) * | 2017-04-21 | 2017-09-01 | 东莞中国科学院云计算产业技术创新与育成中心 | Text code recognition methods and device |
| CN107203606A (en) * | 2017-05-17 | 2017-09-26 | 西北工业大学 | Text detection and recognition methods under natural scene based on convolutional neural networks |
| US20180365560A1 (en) * | 2017-06-19 | 2018-12-20 | International Business Machines Corporation | Context aware sensitive information detection |
| CN107622267A (en) * | 2017-10-16 | 2018-01-23 | 天津师范大学 | A kind of scene character recognition method based on embedded bilateral convolution activation |
| CN108288088A (en) * | 2018-01-17 | 2018-07-17 | 浙江大学 | A kind of scene text detection method based on end-to-end full convolutional neural networks |
| CN108399419A (en) * | 2018-01-25 | 2018-08-14 | 华南理工大学 | Chinese text recognition methods in natural scene image based on two-dimentional Recursive Networks |
| CN108304835A (en) * | 2018-01-30 | 2018-07-20 | 百度在线网络技术（北京）有限公司 | character detecting method and device |
| CN108427665A (en) * | 2018-03-15 | 2018-08-21 | 广州大学 | A kind of text automatic generation method based on LSTM type RNN models |
| CN108549893A (en) * | 2018-04-04 | 2018-09-18 | 华中科技大学 | A kind of end-to-end recognition methods of the scene text of arbitrary shape |
| CN108764133A (en) * | 2018-05-25 | 2018-11-06 | 北京旷视科技有限公司 | Image-recognizing method, apparatus and system |
| CN109344824A (en) * | 2018-09-21 | 2019-02-15 | 泰康保险集团股份有限公司 | A kind of line of text method for detecting area, device, medium and electronic equipment |
| CN109299274A (en) * | 2018-11-07 | 2019-02-01 | 南京大学 | A kind of natural scene Method for text detection based on full convolutional neural networks |
| CN109492638A (en) * | 2018-11-07 | 2019-03-19 | 北京旷视科技有限公司 | Method for text detection, device and electronic equipment |
Non-Patent Citations (2)
| Title |
|---|
| YU SONG ETC.: "Scene Text Detection via Deep Semantic Feature Fusion and Attention-based Refinement", 《2018 24TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION(ICPR)》 * |
| 蔡华杰等: "基于 WT-BTC 特征和 SVM 组合分类的场景文本检测", 《科学技术与工程》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020221298A1 (en) * | 2019-04-30 | 2020-11-05 | 北京金山云网络技术有限公司 | Text detection model training method and apparatus, text region determination method and apparatus, and text content determination method and apparatus |
| CN110766020A (en) * | 2019-10-30 | 2020-02-07 | 哈尔滨工业大学 | System and method for detecting and identifying multi-language natural scene text |
| CN110807422A (en) * | 2019-10-31 | 2020-02-18 | 华南理工大学 | Natural scene text detection method based on deep learning |
| CN110807422B (en) * | 2019-10-31 | 2023-05-23 | 华南理工大学 | Natural scene text detection method based on deep learning |
| CN111753714A (en) * | 2020-06-23 | 2020-10-09 | 中南大学 | Multidirectional natural scene text detection method based on character segmentation |
| CN111753714B (en) * | 2020-06-23 | 2023-09-01 | 中南大学 | Multidirectional natural scene text detection method based on character segmentation |
| CN113591829A (en) * | 2021-05-25 | 2021-11-02 | 上海一谈网络科技有限公司 | Character recognition method, device, equipment and storage medium |
| CN113591829B (en) * | 2021-05-25 | 2024-02-13 | 上海一谈网络科技有限公司 | Character recognition method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110135248A (en) | A kind of natural scene Method for text detection based on deep learning | |
| Yang et al. | Real-time face detection based on YOLO | |
| CN108428229B (en) | Lung texture recognition method based on appearance and geometric features extracted by deep neural network | |
| CN110807422B (en) | Natural scene text detection method based on deep learning | |
| CN106599854B (en) | Automatic facial expression recognition method based on multi-feature fusion | |
| CN104463250B (en) | A kind of Sign Language Recognition interpretation method based on Davinci technology | |
| CN109034210A (en) | Object detection method based on super Fusion Features Yu multi-Scale Pyramid network | |
| CN103049763B (en) | Context-constraint-based target identification method | |
| CN109272500B (en) | Fabric classification method based on adaptive convolutional neural network | |
| CN108288271A (en) | Image detecting system and method based on three-dimensional residual error network | |
| CN108288075A (en) | A kind of lightweight small target detecting method improving SSD | |
| CN106610969A (en) | Multimodal information-based video content auditing system and method | |
| CN110135296A (en) | Airfield runway FOD detection method based on convolutional neural networks | |
| CN107016405A (en) | A kind of insect image classification method based on classification prediction convolutional neural networks | |
| CN109446982B (en) | AR glasses-based electric power cabinet pressing plate state identification method and system | |
| CN107133622A (en) | The dividing method and device of a kind of word | |
| CN104992223A (en) | Dense population estimation method based on deep learning | |
| CN104680127A (en) | Gesture identification method and gesture identification system | |
| CN111680706A (en) | Double-channel output contour detection method based on coding and decoding structure | |
| CN104268590B (en) | The blind image quality evaluating method returned based on complementary combination feature and multiphase | |
| CN106557740B (en) | The recognition methods of oil depot target in a kind of remote sensing images | |
| CN111445459A (en) | Image defect detection method and system based on depth twin network | |
| CN106228528A (en) | A kind of multi-focus image fusing method based on decision diagram Yu rarefaction representation | |
| Huang et al. | A multi-feature fusion-based deep learning for insulator image identification and fault detection | |
| CN114972312A (en) | Improved insulator defect detection method based on YOLOv4-Tiny |
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: 20190816 |