CN108694393A - A kind of certificate image text area extraction method based on depth convolution - Google Patents
A kind of certificate image text area extraction method based on depth convolution Download PDFInfo
- Publication number
- CN108694393A CN108694393A CN201810536528.9A CN201810536528A CN108694393A CN 108694393 A CN108694393 A CN 108694393A CN 201810536528 A CN201810536528 A CN 201810536528A CN 108694393 A CN108694393 A CN 108694393A
- Authority
- CN
- China
- Prior art keywords
- image
- text
- certificate
- feature
- 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
Classifications
-
- 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
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/245—Aligning, centring, orientation detection or correction of the image by locating a pattern; Special marks for positioning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/146—Aligning or centring of the image pick-up or image-field
- G06V30/1475—Inclination or skew detection or correction of characters or of image to be recognised
- G06V30/1478—Inclination or skew detection or correction of characters or of image to be recognised of characters or characters lines
-
- 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
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The certificate image text area extraction method based on depth convolution that the invention discloses a kind of,It is related to image processing field,A kind of more particularly to certificate image text area extraction method based on depth convolution,Multiple certificate images that can not be irregular to polymorphic type certificate or the space of a whole page to the image recognition technology solved the problems, such as under the prior art carry out text filed positioning,The present invention includes pre-process simultaneously level correction to target image,The line of text image in horizontal direction is carried out to the image after level correction using depth convolutional neural networks to be accurately positioned,By the line of text image zooming-out of positioning and export,The present invention uses the certificate image text area extraction method based on depth convolution,It can be irregular to polymorphic type certificate or the space of a whole page,The certificate image that the space of a whole page of multiple certificate images is inconsistent carries out text filed be accurately positioned,With accurate positioning,The good advantage of real-time.
Description
Technical field
The present invention relates to image processing field more particularly to a kind of certificate image text area extractions based on depth convolution
Method.
Background technology
The final steps that the text area extraction of certificate image is identified as certificate go out the Text Feature Extraction in image
Come, in order to the extraction and identification for character, with the more acurrate better process for obtaining the text message on certificate image.
The text area extraction of certificate image be exactly by the image comprising certificate carry out it is text filed be accurately positioned, and base
Text is accurately extracted in positioning.Text filed extraction is generally divided into two ways, directly to the text in image
Carry out positioning analysis;First image is corrected, then carries out text filed positioning based on the image after correction.
Traditional character recognition (OCR) is generally divided into image preprocessing, word area detection, character cutting, character recognition
Four modules.Image preprocessing mainly with to image carry out image enhancing, noise reduction and correction etc..The pretreatment of image is figure
As the first step of processing, can significantly improve to extraction text filed in image and recognition accuracy.In image
The detection of character area is divided into two methods, a kind of to use printed page analysis, by being carried out on the space of a whole page to specific certificate image
Analysis requires the space of a whole page of certificate image to return to directly carry out text filed positioning to the image of certificate using printed page analysis
It is whole, first certificate target image is extracted, then only positioned with relative position, accuracy rate by the space of a whole page alignment degree with
The extraction correctness of target image influences;Another method is the method using neural network learning, directly to image into style of writing
The selection of one's respective area is not influenced by space of a whole page etc. using network learning method, need not also be extracted to target image, directly
It is text filed to connect the intelligent positioning extraction from image.
In the application of depth convolutional neural networks OCR, there are several forms:Text filed, use is extracted using method for distinguishing
Character picture is identified in depth convolutional neural networks;Using depth convolutional neural networks to the text filed progress in image
Extraction, is then identified again;Using deep learning method, a system end to end is designed, directly by the text in image
Region recognition returns to the result of identification.Wherein optimal mode certainly system end to end, but it realizes that difficulty is also
Highest, to reach ideal accuracy rate is difficult.And the research of image recognition is mainly concentrated in OCR for image Chinese in recent years
The research of the positioning of one's respective area, the proposition of Faster R-CNN, Yolo methods etc., can quickly determine target in image
Position and identification, based on this and improved and propose much for natural scene character area extraction and recognition methods,
Also reach preferable effect.From the successive proposition of Faster R-CNN and Yolo algorithms, to the text filed inspection under natural scene
Survey is also paid close attention to, and has the method much based on depth convolutional neural networks to the text filed inspection under natural scene in succession
It surveys, can reach 85% or so to the verification and measurement ratio of text under natural scene.
Traditional printed page analysis method is higher to text filed locating accuracy, but by the type shadow of certificate image
It rings, needs to carry out printed page analysis to the image of every one kind, and it also requires image is regular, and some irregular certificates are then
It cannot carry out printed page analysis;Although the text filed positioning based on depth convolutional neural networks intelligent can be determined text
Position, but it is inadequate to the setting accuracy in region during the study of the method for one side neural network, and most of all, being based on
The model of convolutional neural networks needs largely to calculate to need longer time in identification positions, it is difficult to accomplish in real time
Effect.
Invention content
It is an object of the invention to:In order to which the image recognition technology solved under the prior art can not be to polymorphic type certificate or version
The problem of irregular multiple certificate images in face carry out text filed positioning, the present invention provides a kind of cards based on depth convolution
Part image text method for extracting region carries out in horizontal direction pretreated target image using depth convolutional neural networks
Line of text image be accurately positioned, the text in image can be accurately positioned, and have the advantages that real-time is good.
The technical solution adopted by the present invention is as follows:
A kind of certificate image text area extraction method based on depth convolution, includes the following steps:
S1:Target image is pre-processed, pretreated target image obtains horizontal image after level correction;
S2:The horizontal image obtained in S1 is substituted into the network model based on convolutional neural networks and is positioned, is obtained
Output image as unit of line of text.
Further, the particular content in the S1 is as follows:
S101:Image outline is obtained after being pre-processed to target image, is connected image outline using closed operation,
And profile lookup algorithm is combined to take out profile;
S102:The profile extracted in S101 is screened using geometric proportion, chooses target certificate profile;
S103:Based on target certificate profile selected in S102, according to the position pair of the slant characteristic of profile and profile
Image is rotated;
S104:Its angle point is monitored using Hough transformation to postrotational image in S103 and combines perspective transform by image calibration
Just obtain horizontal image.
Further, the pretreated method includes following two:
Method 1:It first uses Gaussian mode gelatinization processing to carry out noise reduction to target image, then carries out gray processing processing, gray processing
Treated, and image uses Sobel operator edge treateds, obtains image outline;
Method 2:First Edge contrast is used to enhance target image details, reuses Canny operator edge treateds and obtain image
Profile.
Further, the standard of the screening is according to the feature of specific certificate and the standard length-breadth ratio of specific certificate
Choose target certificate profile.
Further, the particular content in the S2 is as follows:
S201:Convolution feature extraction is carried out to the horizontal image obtained in S1, is carried using the depth convolution model of VGG16
The data taken extract feature as basic network, and it is big as characteristic pattern to obtain the 3rd sublayer feature in VGG layer 5 convolutional layers
The small width and height for being respectively input picture for W × H × C, W and H, C are convolution nuclear volume, and VGG16 is generated the is used only herein
Five layer datas are handled.
S202:Convolution is made using the sliding window of 512 3 × 3 sizes to the characteristic pattern obtained in S201, each point is rolled up
Product obtains the feature vector of 512 sizes;
S203:For each window center, the text box that 10 height are 13~273 can be all generated, and to each text
Frame uses the probability that the feature vector zoning extracted in S202 is text;
S204:Using the feature vector obtained in S202 as the input of a two-way LSTM, and provide its output be W ×
256 length connect the output of 512 full articulamentum progress result behind two-way LSTM;
S205:Output result is integrated based on line of text construction algorithm, sequence or the text box mutually closed on are carried out
It integrates, and non-maxima suppression algorithm is carried out to extra text box and is filtered, the line of text finally integrated is target figure
The line of text of picture.
Further, feature of the output result comprising text box position, text box judges and for adjusting line of text
The predictive information of endpoint location, the text box are indicated by two values of height of center and rectangle frame.
Further, the feature to text box is judged as two, and one is the probability for being judged as text, another is
It is judged as non-textual probability.
In conclusion by adopting the above-described technical solution, the beneficial effects of the invention are as follows:
1. the present invention uses the certificate image text area extraction method based on depth convolution, can to polymorphic type certificate or
The certificate image that the space of a whole page of irregular, the multiple certificate image of the space of a whole page is inconsistent carries out text filed be accurately positioned.
2. the present invention can carry out String localization extraction under the premise of being calculated using GPU in 1s, recognition speed is fast, real
When property is good.
3. the present invention breaks through the weakness of printed page analysis in tradition, to that cannot carry out the certificate image and multiple types of printed page analysis
The certificate image of type can also carry out text filed positioning.
4. by the present invention in that carrying out level correction to image with image preprocessing, can filter to being caused by photo environment
The illumination effect to image, the influences such as angular transformation and rotation transformation interference, so as to realize to it is text filed carry out it is accurate
Positioning.
Description of the drawings
Examples of the present invention will be described by way of reference to the accompanying drawings, in the accompanying drawings:
Fig. 1 is overall flow schematic diagram of the present invention;
Fig. 2 is the neural network model figure the present invention is based on depth convolution;
Fig. 3 is result figure after present invention pretreatment, and left side is using Sobel operator handling result figures, and right side is to use
Canny operator handling result figures.
Fig. 4 is the design sketch that the present invention carries out image rotation correction;
Fig. 5 is the positioning effect to ID Card Image the present invention is based on the text filed location model of depth convolutional neural networks
Fruit is schemed.
Specific implementation mode
In order to which those skilled in the art are better understood from invention, the present invention is carried out with reference to the accompanying drawings and examples detailed
It describes in detail bright.
A kind of certificate image text area extraction method based on depth convolution, includes the following steps:
S1:Target image is pre-processed, pretreated target image obtains horizontal image after level correction;
S2:The horizontal image obtained in S1 is substituted into the network model based on convolutional neural networks and is positioned, is obtained
Output image as unit of line of text.
As a preferred embodiment, the particular content in the S1 is as follows:
S101:Image outline is obtained after being pre-processed to target image on the left of Fig. 4, is connected image outline using closed operation
It picks up and, and profile lookup algorithm is combined to take out profile;
S102:The profile extracted in S101 is screened using geometric proportion, chooses target certificate profile;
S103:Based on target certificate profile selected in S102, according to the position pair of the slant characteristic of profile and profile
Image is rotated;
S104:Its angle point is monitored using Hough transformation to postrotational image in S103 and combines perspective transform by image calibration
Horizontal image shown in just obtaining on the right side of Fig. 4,
As a preferred embodiment, the pretreated method includes following two:
Method 1:It first uses Gaussian mode gelatinization processing to carry out noise reduction to target image, then carries out gray processing processing, gray processing
Treated, and image uses Sobel operator edge treateds, obtains image outline, is obtained as described in the left sides Fig. 3 using Sobel operators
Profile, the Sobel operator sizes used are 5;
Method 2:First Edge contrast is used to enhance target image details, reuses Canny operator edge treateds and obtain image
Profile obtains the profile as shown in the right sides Fig. 3 using Canny operators, and Canny operators are 3 using size, and upper lower threshold value is respectively 89
With 40.
As a preferred embodiment, the method screened described in S102:By the profile combination identity card sheet of extraction
The length-width ratio of body is closest to 108:66 ratio and the feature of identity card are screened, and are chosen identity card certificate profile, are passed through length
For width than carrying out coarse sizing, whether the classification of retraining one is that the SVM classifier of ID Card Image is chosen.
As a preferred embodiment, being set up in the S2 as shown in Figure 2 based on depth convolutional neural networks
Text filed location model and carry out target image String localization analysis using the model, particular content is as follows:
S201:Convolution feature extraction is carried out to the horizontal image obtained in S1, is carried using the depth convolution model of VGG16
The data taken extract feature as basic network, and it is big as characteristic pattern to obtain the 3rd sublayer feature in VGG layer 5 convolutional layers
The small width and height for being respectively input picture for W × H × C, W and H, C are convolution nuclear volume, and VGG16 is generated the is used only herein
Five layer datas are handled.
S202:Convolution is made using the sliding window of 512 3 × 3 sizes to the characteristic pattern obtained in S201, each point is rolled up
Product obtains the feature vector of 512 sizes;
S203:For each window center, the text box that 10 height are 13~273 can be all generated, and to each text
Frame uses the probability that the feature vector zoning extracted in S202 is text;
S204:Using the feature vector obtained in S202 as the input of a two-way LSTM, and provide its output be W ×
256 length connect the output of 512 full articulamentum progress result behind two-way LSTM;
S205:Output result is integrated based on line of text construction algorithm, sequence or the text box mutually closed on are carried out
Integrate, and non-maxima suppression algorithm carried out to extra text box and is filtered, the line of text finally integrated as shown in figure 5,
The as line of text of target image.
As a preferred embodiment, it is described output result include text box position, text box feature judge and
Predictive information for adjusting line of text endpoint location, the text box are worth tables by the height two of center and rectangle frame
Show.
As a preferred embodiment, the feature to text box is judged as two, one is to be judged as text
Probability, another is to be judged as non-textual probability.
The above is only the preferred embodiment of the present invention, it is noted that is come for those of ordinary skill in the art
It says, without departing from the inventive concept of the premise, can also make several modifications and improvements, these belong to the protection model of invention
It encloses.
Claims (7)
1. a kind of certificate image text area extraction method based on depth convolution, which is characterized in that include the following steps:
S1:Target image is pre-processed, pretreated target image obtains horizontal image after level correction;
S2:The horizontal image obtained in S1 is substituted into the network model based on convolutional neural networks and is positioned, is obtained with text
One's own profession is the output image of unit.
2. a kind of certificate image text area extraction method based on depth convolution according to claim 1, feature exist
In the particular content in the S1 is as follows:
S101:Image outline is obtained after being pre-processed to target image, is connected image outline using closed operation, and tie
It closes profile lookup algorithm and takes out profile;
S102:The profile extracted in S101 is screened using geometric proportion, chooses target certificate profile;
S103:Based on target certificate profile selected in S102, according to the position of the slant characteristic of profile and profile to image
It is rotated;
S104:Its angle point is monitored using Hough transformation to postrotational image in S103 and obtains image rectification in conjunction with perspective transform
To horizontal image.
3. a kind of certificate image text area extraction method based on depth convolution according to claim 2, feature exist
In the pretreated method includes following two:
Method 1:It first uses Gaussian mode gelatinization processing to carry out noise reduction to target image, then carries out gray processing processing, gray processing processing
Image afterwards uses Sobel operator edge treateds, obtains image outline;
Method 2:First Edge contrast is used to enhance target image details, reuses Canny operator edge treateds and obtain image outline.
4. a kind of certificate image text area extraction method based on depth convolution according to claim 2, feature exist
In the standard of the screening is to choose target certificate wheel according to the feature of specific certificate and the standard length-breadth ratio of specific certificate
It is wide.
5. a kind of certificate image text area extraction method based on depth convolution according to claim 1, feature exist
In the particular content in the S2 is as follows:
S201:Convolution feature extraction is carried out to the horizontal image obtained in S1, is extracted using the depth convolution model of VGG16
Data extract feature as basic network, and the 3rd sublayer feature is as characteristic pattern, size W in acquisition VGG layer 5 convolutional layers
× H × C, W and H are respectively the width and height of input picture, and C is convolution nuclear volume;
S202:Convolution is made using the sliding window of 512 3 × 3 sizes to the characteristic pattern obtained in S201, each convolution is obtained
To the feature vector of 512 sizes;
S203:For each window center, it can all generate 10 height and be 13~273 text box, and each text box is made
With the probability that the feature vector zoning extracted in S202 is text;
S204:Using the feature vector obtained in S202 as the input of a two-way LSTM, and provide that its output is that W × 256 is grown
Degree connects the output of 512 full articulamentum progress result behind two-way LSTM;
S205:Output result is integrated based on line of text construction algorithm, sequence or the text box mutually closed on are integrated,
And non-maxima suppression algorithm is carried out to extra text box and is filtered, the line of text finally integrated is the text of target image
One's own profession.
6. a kind of certificate image text area extraction method based on depth convolution according to claim 5, feature exist
In the output result includes text box position, the feature judgement of text box and the prediction for adjusting line of text endpoint location
Information, the text box are indicated by two values of height of center and rectangle frame.
7. a kind of certificate image text area extraction method based on depth convolution according to claim 6, feature exist
In, the feature to text box is judged as two, and one is the probability for being judged as text, another be judged as it is non-textual
Probability.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810536528.9A CN108694393A (en) | 2018-05-30 | 2018-05-30 | A kind of certificate image text area extraction method based on depth convolution |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810536528.9A CN108694393A (en) | 2018-05-30 | 2018-05-30 | A kind of certificate image text area extraction method based on depth convolution |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108694393A true CN108694393A (en) | 2018-10-23 |
Family
ID=63849295
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810536528.9A Pending CN108694393A (en) | 2018-05-30 | 2018-05-30 | A kind of certificate image text area extraction method based on depth convolution |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108694393A (en) |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109376658A (en) * | 2018-10-26 | 2019-02-22 | 信雅达系统工程股份有限公司 | A kind of OCR method based on deep learning |
CN109800757A (en) * | 2019-01-04 | 2019-05-24 | 西北工业大学 | A kind of video text method for tracing based on layout constraint |
CN109886978A (en) * | 2019-02-20 | 2019-06-14 | 贵州电网有限责任公司 | A kind of end-to-end warning information recognition methods based on deep learning |
CN110033469A (en) * | 2019-04-01 | 2019-07-19 | 北京科技大学 | A kind of sub-pixel edge detection method and system |
CN110222752A (en) * | 2019-05-28 | 2019-09-10 | 北京金山数字娱乐科技有限公司 | Image processing method, system, computer equipment, storage medium and chip |
CN110363199A (en) * | 2019-07-16 | 2019-10-22 | 济南浪潮高新科技投资发展有限公司 | Certificate image text recognition method and system based on deep learning |
CN110554991A (en) * | 2019-09-03 | 2019-12-10 | 浙江传媒学院 | Method for correcting and managing text picture |
CN110866871A (en) * | 2019-11-15 | 2020-03-06 | 深圳市华云中盛科技股份有限公司 | Text image correction method and device, computer equipment and storage medium |
CN110889404A (en) * | 2019-11-21 | 2020-03-17 | 山东浪潮人工智能研究院有限公司 | Irregular text recognition system and method based on correction network |
CN111325194A (en) * | 2018-12-13 | 2020-06-23 | 杭州海康威视数字技术股份有限公司 | Character recognition method, device and equipment and storage medium |
CN111368820A (en) * | 2020-03-06 | 2020-07-03 | 腾讯科技(深圳)有限公司 | Text labeling method and device and storage medium |
CN111783761A (en) * | 2020-06-30 | 2020-10-16 | 苏州科达科技股份有限公司 | Certificate text detection method and device and electronic equipment |
CN111797922A (en) * | 2020-07-03 | 2020-10-20 | 泰康保险集团股份有限公司 | Text image classification method and device |
CN112016438A (en) * | 2020-08-26 | 2020-12-01 | 北京嘀嘀无限科技发展有限公司 | Method and system for identifying certificate based on graph neural network |
CN112200191A (en) * | 2020-12-01 | 2021-01-08 | 北京京东尚科信息技术有限公司 | Image processing method, image processing device, computing equipment and medium |
CN112396082A (en) * | 2019-08-19 | 2021-02-23 | 北京中关村科金技术有限公司 | Image authentication method, device and storage medium |
CN112686247A (en) * | 2020-12-10 | 2021-04-20 | 广州广电运通金融电子股份有限公司 | Identification card number detection method and device, readable storage medium and terminal |
CN113051901A (en) * | 2021-03-26 | 2021-06-29 | 重庆紫光华山智安科技有限公司 | Identification card text recognition method, system, medium and electronic terminal |
CN113784009A (en) * | 2021-09-10 | 2021-12-10 | 北京航星永志科技有限公司 | Paper text image processing method and device and electronic equipment |
US11379534B2 (en) | 2019-11-19 | 2022-07-05 | International Business Machines Corporation | Document feature repository management |
WO2022179471A1 (en) * | 2021-02-25 | 2022-09-01 | 华为技术有限公司 | Card text recognition method and apparatus, and storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106570497A (en) * | 2016-10-08 | 2017-04-19 | 中国科学院深圳先进技术研究院 | Text detection method and device for scene image |
CN107247950A (en) * | 2017-06-06 | 2017-10-13 | 电子科技大学 | A kind of ID Card Image text recognition method based on machine learning |
CN107545239A (en) * | 2017-07-06 | 2018-01-05 | 南京理工大学 | A kind of deck detection method matched based on Car license recognition with vehicle characteristics |
CN107609549A (en) * | 2017-09-20 | 2018-01-19 | 北京工业大学 | The Method for text detection of certificate image under a kind of natural scene |
-
2018
- 2018-05-30 CN CN201810536528.9A patent/CN108694393A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106570497A (en) * | 2016-10-08 | 2017-04-19 | 中国科学院深圳先进技术研究院 | Text detection method and device for scene image |
CN107247950A (en) * | 2017-06-06 | 2017-10-13 | 电子科技大学 | A kind of ID Card Image text recognition method based on machine learning |
CN107545239A (en) * | 2017-07-06 | 2018-01-05 | 南京理工大学 | A kind of deck detection method matched based on Car license recognition with vehicle characteristics |
CN107609549A (en) * | 2017-09-20 | 2018-01-19 | 北京工业大学 | The Method for text detection of certificate image under a kind of natural scene |
Non-Patent Citations (1)
Title |
---|
ZHI TIAN 等: "Detecting Text in Natural Image with Connectionist Text Proposal Network", 《SPRINGER INTERNATIONAL PUBLISHING AG 2016》 * |
Cited By (27)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109376658A (en) * | 2018-10-26 | 2019-02-22 | 信雅达系统工程股份有限公司 | A kind of OCR method based on deep learning |
CN111325194A (en) * | 2018-12-13 | 2020-06-23 | 杭州海康威视数字技术股份有限公司 | Character recognition method, device and equipment and storage medium |
CN111325194B (en) * | 2018-12-13 | 2023-12-29 | 杭州海康威视数字技术股份有限公司 | Character recognition method, device and equipment and storage medium |
CN109800757A (en) * | 2019-01-04 | 2019-05-24 | 西北工业大学 | A kind of video text method for tracing based on layout constraint |
CN109800757B (en) * | 2019-01-04 | 2022-04-19 | 西北工业大学 | Video character tracking method based on layout constraint |
CN109886978A (en) * | 2019-02-20 | 2019-06-14 | 贵州电网有限责任公司 | A kind of end-to-end warning information recognition methods based on deep learning |
CN110033469A (en) * | 2019-04-01 | 2019-07-19 | 北京科技大学 | A kind of sub-pixel edge detection method and system |
CN110222752A (en) * | 2019-05-28 | 2019-09-10 | 北京金山数字娱乐科技有限公司 | Image processing method, system, computer equipment, storage medium and chip |
CN110222752B (en) * | 2019-05-28 | 2021-11-16 | 北京金山数字娱乐科技有限公司 | Image processing method, system, computer device, storage medium and chip |
CN110363199A (en) * | 2019-07-16 | 2019-10-22 | 济南浪潮高新科技投资发展有限公司 | Certificate image text recognition method and system based on deep learning |
CN112396082A (en) * | 2019-08-19 | 2021-02-23 | 北京中关村科金技术有限公司 | Image authentication method, device and storage medium |
CN110554991A (en) * | 2019-09-03 | 2019-12-10 | 浙江传媒学院 | Method for correcting and managing text picture |
CN110866871A (en) * | 2019-11-15 | 2020-03-06 | 深圳市华云中盛科技股份有限公司 | Text image correction method and device, computer equipment and storage medium |
US11379534B2 (en) | 2019-11-19 | 2022-07-05 | International Business Machines Corporation | Document feature repository management |
CN110889404A (en) * | 2019-11-21 | 2020-03-17 | 山东浪潮人工智能研究院有限公司 | Irregular text recognition system and method based on correction network |
CN111368820A (en) * | 2020-03-06 | 2020-07-03 | 腾讯科技(深圳)有限公司 | Text labeling method and device and storage medium |
CN111783761A (en) * | 2020-06-30 | 2020-10-16 | 苏州科达科技股份有限公司 | Certificate text detection method and device and electronic equipment |
CN111797922A (en) * | 2020-07-03 | 2020-10-20 | 泰康保险集团股份有限公司 | Text image classification method and device |
CN111797922B (en) * | 2020-07-03 | 2023-11-28 | 泰康保险集团股份有限公司 | Text image classification method and device |
CN112016438A (en) * | 2020-08-26 | 2020-12-01 | 北京嘀嘀无限科技发展有限公司 | Method and system for identifying certificate based on graph neural network |
CN112200191B (en) * | 2020-12-01 | 2021-07-20 | 北京京东尚科信息技术有限公司 | Image processing method, image processing device, computing equipment and medium |
CN112200191A (en) * | 2020-12-01 | 2021-01-08 | 北京京东尚科信息技术有限公司 | Image processing method, image processing device, computing equipment and medium |
CN112686247A (en) * | 2020-12-10 | 2021-04-20 | 广州广电运通金融电子股份有限公司 | Identification card number detection method and device, readable storage medium and terminal |
WO2022179471A1 (en) * | 2021-02-25 | 2022-09-01 | 华为技术有限公司 | Card text recognition method and apparatus, and storage medium |
CN113051901A (en) * | 2021-03-26 | 2021-06-29 | 重庆紫光华山智安科技有限公司 | Identification card text recognition method, system, medium and electronic terminal |
CN113784009A (en) * | 2021-09-10 | 2021-12-10 | 北京航星永志科技有限公司 | Paper text image processing method and device and electronic equipment |
CN113784009B (en) * | 2021-09-10 | 2022-10-18 | 北京航星永志科技有限公司 | Paper text image processing method and device and electronic equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108694393A (en) | A kind of certificate image text area extraction method based on depth convolution | |
CN104751142B (en) | A kind of natural scene Method for text detection based on stroke feature | |
CN104112128B (en) | Digital image processing system and method applied to bill image character recognition | |
CN110119741B (en) | Card image information identification method with background | |
CN104299008B (en) | Vehicle type classification method based on multi-feature fusion | |
CN103049763B (en) | Context-constraint-based target identification method | |
CN107563377A (en) | It is a kind of to detect localization method using the certificate key area of edge and character area | |
CN104463877B (en) | A kind of water front method for registering based on radar image Yu electronic chart information | |
CN106156684B (en) | A kind of two-dimensional code identification method and device | |
CN106909941A (en) | Multilist character recognition system and method based on machine vision | |
CN109426814A (en) | A kind of positioning of the specific plate of invoice picture, recognition methods, system, equipment | |
CN107563380A (en) | A kind of vehicle license plate detection recognition method being combined based on MSER and SWT | |
CN108734235A (en) | A kind of personal identification method and system for electronic prescription | |
CN110555382A (en) | Finger vein identification method based on deep learning and Wasserstein distance measurement | |
CN107273608A (en) | A kind of reservoir geology profile vectorization method | |
CN106530281A (en) | Edge feature-based unmanned aerial vehicle image blur judgment method and system | |
CN112734729B (en) | Water gauge water level line image detection method and device suitable for night light supplement condition and storage medium | |
CN109886935A (en) | A kind of road face foreign matter detecting method based on deep learning | |
CN110738216A (en) | Medicine identification method based on improved SURF algorithm | |
CN105404868A (en) | Interaction platform based method for rapidly detecting text in complex background | |
CN107195069A (en) | A kind of RMB crown word number automatic identifying method | |
CN109766750A (en) | A kind of table line position finding and detection method of financial statement | |
CN112766184A (en) | Remote sensing target detection method based on multi-level feature selection convolutional neural network | |
CN109766892A (en) | A kind of ray detection image tagged information character dividing method based on edge detection | |
CN106778766B (en) | Positioning point-based rotating number identification method and system |
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: 20181023 |
|
RJ01 | Rejection of invention patent application after publication |