CN109670495A - A kind of method and system of the length text detection based on deep neural network - Google Patents
A kind of method and system of the length text detection based on deep neural network Download PDFInfo
- Publication number
- CN109670495A CN109670495A CN201811528135.XA CN201811528135A CN109670495A CN 109670495 A CN109670495 A CN 109670495A CN 201811528135 A CN201811528135 A CN 201811528135A CN 109670495 A CN109670495 A CN 109670495A
- Authority
- CN
- China
- Prior art keywords
- text
- frame
- class
- region
- length
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or 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
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The method and system for the length text detection based on deep neural network that the invention discloses a kind of, packet are included: extracting characteristic pattern to original image；The first kind rectangular area of several default length-width ratios is enumerated from characteristic pattern based on Faster RCNN, the prospect or background, the real estate for predicting first kind rectangular area of first kind rectangular area is predicted, obtains all first kind text boxes；Based on CTPN from several predetermined widths of feature enumeration of graph, the second class rectangle region of indefinite length, if the second class rectangle region be it is text filed, access RNN recirculating network and obtain the second class text frame；Merge first kind text box and the second class text frame based on non-maxima suppression.The advantages of present invention incorporates Faster RCNN and CTPN, and merged the detection content of the two based on non-maxima suppression logic by certain rule, so that the recall rate and accuracy rate of text detection are all improved.
Description
Technical field
The present invention relates to text detection techniques fields, and in particular to a kind of length text detection based on deep neural network
Method and system.
Background technique
Currently, there is the demand for the Pictures Electronics with text for largely providing user in the market, such demand needs
Text detection in picture is come out, and is identified as digitized text.The work of this part needs a large amount of manpower in the past, and present
It is then this technology that electronic data is converted picture into using OCR technique.Current OCR technique is divided into detection and identification two
Module, this technology are mainly used to improve the performance of text detection module, the accuracy rate that the boundary for promoting text is returned.
Text detection techniques, what is be based primarily upon is to develop preferable depth convolutional neural networks at present to be detected, packet
Include Faster RCNN frame and CTPN detection framework；Wherein:
The basic version of Faster RCNN frame has biggish object relatively high accuracy rate.The process of frame
Are as follows: I. extracts feature to picture, and II. enumerates a large amount of rectangle to attempt to return out corresponding object, the square that III. will be enumerated
Shape is divided into 2 classes: comprising target and the biggish positive sample of intersection and other negative samples, IV. cuts out positive sample from characteristic pattern
Come, the boundary of regressive object is then gone to according to characteristic pattern；
Occur for the CTPN testing stand of text class inspection optimization for text is always horizontal, and length is unfixed existing
As use following process: I. extracts feature to picture, and II. enumerates a large amount of small rectangle, is different from Faster RCNN, American series
Small rectangle can fixed width, and height will use different scales, such as: (11 pixels to 273 pixels, every time multiplied by fixation
Ratio, totally 10 different scales), III is attached using the small scale text that RNN recirculating network will test, obtains text
Row.IV. using CNN+RNN, training method, support are multiple dimensioned end to end.
Above-mentioned two detection model there are the shortcomings that are as follows:
Faster RCNN detection model is in context of detection, because text has that gap different in size is larger, it is difficult to
It determines that the height for enumerating anchor is wide, it is poor to there is a problem of that boundary returns to the great text of length-width ratio gap.
CTPN detection model returns preferably boundary in context of detection, but for there are in the case where text overlapping,
There can be the case where text box loss, it is difficult to accomplish to give all texts for change.
Summary of the invention
Aiming at the shortcomings existing in the above problems, the present invention provides a kind of long short text based on deep neural network
The method and system of detection.
The method for the length text detection based on deep neural network that the invention discloses a kind of, comprising:
Original image is selected, characteristic pattern is extracted to original image；
The first kind rectangular area of several default length-width ratios, prediction first are enumerated from characteristic pattern based on Faster RCNN
The prospect or background in class rectangle region, the real estate for predicting first kind rectangular area, obtain all first kind text boxes；
Based on CTPN from several predetermined widths of feature enumeration of graph, the second class rectangle region of indefinite length, the second class is judged
Rectangular area is text or non-textual region；If the second class rectangle region be it is text filed, access RNN recirculating network and obtain the
Two class text frames；
Merge the first kind text box and the second class text frame based on non-maxima suppression.
As a further improvement of the present invention, described that several default length and width are enumerated from characteristic pattern based on Faster RCNN
The first kind rectangular area of ratio, comprising:
Original image abstract characteristics are extracted based on depth convolutional network；
The candidate region of using area candidate network recommendation original image；
The accurate region of text is returned from candidate region.
As a further improvement of the present invention, the CTPN that is based on is from several predetermined widths of feature enumeration of graph, indefinite length
The second class rectangle region, comprising:
Characteristic image is generated using network；
Candidate the second class rectangle region enumerated is generated using fixed width and indefinite length；
The non-textual prediction of text is carried out to the second class rectangle region.
As a further improvement of the present invention, described that the first kind text box and second are merged based on non-maxima suppression
Class text frame, comprising:
Framed score is sorted, best result and its corresponding frame is chosen, is set as valid frame；
Remaining frame is traversed, if the overlapping area with current best result frame is greater than certain threshold value, deletes valid frame；
The frame for continuing to select a highest scoring from untreated frame, repeats the above process.
The system for the length text detection based on deep neural network that the present invention also provides a kind of, comprising:
Extraction module extracts characteristic pattern to original image for selecting original image；
Faster RCNN processing module, for enumerating several default length-width ratios from characteristic pattern based on Faster RCNN
First kind rectangular area is predicted the prospect or background, the real estate for predicting first kind rectangular area of first kind rectangular area, is obtained
To all first kind text boxes；
CTPN processing module, for based on CTPN from several predetermined widths of feature enumeration of graph, the second class square of indefinite length
Shape region judges the second class rectangle region for text or non-textual region；If the second class rectangle region be it is text filed, access
RNN recirculating network obtains the second class text frame；
Merging module, for merging the first kind text box and the second class text frame based on non-maxima suppression.
As a further improvement of the present invention, described that several default length and width are enumerated from characteristic pattern based on Faster RCNN
The first kind rectangular area of ratio, comprising:
Original image abstract characteristics are extracted based on depth convolutional network；
The candidate region of using area candidate network recommendation original image；
The accurate region of text is returned from candidate region.
As a further improvement of the present invention, the CTPN that is based on is from several predetermined widths of feature enumeration of graph, indefinite length
The second class rectangle region, comprising:
Characteristic image is generated using network；
Candidate the second class rectangle region enumerated is generated using fixed width and indefinite length；
The non-textual prediction of text is carried out to the second class rectangle region.
As a further improvement of the present invention, described that the first kind text box and second are merged based on non-maxima suppression
Class text frame, comprising:
Framed score is sorted, best result and its corresponding frame is chosen, is set as valid frame；
Remaining frame is traversed, if the overlapping area with current best result frame is greater than certain threshold value, deletes valid frame；
The frame for continuing to select a highest scoring from untreated frame, repeats the above process.
Compared with prior art, the invention has the benefit that
The Faster RCNN and CTPN that the present invention uses are all based on deep learning algorithm, are the spies in original image
The text detection done on sign figure can use under various complex environments；
The advantages of present invention incorporates Faster RCNN: still there is higher recall rate for overlay target.And CTPN
Advantage: still having preferable boundary accurate rate for longer text, has better compatibility simultaneously for the text of different scale
Property.The detection content of the two is merged based on non-maxima suppression logic by certain rule, so that text detection is recalled
Rate and accuracy rate are all improved.
Detailed description of the invention
Fig. 1 is the process of the method for the length text detection based on deep neural network disclosed in an embodiment of the present invention
Figure；
Fig. 2 is the frame of the system of the length text detection based on deep neural network disclosed in an embodiment of the present invention
Figure；
Fig. 3 is the schematic diagram of first kind text box disclosed in an embodiment of the present invention；
Fig. 4 is that RNN recirculating network disclosed in an embodiment of the present invention obtains the schematic diagram of the second class text frame.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiments of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill people
Member's every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
The present invention is described in further detail with reference to the accompanying drawing:
As shown in Figure 1, the method for the invention discloses a kind of length text detection based on deep neural network, comprising:
Step 1, selection original image, carry out feature extraction to original image, obtain characteristic pattern；Wherein original image can be
Bill picture；
Step 2, the first kind rectangular area for enumerating several default length-width ratios from characteristic pattern based on Faster RCNN, in advance
The prospect or background, the real estate for predicting first kind rectangular area for surveying first kind rectangular area, obtain all first kind texts
This frame, first kind text box are as shown in Figure 3；Wherein:
The principle of Faster RCNN are as follows:
Original image abstract characteristics are extracted based on depth convolutional network；
The candidate region of using area candidate network recommendation original image；
The accurate region of text is returned from candidate region；
Step 3, based on CTPN from several predetermined widths of feature enumeration of graph, the second class rectangle region of indefinite length, judgement
Second class rectangle region is text or non-textual region；If the second class rectangle region be it is text filed, access RNN recirculating network
The second class text frame is obtained, the second class text frame is as shown in Figure 4；Wherein:
The principle of CTPN are as follows:
Characteristic image is generated using network；
Candidate the second class rectangle region enumerated is generated using fixed width and indefinite length；
The non-textual prediction of text is carried out to the second class rectangle region.
Step 4 merges first kind text box and the second class text frame based on non-maxima suppression (NMS)；Wherein:
Merge first kind text box and the second class text frame based on non-maxima suppression, comprising:
Framed score is sorted, best result and its corresponding frame is chosen, is set as valid frame；
Remaining frame is traversed, if the overlapping area with current best result frame is greater than certain threshold value, deletes valid frame；
The frame for continuing to select a highest scoring from untreated frame, repeats the above process.
Such as:
Assuming that object detects that 4 Boxes, each Box respectively correspond a classification Score, according to Score from it is small to
Longer spread is followed successively by, (B1, S1), (B2, S2), (B3, S3), (B4, S4) .S4 > S3 > S2 > S1；
Step 1. is according to Score size, since Box B4 frame；
Step 2. calculates separately the overlapping degree IoU of B1, B2, B3 and B4, judges whether to be greater than presetting threshold value；Such as
Fruit is greater than given threshold, then gives up the Box；The Box. of reservation is marked to assume that the threshold value of B3 and B4 is more than given threshold simultaneously, then
Give up B3, label B4 is the Box to be retained；
The maximum B2 of Score is chosen in Step 3. B1 from remaining Boxes, B2, then calculates B2's and remaining B1
Overlapping degree IoU；If it is greater than given threshold, the Box is equally abandoned；The Box of reservation is marked simultaneously；
Above procedure is repeated, until finding whole reservation Boxes.
As shown in Fig. 2, the present invention provides a kind of system of length text detection based on deep neural network, comprising:
Extraction module carries out feature extraction to original image, obtains characteristic pattern for selecting original image；It is wherein original
Picture can be bill picture；
Faster RCNN processing module, for enumerating several default length-width ratios from characteristic pattern based on Faster RCNN
First kind rectangular area is predicted the prospect or background, the real estate for predicting first kind rectangular area of first kind rectangular area, is obtained
To all first kind text boxes, first kind text box is as shown in Figure 3；Wherein:
The principle of Faster RCNN are as follows:
Original image abstract characteristics are extracted based on depth convolutional network；
The candidate region of using area candidate network recommendation original image；
The accurate region of text is returned from candidate region；
CTPN processing module, for based on CTPN from several predetermined widths of feature enumeration of graph, the second class square of indefinite length
Shape region judges the second class rectangle region for text or non-textual region；If the second class rectangle region be it is text filed, access
RNN recirculating network obtains the second class text frame, and the second class text frame is as shown in Figure 4；Wherein:
The principle of CTPN are as follows:
Characteristic image is generated using network；
Candidate the second class rectangle region enumerated is generated using fixed width and indefinite length；
The non-textual prediction of text is carried out to the second class rectangle region；
Merging module, for merging first kind text box and the second class text frame based on non-maxima suppression (NMS)；Its
In:
Merge first kind text box and the second class text frame based on non-maxima suppression, comprising:
Framed score is sorted, best result and its corresponding frame is chosen, is set as valid frame；
Remaining frame is traversed, if the overlapping area with current best result frame is greater than certain threshold value, deletes valid frame；
The frame for continuing to select a highest scoring from untreated frame, repeats the above process.
Such as:
Assuming that object detects that 4 Boxes, each Box respectively correspond a classification Score, according to Score from it is small to
Longer spread is followed successively by, (B1, S1), (B2, S2), (B3, S3), (B4, S4) .S4 > S3 > S2 > S1；
Step 1. is according to Score size, since Box B4 frame；
Step 2. calculates separately the overlapping degree IoU of B1, B2, B3 and B4, judges whether to be greater than presetting threshold value；Such as
Fruit is greater than given threshold, then gives up the Box；The Box. of reservation is marked to assume that the threshold value of B3 and B4 is more than given threshold simultaneously, then
Give up B3, label B4 is the Box to be retained；
The maximum B2 of Score is chosen in Step 3. B1 from remaining Boxes, B2, then calculates B2's and remaining B1
Overlapping degree IoU；If it is greater than given threshold, the Box is equally abandoned；The Box of reservation is marked simultaneously；
Above procedure is repeated, until finding whole reservation Boxes.
Further, the merging of text filed frame can be NMS (non-maxima suppression) method there are many mode,
It can be region and merging etc. be completely covered.
The invention has the benefit that
The Faster RCNN and CTPN that the present invention uses are all based on deep learning algorithm, are the spies in original image
The text detection done on sign figure can use under various complex environments；The advantages of present invention incorporates Faster RCNN: right
Still there is higher recall rate in overlay target.And the advantages of CTPN: still there is preferable boundary accurate for longer text
Rate has better compatibility simultaneously for the text of different scale.By certain rule by the detection content of the two based on non-
Maximum inhibits logic to merge, so that the recall rate and accuracy rate of text detection are all improved.
These are only the preferred embodiment of the present invention, is not intended to restrict the invention, for those skilled in the art
For member, the invention may be variously modified and varied.All within the spirits and principles of the present invention, it is made it is any modification,
Equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (8)
1. a kind of method of the length text detection based on deep neural network characterized by comprising
Original image is selected, characteristic pattern is extracted to original image；
The first kind rectangular area of several default length-width ratios is enumerated from characteristic pattern based on Faster RCNN, predicts first kind square
The prospect or background in shape region, the real estate for predicting first kind rectangular area, obtain all first kind text boxes；
Based on CTPN from several predetermined widths of feature enumeration of graph, the second class rectangle region of indefinite length, the second class rectangle is judged
Region is text or non-textual region；If the second class rectangle region be it is text filed, access RNN recirculating network and obtain the second class
Text box；
Merge the first kind text box and the second class text frame based on non-maxima suppression.
2. the method for the length text detection based on deep neural network as described in claim 1, which is characterized in that the base
The first kind rectangular area of several default length-width ratios is enumerated from characteristic pattern in Faster RCNN, comprising:
Original image abstract characteristics are extracted based on depth convolutional network；
The candidate region of using area candidate network recommendation original image；
The accurate region of text is returned from candidate region.
3. the method for the length text detection based on deep neural network as described in claim 1, which is characterized in that the base
In CTPN from several predetermined widths of feature enumeration of graph, the second class rectangle region of indefinite length, comprising:
Characteristic image is generated using network；
Candidate the second class rectangle region enumerated is generated using fixed width and indefinite length；
The non-textual prediction of text is carried out to the second class rectangle region.
4. the method for the length text detection based on deep neural network as described in claim 1, which is characterized in that the base
Merge the first kind text box and the second class text frame in non-maxima suppression, comprising:
Framed score is sorted, best result and its corresponding frame is chosen, is set as valid frame；
Remaining frame is traversed, if the overlapping area with current best result frame is greater than certain threshold value, deletes valid frame；
The frame for continuing to select a highest scoring from untreated frame, repeats the above process.
5. a kind of system of the length text detection based on deep neural network characterized by comprising
Extraction module extracts characteristic pattern to original image for selecting original image；
Faster RCNN processing module, for enumerating the first of several default length-width ratios from characteristic pattern based on Faster RCNN
Class rectangle region predicts the prospect or background, the real estate for predicting first kind rectangular area of first kind rectangular area, obtains institute
Some first kind text boxes；
CTPN processing module, for based on CTPN from several predetermined widths of feature enumeration of graph, the second class rectangle area of indefinite length
Domain judges the second class rectangle region for text or non-textual region；If the second class rectangle region be it is text filed, access RNN
Recirculating network obtains the second class text frame；
Merging module, for merging the first kind text box and the second class text frame based on non-maxima suppression.
6. the system of the length text detection based on deep neural network as claimed in claim 5, which is characterized in that the base
The first kind rectangular area of several default length-width ratios is enumerated from characteristic pattern in Faster RCNN, comprising:
Original image abstract characteristics are extracted based on depth convolutional network；
The candidate region of using area candidate network recommendation original image；
The accurate region of text is returned from candidate region.
7. the system of the length text detection based on deep neural network as claimed in claim 5, which is characterized in that the base
In CTPN from several predetermined widths of feature enumeration of graph, the second class rectangle region of indefinite length, comprising:
Characteristic image is generated using network；
Candidate the second class rectangle region enumerated is generated using fixed width and indefinite length；
The non-textual prediction of text is carried out to the second class rectangle region.
8. the system of the length text detection based on deep neural network as claimed in claim 5, which is characterized in that the base
Merge the first kind text box and the second class text frame in non-maxima suppression, comprising:
Framed score is sorted, best result and its corresponding frame is chosen, is set as valid frame；
Remaining frame is traversed, if the overlapping area with current best result frame is greater than certain threshold value, deletes valid frame；
The frame for continuing to select a highest scoring from untreated frame, repeats the above process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811528135.XA CN109670495A (en) | 2018-12-13 | 2018-12-13 | A kind of method and system of the length text detection based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811528135.XA CN109670495A (en) | 2018-12-13 | 2018-12-13 | A kind of method and system of the length text detection based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109670495A true CN109670495A (en) | 2019-04-23 |
Family
ID=66143857
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811528135.XA Pending CN109670495A (en) | 2018-12-13 | 2018-12-13 | A kind of method and system of the length text detection based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109670495A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110738238A (en) * | 2019-09-18 | 2020-01-31 | 平安科技（深圳）有限公司 | certificate information classification positioning method and device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107346420A (en) * | 2017-06-19 | 2017-11-14 | 中国科学院信息工程研究所 | Text detection localization method under a kind of natural scene based on deep learning |
| CN108108731A (en) * | 2016-11-25 | 2018-06-01 | 中移（杭州）信息技术有限公司 | Method for text detection and device based on generated data |
| CN108520229A (en) * | 2018-04-04 | 2018-09-11 | 北京旷视科技有限公司 | Image detecting method, device, electronic equipment and computer-readable medium |
| CN108921166A (en) * | 2018-06-22 | 2018-11-30 | 深源恒际科技有限公司 | Medical bill class text detection recognition method and system based on deep neural network |
| CN108960229A (en) * | 2018-04-23 | 2018-12-07 | 中国科学院信息工程研究所 | One kind is towards multidirectional character detecting method and device |
-
2018
- 2018-12-13 CN CN201811528135.XA patent/CN109670495A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108108731A (en) * | 2016-11-25 | 2018-06-01 | 中移（杭州）信息技术有限公司 | Method for text detection and device based on generated data |
| CN107346420A (en) * | 2017-06-19 | 2017-11-14 | 中国科学院信息工程研究所 | Text detection localization method under a kind of natural scene based on deep learning |
| CN108520229A (en) * | 2018-04-04 | 2018-09-11 | 北京旷视科技有限公司 | Image detecting method, device, electronic equipment and computer-readable medium |
| CN108960229A (en) * | 2018-04-23 | 2018-12-07 | 中国科学院信息工程研究所 | One kind is towards multidirectional character detecting method and device |
| CN108921166A (en) * | 2018-06-22 | 2018-11-30 | 深源恒际科技有限公司 | Medical bill class text detection recognition method and system based on deep neural network |
Non-Patent Citations (8)
| Title |
|---|
| AIHGF: "目标检测-非极大值抑制(Non Maximum Suppression)", 《HTTPS://BLOG.CSDN.NET/ZZIAHGF/ARTICLE/DETAILS/78525279》 * |
| AIHGF: "目标检测-非极大值抑制(Non Maximum Suppression)", 《HTTPS://BLOG.CSDN.NET/ZZIAHGF/ARTICLE/DETAILS/78525279》, 13 November 2017 (2017-11-13), pages 1 * |
| JIANG YINGYING 等: "R2CNN: Rotational Region CNN for Orientation Robust Scene Text Detection", 《ARXIV》 * |
| JIANG YINGYING 等: "R2CNN: Rotational Region CNN for Orientation Robust Scene Text Detection", 《ARXIV》, 30 June 2017 (2017-06-30), pages 1 - 8 * |
| TIAN ZHI 等: "Detecting Text in Natural Image with Connectionist Text Proposal Network", 《ECCV 2016》, 12 November 2016 (2016-11-12), pages 1 - 16 * |
| TIAN ZHI 等: "Detecting Text in Natural Image with Connectionist Text Proposal Network", 《ECCV 2016》, pages 1 - 16 * |
| 郑阳: "自然场景图像中文本检测技术研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
| 郑阳: "自然场景图像中文本检测技术研究", 《中国博士学位论文全文数据库 信息科技辑》, no. 02, 15 February 2018 (2018-02-15) * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110738238A (en) * | 2019-09-18 | 2020-01-31 | 平安科技（深圳）有限公司 | certificate information classification positioning method and device |
| WO2021051553A1 (en) * | 2019-09-18 | 2021-03-25 | 平安科技（深圳）有限公司 | Certificate information classification and positioning method and apparatus |
| CN110738238B (en) * | 2019-09-18 | 2023-05-26 | 平安科技（深圳）有限公司 | Classification positioning method and device for certificate information |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107093172B (en) | Character detection method and system | |
| CN109117836B (en) | Method and device for detecting and positioning characters in natural scene based on focus loss function | |
| CN105608456B (en) | A kind of multi-direction Method for text detection based on full convolutional network | |
| CN108805116B (en) | Image text detection method and system | |
| US8315465B1 (en) | Effective feature classification in images | |
| JPH0737103A (en) | Inclination angle detection device | |
| CN101833664A (en) | Video image character detecting method based on sparse expression | |
| CN110414538A (en) | Defect classification method, defect classification based training method and device thereof | |
| KR20070045445A (en) | Method for discriminating a obscene video using visual features and apparatus thereof | |
| CN110503054A (en) | The processing method and processing device of text image | |
| KR101177626B1 (en) | Object checking apparatus and method | |
| CN103093185B (en) | Character recognition device, image processing apparatus and method thereof | |
| CN109741551A (en) | A kind of commodity identification settlement method, apparatus and system | |
| CN109035256A (en) | User interface image cutting method, device, server and storage medium | |
| CN106778717A (en) | A kind of test and appraisal table recognition methods based on image recognition and k nearest neighbor | |
| CN111382766A (en) | Equipment fault detection method based on fast R-CNN | |
| CN114821620A (en) | Text content extraction and identification method based on longitudinal combination of line text boxes | |
| CN108647703A (en) | A kind of type judgement method of the classification image library based on conspicuousness | |
| Hu et al. | Generalized image recognition algorithm for sign inventory | |
| JP3471578B2 (en) | Line direction determining device, image tilt detecting device, and image tilt correcting device | |
| Giri | Text information extraction and analysis from images using digital image processing techniques | |
| Fleuret et al. | Fast face detection with precise pose estimation | |
| CN114359917A (en) | Handwritten Chinese character detection and recognition and font evaluation method | |
| CN109670495A (en) | A kind of method and system of the length text detection based on deep neural network | |
| Ovodov | Optical braille recognition using object detection neural network |
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 |