CN109447015A - A kind of method and device handling form Image center selection word - Google Patents
A kind of method and device handling form Image center selection word Download PDFInfo
- Publication number
- CN109447015A CN109447015A CN201811317237.7A CN201811317237A CN109447015A CN 109447015 A CN109447015 A CN 109447015A CN 201811317237 A CN201811317237 A CN 201811317237A CN 109447015 A CN109447015 A CN 109447015A
- Authority
- CN
- China
- Prior art keywords
- brief note
- form image
- position coordinates
- lamella
- word
- 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
- 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
- G06V30/148—Segmentation of character regions
- G06V30/153—Segmentation of character regions using recognition of characters or words
-
- 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)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention provides a kind of method for handling form Image center selection word, comprising: the disturbance ecology item in removal form Image；Using joint training model, brief note is selected in form Image centre circle, obtain brief note position coordinates of the brief note in form Image and identifies word content corresponding with brief note；Table reduction is carried out to brief note, brief note position coordinates and word content using table characteristic.Implement the device of the above method, comprising: for removing the preprocessing module of disturbance ecology item in form Image；Using joint training model, brief note is selected in form Image centre circle, obtain brief note position coordinates of the brief note in form Image and identifies the identification module of word content corresponding with brief note；The table recovery module of table reduction is carried out to brief note, brief note position coordinates and word content using table characteristic.The present invention can promote Text region and reduction accuracy rate in form Image.
Description
Technical field
The present invention relates to a kind of form processing method, especially a kind of method and dress for handling form Image center selection word
It sets.
Background technique
It is higher for the accuracy rate of the big section Text region of similar A4 paper in OCR identification field.But it is directed to table
Identification, industry accuracy rate is not very high at present.It cuts word knowledge because original and will cause be difficult to carry out layout reversion otherwise,
And the information in table can not be utilized.
Summary of the invention
Aiming at the shortcomings existing in the above problems, the present invention, which provides one kind, can promote Text region in form Image
With a kind of method and device of processing form Image center selection word of reduction accuracy rate.
To achieve the above object, the present invention provides a kind of method for handling form Image center selection word, including following step
It is rapid:
Step 1, to remove the disturbance ecology item in form Image；
Step 2, using joint training model, select brief note in form Image centre circle, obtain brief note in form Image
Brief note position coordinates simultaneously identify word content corresponding with brief note；
Step 3 carries out table reduction to brief note, brief note position coordinates and word content using table characteristic.
The method of above-mentioned a kind of processing form Image center selection word, wherein in step 1, form Image is carried out
Pretreatment includes picture angle correction behaviour to the pretreatment that form Image carries out to remove the disturbance ecology item in form Image
Make or the removal of watermark seal operates.
A kind of method of above-mentioned processing form Image center selection word, wherein in step 2, including following sub-step:
Step 21 carries out RGB three-channel processing to the form Image of removal disturbance ecology item, to form at least two tables
Picture layer；
Step 22 carries out feature extraction to each tabular drawing lamella by convolution transform；
Step 23, in the first tabular drawing lamella, predict brief note position coordinates of the brief note in the first tabular drawing lamella；
Step 24, in the second tabular drawing lamella, obtained by image information and applicational language model corresponding with brief note
Word content.
A kind of method of above-mentioned processing form Image center selection word, wherein in step 23, brief note position coordinates packet
Include top-left coordinates (x0, y0), upper right coordinate (x1, y1), lower right coordinate (x2, y2), lower-left coordinate (x3, y3).
The method of above-mentioned a kind of processing form Image center selection word, wherein in step 3, sat according to brief note position
Mark carries out the cutting of table row and grid column, and word content is imported in brief note position, carries out cell according to Semantic judgement
In conjunction with to complete the reduction of whole table.
The method of above-mentioned a kind of processing form Image center selection word, wherein after further including step 4, reduction being presented
Table.
The present invention also provides a kind of devices for handling form Image center selection word, comprising: preprocessing module, identification module
With table recovery module；
Preprocessing module, for removing the disturbance ecology item in form Image；
Identification module selects brief note in form Image centre circle, obtains brief note in form Image using joint training model
Brief note position coordinates and identify word content corresponding with brief note；
Table recovery module carries out table reduction to brief note, brief note position coordinates and word content using table characteristic.
Above-mentioned device, wherein the pretreatment that the preprocessing module carries out form Image includes picture angle correction
Operation or the removal operation of watermark seal.
Above-mentioned device, wherein the implementation steps of the identification module are as follows:
RGB three-channel processing is carried out to form Image, to form at least two tabular drawing lamellas；
Feature extraction is carried out to each tabular drawing lamella by convolution transform；
In the first tabular drawing lamella, brief note position coordinates of the brief note in the first tabular drawing lamella are predicted；
In the second tabular drawing lamella, obtained in text corresponding with brief note by image information and applicational language model
Hold.
Above-mentioned device, wherein the table recovery module carries out table row and grid column according to brief note position coordinates
Cutting imports word content in brief note position, and the combination of cell is carried out according to Semantic judgement, to complete going back for whole table
It is former.
Compared with prior art, the invention has the following advantages that
It by the textbox choosing based on table and identifies progress joint training deep learning model, frame is made to select and identify two
Task can keep final table Text region more accurate, and do not lose the space of a whole page of table itself with the image information of public table
Information promotes the accuracy rate of table layout reversion.
Detailed description of the invention
Fig. 1 is the flow chart of method part in the present invention；
Fig. 2 is the structural block diagram of device part in the present invention.
Main appended drawing reference is described as follows:
1- preprocessing module；2- identification module；3- table recovery module；Module is presented in 4-
Specific embodiment
As shown in Figure 1, the present invention provides a kind of method for handling form Image center selection word, comprising the following steps:
Disturbance ecology item in step 1, removal form Image.
In step 1, form Image is pre-processed, to remove the disturbance ecology item in form Image, to tabular drawing
The pretreatment that piece carries out includes picture angle correction operation or the removal operation of watermark seal.
Step 2, using joint training model, select brief note in form Image centre circle, obtain brief note in form Image
Brief note position coordinates simultaneously identify word content corresponding with brief note.
In step 2, including following sub-step:
Step 21 carries out RGB three-channel processing to the form Image of removal disturbance ecology item, to form at least two tables
Picture layer；
Step 22 carries out feature extraction to each tabular drawing lamella by convolution transform；
Step 23, in the first tabular drawing lamella, predict brief note position coordinates of the brief note in the first tabular drawing lamella；
Wherein, brief note position coordinates include top-left coordinates (x0, y0), upper right coordinate (x1, y1), lower right coordinate (x2, y2),
Lower-left coordinate (x3, y3).
Step 24, in the second tabular drawing lamella, obtained by image information and applicational language model corresponding with brief note
Word content.
Step 3 carries out table reduction to brief note, brief note position coordinates and word content using table characteristic.
In step 3, the cutting that table row and grid column are carried out according to brief note position coordinates, imports brief note for word content
In position, the combination of cell is carried out according to Semantic judgement, to complete the reduction of whole table.
The table after reduction is presented in step 4.
The training process of joint training model is as follows:
1. generating table, and enclose the corresponding informance of brief note, text for different fonts, different form types；
2. a pair generated table adds noise, guarantee the robustness of model；
3. sample is sent to training in joint training model；
4. the model after being trained is identified for OCR.
As shown in Fig. 2, the present invention provides a kind of device for handling form Image center selection word, comprising: preprocessing module
1, identification module 2 and table recovery module 3.
Preprocessing module 1, for removing the disturbance ecology item in form Image.
Preprocessing module pre-processes form Image, to remove the disturbance ecology item in form Image, to tabular drawing
The pretreatment that piece carries out includes picture angle correction operation or the removal operation of watermark seal.
Identification module 2 selects brief note in form Image centre circle, obtains brief note in form Image using joint training model
In brief note position coordinates and identify word content corresponding with brief note.
The implementation steps of identification module are as follows:
RGB three-channel processing is carried out to the form Image of removal disturbance ecology item, to form at least two tabular drawing lamellas；
Feature extraction is carried out to each tabular drawing lamella by convolution transform；
In the first tabular drawing lamella, brief note position coordinates of the brief note in the first tabular drawing lamella are predicted；
In the second tabular drawing lamella, obtained in text corresponding with brief note by image information and applicational language model
Hold.
Wherein, brief note position coordinates include top-left coordinates (x0, y0), upper right coordinate (x1, y1), lower right coordinate (x2, y2),
Lower-left coordinate (x3, y3).
Table recovery module 3 carries out table reduction to brief note, brief note position coordinates and word content using table characteristic.
Wherein, table recovery module carries out the cutting of table row and grid column according to brief note position coordinates, by word content
It imports in brief note position, the combination of cell is carried out according to Semantic judgement, to complete the reduction of whole table.
It further include that module 4 is presented, the table after restoring for rendering.
Whole CTPN model of the joint training model based on deep learning, while creative on CTPN model connecing
Enter CTC and identify the feature around brief note, so that identification process is can use table characteristic, accuracy is substantially improved.
The foregoing is merely presently preferred embodiments of the present invention, is merely illustrative and not restrictive for the invention.
Those skilled in the art understand that many changes can be carried out in the spirit and scope defined by invention claim to it, modify,
It is even equivalent, but fall in protection scope of the present invention.
Claims (10)
1. a kind of method for handling form Image center selection word, comprising the following steps:
Step 1, to remove the disturbance ecology item in form Image；
Step 2, using joint training model, select brief note in form Image centre circle, obtain brief note of the brief note in form Image
Position coordinates simultaneously identify word content corresponding with brief note；
Step 3 carries out table reduction to brief note, brief note position coordinates and word content using table characteristic.
2. a kind of method for handling form Image center selection word according to claim 1, which is characterized in that in step 1
In, form Image is pre-processed, to remove the disturbance ecology item in form Image, to the pretreatment packet of form Image progress
Include picture angle correction operation or the removal operation of watermark seal.
3. a kind of method for handling form Image center selection word according to claim 1, which is characterized in that in step 2
In, including following sub-step:
Step 21 carries out RGB three-channel processing to the form Image of removal disturbance ecology item, to form at least two form Images
Layer；
Step 22 carries out feature extraction to each tabular drawing lamella by convolution transform；
Step 23, in the first tabular drawing lamella, predict brief note position coordinates of the brief note in the first tabular drawing lamella；
Step 24, in the second tabular drawing lamella, text corresponding with brief note is obtained by image information and applicational language model
Word content.
4. a kind of method for handling form Image center selection word according to claim 3, which is characterized in that in step 23
In, brief note position coordinates include top-left coordinates (x0, y0), upper right coordinate (x1, y1), lower right coordinate (x2, y2), lower-left coordinate
(x3, y3).
5. a kind of method for handling form Image center selection word according to claim 1, which is characterized in that in step 3
In, the cutting of table row and grid column is carried out according to brief note position coordinates, word content is imported in brief note position, according to semanteme
Judgement carries out the combination of cell, to complete the reduction of whole table.
6. a kind of method for handling form Image center selection word according to claim 1, which is characterized in that further include step
Rapid 4, the table after reduction is presented.
7. a kind of device of the method for processing form Image center selection word described in a kind of implementation claim 1, feature
It is, comprising: preprocessing module, identification module and table recovery module；
Preprocessing module, for removing the disturbance ecology item in form Image；
Identification module selects brief note in form Image centre circle, obtains word of the brief note in form Image using joint training model
Position coordinates simultaneously identify word content corresponding with brief note；
Table recovery module carries out table reduction to brief note, brief note position coordinates and word content using table characteristic.
8. device according to claim 7, which is characterized in that the pretreatment that the preprocessing module carries out form Image
Including picture angle correction operation or the removal operation of watermark seal.
9. device according to claim 7, which is characterized in that the implementation steps of the identification module are as follows:
RGB three-channel processing is carried out to form Image, to form at least two tabular drawing lamellas；
Feature extraction is carried out to each tabular drawing lamella by convolution transform；
In the first tabular drawing lamella, brief note position coordinates of the brief note in the first tabular drawing lamella are predicted；
In the second tabular drawing lamella, word content corresponding with brief note is obtained by image information and applicational language model.
10. device according to claim 7, which is characterized in that the table recovery module according to brief note position coordinates into
The cutting of row table row and grid column imports word content in brief note position, and the combination of cell is carried out according to Semantic judgement,
To complete the reduction of whole table.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811317237.7A CN109447015A (en) | 2018-11-03 | 2018-11-03 | A kind of method and device handling form Image center selection word |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811317237.7A CN109447015A (en) | 2018-11-03 | 2018-11-03 | A kind of method and device handling form Image center selection word |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109447015A true CN109447015A (en) | 2019-03-08 |
Family
ID=65551246
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811317237.7A Pending CN109447015A (en) | 2018-11-03 | 2018-11-03 | A kind of method and device handling form Image center selection word |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109447015A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109948507A (en) * | 2019-03-14 | 2019-06-28 | 北京百度网讯科技有限公司 | Method and apparatus for detecting table |
| CN111340023A (en) * | 2020-02-24 | 2020-06-26 | 创新奇智(上海)科技有限公司 | Text recognition method and device, electronic equipment and storage medium |
| CN112861736A (en) * | 2021-02-10 | 2021-05-28 | 上海大学 | Document table content identification and information extraction method based on image processing |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH07220023A (en) * | 1994-01-31 | 1995-08-18 | Hitachi Ltd | Method and device for table recognition |
| CN105574486A (en) * | 2015-11-25 | 2016-05-11 | 成都数联铭品科技有限公司 | Image table character segmenting method |
| CN107704857A (en) * | 2017-09-25 | 2018-02-16 | 北京邮电大学 | A kind of lightweight licence plate recognition method and device end to end |
| CN108416279A (en) * | 2018-02-26 | 2018-08-17 | 阿博茨德（北京）科技有限公司 | Form analysis method and device in file and picture |
-
2018
- 2018-11-03 CN CN201811317237.7A patent/CN109447015A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH07220023A (en) * | 1994-01-31 | 1995-08-18 | Hitachi Ltd | Method and device for table recognition |
| CN105574486A (en) * | 2015-11-25 | 2016-05-11 | 成都数联铭品科技有限公司 | Image table character segmenting method |
| CN107704857A (en) * | 2017-09-25 | 2018-02-16 | 北京邮电大学 | A kind of lightweight licence plate recognition method and device end to end |
| CN108416279A (en) * | 2018-02-26 | 2018-08-17 | 阿博茨德（北京）科技有限公司 | Form analysis method and device in file and picture |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109948507A (en) * | 2019-03-14 | 2019-06-28 | 北京百度网讯科技有限公司 | Method and apparatus for detecting table |
| CN109948507B (en) * | 2019-03-14 | 2021-05-07 | 北京百度网讯科技有限公司 | Method and device for detecting table |
| CN111340023A (en) * | 2020-02-24 | 2020-06-26 | 创新奇智(上海)科技有限公司 | Text recognition method and device, electronic equipment and storage medium |
| CN111340023B (en) * | 2020-02-24 | 2022-09-09 | 创新奇智(上海)科技有限公司 | Text recognition method and device, electronic equipment and storage medium |
| CN112861736A (en) * | 2021-02-10 | 2021-05-28 | 上海大学 | Document table content identification and information extraction method based on image processing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| KR102107887B1 (en) | Method and apparatus for generating a composite image | |
| CN109634961B (en) | Test paper sample generation method and device, electronic equipment and storage medium | |
| CN109447015A (en) | A kind of method and device handling form Image center selection word | |
| CN110390269A (en) | PDF document table extracting method, device, equipment and computer readable storage medium | |
| SG11201900263SA (en) | Method, device and server for recognizing characters of claim document, and storage medium | |
| CN104636742B (en) | A kind of method by imaging automatic lock onto target topic and transmitting | |
| US20170139575A1 (en) | Data entering method and terminal | |
| CN105260751B (en) | A kind of character recognition method and its system | |
| CN102156865A (en) | Handwritten text line character segmentation method and identification method | |
| JPH04104324A (en) | Program generating device | |
| CN105574486A (en) | Image table character segmenting method | |
| CN112036406B (en) | Text extraction method and device for image document and electronic equipment | |
| JP2021039424A5 (en) | ||
| CN103440239A (en) | Functional region recognition-based webpage segmentation method and device | |
| CN111240669A (en) | Interface generation method and device, electronic equipment and computer storage medium | |
| CN113191309A (en) | Method and system for recognizing, scoring and correcting handwritten Chinese characters | |
| CN109816743A (en) | Generate the method and terminal device of identification pattern | |
| CN105868240A (en) | Method and device for self-adaptive display of page | |
| CN107436931B (en) | Webpage text extraction method and device | |
| CN103186880B (en) | Generate the method and apparatus of thumbnail | |
| CN106297477A (en) | A kind of method and device generating digitized copybook | |
| CN108021423B (en) | Multilingual website generation method and system and computer readable storage medium | |
| CN112416340A (en) | Webpage generation method and system based on sketch | |
| CN108062548B (en) | Braille square self-adaptive positioning method and system | |
| CN102200966A (en) | Method for extracting and processing layout information |
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: 20190308 |
|
| RJ01 | Rejection of invention patent application after publication |