CN104966051B - A kind of Layout Recognition method of file and picture - Google Patents
A kind of Layout Recognition method of file and picture Download PDFInfo
- Publication number
- CN104966051B CN104966051B CN201510297257.2A CN201510297257A CN104966051B CN 104966051 B CN104966051 B CN 104966051B CN 201510297257 A CN201510297257 A CN 201510297257A CN 104966051 B CN104966051 B CN 104966051B
- Authority
- CN
- China
- Prior art keywords
- text
- picture
- format
- region
- line
- 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.)
- Active
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
- G06V30/41—Analysis of document content
- G06V30/416—Extracting the logical structure, e.g. chapters, sections or page numbers; Identifying elements of the document, e.g. authors
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Character Input (AREA)
Abstract
The invention discloses a kind of Layout Recognition methods of file and picture, a format is devised first enters library facility, format content can be preserved in library, and by the format content format sequence number that high, alignment thereof generates with respect to word, if a unknown picture is analyzed by format, obtained format sequence number is as some format sequence number in library, then the layout information that will remove to extract the unknown picture by the prompt message in library.The present invention identifies document picture by efficient and accurate printed page analysis method, is particularly suitable for the Layout Recognition of Chinese official document file and picture.
Description
Technical field
The invention belongs to area of pattern recognition, are a kind of Layout Recognition methods proposed for file scanned image.
Background technology
In recent years, with China's economic fast development, government department guidance and formulate policy it is more and more, country and
Local policy is issued in the form of official document, with the development of science and technology, the documents such as more and more official documents are preserved with the format of image.
The official document different in face of enormous amount, format, it would be desirable to it can go out the format of official document to its automatic distinguishing, and impersonal force.
Official document, that is, Party and government offices' official document.The type abbreviation language of official document, General Office of the State Council's publication《State administrative organs
Document treatment Tentative Measures》The official document of state administrative organs is summarized as 13 kinds of nine class, order determines, bulletin, notice, leads to
Report proposal, is reported, is asked for instructions, giving an written reply, opinion, letter, meeting summary.Include part number, level of confidentiality and security deadline, urgent journey in official document
The attributes such as degree, issued organ's mark, documment number, signed by, the cut-off rule in version head, title, Zhu Song organs, text.Having
In body implementation procedure, a official document includes not necessarily above-mentioned all properties, and with the increase of official document quantity, the electronics such as scanner are set
Standby extensive use, official document are able to preserve with the format of scan image, therefore how effectively to carry out format knowledge to pictures such as official documents
It is not very necessary.
How particular document picture, and correct extraction document picture corresponding information are detected from a large amount of pictures, so far
Until the present, still without what good method.Currently, printed page analysis technology has had evolved to uses difference for different documents
Technology.Ma Zhuan, State of Zhao's power, Ren Zhanpeng et al. propose the research of the automatic marking papers system based on OCR identification technologies.This is a kind of
Top-down analysis method refers to the entirety from the page, payes attention to global information, general image is divided into several areas
Main region is continued to divide further according to the hierarchical structure information of text image in domain.Wu Yukun proposes the business card based on OCR
The bottom-up analysis method of printed page analysis has been used in system research in the research, from the pixel of image, pay attention to part
Image zonule is gradually synthesized big region, word by information, and --- word --- line of text --- paragraph etc. is schemed until covering is entire
Picture.For these methods both for the format of the similar size of font, the algorithm of use is template matching algorithm, connected domain algorithm
Deng the disadvantage is that operand is big, speed is slow.Current existing line of text, character cutting method Chinese, digital mixing environment and
Cutting can not accurately be carried out in the case of different font size word mixings, in official document identifying system, about dispatch for word and dispatch
Department, title etc. are all that font size differs.Therefore, it is necessary to an efficient and accurate printed page analysis methods to identify text
Shelves picture.
Invention content
In view of the above-mentioned problems, the object of the present invention is to provide a kind of Layout Recognition method of file and picture, by efficiently with
And accurately printed page analysis method identifies document picture, is particularly suitable for the Layout Recognition of Chinese official document file and picture.
To achieve the goals above, the present invention uses following technical scheme：
A kind of Layout Recognition method of file and picture, includes the following steps：
1) according to the format picture of different document sample, format feature database is generated.
Further, the format content of different document sample is preserved in the format feature database and by format content with respect to word
The format sequence number that high, alignment thereof generates.
In order to which more accurately extraction layout information, the present invention devise a format and enter library facility, exactly pass through first
User interface draws rectangle frame by user and goes to indicate which block is title, which block is dispatch department, which block to the format picture of input
It is to send the documents for word etc., is then put in storage, format content can be preserved in library, and high, alignment thereof generates with respect to word by format content
Format sequence number, the format sequence number layout information extraction in it is extremely important.It is the serial number by sequence, and alignment
What the numeric sequence number that mode generates generated.If having 3 pieces in format, first generated after ranking results sequence is
001221, first 0 indicates first piece, and second 0 indicates first piece and indicate second piece for maximum, 1, and 2 indicate that second piece is the
Three is big, and so on.Second sequence that alignment thereof generates is 212, wherein 2 indicate align center, 1 indicates Right Aligns.That
Its Serial No. 001221212.
In the format analysis phase, only there are one sequence numbers, if a unknown picture is analyzed by format, obtained format
Sequence number is as some format sequence number in library, then the version that will remove to extract the unknown picture by the prompt message in library
Formula information.This format feature database generated can improve the accuracy of layout information extraction.
2) document to be identified is scanned, scan image is obtained.
This step can also include being pre-processed to scan image, and the pretreatment includes that (removal ink goes to print for denoising
Chapter), Slant Rectify etc..
Some documents may will produce pad-ink in print procedure, other may be will produce in scanning process and is made an uproar
Sound, especially salt-pepper noise.Secondly, some document pictures have been capped some seals, it can generate normal format region dry
It disturbs, this also results in subsequent OCR (Optical Character Recognition, optical character identification) Recognition feedback knot
Fruit is a piece of mess code.Again, the inclination of document picture, which can divide line of text, generates interference.Therefore the invention system is needed to provide
The denoising function of picture, to enhance the robustness and accuracy of this invention.
3) region division is carried out to scan image, determines the text of document to be identified.
Line of text segmentation is carried out to scan image according to projection information, mainly by the textural characteristics of monochrome pixels point come really
Determine cutting position.Find out the minimum font size of line of text, the bottom-up end of text row for finding text, then top-down searching
It can be with the matched text initial row of end line.If can not find start of text row or end of text row, by start of text rower
It is denoted as 0, end of text rower is denoted as the ending of line of text.It is the text of document between start of text row and end of text row.
4) region division is carried out to part more than document text to be identified, and obtains the layout information in each region.
To part more than text, the row of word height having the same, line space, alignment thereof is put into the same region.
And if there are multiple line of text in left side inside the same region, only there are one line of text on right side, need to draw region again
Point, using a line of text on right side as the subregion in the region.
Ready-portioned region will generate a format sequence number, which is the opposite word Gao Sheng by alignment thereof
At.
The layout information includes：The alignment side of font size size, sequence, region relative to entire scan image in region
Formula.
5) layout information that step 4) obtains is matched with the layout information in format feature database, if matched,
Corresponding layout information is then extracted from format feature database；If do not matched, by the layout information in each region and in advance
The format word set integrates that (when document is official document document, which includes lemma collection, and department's word collection and dispatch are for word word
Collection) matching, obtain Layout Recognition result information.
Specifically, the layout information that step 4) obtains is primarily directed to document picture to be identified, mainly format sequence
Number, and each OCR result in region.Layout information in format feature database is mainly：Each corresponding rule of storage picture,
Namely：1) format sequence number；2) information labels (the corresponding regional number of information belonging to i.e.), for example title is which block, dispatch portion
Which block door is, which block dispatch is for word.If some pending picture match has arrived sequence number, corresponded to by information labels
Information is extracted to pending picture, such as the sequence number of title：1,1 indicates that first region is title.
By above step, the analysis to picture format can be completed, finally correctly extracts corresponding layout information.Wherein
It finds the text of file and picture and determines that the format region of text above section is core of the invention.
The beneficial effects of the present invention are：
Compared with prior art, Layout Recognition method provided by the invention has higher recognition accuracy, precision and effect
Rate, and there is larger practicability and application value.
Description of the drawings
Fig. 1 is the overall flow figure of Layout Recognition method of the present invention.
Fig. 2 is official document schematic diagram in the embodiment of the present invention 1.
Fig. 3 is the layout information schematic diagram extracted in the embodiment of the present invention 1.
Fig. 4 is official document schematic diagram in the embodiment of the present invention 2.
Fig. 5 is the layout information schematic diagram extracted in the embodiment of the present invention 2.
Specific implementation mode
It will elaborate below to embodiments of the present invention in conjunction with attached drawing by taking Chinese official document document as an example.
The overall flow of Layout Recognition method of the present invention is as shown in Figure 1, specifically include five steps：
1. a pair official document scan image pre-processes, the behaviour such as size adjusting, the fuzzy, slant correction of removal are carried out to image
Make, in favor of the Layout Recognition of official document.Concrete processing procedure is as follows：
(1) for removing salt-pepper noise, according to switch filtering thought, present invention preparation uses max-min operators as green pepper
Salt noise detector carries out progressive scan from left to right using adaptive neighborhood window to image, while to being located in window
The pixel of the heart carries out noise differentiation.If the gray value of the point is between maximum and minimum, then it is assumed that the point is quilt
Noise pollution；If the gray value of the point is equal to extreme value, then it is assumed that the point may be polluted by salt-pepper noise, then be recycled improved
Method differentiated, and using operation result as the substitution value of the point.
(2) seal for removing part on title finds profile, according to the training of some samples using canny edge detections
Value, when the contour area at edge be more than a certain threshold value when, then it be seal possibility it is very big, it can be removed.
(3) Slant Rectify is the statistical chart by adding up black pixel number in image, to line direction project
To horizontal and vertical projection.It is maximum according to the side of the perspective view along text inclined direction for inclined image, in certain angle
File and picture is rotated as interval using specific resolution ratio respectively in range, obtains the perspective view of rotated image, then will be made
The maximum rotation angle of perspective view mean square error is as angle of inclination.
2. according to projection information, line of text segmentation is carried out to official document.For determining character area, count black per a line
Point number.Find the initial row that continuous three rows stain number is more than 3, initial row of the label current line as text.From starting text
Row starts to count the average stain number of the first eight row, counts the stain number of each column in this eight row, and first stain number is more than etc.
Row in 5 originate row as text.Row of the last one stain number more than or equal to 5 are arranged as the end of text.Text is originated
It is equally divided into 5 regions between row and text starting row.If the stain number in two regions is less than 3, current text row is marked
For end of text row, next line of text is otherwise continued to scan on.It is line of text segmentation between text initial row and end of text row
Result.
3. the row calculated per a line is high, according to the alignment thereof of text and the high information of row, where determining text
Row.Find out the minimum font size of line of text, bottom-up ground scan text row.The line of text for meeting the following conditions is found as text
End line：Font size is differed with minimum font size within two pixels；Both ends are aligned or left-justify；Section after away from minimum text line number
After the section of the line of text at place away from difference two pixels within.Top-down scan text row, finds and meets the following conditions
Line of text is as start of text row：Font size is differed with minimum font size within two pixels；Both ends are aligned or Right Aligns；Section after away from
With after the section of the line of text where minimum text line number away from differing within two pixels.If can not find start of text row or just
Start of text rower is denoted as 0 by literary end line, and end of text rower is denoted as the ending of line of text.More than text it is us in this way
Carry out the region of Layout Recognition.
4. determining each region according to communication information, and to carrying out line of text segmentation in each region, preserve the area
The information such as line of text height, the alignment thereof of line number, region initial position, region relative to entire scan image in domain.Specifically
Steps are as follows：
(1) floor projection is carried out to text area above, forms line of text, region divides in advance.
A) denoising is carried out to floor projection, deletes the influence of some straight lines and discrete point.(filter continuous line number be less than etc.
In 7 successive projection row；It filters continuous line number and is less than or equal to 10 more than 7, and floor projection result mean value is less than or equal to 20
Successive projection row) merge projection line of text as region.(horizontal scan projection result from top to bottom, continuous two projections text
One's own profession font size is identical (criterion is that absolute value of the difference is less than or equal to 2), and (1) judges whether line-spacing is less than or equal to 2 times of font sizes, small
In equal to 2 times font sizes, merging two projection rows becomes a region；(2) continuous two rows font size it is close (criterion be difference it is exhausted
It is more than 2 to value to be less than or equal to 4), judge whether line-spacing is less than or equal to 1 times of font size, if it is less than equal to 1 times font size, merges two
Projection row becomes a region；(3) a line is bigger than upper row font size below, and difference is less than or equal to 10, and line-spacing is less than etc.
In 1 times of font size, while the line-spacing and the third line of the third line and the second row and the font size of the first row meet preceding two rule.)
(2) division determination is carried out to each pre- division region.
A) to region progress upright projection and to projection result denoising, storage zone row initial position, end position and width
Degree.
B) region line of text divides, text message record.(floor projection is carried out to region, and projection result is gone
It makes an uproar operation, redefines text row information, the details of line of text in posting field.)
C) judge that (a large amount of blank refer to that continuous white point number is more than or equal to 10 times of areas with the presence or absence of a large amount of blank in upright projection
The row in domain is high).In the presence of jumping to d), there is no jump to e).
D) it is several regions by region division according to a large amount of blank.
I. initial position and the end position of the row, column in the region after each segmentation, height, width are determined.
Ii. floor projection is carried out to the region after each segmentation, and denoising operation is carried out to projection result, redefine text
This row information, the details of line of text in posting field.
E) line of text in region is judged, judges whether the region is that multiple line of text correspond to a line of text
Situation.
Iii. region is presorted as three sub-spaces (left subspace, sub-spaces, right subspace).(Subspace partition
It is defined as, left subspace：Initial position on the left of region, at the 1/3 of zone length；Sub-spaces：To at 2/3 at 1/3；Right son
Space：The end position in region is arrived at 2/3).
Iv. floor projection is carried out to three sub-spaces respectively, and denoising operation is carried out to projection result.
V. the text row information of record subspace (text line number, initial position and end position, row is high, line-spacing)
Vi. judge the correlation of the line of text of 3 sub-spaces and whole region.Right subspace there are a line of text,
There are two and more line of text for left subspace or at least one space of sub-spaces.And the line of text of right subspace
Row height occupy whole region height (95% or more) or line of text be present in region floor projection part centre.It is such
Situation needs specially treated to go to f), otherwise terminates.
F) multiple line of text correspond to the case where line of text.
I. the part of multiple line of text is divided into region, the part of a remaining line of text is as the attached of the region
Subregion.Determine current region and attached subregion.(according to upright projection)
Ii. whether detection current region can merge with previous region, combination principle and b in (1)) it is similar.If can if close
And cannot then it continue.
Iii. whether detection current region can merge with latter area, combination principle and b in (1)) it is similar.Can, merge；
Cannot, continue.
Iv. initial position and the end position of the row, column in the region after merging or detecting, height, width are determined.
V. floor projection is carried out to region, and denoising operation is carried out to projection result, redefined text row information, record
The details of line of text in region.
It determines the region for finishing current official document, traverses each region and obtain layout information, extract font size size in region, row
Sequence, the alignment thereof in region is as layout information.
5. matched using the rule in the information and format feature database retained above (including location matches and keyword
Matching), it has matched and has then extracted layout information by format feature database.If not being matched to format sequence number, pass through setting
Lemma collection, department's word collection, the word collection sent the documents for word, each region that will identify that are matched with word collection, obtain Layout Recognition knot
Fruit information.
Embodiment 1
The official document in the one environmental protection Room, width Anhui Province is as shown in Fig. 2, carry out layout information such as Fig. 3 institutes of format Detection and Extraction
Show,
Region division is carried out to picture first, by the OCR result for obtaining sequence number and each region after division.
It is gone and format storehouse matching according to the method provided in text.First sample figure in format has been hit after matching (id is hit in Fig. 3
=0), information extraction is carried out according to hit format rule.
Embodiment 2
The official document of one width the National Audit Office is as shown in figure 4, the layout information for carrying out format Detection and Extraction is as shown in Figure 5.
Claims (10)
1. a kind of Layout Recognition method of file and picture, includes the following steps：
1) according to the format picture of different document sample, format feature database is generated；
2) document to be identified is scanned, scan image is obtained；
3) line of text segmentation is carried out to scan image, determines the text of document to be identified；
4) region division is carried out to part more than document text to be identified, and obtains the layout information in each region；
5) layout information that step 4) obtains is matched with the layout information in format feature database, if matched, from
Corresponding layout information is extracted in format feature database；If do not matched, by the layout information in each region with preset
Format word collection matching, obtain Layout Recognition result information.
2. the Layout Recognition method of file and picture as described in claim 1, which is characterized in that preserved in the format feature database
The format content of different document sample and the format sequence number that high, alignment thereof generates by the opposite word in format content.
3. the Layout Recognition method of file and picture as described in claim 1, which is characterized in that further include to sweeping in step 2)
Tracing is as being pre-processed.
4. the Layout Recognition method of file and picture as claimed in claim 3, which is characterized in that the pretreatment include denoising with
Slant Rectify.
5. the Layout Recognition method of file and picture as claimed in claim 4, which is characterized in that the denoising include removal ink and
Remove seal.
6. the Layout Recognition method of file and picture as described in claim 1, which is characterized in that according to projection information in step 3)
Line of text segmentation is carried out to scan image, cutting position is determined by the textural characteristics of monochrome pixels point.
7. the Layout Recognition method of file and picture as claimed in claim 6, which is characterized in that the bottom-up text for finding text
This end line, then top-down searching can be with the matched text initial row of end line；If can not find start of text row or
Start of text rower is denoted as 0 by end of text row, and end of text rower is denoted as the ending of line of text；Text initial row and text
It is the result of line of text segmentation between end line.
8. the Layout Recognition method of file and picture as described in claim 1, which is characterized in that having the same in step 4)
Word height, line space, alignment thereof row be put into the same region, and if there are multiple texts in left side inside the same region
Row, only there are one line of text on right side, need to divide region again, using a line of text on right side as the sub-district in the region
Domain.
9. the Layout Recognition method of file and picture as described in claim 1, which is characterized in that ready-portioned region in step 4)
A format sequence number is generated, which is by alignment thereof, and word height relatively generates.
10. the Layout Recognition method of file and picture as described in claim 1, which is characterized in that in step 4), the format letter
Breath includes：The alignment thereof of font size size, sequence, region relative to entire scan image in region.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510297257.2A CN104966051B (en) | 2015-06-03 | 2015-06-03 | A kind of Layout Recognition method of file and picture |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510297257.2A CN104966051B (en) | 2015-06-03 | 2015-06-03 | A kind of Layout Recognition method of file and picture |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104966051A CN104966051A (en) | 2015-10-07 |
| CN104966051B true CN104966051B (en) | 2018-07-17 |
Family
ID=54220089
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510297257.2A Active CN104966051B (en) | 2015-06-03 | 2015-06-03 | A kind of Layout Recognition method of file and picture |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104966051B (en) |
Families Citing this family (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106203454B (en) * | 2016-07-25 | 2019-05-21 | 重庆中科云从科技有限公司 | The method and device of certificate format analysis |
| CN107180239B (en) * | 2017-06-09 | 2020-09-11 | 科大讯飞股份有限公司 | Text line identification method and system |
| CN107798355B (en) * | 2017-11-17 | 2021-12-07 | 山西同方知网数字出版技术有限公司 | Automatic analysis and judgment method based on document image format |
| CN108830133B (en) * | 2018-04-17 | 2020-02-21 | 平安科技（深圳）有限公司 | Contract image picture identification method, electronic device and readable storage medium |
| CN108717544B (en) * | 2018-05-21 | 2022-11-25 | 天津科技大学 | Newspaper sample manuscript text automatic detection method based on intelligent image analysis |
| CN110969056B (en) * | 2018-09-29 | 2023-08-08 | 杭州海康威视数字技术股份有限公司 | Document layout analysis method, device and storage medium for document image |
| CN110414497A (en) * | 2019-06-14 | 2019-11-05 | 拉扎斯网络科技（上海）有限公司 | Method, apparatus, server and the storage medium of subject electronic |
| CN111062258B (en) * | 2019-11-22 | 2023-10-24 | 华为技术有限公司 | Text region identification method, device, terminal equipment and readable storage medium |
| CN111340031A (en) * | 2020-02-25 | 2020-06-26 | 杭州测质成科技有限公司 | Equipment almanac target information extraction and identification system based on image identification and method thereof |
| CN111428067B (en) * | 2020-03-20 | 2023-09-01 | 南京中孚信息技术有限公司 | Document picture acquisition method and device and electronic equipment |
| CN111539412B (en) * | 2020-04-21 | 2021-02-26 | 上海云从企业发展有限公司 | Image analysis method, system, device and medium based on OCR |
| CN111710379A (en) * | 2020-05-25 | 2020-09-25 | 广东百慧科技有限公司 | Personal medical information processing method, system, equipment and storage medium |
| CN111710437A (en) * | 2020-05-25 | 2020-09-25 | 广东百慧科技有限公司 | Intelligent inquiry method, system and storage medium based on image processing |
| CN111491069B (en) * | 2020-06-29 | 2020-10-02 | 北京灵伴即时智能科技有限公司 | Automatic setting method for color mode of document image |
| CN112861865B (en) * | 2021-01-29 | 2024-03-29 | 国网内蒙古东部电力有限公司 | Auxiliary auditing method based on OCR technology |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102081732A (en) * | 2010-12-29 | 2011-06-01 | 方正国际软件有限公司 | Method and system for recognizing format template |
| CN102880857A (en) * | 2012-08-29 | 2013-01-16 | 华东师范大学 | Method for recognizing format information of document image based on support vector machine (SVM) |
| CN104517106A (en) * | 2013-09-29 | 2015-04-15 | 北大方正集团有限公司 | List recognition method and system |
-
2015
- 2015-06-03 CN CN201510297257.2A patent/CN104966051B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102081732A (en) * | 2010-12-29 | 2011-06-01 | 方正国际软件有限公司 | Method and system for recognizing format template |
| CN102880857A (en) * | 2012-08-29 | 2013-01-16 | 华东师范大学 | Method for recognizing format information of document image based on support vector machine (SVM) |
| CN104517106A (en) * | 2013-09-29 | 2015-04-15 | 北大方正集团有限公司 | List recognition method and system |
Non-Patent Citations (3)
| Title |
|---|
| 基于纹理特征的版式识别研究;田学东等;《中国中文信息学会二十周年学术会议论文集》;20030908;第280-283页 * |
| 模板化网页主题信息的提取方法;欧健文等;《清华大学学报（自然科学版）》;20050930;第45卷(第9期);第1743-1747页 * |
| 电子病历版式化归档与信息抽取的研究;潘其明;《中国数字医学》;20150228(第2期);第107-109、117页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104966051A (en) | 2015-10-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104966051B (en) | A kind of Layout Recognition method of file and picture | |
| CN102332096B (en) | Video caption text extraction and identification method | |
| Zhou et al. | Bangla/English script identification based on analysis of connected component profiles | |
| CN1276384C (en) | Video stream classifiable symbol isolation method and system | |
| CN107093172B (en) | Character detection method and system | |
| Guo et al. | Separating handwritten material from machine printed text using hidden markov models | |
| CN103258198B (en) | Character extracting method in a kind of form document image | |
| US5390259A (en) | Methods and apparatus for selecting semantically significant images in a document image without decoding image content | |
| dos Santos et al. | Text line segmentation based on morphology and histogram projection | |
| CN102081731B (en) | Method and device for extracting text from image | |
| CN101102419B (en) | A method for caption area of positioning video | |
| Kumar et al. | Segmentation of isolated and touching characters in offline handwritten Gurmukhi script recognition | |
| Pal et al. | Automatic identification of english, chinese, arabic, devnagari and bangla script line | |
| CN105654072A (en) | Automatic character extraction and recognition system and method for low-resolution medical bill image | |
| CN101719142B (en) | Method for detecting picture characters by sparse representation based on classifying dictionary | |
| Chen et al. | An efficient algorithm for form structure extraction using strip projection | |
| CN101115151A (en) | Method for extracting video subtitling | |
| Chamchong et al. | Character segmentation from ancient palm leaf manuscripts in Thailand | |
| CN106778717A (en) | A kind of test and appraisal table recognition methods based on image recognition and k nearest neighbor | |
| CN1181446C (en) | Character recognition device | |
| CN110516673A (en) | Ancient Books in Yi Language character detection method based on connected component and regression equation character segmentation | |
| CN106203397A (en) | Differentiate and localization method based on the form of tabular analysis technology in image | |
| CN108052955B (en) | High-precision Braille identification method and system | |
| Chanda et al. | English, Devanagari and Urdu text identification | |
| Boukerma et al. | A novel Arabic baseline estimation algorithm based on sub-words treatment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |