CN109800750A - A kind of character area mask method of the text based on morphological image for rule composing - Google Patents
A kind of character area mask method of the text based on morphological image for rule composing Download PDFInfo
- Publication number
- CN109800750A CN109800750A CN201910072288.6A CN201910072288A CN109800750A CN 109800750 A CN109800750 A CN 109800750A CN 201910072288 A CN201910072288 A CN 201910072288A CN 109800750 A CN109800750 A CN 109800750A
- Authority
- CN
- China
- Prior art keywords
- text
- image
- connected domain
- threshold value
- rectangle
- 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
Landscapes
- Character Input (AREA)
Abstract
The present invention provides a kind of character area mask method of text based on morphological image for rule composing, belong to OCR image identification technical field, the present invention is by the binaryzation of image, burn into expands, opens the comprehensive use of the operations such as operation, closed operation and connected domain analysis in morphological image, extracts text region and is labeled.The text box mark of specification typesetting is simplified into flexibility, with lesser consumption and cost obtains and the essentially identical effect of complicated approach.
Description
Technical field
The present invention relates to OCR image recognition technology more particularly to a kind of texts that rule composing is used for based on morphological image
This character area mask method.
Background technique
The text box label technology of mainstream is the supervised learning side using deep learning and Fast RCNN series at present
Method, but need to consume manpower in practical operation and training data is labeled.It is suitable for the String localization and text of complex scene
This collimation mark note.In simple scene, if business card OCR is identified, the simple and text such as invoice OCR identification compared in the scene of specification,
Using the methods of deep learning excessively complexity and it is easy to produce unstable mark.
In the tide of Artificial Intelligence Development, it will repeat and the work working machine of the machinery substitution mankind be a certainty
Trend.In current existing artificial intelligence application, Text region is highly developed, and is not limited only to the knowledge of type fount
It not, further include the identification of handwritten form.But if it is desired to accomplishing more intelligent, automation, before Text region, also to there is one
Walk the operation of String localization and text box mark.Want to identify single text, need accurately to mark text and divides
Cut
Traditional method is mostly based on Fast RCNN and its deriving method, these method belongs to supervised learning, into
It needs manually to be labeled text box before row training, needs a large amount of manpowers, and have certain consumption to hardware resource.In addition,
It is excessively complicated using study and neural network and will not be than simple image processing method in the text identification scene of comparison rule
Generate better effect.
The deficiency of traditional method:
1) training caused by supervised learning is complicated, and data set building is spent human and material resources, and preferable hardware is needed to be propped up
It holds.
2) after complicated step and model training, the promotion of matter can not be obtained on text box mark.
Therefore, a unsupervised method based on image procossing is widely used scene.
Summary of the invention
In order to solve the above technical problems, the invention proposes a kind of texts for being used for rule composing based on morphological image
Character area mask method, using to technology mainly have the binaryzation of image, behaviour is opened in burn into expansion in morphological image
Work, closed operation and connected domain analysis.
The technical scheme is that
A kind of character area mask method of the text based on morphological image for rule composing, can be used for typesetting specification
Text identification scene in text box mark.By corroding (Erosion) in morphological image, expansion (Dilate), opening
(Open)/and (Close) operation is closed, the target area of the image of the binaryzation containing target text is become into single connected domain, it is fixed
Text box can be marked out and behind position to the boundary of the connected domain.
Generally speaking, input is the text identification scene of a typesetting specification, the image arrived such as the photo of invoice or scanning.
Output is the text box marked.
If because visual angle difference leads to the invoice region photographed not and is the rectangle of specification to overlook visual angle, first using singly answering
Property transformation (Homography) by geographic norms to be marked become standard rectangular.
After the region standardized, binarization operation is carried out to image, the region other than text is schemed by 3 channel rgb
It is white (1) by text conversion as being converted to the black (0) in two bit image of single channel.
Histogram analytic approach may be used herein, obtain a suitable threshold value and come out text and background separation.?
After the binary image gone out to text and background separation, operated using morphological image by large stretch of character area and the equal structure of noise
Connected domain is built up, noise removal is regarded as lesser connected domain, biggish connected domain is considered that text retains.It obtains respectively
The boundary up and down of text connected domain can draw the rectangular text frame an of standard.Since the height of line of text has centainly
Range, therefore a threshold value is manually set and falls noise filtering, the connected domain less or greater than threshold value regards as noise removal,
Connected domain within the set range is considered that character area is retained.This usually upper range is set as all effective text boxes
0.5-1.5 times of height median.
Specifically used technology such as homography conversion, Histogram analysis, morphological image operation etc. are mature and logical
Technology, details are not described herein.
Specific step is as follows.
The first step obtains input picture, it is ensured that it only includes target area and is rectangle；If not rectangle, by singly answering
Property transformation standardize be standard quadrilateral；
Second step sets the threshold value of image binaryzation, distinguishes background and word segment by histogram analytic approach, and
A threshold value is obtained, by the way that threshold value is reasonably arranged, reduces noise as far as possible；Background and word segment are finally used 0 respectively,
1 is expressed as a bianry image；
Third step is operated using morphological image word segment connecting into an entire connected domain；
4th step obtains a rectangle connected domain after third step, extracts the seat on this connected domain four sides up and down
Target maximum value, which can mark out text box region, to be come.
Further, in the third step, closed operation is selected, structural element selects rectangle, for the text area that will do not gone together
It branches away, the height of structural element is no more than this number of pixels of two styles of writing.
Further, at the same time, in order to which the connection of the text of same a line is got up, the wide of structural element is not less than two words
Lateral separation cuts extra pixel when finally calculating text box position if setting is larger.
The beneficial effects of the invention are as follows
Flexible utilization prior art evades its disadvantage, play its strong point, proposes unsupervised based on morphological image
Text box mask method has following advantage:
1) method is simple and clear, has stronger versatility for the Text Feature Extraction scene of specification typesetting；
2) unsupervised learning does not need manpower and material resources and is trained collection mark and time-consuming training step；
3) calculation amount is smaller, does not need powerful hardware supported.
Detailed description of the invention
Fig. 1 is workflow schematic diagram of the invention.
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 embodiment of the present invention, instead of all the embodiments, based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
A kind of character area mask method of text based on morphological image for rule composing of the invention, can be used for
Text box mark in the text identification scene of typesetting specification.By corroding (Erosion), expansion in morphological image
(Dilate), (Open)/close (Close) operation is opened, the target area of the image of the binaryzation containing target text is become single
A connected domain can mark out text box to come after navigating to the boundary of the connected domain.
The specific steps of which are as follows:
The first step obtains input picture, takes pictures or scan, it is ensured that it only includes target area and is rectangle.If
It is not rectangle, can be standardized by homography conversion is standard quadrilateral
Second step sets the threshold value of image binaryzation, can distinguish background and word segment by histogram analytic approach,
And a suitable threshold value is obtained, by the way that threshold value is reasonably arranged, reduce noise as far as possible.Finally by background and word segment
A bianry image is expressed as with 0,1 respectively.
Third step is operated using morphological image word segment connecting into an entire connected domain.Here selection is closed
(close) it operating, structural element (structural element) selects rectangle, in order to which the text that do not go together is distinguished,
The high of structural element cannot be greater than this number of pixels of two styles of writing.It can be set to 1 or 2.Meanwhile in order to by same a line
Text connection is got up, and the wide of structural element cannot be less than the lateral separation of two words, and what be can be set is larger, but will be most
Extra pixel is cut when calculating text box position afterwards.
4th step should be able to theoretically obtain a rectangle connected domain after third step, and it is left up and down to extract this connected domain
The maximum value of the coordinate on right four sides, which can mark out text box region, to be come.
The present invention is flexibly used the prior art, is maximized favourable factors and minimized unfavourable ones using general at present Open Framework and language, and specification is arranged
The text box mark of version simplifies flexibility, and with lesser consumption and cost obtains and the essentially identical effect of complicated approach
The foregoing is merely presently preferred embodiments of the present invention, is only used to illustrate the technical scheme of the present invention, and is not intended to limit
Determine protection scope of the present invention.Any modification, equivalent substitution, improvement and etc. done all within the spirits and principles of the present invention,
It is included within the scope of protection of the present invention.
Claims (9)
1. a kind of character area mask method of the text based on morphological image for rule composing, which is characterized in that
By burn into expansion, opening and closing operations in morphological image, by the target area of the image of the binaryzation containing target text
Become single connected domain, text box can be marked out after navigating to the boundary of the connected domain.
2. the method according to claim 1, wherein
If become first using homography because it is the rectangle vertical view visual angle of specification that visual angle difference, which leads to the invoice region photographed not,
Geographic norms to be marked of changing commanders become standard rectangular.
3. according to the method described in claim 2, it is characterized in that,
After the region standardized, binarization operation is carried out to image, the region other than text is turned by 3 channel rgb images
Text conversion is white by the black being changed in two bit image of single channel.
4. according to the method described in claim 3, it is characterized in that,
Using histogram analytic approach, obtains a threshold value and come out text and background separation.
5. according to the method described in claim 4, it is characterized in that,
After obtaining the binary image of text and background separation out, by large stretch of character area and made an uproar using morphological image operation
Sound is built into connected domain, since the height of line of text has a certain range, a threshold value is manually set and falls noise filtering,
Connected domain less or greater than threshold value regards as noise removal, and connected domain within the set range is considered that character area carries out
Retain；This range is set as 0.5-1.5 times of all effective TextField._height medians.
6. according to the method described in claim 5, it is characterized in that,
The boundary up and down for obtaining text connected domain respectively can draw the rectangular text frame an of standard.
7. according to the method described in claim 5, it is characterized in that,
Concrete operation step is as follows.
The first step obtains input picture, it is ensured that it only includes target area and is rectangle；If not rectangle is become by homography
Its specification of changing commanders is standard quadrilateral；
Second step sets the threshold value of image binaryzation, distinguishes background and word segment by histogram analytic approach, and obtain
One threshold value reduces noise by the way that threshold value is arranged as far as possible；Finally 0,1 is used to be expressed as one respectively background and word segment
A bianry image；
Third step is operated using morphological image word segment connecting into an entire connected domain；
4th step obtains a rectangle connected domain after third step, extracts this connected domain coordinate on four sides up and down
Maximum value, which can mark out text box region, to be come.
8. the method according to the description of claim 7 is characterized in that
In the third step, closed operation is selected, structural element selects rectangle, in order to distinguish the text that do not go together, structural elements
The height of element is no more than two style of writing this number of pixels.
9. according to the method described in claim 8, it is characterized in that,
Meanwhile in order to which the connection of the text of same a line is got up, the wide lateral separation for being not less than two words of structural element, if set
It sets larger, cuts extra pixel when finally calculating text box position.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910072288.6A CN109800750A (en) | 2019-01-25 | 2019-01-25 | A kind of character area mask method of the text based on morphological image for rule composing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910072288.6A CN109800750A (en) | 2019-01-25 | 2019-01-25 | A kind of character area mask method of the text based on morphological image for rule composing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109800750A true CN109800750A (en) | 2019-05-24 |
Family
ID=66560335
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910072288.6A Pending CN109800750A (en) | 2019-01-25 | 2019-01-25 | A kind of character area mask method of the text based on morphological image for rule composing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109800750A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110175563A (en) * | 2019-05-27 | 2019-08-27 | 上海交通大学 | The recognition methods of metal cutting tool drawings marked and system |
| CN110989947A (en) * | 2020-01-02 | 2020-04-10 | 大连中盈科技股份有限公司 | Electronic invoice printer and printing method |
| CN112801923A (en) * | 2021-04-14 | 2021-05-14 | 江西博微新技术有限公司 | Word processing method, system, readable storage medium and computer equipment |
| CN112836696A (en) * | 2019-11-22 | 2021-05-25 | 搜狗（杭州）智能科技有限公司 | Text data detection method and device and electronic equipment |
| CN113033380A (en) * | 2021-03-23 | 2021-06-25 | 建信览智科技(北京)有限公司 | Text labeling method |
| CN113657407A (en) * | 2021-07-26 | 2021-11-16 | 扆亮海 | High-recall-rate accurate positioning method for large-amplitude picture characters |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103077388A (en) * | 2012-10-31 | 2013-05-01 | 浙江大学 | Rapid text scanning method oriented to portable computing equipment |
| US20140050396A1 (en) * | 2012-08-17 | 2014-02-20 | Evernote Corporation | Using surfaces with printed patterns for identification of imaging and data processing activities |
| CN105701488A (en) * | 2016-01-01 | 2016-06-22 | 广州恒巨信息科技有限公司 | Identity card identification method |
| CN106250831A (en) * | 2016-07-22 | 2016-12-21 | 北京小米移动软件有限公司 | Image detecting method, device and the device for image detection |
| CN108806059A (en) * | 2018-05-08 | 2018-11-13 | 中山大学 | The text filed localization method of the bill alignment and eight neighborhood connected component offset correction of feature based point |
-
2019
- 2019-01-25 CN CN201910072288.6A patent/CN109800750A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140050396A1 (en) * | 2012-08-17 | 2014-02-20 | Evernote Corporation | Using surfaces with printed patterns for identification of imaging and data processing activities |
| CN103077388A (en) * | 2012-10-31 | 2013-05-01 | 浙江大学 | Rapid text scanning method oriented to portable computing equipment |
| CN105701488A (en) * | 2016-01-01 | 2016-06-22 | 广州恒巨信息科技有限公司 | Identity card identification method |
| CN106250831A (en) * | 2016-07-22 | 2016-12-21 | 北京小米移动软件有限公司 | Image detecting method, device and the device for image detection |
| CN108806059A (en) * | 2018-05-08 | 2018-11-13 | 中山大学 | The text filed localization method of the bill alignment and eight neighborhood connected component offset correction of feature based point |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110175563A (en) * | 2019-05-27 | 2019-08-27 | 上海交通大学 | The recognition methods of metal cutting tool drawings marked and system |
| CN110175563B (en) * | 2019-05-27 | 2023-03-24 | 上海交通大学 | Metal cutting tool drawing mark identification method and system |
| CN112836696A (en) * | 2019-11-22 | 2021-05-25 | 搜狗（杭州）智能科技有限公司 | Text data detection method and device and electronic equipment |
| CN110989947A (en) * | 2020-01-02 | 2020-04-10 | 大连中盈科技股份有限公司 | Electronic invoice printer and printing method |
| CN113033380A (en) * | 2021-03-23 | 2021-06-25 | 建信览智科技(北京)有限公司 | Text labeling method |
| CN113033380B (en) * | 2021-03-23 | 2024-03-26 | 金科览智科技(北京)有限公司 | Text labeling method |
| CN112801923A (en) * | 2021-04-14 | 2021-05-14 | 江西博微新技术有限公司 | Word processing method, system, readable storage medium and computer equipment |
| CN113657407A (en) * | 2021-07-26 | 2021-11-16 | 扆亮海 | High-recall-rate accurate positioning method for large-amplitude picture characters |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109800750A (en) | A kind of character area mask method of the text based on morphological image for rule composing | |
| Xu et al. | Tell me what you see and i will show you where it is | |
| CN104182750B (en) | A kind of Chinese detection method based on extreme value connected domain in natural scene image | |
| JP4646797B2 (en) | Image processing apparatus, control method therefor, and program | |
| CN110969129A (en) | End-to-end tax bill text detection and identification method | |
| CN107527054B (en) | Automatic foreground extraction method based on multi-view fusion | |
| CN101901484A (en) | Document image segmentation system | |
| CN112712273B (en) | Handwriting Chinese character aesthetic degree judging method based on skeleton similarity | |
| CN108492296B (en) | Wheat ear intelligent counting system and method based on superpixel segmentation | |
| CN105825216A (en) | Method of locating text in complex background image | |
| CN105701489A (en) | Novel digital extraction and identification method and system thereof | |
| CN105260428A (en) | Picture processing method and apparatus | |
| CN108898092A (en) | Multi-spectrum remote sensing image road network extracting method based on full convolutional neural networks | |
| CN107122775A (en) | A kind of Android mobile phone identity card character identifying method of feature based matching | |
| CN112381082A (en) | Table structure reconstruction method based on deep learning | |
| CN112257629A (en) | Text information identification method and device for construction drawing | |
| Shah et al. | Efficient portable camera based text to speech converter for blind person | |
| Schnitman et al. | Inducing semantic segmentation from an example | |
| CN104951749A (en) | Image content recognition device and image content recognition method | |
| CN114581932A (en) | Picture table line extraction model construction method and picture table extraction method | |
| CN110533049A (en) | The method and apparatus for extracting seal image | |
| CN103927533A (en) | Intelligent processing method for graphics and text information in early patent document scanning copy | |
| Palaniappan et al. | Deep learning the indus script | |
| CN110414497A (en) | Method, apparatus, server and the storage medium of subject electronic | |
| CN111444903A (en) | Method, device and equipment for positioning characters in cartoon bubbles and readable storage medium |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190524 |