CN106127118A - A kind of English word recognition methods and device - Google Patents
A kind of English word recognition methods and device Download PDFInfo
- Publication number
- CN106127118A CN106127118A CN201610430159.6A CN201610430159A CN106127118A CN 106127118 A CN106127118 A CN 106127118A CN 201610430159 A CN201610430159 A CN 201610430159A CN 106127118 A CN106127118 A CN 106127118A
- Authority
- CN
- China
- Prior art keywords
- text
- connected domain
- line
- image
- stroke width
- 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
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/30—Writer recognition; Reading and verifying signatures
- G06V40/37—Writer recognition; Reading and verifying signatures based only on signature signals such as velocity or pressure, e.g. dynamic signature recognition
- G06V40/382—Preprocessing; Feature extraction
- G06V40/388—Sampling; Contour coding; Stroke extraction
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Human Computer Interaction (AREA)
- Character Discrimination (AREA)
Abstract
This application discloses a kind of English word recognition methods and device, the method includes: the video image of input is carried out stroke width conversion;The image of output after stroke width converts is carried out connected domain analysis, and to filter out from analysis result be text filed connected domain;The connected domain filtered out is merged, obtains line of text;Utilizing optical character recognition model to be identified described line of text, wherein, the training data of described optical character recognition model is English alphabet, and each English alphabet has the template of multiple different degree of corrosion;The line of text identified is carried out semantic analysis, selects the line of text meeting semanteme.The application improves English word identification accuracy under complex scene.
Description
Technical field
The present invention relates to technical field of character recognition, more particularly, it relates to a kind of English word recognition methods and device.
Background technology
Text is a key character in many application of computer vision, and the text in video image usually contains
Abundant information, extracts the text in video image and identifies, for the analysis of video image content, understanding, information
The aspects such as retrieval have great importance.
Extracting the contour feature of word from video image is the important ring during Text region, such as, at English
During language word identification, the contour feature first extracting each English alphabet is needed to be merged to identify whole English list again
Word.But owing to video image is natural scene, under complex scene, its background noise is overweight, letter profile disappearance can be made to be difficult to
Identify thus English word missing inspection occurs and identifies mistake, affect English word identification accuracy.
Summary of the invention
In view of this, the present invention provides a kind of English word recognition methods and device, to improve English list under complex scene
Word identification accuracy.
A kind of English word recognition methods, including:
The video image of input is carried out stroke width conversion;
The image of output after stroke width converts is carried out connected domain analysis, and to filter out from analysis result be text
The connected domain in region;
The connected domain filtered out is merged, obtains line of text;
Utilize optical character recognition model that described line of text is identified, wherein, described optical character recognition model
Training data is English alphabet, and each English alphabet has the template of multiple different degree of corrosion;
The line of text identified is carried out semantic analysis, selects the line of text meeting semanteme.
Wherein, described input picture is carried out stroke width conversion, including:
It is RGB image by the video image decoding of input;
Described RGB image is changed into gray-scale map;
Described gray-scale map is changed into normal window widget workbox image;
Utilize Canny edge detection operator that described normal window widget workbox image is carried out rim detection, obtain
All edge pixel points;
Sobel operator is utilized to be calculated the gradient direction of each edge pixel point respectively;
Find the edge pixel point contrary with its gradient direction for each described edge pixel point, form edge pixel point
Right;
Calculating respectively by the edge pixel point stroke width value to determining each described, the size of its stroke width value is
This edge pixel between Euclidean distance.
Wherein, described filtering out from analysis result is text filed connected domain, including:
Filtering out from analysis result is text filed connected domain, and screening conditions include: the stroke width one of connected domain
Cause;And pixel in connected domain the proportion identical with the color of English word to be identified be not less than first preset
Value.
Wherein, described filtering out from analysis result is text filed connected domain, including:
Filtering out from analysis result is text filed connected domain, and screening conditions include: the stroke width one of connected domain
Cause;And the stroke variance of connected domain is not less than the second preset value, stroke average is not less than the 3rd preset value and connection field width is high
Ratio is less than the 4th preset value.
Alternatively, described utilize optical character recognition model that described line of text is identified before, also include: utilize maximum
Inter-class variance binaryzation filters the background noise of described line of text;
Corresponding, described utilize optical character recognition model that described line of text is identified, for: utilize optical character to know
Line of text after background noise is filtered by other model is identified.
A kind of English word identification device, including:
Stroke width conversion module, for carrying out stroke width conversion to the video image of input;
Connected domain analysis screening unit, for the image of output after stroke width converts is carried out connected domain analysis, and
Filtering out from analysis result is text filed connected domain;
Line of text combining unit, for merging the connected domain filtered out, obtains line of text;
OCR recognition unit, is used for utilizing optical character recognition model to be identified described line of text, wherein, and described light
The training data learning character recognition model is English alphabet, and each English alphabet has the template of multiple different degree of corrosion;
Semantic analysis unit, carries out semantic analysis to the line of text identified, and selects the line of text meeting semanteme.
Wherein, described stroke width conversion module specifically includes:
RGB image conversion unit, being used for the video image decoding of input is RGB image;
Gray-scale map conversion unit, for changing into gray-scale map described RGB image;
SWT image conversion unit, for changing into SWT image described gray-scale map;
Edge detection unit, is used for utilizing Canny edge detection operator that described SWT image is carried out rim detection, obtains
All edge pixel points;
Gradient direction computing unit, for utilizing sobel operator to be calculated the gradient direction of each edge pixel point respectively;
Stroke width computing unit, for finding the edge contrary with its gradient direction for each described edge pixel point
Pixel, forms edge pixel point pair;Calculating by each edge pixel point stroke width value to determining respectively, size is this
Edge pixel between Euclidean distance.
Wherein, described connected domain analysis screening unit is specifically for carrying out even the image of output after stroke width converts
Logical domain analysis, and it is consistent therefrom to filter out stroke width, and also the pixel identical with the color of English word to be identified exists
In connected domain, proportion is not less than the connected domain of the first preset value.
Wherein, described connected domain analysis screening unit is specifically for carrying out even the image of output after stroke width converts
Logical domain analysis, and the stroke width therefrom filtering out connected domain is consistent, and the stroke variance of connected domain is not less than second presets
Value, stroke average are not less than the 3rd preset value and connected domain the ratio of width to height connected domain less than the 4th preset value.
Alternatively, described device also includes: background noise filter element, for utilizing optical character recognition model to institute
State before line of text is identified, filter the background noise of described line of text first with maximum between-cluster variance binaryzation.
From above-mentioned technical scheme it can be seen that the present invention is by corroding optical character recognition Model Identification difference in advance
The English alphabet of degree is trained study, increases the discrimination under letter profile damage situations, reduces English word missing inspection
Rate;And this present invention also carries out semantic analysis screening to the line of text identified, to select the line of text meeting semanteme, reduce
English word fallout ratio, thus improve English word identification accuracy under complex scene.
Accompanying drawing explanation
In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
In having technology to describe, the required accompanying drawing used is briefly described, it should be apparent that, the accompanying drawing in describing below is only this
Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to
Other accompanying drawing is obtained according to these accompanying drawings.
Fig. 1 is a kind of English word recognition methods flow chart disclosed by the invention;
Fig. 2 is a kind of stroke width alternative approach flow chart disclosed by the invention;
Fig. 3 is a kind of English word identification apparatus structure schematic diagram disclosed by the invention;
Fig. 4 is another English word identification apparatus structure schematic diagram disclosed by the invention;
Fig. 5 is another English word identification apparatus structure schematic diagram disclosed by the invention.
Detailed description of the invention
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Describe, it is clear that described embodiment is only a part of embodiment of the present invention rather than whole embodiments wholely.Based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under not making creative work premise
Embodiment, broadly falls into the scope of protection of the invention.
See Fig. 1, the embodiment of the invention discloses a kind of English word recognition methods, to improve English list under complex scene
Word identification accuracy, including:
Step 100: the video image of input is carried out stroke width conversion;
The purpose that the video image inputted carries out stroke width conversion is to obtain connected domain information.Stroke width converts
Thinking as follows: first to input video image carry out rim detection, obtain marginal information;Then from each edge pixel
Point sets out, and finds the edge pixel point that gradient direction therewith is contrary, forms an edge pixel point pair;Calculate each limit respectively
Edge pixel between Euclidean distance, and this value is given this edge pixel point between all of pixel.Through stroke
After width conversion, the image slices vegetarian refreshments of output represents possible stroke width.Utilize stroke width information can obtain possible literary composition
This information, because the consistent connected domain of stroke width is likely to be text filed.
The detailed process of stroke width conversion is as shown in Figure 2.Including:
Step 101: be RGB image by the video image decoding of input;
Step 102: described RGB image is changed into gray-scale map;
Step 103: described gray-scale map is changed into SWT (Standard Widget Toolkit, normal window widget
Workbox) image;
Step 104: utilize Canny edge detection operator that described SWT image is carried out rim detection, obtain all edges picture
Vegetarian refreshments;Wherein, Canny edge detection operator is that the multistage rim detection that John F.Canny developed in 1986 is calculated
Method;
Step 105: utilize sobel operator (Sobel operator, Sobel Operator) to be calculated each edge picture respectively
The gradient direction of vegetarian refreshments;
Step 106: find the edge pixel point contrary with its gradient direction for each described edge pixel point, forms limit
Edge pixel pair;
Step 107: calculate respectively by the edge pixel point stroke width value to determining each described, its stroke width value
Size be this edge pixel between Euclidean distance.
Step 200: the image of output after stroke width converts is carried out connected domain analysis, and screens from analysis result
Going out is text filed connected domain;
Connected domain refers to have in the image of output the adjacent prospect of same pixel value and position after stroke width converts
The image-region of pixel composition.Connected domain analysis refers to each connection in the image of the output after stroke width converts
Territory is found out and labelling.Prior art during English word identification, therefrom filter out be text filed connected domain time logical
The most only consider that the stroke width of connected domain is the most consistent, but the interference of background color is easily caused English word false retrieval, therefore originally
Embodiment increases by screening conditions: pixel in connected domain the proportion identical with the color of English word is not less than first
Preset value, such as, English word to be identified is black, then may call for black pixel point proportion in connected domain the lowest
In 60%.Additionally, for avoiding, because English word is too small, false retrieval occurs, it is also possible to it is further added by screening conditions: stroke variance is the lowest
It is not less than the 3rd preset value and connected domain the ratio of width to height less than the 4th preset value in the second preset value, stroke average.
Step 300: the connected domain filtered out is merged, obtains line of text;
Such as, the several connected domains filtered out are to show that connected domain that content is l, display content are u the most successively
Connected domain, display content be c connected domain, display content be k connected domain, display content be the connected domain of y, then merge it
Rear available line of text lucky.
Step 400: utilize OCR (Optical Character Recognition, optical character recognition) model to described
Line of text is identified, and wherein, the training data of described OCR model is that (described 26 English alphabets include 26 to English alphabet
Capitalization English letter A~Z and/or 26 small English alphabet a~z), each English alphabet has multiple different degree of corrosion
Template;
The present embodiment English alphabet to described OCR Model Identification difference degree of corrosion in advance is trained study, adds
Discrimination under English alphabet profile damage situations, training pattern can use existing SVM (Support Vector
Machine, support vector machine) algorithm, but do not limit to.Wherein, the template of described multiple different degree of corrosions, may is that complete
The template of free from corrosion template, low degree corrosion, the template of middle degree corrosion and the template of high level corrosion.
Step 500: the line of text identified is carried out semantic analysis, selects the line of text meeting semanteme.
English word occurs in the number of times in video image and has regular hour regularity, and therefore the present embodiment is to
The English word identified carries out semantic statistics, and the quantity of statistics is the most, and semantic statistical result is the most accurate, if this identifies
Line of text do not meet semanteme, be not i.e. inconsistent with the semantic statistical result previously obtained, then the line of text that this identified is got rid of,
To reduce fallout ratio, this is the basic thought that this line of text identified carries out semantic analysis.Such as, identified
English word include happy, happiness, joy, relaxed etc. of repeatedly occurring in video image, its semantic similarity,
If the text behavior pain that this identifies, then semantic contrary with the former due to it, it is known that it does not meets semanteme, is a false retrieval list
Word, needs to get rid of.The present embodiment can use HMM (Hidden Markov Model, hidden Markov model) to carry out line of text
Semantic analysis is added up, but does not limit to.
From the foregoing, it will be observed that the video image of input is carried out stroke width conversion, then to output after stroke width converts
Image carries out connected domain analysis, and therefrom to filter out be text filed connected domain, then merges the connected domain filtered out,
Obtain line of text, more described line of text is carried out OCR identification, be the routine techniques hands that word in video image is identified
Section.But the interference that in video image, background noise is overweight can make letter profile disappearance be difficult to thus English word missing inspection occurs
With identification mistake, affect English word identification accuracy.To this, the present embodiment is in advance to OCR Model Identification difference degree of corrosion
English alphabet is trained study, increases the discrimination under letter profile damage situations, reduces English word loss;And
And the present embodiment also carries out semantic analysis to the line of text identified, select the line of text meeting semanteme, reduce English word
Fallout ratio, thus improve English word identification accuracy under complex scene.
Additionally, before utilizing OCR model that described line of text is identified, also can be first with OSTU (maximum between-cluster variance)
Binaryzation filters the background noise of described line of text, and the line of text after background noise is filtered by recycling OCR model afterwards is carried out
Identify.It has the beneficial effects that: by filter background noise, and line of text can be made clear-cut, reduces background noise and treats knowledge
The interference corrosion of other English word, reduces false retrieval situation further.
Additionally, see Fig. 3, the embodiment of the invention also discloses a kind of English word identification device, to improve complex scene
Lower English word identification accuracy, including:
Stroke width conversion module 100, for carrying out stroke width conversion to the video image of input;
Connected domain analysis screening unit 200, for the image of output after stroke width converts is carried out connected domain analysis,
And to filter out from analysis result be text filed connected domain;
Line of text combining unit 300, for merging the connected domain filtered out, obtains line of text;
OCR recognition unit 400, is used for utilizing optical character recognition model to be identified described line of text, wherein, described
The training data of optical character recognition model is English alphabet, and each English alphabet has the template of multiple different degree of corrosion;
Semantic analysis unit 500, carries out semantic analysis to the line of text identified, and selects the line of text meeting semanteme.
Wherein, seeing Fig. 4, stroke width conversion module 100 specifically includes:
RGB image conversion unit 101, being used for the video image decoding of input is RGB image one by one;
Gray-scale map conversion unit 102, for changing into gray-scale map described RGB image;
SWT image conversion unit 103, for changing into SWT image described gray-scale map;
Edge detection unit 104, is used for utilizing Canny edge detection operator that described SWT image is carried out rim detection,
To all edge pixel points;
Gradient direction computing unit 105, for utilizing sobel operator to be calculated the gradient side of each edge pixel point respectively
To;
Stroke width computing unit 106, for finding contrary with its gradient direction for each described edge pixel point
Edge pixel point, forms edge pixel point pair;Calculate respectively by each edge pixel point stroke width value to determining, size
For this edge pixel between Euclidean distance.
Wherein, connected domain analysis screening unit 200 is specifically for carrying out even the image of output after stroke width converts
Logical domain analysis, and it is consistent therefrom to filter out stroke width, and also the pixel identical with the color of English word to be identified exists
In connected domain, proportion is not less than the connected domain of the first preset value.
Or, connected domain analysis screening unit 200 is specifically for carrying out even the image of output after stroke width converts
Logical domain analysis, and the stroke width therefrom filtering out connected domain is consistent, and the stroke variance of connected domain is not less than second presets
Value, stroke average are not less than the 3rd preset value and connected domain the ratio of width to height connected domain less than the 4th preset value.
Alternatively, as it is shown in figure 5, described English word identification device also includes: background noise filter element 600, it is used for
Before utilizing optical character recognition model that described line of text is identified, filter described first with maximum between-cluster variance binaryzation
The background noise of line of text.
In sum, the present invention is by carrying out the English alphabet of optical character recognition Model Identification difference degree of corrosion in advance
Training study, increases the discrimination under letter profile damage situations, reduces English word loss;And this present invention is also
The line of text identified is carried out semantic analysis screening, to select the line of text meeting semanteme, reduces English word fallout ratio,
Thus improve English word identification accuracy under complex scene.
In this specification, each embodiment uses the mode gone forward one by one to describe, and what each embodiment stressed is and other
The difference of embodiment, between each embodiment, identical similar portion sees mutually.For device disclosed in embodiment
For, owing to it corresponds to the method disclosed in Example, so describe is fairly simple, relevant part sees method part and says
Bright.
Described above to the disclosed embodiments, makes professional and technical personnel in the field be capable of or uses the present invention.
Multiple amendment to these embodiments will be apparent from for those skilled in the art, as defined herein
General Principle can realize in the case of without departing from the spirit or scope of the embodiment of the present invention in other embodiments.Therefore,
The embodiment of the present invention is not intended to be limited to the embodiments shown herein, and be to fit to principles disclosed herein and
The widest scope that features of novelty is consistent.
Claims (10)
1. an English word recognition methods, it is characterised in that including:
The video image of input is carried out stroke width conversion;
The image of output after stroke width converts is carried out connected domain analysis, and to filter out from analysis result be text filed
Connected domain;
The connected domain filtered out is merged, obtains line of text;
Utilize optical character recognition model that described line of text is identified, wherein, the training of described optical character recognition model
Data are English alphabet, and each English alphabet has the template of multiple different degree of corrosion;
The line of text identified is carried out semantic analysis, selects the line of text meeting semanteme.
Method the most according to claim 1, it is characterised in that described input picture is carried out stroke width conversion, including:
It is RGB image by the video image decoding of input;
Described RGB image is changed into gray-scale map;
Described gray-scale map is changed into normal window widget workbox image;
Utilize Canny edge detection operator that described normal window widget workbox image is carried out rim detection, owned
Edge pixel point;
Sobel operator is utilized to be calculated the gradient direction of each edge pixel point respectively;
Find the edge pixel point contrary with its gradient direction for each described edge pixel point, form edge pixel point pair;
Calculating by the edge pixel point stroke width value to determining each described respectively, the size of its stroke width value is this limit
Edge pixel between Euclidean distance.
Method the most according to claim 1, it is characterised in that described filtering out from analysis result is text filed company
Logical territory, including:
Filtering out from analysis result is text filed connected domain, and screening conditions include: the stroke width of connected domain is consistent;And
And pixel in connected domain the proportion identical with the color of English word to be identified is not less than the first preset value.
Method the most according to claim 1, it is characterised in that described filtering out from analysis result is text filed company
Logical territory, including:
Filtering out from analysis result is text filed connected domain, and screening conditions include: the stroke width of connected domain is consistent;And
And the stroke variance of connected domain is not less than the second preset value, stroke average is not less than the 3rd preset value and connected domain the ratio of width to height does not surpasses
Cross the 4th preset value.
5. according to the method according to any one of claim 1-4, it is characterised in that described utilize optical character recognition model pair
Before described line of text is identified, also include: utilize maximum between-cluster variance binaryzation to filter the background noise of described line of text;
Corresponding, described utilize optical character recognition model that described line of text is identified, for: utilize optical character recognition mould
Line of text after background noise is filtered by type is identified.
6. an English word identification device, it is characterised in that including:
Stroke width conversion module, for carrying out stroke width conversion to the video image of input;
Connected domain analysis screening unit, for carrying out connected domain analysis, and from dividing to the image of output after stroke width converts
Filtering out in analysis result is text filed connected domain;
Line of text combining unit, for merging the connected domain filtered out, obtains line of text;
OCR recognition unit, is used for utilizing optical character recognition model to be identified described line of text, wherein, and described optics word
Symbol identifies that the training data of model is English alphabet, and each English alphabet has the template of multiple different degree of corrosion;
Semantic analysis unit, carries out semantic analysis to the line of text identified, and selects the line of text meeting semanteme.
Device the most according to claim 6, it is characterised in that described stroke width conversion module specifically includes:
RGB image conversion unit, being used for the video image decoding of input is RGB image;
Gray-scale map conversion unit, for changing into gray-scale map described RGB image;
SWT image conversion unit, for changing into SWT image described gray-scale map;
Edge detection unit, is used for utilizing Canny edge detection operator that described SWT image is carried out rim detection, is owned
Edge pixel point;
Gradient direction computing unit, for utilizing sobel operator to be calculated the gradient direction of each edge pixel point respectively;
Stroke width computing unit, for finding the edge pixel contrary with its gradient direction for each described edge pixel point
Point, forms edge pixel point pair;Calculating respectively by each edge pixel point stroke width value to determining, size is this edge
Pixel between Euclidean distance.
Device the most according to claim 6, it is characterised in that described connected domain analysis screening unit is specifically for through pen
After drawing width conversion, the image of output carries out connected domain analysis, and it is consistent therefrom to filter out stroke width, and with to be identified
Pixel proportion in connected domain that the color of English word is identical is not less than the connected domain of the first preset value.
Device the most according to claim 6, it is characterised in that described connected domain analysis screening unit is specifically for through pen
After drawing width conversion, the image of output carries out connected domain analysis, and the stroke width therefrom filtering out connected domain is consistent, Er Qielian
The stroke variance in logical territory is not less than the second preset value, stroke average is not less than the 3rd preset value and connected domain the ratio of width to height is less than the
The connected domain of four preset values.
10. according to the device according to any one of claim 6-9, it is characterised in that described device also includes: background noise mistake
Filter unit, for before utilizing optical character recognition model to be identified described line of text, first with maximum between-cluster variance two
The background noise of described line of text is filtered in value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610430159.6A CN106127118A (en) | 2016-06-15 | 2016-06-15 | A kind of English word recognition methods and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610430159.6A CN106127118A (en) | 2016-06-15 | 2016-06-15 | A kind of English word recognition methods and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106127118A true CN106127118A (en) | 2016-11-16 |
Family
ID=57469919
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610430159.6A Pending CN106127118A (en) | 2016-06-15 | 2016-06-15 | A kind of English word recognition methods and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106127118A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108038481A (en) * | 2017-12-11 | 2018-05-15 | 江苏科技大学 | A kind of combination maximum extreme value stability region and the text positioning method of stroke width change |
CN110929647A (en) * | 2019-11-22 | 2020-03-27 | 科大讯飞股份有限公司 | Text detection method, device, equipment and storage medium |
CN112488107A (en) * | 2020-12-04 | 2021-03-12 | 北京华录新媒信息技术有限公司 | Video subtitle processing method and processing device |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101777124A (en) * | 2010-01-29 | 2010-07-14 | 北京新岸线网络技术有限公司 | Method for extracting video text message and device thereof |
US20140023275A1 (en) * | 2012-07-19 | 2014-01-23 | Qualcomm Incorporated | Redundant aspect ratio decoding of devanagari characters |
CN104112130A (en) * | 2014-06-26 | 2014-10-22 | 小米科技有限责任公司 | Optical character recognition method and device |
CN104268512A (en) * | 2014-09-17 | 2015-01-07 | 清华大学 | Method and device for recognizing characters in image on basis of optical character recognition |
CN104408449A (en) * | 2014-10-27 | 2015-03-11 | 西安电子科技大学宁波信息技术研究院 | Intelligent mobile terminal scene character processing method |
US20150070373A1 (en) * | 2012-08-23 | 2015-03-12 | Google Inc. | Clarification of Zoomed Text Embedded in Images |
-
2016
- 2016-06-15 CN CN201610430159.6A patent/CN106127118A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101777124A (en) * | 2010-01-29 | 2010-07-14 | 北京新岸线网络技术有限公司 | Method for extracting video text message and device thereof |
US20140023275A1 (en) * | 2012-07-19 | 2014-01-23 | Qualcomm Incorporated | Redundant aspect ratio decoding of devanagari characters |
US20150070373A1 (en) * | 2012-08-23 | 2015-03-12 | Google Inc. | Clarification of Zoomed Text Embedded in Images |
CN104112130A (en) * | 2014-06-26 | 2014-10-22 | 小米科技有限责任公司 | Optical character recognition method and device |
CN104268512A (en) * | 2014-09-17 | 2015-01-07 | 清华大学 | Method and device for recognizing characters in image on basis of optical character recognition |
CN104408449A (en) * | 2014-10-27 | 2015-03-11 | 西安电子科技大学宁波信息技术研究院 | Intelligent mobile terminal scene character processing method |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108038481A (en) * | 2017-12-11 | 2018-05-15 | 江苏科技大学 | A kind of combination maximum extreme value stability region and the text positioning method of stroke width change |
CN110929647A (en) * | 2019-11-22 | 2020-03-27 | 科大讯飞股份有限公司 | Text detection method, device, equipment and storage medium |
CN112488107A (en) * | 2020-12-04 | 2021-03-12 | 北京华录新媒信息技术有限公司 | Video subtitle processing method and processing device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Yi et al. | Localizing text in scene images by boundary clustering, stroke segmentation, and string fragment classification | |
CN104182750B (en) | A kind of Chinese detection method based on extreme value connected domain in natural scene image | |
CN104408449B (en) | Intelligent mobile terminal scene literal processing method | |
CN106384112A (en) | Rapid image text detection method based on multi-channel and multi-dimensional cascade filter | |
CN103093228A (en) | Chinese detection method in natural scene image based on connected domain | |
CN108986125B (en) | Object edge extraction method and device and electronic equipment | |
CN103295009B (en) | Based on the license plate character recognition method of Stroke decomposition | |
EP3846122B1 (en) | Method and apparatus for generating background-free image, device, and medium | |
Darab et al. | A hybrid approach to localize farsi text in natural scene images | |
CN105117740A (en) | Font identification method and device | |
Mishchenko et al. | Chart image understanding and numerical data extraction | |
CN106127118A (en) | A kind of English word recognition methods and device | |
Roy et al. | Date-field retrieval in scene image and video frames using text enhancement and shape coding | |
Kesiman et al. | Southeast Asian palm leaf manuscript images: a review of handwritten text line segmentation methods and new challenges | |
Ayesh et al. | A robust line segmentation algorithm for Arabic printed text with diacritics | |
Owamoyo et al. | Number plate recognition for Nigerian vehicles | |
CN104281850A (en) | Character area identification method and device | |
CN104966109A (en) | Medical laboratory report image classification method and apparatus | |
Jain et al. | A hybrid approach for detection and recognition of traffic text sign using MSER and OCR | |
Mukhiddinov | Scene text detection and localization using fully convolutional network | |
Tran et al. | A novel approach for text detection in images using structural features | |
CN109800758A (en) | A kind of natural scene character detecting method of maximum region detection | |
Romic et al. | Character recognition based on region pixel concentration for license plate identification | |
CN104504385A (en) | Recognition method of handwritten connected numerical string | |
Sun et al. | Contextual models for automatic building extraction in high resolution remote sensing image using object-based boosting method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20161116 |