CN110503105A - Character identifying method, training data acquisition methods, device and medium - Google Patents
Character identifying method, training data acquisition methods, device and medium Download PDFInfo
- Publication number
- CN110503105A CN110503105A CN201910821315.5A CN201910821315A CN110503105A CN 110503105 A CN110503105 A CN 110503105A CN 201910821315 A CN201910821315 A CN 201910821315A CN 110503105 A CN110503105 A CN 110503105A
- Authority
- CN
- China
- Prior art keywords
- character
- size
- picture
- shared
- sample text
- 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
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000001514 detection method Methods 0.000 claims abstract description 37
- 238000007689 inspection Methods 0.000 claims abstract description 20
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 230000000694 effects Effects 0.000 abstract description 5
- 230000011218 segmentation Effects 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000012015 optical character recognition Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/148—Segmentation of character regions
- G06V30/153—Segmentation of character regions using recognition of characters or words
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Abstract
This application involves a kind of character identifying method, training data acquisition methods, device and media, belong to image identification technical field, this method comprises: obtaining Target Photo to be identified；Pass through size shared by the location of character in Target Photo described in target detection model inspection and character；Size shared by the location of character obtained by character recognition model according to the target detection model inspection and character, identification obtain the character in the Target Photo.Go out the location of character and size shared by character in Target Photo by target detection model inspection, and then the location of character obtained subsequently through character recognition model according to detection identifies to obtain the character in Target Photo with size shared by character；It solves the problems, such as that accuracy of identification is poor in the prior art, has reached without doing Character segmentation in identification process, have ambiguity and then the higher effect of accuracy of identification without the output of CTC model is solved.
Description
Technical field
This application involves character identifying method, training data acquisition methods, device and media, belong to image recognition technology neck
Domain.
Background technique
Currently, needing to identify the character in paper in many scenes.For example, for paper document, when needing its turn
It then needs to identify the character in paper document when being changed to electronics shelves.
Currently used character identifying method is to pass through OCR (Optical Character Recognition, optics word
Symbol identification) Lai Shixian, the specific scheme is that (1) detects the size and location of line of text；(2) based on the figure where line of text
Panel region identifies text.
Summary of the invention
This application provides a kind of character identifying method, training data acquisition methods, device and media, can solve existing
The problem of scheme.The application provides the following technical solutions:
In a first aspect, providing a kind of symbol recognition methods, which comprises
Obtain Target Photo to be identified；
Pass through size shared by the location of character in Target Photo described in target detection model inspection and character；
The location of character obtained by character recognition model according to the target detection model inspection and character institute
Size is accounted for, identification obtains the character in the Target Photo.
Optionally, described to pass through character recognition model position according to locating for the character that the target detection model inspection obtains
It sets and obtains the character in the Target Photo with size shared by character, identification, comprising:
By the picture of character scaled as shared by the character detected to default size；
Using the picture of the location of described character and the character of the default size as the character recognition
The input of model, and then identify and obtain the character in the Target Photo.
Optionally, described by shared by the location of character in Target Photo described in target detection model inspection and character
Size, comprising:
The Target Photo is cut into n picture, the size of the n picture is identical, and the n is more than or equal to 2
Integer；
By n picture described in the target detection model inspection, and detects and obtain the position of the character in the n picture
It sets and size shared by character.
Second aspect provides a kind of training data acquisition methods, which comprises
Background picture is obtained, the background picture is mixed to get by least two pictures；
It obtains sample text and records the character in the sample text；
The sample text is plotted in the background picture, and record each character position and shared size；
According to the character, each in the sample text for drawing the background picture after the sample text, record
The position of character and shared size, generate training data, and the training data is known for training objective detection model and character
Other model, the target detection model is used to detect the position of the character in Target Photo and shared size, the character are known
Target figure described in the position for the character that other model is used to be obtained according to the target detection model inspection and shared size identification
Character in piece.
Optionally, the acquisition background picture, comprising:
Obtain the background picture of content blank；
Obtain the random background picture of content.
Optionally, the acquisition sample text, comprising:
Text is randomly selected from internet, using the text randomly selected as the sample text.
Optionally, in the sample text of the background picture after the sample text according to drafting, record
Character, each character position and shared size, generate training data, comprising:
The background picture after the sample text will be drawn to convert, the transformed background picture is obtained, remember
Record size shared in the picture of character after the conversion；
By the transformed background picture, record the sample text in character, the position of each character and every
Shared size in the picture of a character after the conversion, as the training data.
Optionally, the method also includes:
Noise is added at random in the background picture；
According to the position of character, each character in the sample text of the background picture, record after addition noise
It sets and shared size, generation training data
The third aspect provides a kind of computer storage medium, is stored at least one in the computer storage medium
Program instruction, at least one program instruction are loaded and executed by processors to realize character recognition side described in first aspect
Method, or realize training data acquisition methods described in second aspect.
Fourth aspect provides a kind of device, and described device includes memory and processor, stores in the memory right
At least one program instruction, the processor is by loading and executing above-metioned instruction to realize character recognition described in first aspect
Method, or realize training data acquisition methods described in second aspect.
The beneficial effects of the present application are as follows:
Go out the location of character and size shared by character, Jin Erhou in Target Photo by target detection model inspection
The location of continuous character obtained by character recognition model according to detection identifies to obtain target with size shared by character
Character in picture；It solves the problems, such as that accuracy of identification is poor in the prior art, has reached and cut in identification process without doing character
It cuts, has ambiguity and then the higher effect of accuracy of identification without the output of CTC model is solved.
By using the picture of the character of default size as the input of character recognition model, so that character recognition model can be with
The matching characteristic on fixed size improves the precision of identification character.
By the way that the bigger Target Photo of size to be cut into the picture of n fixed sizes, and then n picture is made respectively
For the input of target detection model, the processing pressure of CPU or GPU are alleviated, and since the size of character is smaller,
It will not influence training effect.
Many enchancement factors and interference is added by automatically generating training data, and when generating training data, reaches
The training data used in model training has more challenge than real scene, improves the robustness of model, namely improve
The accuracy of character in target detection model and character recognition model the identification picture for using training to obtain.
Above description is only the general introduction of technical scheme, in order to better understand the technological means of the application,
And can be implemented in accordance with the contents of the specification, with the preferred embodiment of the application and cooperate attached drawing below detailed description is as follows.
Detailed description of the invention
Fig. 1 is the method flow diagram for the character identifying method that the application one embodiment provides；
Fig. 2 is the method flow diagram for the training data acquisition methods that the application one embodiment provides.
Specific embodiment
With reference to the accompanying drawings and examples, the specific embodiment of the application is described in further detail.Implement below
Example is not limited to scope of the present application for illustrating the application.
Referring to FIG. 1, the method flow diagram of the character identifying method provided it illustrates the application one embodiment, such as schemes
Shown in 1, which comprises
Step 101, Target Photo to be identified is obtained；
Target Photo can be the picture being locally stored, and be also possible to the picture acquired from other equipment, certainly
It can also be the picture acquired from internet.
Step 102, by shared by the location of character in Target Photo described in target detection model inspection and character
Size；
Step 103, the location of the character obtained by character recognition model according to the target detection model inspection
With size shared by character, identification obtains the character in the Target Photo.
Above-mentioned described target detection model and character recognition model can be CNN (Convolutional Neural
Network, convolutional neural networks), certainly in actual implementation, it can also be other kinds of network model, this is not limited
It is fixed.
In conclusion being gone out in Target Photo by target detection model inspection big shared by the location of character and character
It is small, and then the location of character obtained subsequently through character recognition model according to detection is identified with size shared by character
Obtain the character in Target Photo；Solve the problems, such as that accuracy of identification is poor in the prior art, has reached and has been not necessarily in identification process
Character segmentation is done, has ambiguity and then the higher effect of accuracy of identification without the output of CTC model is solved.
Optionally, step 103, comprising:
First, by the picture of character scaled as shared by the character detected to default size；
May be different as detecting size shared by obtained each character, in order to improve precision, the present embodiment
Scaled shared by obtained character can will be identified to default size by panntographic system.Wherein, according to default size
The size of empirical value setting, which is the numerical value greater than ordinary symbol, and the numerical value of default size is less than threshold value.
Second, using the picture of the location of described character and the character of the default size as the character
The input of identification model, and then identify and obtain the character in the Target Photo.
By using the picture of the character of default size as the input of character recognition model, so that character recognition model can be with
The matching characteristic on fixed size improves the precision of identification character.
Optionally, step 102, comprising:
First, the Target Photo is cut into n picture, the size of the n picture is identical, the n be greater than etc.
In 2 integer；
In actual implementation, Target Photo can be cut into the picture of fixed size, for example Target Photo is cut into n
Picture.
Second, by n picture described in the target detection model inspection, and detects and obtain the word in the n picture
Size shared by the position of symbol and character.
By the way that the bigger Target Photo of size to be cut into the picture of n fixed sizes, and then n picture is made respectively
For the input of target detection model, the processing pressure of CPU or GPU are alleviated, and since the size of character is smaller,
It will not influence training effect.
It should be noted that the target detection model and character recognition model in above-described embodiment are pre- according to training data
First training obtains, and training data can be and directly acquire, and is also possible to be generated according to preset rules, the present embodiment with
Training data is to be illustrated according to what preset rules generated.
Referring to FIG. 2, such as scheming it illustrates a kind of method flow diagram of training data acquisition methods provided in this embodiment
Shown in 2, which comprises
Step 201, background picture is obtained, the background picture is mixed to get by least two pictures；
The present embodiment is mixed to get with background picture by two pictures come for example, this step includes that the following two kinds is realized
At least one of mode:
First, obtain the background picture of content blank；
In actual implementation, paper blank can be scanned to obtain the background picture.
Second, obtain the random background picture of content.
In actual implementation, at the same time include it is above two when, can will acquire above two picture synthesis as back
Scape picture, details are not described herein.
Step 202, it obtains sample text and records the character in the sample text；
Text is randomly selected from internet, using the text randomly selected as the sample text.
For example, text can be randomly selected from the news release of finance or entertainment news, and then obtain sample text.
Step 203, the sample text is plotted in the background picture, and records the position of each character and shared
Size；
Step 204, according to the word in the sample text of the background picture, record after the drafting sample text
Symbol, the position of each character and shared size generate training data.
The training data is used for training objective detection model and character recognition model, and the target detection model is for examining
The position and shared size, the character recognition model for surveying the character in Target Photo are used for according to the target detection model
Character in Target Photo described in the position for the character that detection obtains and shared size identification.
Optionally, in the sample text of the background picture after the sample text according to drafting, record
Character, each character position and shared size, generate training data, comprising:
The background picture after the sample text will be drawn to convert, the transformed background picture is obtained, remember
Record size shared in the picture of character after the conversion；
Transformation described herein may include at least one of rotation and stretching.
By the transformed background picture, record the sample text in character, the position of each character and every
Shared size in the picture of a character after the conversion, as the training data.
Optionally, the method also includes:
Noise is added at random in the background picture；
According to the position of character, each character in the sample text of the background picture, record after addition noise
It sets and shared size, generation training data.
Many enchancement factors and interference is added by automatically generating training data, and when generating training data, reaches
The training data used in model training has more challenge than real scene, improves the robustness of model, namely improve
The accuracy of character in target detection model and character recognition model the identification picture for using training to obtain.
Optionally, the application is also provided with a kind of computer storage medium, be stored in the computer storage medium to
A few program instruction, at least one program instruction are loaded and executed by processors to realize above-mentioned described character recognition
Method, or realize above-mentioned described training data acquisition methods.
Optionally, the application is also provided with a kind of device, and described device includes memory and processor, in the memory
Right at least one program instruction is stored, the processor realizes that above-mentioned described character is known by loading and executing above-metioned instruction
Other method, or realize above-mentioned described training data acquisition methods.
Each technical characteristic of embodiment described above can be combined arbitrarily, for simplicity of description, not to above-mentioned reality
It applies all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited
In contradiction, all should be considered as described in this specification.
The several embodiments of the application above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously
It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art
It says, without departing from the concept of this application, various modifications and improvements can be made, these belong to the protection of the application
Range.Therefore, the scope of protection shall be subject to the appended claims for the application patent.
Claims (10)
1. a kind of character identifying method, which is characterized in that the described method includes:
Obtain Target Photo to be identified；
Pass through size shared by the location of character in Target Photo described in target detection model inspection and character；
It is big shared by the location of character obtained by character recognition model according to the target detection model inspection and character
Small, identification obtains the character in the Target Photo.
2. the method according to claim 1, wherein it is described by character recognition model according to the target detection
Size shared by the location of character that model inspection obtains and character, identification obtain the character in the Target Photo, comprising:
By the picture of character scaled as shared by the character detected to default size；
Using the picture of the location of described character and the character of the default size as the character recognition model
Input, and then identify obtain the character in the Target Photo.
3. the method according to claim 1, wherein described pass through Target Photo described in target detection model inspection
Size shared by the location of middle character and character, comprising:
The Target Photo is cut into n picture, the size of the n picture is identical, and the n is the integer more than or equal to 2；
By n picture described in the target detection model inspection, and detect the position for obtaining the character in the n picture and
Size shared by character.
4. a kind of training data acquisition methods, which is characterized in that the described method includes:
Obtain background picture；
It obtains sample text and records the character in the sample text；
The sample text is plotted in the background picture, and record each character position and shared size；
According to character, each character in the sample text of the background picture, record after the drafting sample text
Position and shared size, generate training data, the training data be used for training objective detection model and character recognition mould
Type, the target detection model be used to detect character in Target Photo position and shared size, the character recognition mould
In Target Photo described in the position for the character that type is used to be obtained according to the target detection model inspection and shared size identification
Character.
5. according to the method described in claim 4, it is characterized in that, the acquisition background picture, comprising:
Obtain the background picture of content blank；And/or
Obtain the random background picture of content.
6. according to the method described in claim 4, it is characterized in that, the acquisition sample text, comprising:
Text is randomly selected from internet, using the text randomly selected as the sample text.
7. according to any method of claim 4 to 6, which is characterized in that after the sample text according to drafting
The background picture, record the sample text in character, each character position and shared size, generate training number
According to, comprising:
The background picture after the sample text will be drawn to convert, the transformed background picture is obtained, record word
Shared size in the picture of symbol after the conversion；
By the transformed background picture, record the sample text in character, each character position and each word
Shared size in the picture of symbol after the conversion, as the training data.
8. according to any method of claim 4 to 6, which is characterized in that the method also includes:
Noise is added at random in the background picture；
According to addition noise after the background picture, record the sample text in character, each character position and
Shared size generates training data.
9. a kind of computer storage medium, be stored at least one program instruction in the computer storage medium, it is described at least
One program instruction is loaded and executed by processor to realize the character identifying method as described in claims 1 to 3 is any, or
Realize the training data acquisition methods as described in claim 4 to 8 is any.
10. a kind of device, which is characterized in that described device includes memory and processor, and storage is right at least in the memory
One program instruction, the processor is by loading and executing above-metioned instruction to realize the word as described in claims 1 to 3 is any
Recognition methods is accorded with, or realizes the training data acquisition methods as described in claim 4 to 8 is any.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910821315.5A CN110503105A (en) | 2019-09-02 | 2019-09-02 | Character identifying method, training data acquisition methods, device and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910821315.5A CN110503105A (en) | 2019-09-02 | 2019-09-02 | Character identifying method, training data acquisition methods, device and medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110503105A true CN110503105A (en) | 2019-11-26 |
Family
ID=68590885
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910821315.5A Pending CN110503105A (en) | 2019-09-02 | 2019-09-02 | Character identifying method, training data acquisition methods, device and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110503105A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111626244A (en) * | 2020-05-29 | 2020-09-04 | 中国工商银行股份有限公司 | Image recognition method, image recognition device, electronic equipment and medium |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103971097A (en) * | 2014-05-15 | 2014-08-06 | 武汉睿智视讯科技有限公司 | Vehicle license plate recognition method and system based on multiscale stroke models |
| CN107292302A (en) * | 2016-03-31 | 2017-10-24 | 高德信息技术有限公司 | Detect the method and system of point of interest in picture |
| CN108052937A (en) * | 2017-12-28 | 2018-05-18 | 百度在线网络技术（北京）有限公司 | Based on Weakly supervised character machining device training method, device, system and medium |
| CN108596180A (en) * | 2018-04-09 | 2018-09-28 | 深圳市腾讯网络信息技术有限公司 | Parameter identification, the training method of parameter identification model and device in image |
| CN109583449A (en) * | 2018-10-29 | 2019-04-05 | 深圳市华尊科技股份有限公司 | Character identifying method and Related product |
| WO2019071660A1 (en) * | 2017-10-09 | 2019-04-18 | 平安科技（深圳）有限公司 | Bill information identification method, electronic device, and readable storage medium |
| CN109766879A (en) * | 2019-01-11 | 2019-05-17 | 北京字节跳动网络技术有限公司 | Generation, character detection method, device, equipment and the medium of character machining model |
-
2019
- 2019-09-02 CN CN201910821315.5A patent/CN110503105A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103971097A (en) * | 2014-05-15 | 2014-08-06 | 武汉睿智视讯科技有限公司 | Vehicle license plate recognition method and system based on multiscale stroke models |
| CN107292302A (en) * | 2016-03-31 | 2017-10-24 | 高德信息技术有限公司 | Detect the method and system of point of interest in picture |
| WO2019071660A1 (en) * | 2017-10-09 | 2019-04-18 | 平安科技（深圳）有限公司 | Bill information identification method, electronic device, and readable storage medium |
| CN108052937A (en) * | 2017-12-28 | 2018-05-18 | 百度在线网络技术（北京）有限公司 | Based on Weakly supervised character machining device training method, device, system and medium |
| CN108596180A (en) * | 2018-04-09 | 2018-09-28 | 深圳市腾讯网络信息技术有限公司 | Parameter identification, the training method of parameter identification model and device in image |
| CN109583449A (en) * | 2018-10-29 | 2019-04-05 | 深圳市华尊科技股份有限公司 | Character identifying method and Related product |
| CN109766879A (en) * | 2019-01-11 | 2019-05-17 | 北京字节跳动网络技术有限公司 | Generation, character detection method, device, equipment and the medium of character machining model |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111626244A (en) * | 2020-05-29 | 2020-09-04 | 中国工商银行股份有限公司 | Image recognition method, image recognition device, electronic equipment and medium |
| CN111626244B (en) * | 2020-05-29 | 2023-09-12 | 中国工商银行股份有限公司 | Image recognition method, device, electronic equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109146892B (en) | Image clipping method and device based on aesthetics | |
| RU2668717C1 (en) | Generation of marking of document images for training sample | |
| WO2019119966A1 (en) | Text image processing method, device, equipment, and storage medium | |
| US8965127B2 (en) | Method for segmenting text words in document images | |
| US8861884B1 (en) | Training classifiers for deblurring images | |
| CN108460396B (en) | Negative sampling method and device | |
| US8538164B2 (en) | Image patch descriptors | |
| US10311099B2 (en) | Method and system for 3D model database retrieval | |
| Dubská et al. | Real-time precise detection of regular grids and matrix codes | |
| US9025889B2 (en) | Method, apparatus and computer program product for providing pattern detection with unknown noise levels | |
| CN108765532B (en) | Child drawing model building method, reading robot and storage device | |
| CN111310759B (en) | Target detection inhibition optimization method and device for dual-mode cooperation | |
| WO2019020061A1 (en) | Video dialogue processing method, video client, video server, and computer readable storage medium | |
| US20150213331A1 (en) | Emotion modification for image and video content | |
| CN111836118B (en) | Video processing method, device, server and storage medium | |
| US20150055866A1 (en) | Optical character recognition by iterative re-segmentation of text images using high-level cues | |
| US20140169684A1 (en) | Distance Metric for Image Comparison | |
| US20220156897A1 (en) | Method and apparatus for classifying fake images | |
| CN115171138A (en) | Method, system and equipment for detecting image text of identity card | |
| CN110503105A (en) | Character identifying method, training data acquisition methods, device and medium | |
| WO2014176746A1 (en) | Method and system for automatically capturing an object using a mobile terminal | |
| JP2016012767A (en) | Image processing system | |
| US10558882B2 (en) | Performing distance-based feature suppression | |
| US10631050B2 (en) | Determining and correlating visual context on a user device with user behavior using digital content on the user device | |
| US10546218B2 (en) | Method for improving quality of recognition of a single frame |
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: 20191126 |