CN105335760A - Image number character recognition method - Google Patents
Image number character recognition method Download PDFInfo
- Publication number
- CN105335760A CN105335760A CN201510784451.3A CN201510784451A CN105335760A CN 105335760 A CN105335760 A CN 105335760A CN 201510784451 A CN201510784451 A CN 201510784451A CN 105335760 A CN105335760 A CN 105335760A
- Authority
- CN
- China
- Prior art keywords
- image
- sample
- character
- numerical character
- algorithm
- 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
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
Abstract
The invention relates to an image number character recognition method. An algorithm structure is optimized, and influences of illumination and inclination are processed in an image preprocessing stage, thereby ensuring that a better binary character image is acquired before character recognition; then, a two-dimensional PCA algorithm is adopted to extract primary characteristics in an image so as to reduce the dimension and operation complexity of the image; a decision tree algorithm and an Adaboost algorithm are combined to optimize a classification algorithm, thereby ensuring the recognition efficiency and reducing the over-fitting risk; and finally, a nine-one division method is used for verifying the generalization ability of a classification model so as to ensure the stability of a classifier. By virtue of algorithm optimization, the accuracy rate and speed of number character recognition are obviously increased.
Description
Technical field
The present invention relates to a kind of image digitization character identifying method, belong to technical field of image processing.
Background technology
Character recognition technologies combines Digital Image Processing and mode identification technology, is an important research direction, has huge economic worth and realistic meaning, can improve administrative authority's work efficiency, save a large amount of manpowers and fund.As far back as the '20s, Germany scientist Taushek proposes optical character identification OCR concept; The fifties, starts the research to printed characters recognition; The late nineteen eighties, character identifies fast becomes possibility; The early 1990s, the paper of a large amount of OCR aspects and system occur; Character recognition technologies progressively tended to ripe and achieved widespread use along with the development of computer technology and artificial intelligence study today.
At present, conventional recognizer has template matches, characteristic matching, neural network sum of subspace method.
Template matching method is that the character picture extracted is zoomed to the consistent size of template, mates one by one with each template, calculates degree of overlapping.Its advantage is that algorithm is simple, and calculated amount is less.But shortcoming is obvious, and the accuracy namely identified is not high, and hold confusable character.
Neural network is a kind of algorithm of simulating biological neural network, can process some environmental informations complexity, the indefinite problems of inference rule, character recognition also achieves important application.The advantage of algorithm is that antijamming capability is strong, and recognition correct rate is high.But its shortcoming also exists, namely need in advance to carry out a large amount of reasoning training to algorithm, at the rational odds for effectiveness of algorithm not obvious under training.Meanwhile, algorithm is also comparatively complicated, and calculated amount is also larger.
SIFT (Scale invariant features transform) algorithm in characteristic matching is a kind of algorithm extracting local feature, can find the unique point of the feature invariants such as position, yardstick, rotation in the metric space of image.This unique point with position, yardstick, invariable rotary characteristic can being detected just because of it, without the need to training algorithm, this algorithm being unanimously used so far.Its shortcoming is that computing velocity is slow, and be not suitable for the more weak platform of arithmetic capability as uses such as smart mobile phones, the dimension of the unique point simultaneously detected is higher, and when unique point quantity is more, committed memory space can be very large.
Usually, the dimension of image is higher, and it is sparse that character distributes on such higher-dimension, and namely make the method for Corpus--based Method analysis to run into small sample problem, calculated amount also increases thereupon.Subspace method a kind ofly can find linear or nonlinear spatial alternation according to certain performance objective, by compressing original data in the subspace of a low-dimensional, make Data distribution8 in this subspace more compact, make data descriptive power stronger, the operand of algorithm also reduces greatly.
Summary of the invention
For above-mentioned technical matters, technical matters to be solved by this invention is to provide a kind of image digitization character identifying method, optimize image recognition algorithm, the recognition accuracy for numerical character is made to obtain effective raising, and accelerate recognition speed, ensure that the work efficiency of image recognition.
The present invention is in order to solve the problems of the technologies described above by the following technical solutions: the present invention devises a kind of image digitization character identifying method, comprises the steps:
Step 001. obtains the sample image containing numerical character, and enters step 002;
Step 002. carries out gray processing process for sample image, obtains sample gray level image, and enters step 003;
Step 003. adopts multi-Scale Retinex Algorithm to carry out illumination pretreatment operation for sample gray level image, obtains illumination pretreatment sample image, and enters step 004;
Step 004. carries out binary conversion treatment for illumination pretreatment sample image, obtains binaryzation sample image, and enters step 005;
Step 005. is corrected for the angle of each numerical character in binaryzation sample image respectively, makes each numerical character be positioned at preset angular positions, and then obtains rectification sample image, and enters step 006;
Step 006., for rectification sample image, obtains the numerical character area image at wherein each numerical character place respectively, and enters step 007;
Step 007. adopts in advance for each sorter that each numerical character of 0-9 trains respectively, identifies for the numerical character in each numerical character area image, obtains each numerical character in each numerical character area image respectively; And according to the position in each numerical character area image place sample image, for obtain each numerical character and combine, obtain the numerical character of corresponding sample image.
As a preferred technical solution of the present invention, described step 003 specifically comprises: adopt multi-Scale Retinex Algorithm, according to following formula, based on image slices vegetarian refreshments, carry out illumination pretreatment operation for sample gray level image;
f(x,y)＝l(x,y)×r(x,y)
For each pixel in sample gray level image, reflecting component r (the x of each pixel in sample gray level image is obtained by the gaussian kernel convolved image algorithm of different angles, y), and for the reflecting component r (x of each pixel in sample gray level image, y), illumination pretreatment sample image is obtained by weighting algorithm; Wherein, f (x, y) represents the actual grey value component of each pixel in sample gray level image, and l (x, y) represents the illumination component of each pixel in sample gray level image.
As a preferred technical solution of the present invention: in described step 004, carry out binary conversion treatment for each pixel in illumination pretreatment sample image, obtain binaryzation sample image.
As a preferred technical solution of the present invention, described step 005 specifically comprises: respectively for each numerical character in binaryzation sample image, obtain the slope on the upper and lower both sides of numerical character, and obtain the mean value of upper and lower both sides slope, as the slope of this numerical character, then according to the slope of this numerical character, carry out rectification for this numerical character and rotate, and then obtain rectification sample image.
As a preferred technical solution of the present invention: in described step 005, respectively for each numerical character in binaryzation sample image, adopt the pixel on the fitting a straight line method difference upper and lower both sides of fitting digital character, in order to obtain the slope on the upper and lower both sides of this numerical character.
As a preferred technical solution of the present invention: in described step 005, respectively for each numerical character in binaryzation sample image, according to the slope of numerical character, adopt bilinearity difference spinning solution, carry out rectification for this numerical character and rotate.
As a preferred technical solution of the present invention, in described step 007, described each sorter trained respectively for each numerical character of 0-9 in advance obtains respectively by following steps:
Step a., for the sample image of predetermined number comprising designation number respectively, extracts the eigenmatrix of each width image, and then obtains the sample characteristics matrix corresponding to this numeral, and enter step b;
Sample characteristics matrix corresponding to this numeral is divided into training sample and test sample book by preset rules by step b., and trains in the default sorter of feeding, obtains the sorter trained for this numeral.
As a preferred technical solution of the present invention: in described step a, for the sample image of predetermined number comprising designation number respectively, two dimensional PCA algorithm is adopted to extract the eigenmatrix of each width image.
As a preferred technical solution of the present invention: in described step b, adopt 91 point-scores, the sample characteristics matrix corresponding to this numeral is divided into nine parts of training samples and a test sample book by preset rules.
As a preferred technical solution of the present invention: in described step b, described default sorter adds the strong classifier that Adaboost algorithm realizes for decision tree.
A kind of image digitization character identifying method of the present invention adopts above technical scheme compared with prior art, there is following technique effect: the image digitization character identifying method of the present invention's design, optimize algorithm structure, the impact of illumination and inclination is placed on image pre-processing phase, ensures before character recognition, obtain good binaryzation character picture; Subsequently, adopt the principal character in two dimensional PCA algorithm extraction image, reduce dimension and the computational complexity of image; And combined by decision tree and Adaboost algorithm, optimize sorting algorithm, both ensure that the efficiency of identification again reduces the risk occurring over-fitting; Finally use the generalization ability of 91 point-score checking disaggregated models, ensure that the stability of sorter; By to above algorithm optimization, the accuracy rate of Number character recognition and speed are all significantly improved.
Accompanying drawing explanation
Fig. 1 is the schematic flow sheet of designed image Number character recognition method of the present invention.
Embodiment
Be described in further detail for the specific embodiment of the present invention below in conjunction with Figure of description.
As shown in Figure 1, a kind of image digitization character identifying method of the present invention's design, in the middle of actual application, is such as applied on the mobile terminal of android system, specifically comprises the steps:
Step 001.Android system provides Camera, SurfaceView and ImageView assembly, and structure is formed by stacking for three, and wherein, Camera is responsible for preview image, takes pictures and intercepts appointed area image, being in the bottom; SurfaceView is responsible for loading preview image and initialization camera parameter, and mediate layer; ImageView is responsible for drawing view-finder, is in the superiors.Obtain the sample image containing numerical character by the Camera that android system provides, SurfaceView and ImageView assembly, and enter step 002.
Step 002. carries out gray processing process for sample image, obtains sample gray level image, and enters step 003.
Step 003. is in algorithm realization, under illumination mainly comes from different light environment for the impact of picture shooting, as burning sun, cloudy day, night etc., cause the situation that captured image domain of the existence uneven illumination is even, the even meeting of uneven illumination causes captured image may erase character information important in image when carrying out binary conversion treatment operation, therefore be necessary to carry out illumination pretreatment operation before binary conversion treatment, then here, design adopts multi-Scale Retinex Algorithm (MSR) to carry out illumination pretreatment operation for sample gray level image, obtain illumination pretreatment sample image, here specifically comprise: adopt multi-Scale Retinex Algorithm (MSR), according to following formula, based on image slices vegetarian refreshments, illumination pretreatment operation is carried out for sample gray level image,
f(x,y)＝l(x,y)×r(x,y)
For each pixel in sample gray level image, reflecting component r (the x of each pixel in sample gray level image is obtained by the gaussian kernel convolved image algorithm of different angles, y), and for the reflecting component r (x of each pixel in sample gray level image, y), obtain illumination pretreatment sample image by weighting algorithm, then enter step 004; Wherein, f (x, y) represents the actual grey value component of each pixel in sample gray level image, and l (x, y) represents the illumination component of each pixel in sample gray level image.
Step 004. carries out binary conversion treatment for each pixel in illumination pretreatment sample image, obtains binaryzation sample image, and enters step 005.
Step 005. is in image shoot process, due to the randomness of shooting, certain angle can be there is between camera and reference object, thus the character run-off the straight in captured image can be caused, character inclination can cause being difficult to be split from image by character exactly, reduces the accuracy of identification.At present, for the image that is taken containing numerical character, wherein, after numerical character run-off the straight, the both sides up and down of numerical character substantially present linear, therefore here, design is corrected for the angle of each numerical character in binaryzation sample image respectively, each numerical character is made to be positioned at preset angular positions, and then obtain rectification sample image, here specifically comprise: respectively for each numerical character in binaryzation sample image, adopt the pixel on the fitting a straight line method difference upper and lower both sides of fitting digital character, in order to obtain the slope on the upper and lower both sides of this numerical character, because angle of inclination is among a small circle about straight slope, therefore, based on obtain the slope on the upper and lower both sides of this numerical character, obtain the mean value of upper and lower both sides slope, as the slope of this numerical character, then according to the slope of this numerical character, adopt bilinearity difference spinning solution, carry out rectification for this numerical character to rotate, and then obtain rectification sample image, then step 006 is entered.
Step 006., for rectification sample image, obtains the numerical character area image at wherein each numerical character place respectively, and enters step 007.
Step 007. adopts in advance for each sorter that each numerical character of 0-9 trains respectively, identifies for the numerical character in each numerical character area image, obtains each numerical character in each numerical character area image respectively; Wherein, each sorter trained respectively for each numerical character of 0-9 is in advance respectively by operating acquisition as follows:
First, use the sample image of many as far as possible designation numbers to train, build one have degree of precision and recognition speed sorter can designation number effectively in recognition image faster, specifically comprise the steps:
Step a. is for the sample image of predetermined number comprising designation number respectively, two dimensional PCA algorithm is adopted to extract the eigenmatrix of each width image, and then the sample characteristics matrix obtained corresponding to this numeral, then step b is entered, the benefit of scheme designed in step a is here the dimension that not only can reduce sample image, be unlikely to the situation occurring over-fitting, have simultaneously and can retain feature main in image; Further, two dimensional PCA algorithm avoids the small sample problem that PCA algorithm there will be.
Step b. adopts 91 point-scores, sample characteristics matrix corresponding to this numeral is divided into nine parts of training samples and a test sample book by preset rules, and send into preset to be added in the strong classifier that Adaboost algorithm realizes by decision tree and train, obtain the sorter trained for this numeral; Here decision tree adds Adaboost algorithm and realizes in the design of strong classifier, the place that decision tree is different from other sorter is, though its operation rule is simple, but error is larger unlike nearest neighbor method uses Euclidean distance, decision tree is generally adopt information entropy as the standard of classification, and objectivity is stronger.But, decision tree is limited as a kind of Weak Classifier classifying quality, want to reach higher classifying quality then also to need to use Adaboost algorithm, the principle of Adaboost algorithm occurs that the sample of mistake increases weight in being classified the last time, subsequently to the process that sample is classified again, the sorter finally generated is the higher strong classifier of an accuracy, and this is the key element ensureing recognition accuracy.
After obtaining the sorter that trains for this numeral, then according to the position in each numerical character area image place sample image, for obtain each numerical character and combine, obtain the numerical character of corresponding sample image.
The image digitization character identifying method of the present invention's design, optimizes algorithm structure, the impact of illumination and inclination is placed on image pre-processing phase, ensure before character recognition, obtain good binaryzation character picture; Subsequently, adopt the principal character in two dimensional PCA algorithm extraction image, reduce dimension and the computational complexity of image; And combined by decision tree and Adaboost algorithm, optimize sorting algorithm, both ensure that the efficiency of identification again reduces the risk occurring over-fitting; Finally use the generalization ability of 91 point-score checking disaggregated models, ensure that the stability of sorter; By to above algorithm optimization, the accuracy rate of Number character recognition and speed are all significantly improved.
Be explained in detail for embodiments of the present invention in conjunction with Figure of description above, but the present invention is not limited to above-mentioned embodiment, in the ken that those of ordinary skill in the art possess, can also make a variety of changes under the prerequisite not departing from present inventive concept.
Claims (10)
1. an image digitization character identifying method, is characterized in that, comprises the steps:
Step 001. obtains the sample image containing numerical character, and enters step 002;
Step 002. carries out gray processing process for sample image, obtains sample gray level image, and enters step 003;
Step 003. adopts multi-Scale Retinex Algorithm to carry out illumination pretreatment operation for sample gray level image, obtains illumination pretreatment sample image, and enters step 004;
Step 004. carries out binary conversion treatment for illumination pretreatment sample image, obtains binaryzation sample image, and enters step 005;
Step 005. is corrected for the angle of each numerical character in binaryzation sample image respectively, makes each numerical character be positioned at preset angular positions, and then obtains rectification sample image, and enters step 006;
Step 006., for rectification sample image, obtains the numerical character area image at wherein each numerical character place respectively, and enters step 007;
Step 007. adopts in advance for each sorter that each numerical character of 0-9 trains respectively, identifies for the numerical character in each numerical character area image, obtains each numerical character in each numerical character area image respectively; And according to the position in each numerical character area image place sample image, for obtain each numerical character and combine, obtain the numerical character of corresponding sample image.
2. a kind of image digitization character identifying method according to claim 1, it is characterized in that, described step 003 specifically comprises: adopt multi-Scale Retinex Algorithm, according to following formula, based on image slices vegetarian refreshments, carry out illumination pretreatment operation for sample gray level image;
f(x,y)＝l(x,y)×r(x,y)
For each pixel in sample gray level image, reflecting component r (the x of each pixel in sample gray level image is obtained by the gaussian kernel convolved image algorithm of different angles, y), and for the reflecting component r (x of each pixel in sample gray level image, y), illumination pretreatment sample image is obtained by weighting algorithm; Wherein, f (x, y) represents the actual grey value component of each pixel in sample gray level image, and l (x, y) represents the illumination component of each pixel in sample gray level image.
3. a kind of image digitization character identifying method according to claim 1, is characterized in that, in described step 004, carry out binary conversion treatment for each pixel in illumination pretreatment sample image, obtains binaryzation sample image.
4. a kind of image digitization character identifying method according to claim 1, it is characterized in that, described step 005 specifically comprises: respectively for each numerical character in binaryzation sample image, obtain the slope on the upper and lower both sides of numerical character, and obtain the mean value of upper and lower both sides slope, as the slope of this numerical character, then according to the slope of this numerical character, carry out rectification for this numerical character to rotate, and then obtain rectification sample image.
5. a kind of image digitization character identifying method according to claim 4, it is characterized in that, in described step 005, respectively for each numerical character in binaryzation sample image, adopt the pixel on the fitting a straight line method difference upper and lower both sides of fitting digital character, in order to obtain the slope on the upper and lower both sides of this numerical character.
6. a kind of image digitization character identifying method according to claim 4, is characterized in that, in described step 005, respectively for each numerical character in binaryzation sample image, according to the slope of numerical character, adopt bilinearity difference spinning solution, carry out rectification for this numerical character and rotate.
7. a kind of image digitization character identifying method according to claim 1, it is characterized in that, in described step 007, described each sorter trained respectively for each numerical character of 0-9 in advance obtains respectively by following steps:
Step a., for the sample image of predetermined number comprising designation number respectively, extracts the eigenmatrix of each width image, and then obtains the sample characteristics matrix corresponding to this numeral, and enter step b;
Sample characteristics matrix corresponding to this numeral is divided into training sample and test sample book by preset rules by step b., and trains in the default sorter of feeding, obtains the sorter trained for this numeral.
8. a kind of image digitization character identifying method according to claim 7, is characterized in that, in described step a, for the sample image of predetermined number comprising designation number respectively, adopts two dimensional PCA algorithm to extract the eigenmatrix of each width image.
9. a kind of image digitization character identifying method according to claim 7, is characterized in that, in described step b, adopts 91 point-scores, the sample characteristics matrix corresponding to this numeral is divided into nine parts of training samples and a test sample book by preset rules.
10. a kind of image digitization character identifying method according to claim 7, is characterized in that, in described step b, described default sorter adds the strong classifier that Adaboost algorithm realizes for decision tree.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510784451.3A CN105335760A (en) | 2015-11-16 | 2015-11-16 | Image number character recognition method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510784451.3A CN105335760A (en) | 2015-11-16 | 2015-11-16 | Image number character recognition method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105335760A true CN105335760A (en) | 2016-02-17 |
Family
ID=55286275
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510784451.3A Pending CN105335760A (en) | 2015-11-16 | 2015-11-16 | Image number character recognition method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105335760A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106886996A (en) * | 2017-02-10 | 2017-06-23 | 九次方大数据信息集团有限公司 | Dividing method and device based on mathematical operation identifying code image |
| CN107024506A (en) * | 2017-03-09 | 2017-08-08 | 深圳市朗驰欣创科技股份有限公司 | A kind of pyrogenicity defect inspection method and system |
| CN107451559A (en) * | 2017-07-31 | 2017-12-08 | 邱宇轩 | Parkinson's people's handwriting automatic identifying method based on machine learning |
| CN108121984A (en) * | 2016-11-30 | 2018-06-05 | 杭州海康威视数字技术股份有限公司 | A kind of character identifying method and device |
| CN108205673A (en) * | 2016-12-16 | 2018-06-26 | 塔塔顾问服务有限公司 | For the method and system of container code identification |
| CN108304842A (en) * | 2018-02-01 | 2018-07-20 | 重庆中陆承大科技有限公司 | Meter reading recognition methods, device and electronic equipment |
| CN108573198A (en) * | 2017-03-14 | 2018-09-25 | 优信互联（北京）信息技术有限公司 | A kind of method and device identifying vehicle information according to Vehicle Identify Number |
| CN110119648A (en) * | 2018-02-05 | 2019-08-13 | 国家计算机网络与信息安全管理中心 | A kind of facsimile signal classification method based on optical character identification |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101923710A (en) * | 2010-06-30 | 2010-12-22 | 青岛海信网络科技股份有限公司 | Image tilt correction method and device |
| US7945101B2 (en) * | 2007-07-26 | 2011-05-17 | Palo Alto Research Center Incorporated | Innovative OCR systems and methods that combine a template based generative model with a discriminative model |
| CN103761531A (en) * | 2014-01-20 | 2014-04-30 | 西安理工大学 | Sparse-coding license plate character recognition method based on shape and contour features |
| CN103886294A (en) * | 2014-03-24 | 2014-06-25 | 北京国网富达科技发展有限责任公司 | Method and system for recognizing number of movement times of arrester for convertor station based on images |
| CN104102912A (en) * | 2013-04-02 | 2014-10-15 | 秦海勇 | Sub-item metering energy consumption data collection device based on video image identification and method thereof |
-
2015
- 2015-11-16 CN CN201510784451.3A patent/CN105335760A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7945101B2 (en) * | 2007-07-26 | 2011-05-17 | Palo Alto Research Center Incorporated | Innovative OCR systems and methods that combine a template based generative model with a discriminative model |
| CN101923710A (en) * | 2010-06-30 | 2010-12-22 | 青岛海信网络科技股份有限公司 | Image tilt correction method and device |
| CN104102912A (en) * | 2013-04-02 | 2014-10-15 | 秦海勇 | Sub-item metering energy consumption data collection device based on video image identification and method thereof |
| CN103761531A (en) * | 2014-01-20 | 2014-04-30 | 西安理工大学 | Sparse-coding license plate character recognition method based on shape and contour features |
| CN103886294A (en) * | 2014-03-24 | 2014-06-25 | 北京国网富达科技发展有限责任公司 | Method and system for recognizing number of movement times of arrester for convertor station based on images |
Non-Patent Citations (3)
| Title |
|---|
| 李晓东 等: "一种改进的模块2DPCA人脸识别新方法", 《系统仿真学报》 * |
| 梁琳 等: "光照不均图像增强方法综述", 《计算机应用研究》 * |
| 邓亚平 等: "基于分块2DPCA的人脸识别方法", 《计算机工程与设计》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108121984A (en) * | 2016-11-30 | 2018-06-05 | 杭州海康威视数字技术股份有限公司 | A kind of character identifying method and device |
| US11003941B2 (en) | 2016-11-30 | 2021-05-11 | Hangzhou Hikvision Digital Technology Co., Ltd. | Character identification method and device |
| CN108121984B (en) * | 2016-11-30 | 2021-09-21 | 杭州海康威视数字技术股份有限公司 | Character recognition method and device |
| CN108205673A (en) * | 2016-12-16 | 2018-06-26 | 塔塔顾问服务有限公司 | For the method and system of container code identification |
| CN108205673B (en) * | 2016-12-16 | 2022-02-25 | 塔塔顾问服务有限公司 | Method and system for container code identification |
| CN106886996A (en) * | 2017-02-10 | 2017-06-23 | 九次方大数据信息集团有限公司 | Dividing method and device based on mathematical operation identifying code image |
| CN107024506A (en) * | 2017-03-09 | 2017-08-08 | 深圳市朗驰欣创科技股份有限公司 | A kind of pyrogenicity defect inspection method and system |
| CN108573198A (en) * | 2017-03-14 | 2018-09-25 | 优信互联（北京）信息技术有限公司 | A kind of method and device identifying vehicle information according to Vehicle Identify Number |
| CN107451559A (en) * | 2017-07-31 | 2017-12-08 | 邱宇轩 | Parkinson's people's handwriting automatic identifying method based on machine learning |
| CN108304842A (en) * | 2018-02-01 | 2018-07-20 | 重庆中陆承大科技有限公司 | Meter reading recognition methods, device and electronic equipment |
| CN110119648A (en) * | 2018-02-05 | 2019-08-13 | 国家计算机网络与信息安全管理中心 | A kind of facsimile signal classification method based on optical character identification |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Long et al. | Scene text detection and recognition: The deep learning era | |
| Afzal et al. | Cutting the error by half: Investigation of very deep cnn and advanced training strategies for document image classification | |
| CN105335760A (en) | Image number character recognition method | |
| Zamberletti et al. | Text localization based on fast feature pyramids and multi-resolution maximally stable extremal regions | |
| US20200134382A1 (en) | Neural network training utilizing specialized loss functions | |
| Ahmed et al. | A novel dataset for English-Arabic scene text recognition (EASTR)-42K and its evaluation using invariant feature extraction on detected extremal regions | |
| CN109685065A (en) | Printed page analysis method, the system of paper automatic content classification | |
| Khan et al. | AUTNT-A component level dataset for text non-text classification and benchmarking with novel script invariant feature descriptors and D-CNN | |
| CN111340022A (en) | Identity card information identification method and device, computer equipment and storage medium | |
| CN112069900A (en) | Bill character recognition method and system based on convolutional neural network | |
| Yang et al. | Instance segmentation and classification method for plant leaf images based on ISC-MRCNN and APS-DCCNN | |
| Cerman et al. | A mobile recognition system for analog energy meter scanning | |
| CN114092938B (en) | Image recognition processing method and device, electronic equipment and storage medium | |
| Qi et al. | A novel joint character categorization and localization approach for character-level scene text recognition | |
| Misgar et al. | Recognition of offline handwritten Urdu characters using RNN and LSTM models | |
| Verma et al. | Enhanced character recognition using surf feature and neural network technique | |
| Adriano et al. | Digital conversion model for hand-filled forms using optical character recognition (OCR) | |
| US11715288B2 (en) | Optical character recognition using specialized confidence functions | |
| CN114842478A (en) | Text area identification method, device, equipment and storage medium | |
| Jia et al. | Grayscale-projection based optimal character segmentation for camera-captured faint text recognition | |
| Chen et al. | Design and Implementation of Second-generation ID Card Number Identification Model based on TensorFlow | |
| Zanwar et al. | English Handwritten Character Recognition Based on Ensembled Machine Learning | |
| CN111611985A (en) | OCR recognition method based on model fusion | |
| Zhu et al. | Chip surface character recognition based on improved LeNet-5 convolutional neural network | |
| Shiravale et al. | Recent advancements in text detection methods from natural scene images |
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 |
Application publication date: 20160217 |
|
| WD01 | Invention patent application deemed withdrawn after publication |