CN106228166B - The recognition methods of character picture - Google Patents
The recognition methods of character picture Download PDFInfo
- Publication number
- CN106228166B CN106228166B CN201610603330.9A CN201610603330A CN106228166B CN 106228166 B CN106228166 B CN 106228166B CN 201610603330 A CN201610603330 A CN 201610603330A CN 106228166 B CN106228166 B CN 106228166B
- Authority
- CN
- China
- Prior art keywords
- image
- feature
- classifier
- layer
- character
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
-
- 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
-
- 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
-
- 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
- G06V30/28—Character recognition specially adapted to the type of the alphabet, e.g. Latin alphabet
- G06V30/287—Character recognition specially adapted to the type of the alphabet, e.g. Latin alphabet of Kanji, Hiragana or Katakana characters
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Character Input (AREA)
Abstract
The present invention provides a kind of recognition methods of character picture.This method, which specifically includes that, to be normalized original character image Image to obtain floating type image array ImageSquare；Convolution algorithm is carried out to floating type image array ImageSquare using multiple filtering cores, obtains one group of image array ImageDst comprising different texture feature；The corresponding feature vector Features of original character image Image is obtained, extracts multilayer feature from one group of image array ImageDst, multilayer feature is arranged into feature vector Features, data sample is obtained；Layer-by-layer classification processing is carried out to data sample using the multi-level classifier with cascade connection, identifies the character in original character image Image.The present invention is based on the thoughts of image classification, a kind of multi-level cascade classifier model is devised using SVM classifier, a kind of recognition methods step by step first sorted out and identified afterwards is realized, reduces the class number of single layer classifier to a certain extent, to improve recognition accuracy.Method of the invention is succinct, strong operability.
Description
Technical field
The present invention relates to character recognition technologies field more particularly to a kind of recognition methods of character picture.
Background technique
Optical character identification (Optical Character Recognition, OCR) is pattern-recognition (Pattern
Recognition) an important research direction in field, main purpose are by the text in hand-written, printing, scanning or image
Machine code is converted to, is a kind of digitized process of text information.In recent years, with the development of science and technology, OCR technique not only exists
Great function is played in the Digitization of literature of ancient book, is also applied in many intelligent scenes, in such as life often
The charge station's Car license recognition and Google glass seen provide wearer according to the text seen suggests accordingly.Character recognition
Process is generally divided into several steps such as image preprocessing, Character segmentation, character recognition and post-processing, and each step requires a variety of
Technology participates in jointly.
Foreign countries are more early to the research work of OCR technique starting, and English character has that total amount is small, single character stroke connection
Equal distinct characteristics, these features are all easy to extract from image.Chinese character recognition is then more challenging in OCR technique
One content, one of main difficulty are that the feature extraction of Chinese character.Chinese character is developed by pictograph, character
Quantity is more, font type is more, structure is complicated, and possesses a certain number of nearly word forms, these features, which both increase extraction Chinese character, to be had
The difficulty for imitating feature, to influence recognition result.
The common method for extracting character feature has statistical nature to extract, structure feature is extracted.Statistical nature extracts main
It is that grayscale information statistics is carried out to each pixel in image by the methods of projection, statistics with histogram, to form character
Feature vector.This method is poor to the discrimination of details, cannot effectively distinguish nearly word form.Structure-based feature extraction side
Method is mainly extracted the stroke information of character, combines the features such as direction, crosspoint, can preferably identify detailed information, still
It correspondingly also can be more sensitive to noise.In recent years due to the burning hot development of deep learning, occur carrying out using neural network
The method of feature extraction, although having relatively high accuracy of identification, technology cost of implementation is high, and Chinese character quantity is various, mind
Design through network becomes sufficiently complex, these are all unfavorable for the exploitation of middle-size and small-size OCR system.Based on the above issues, of the invention
It is proposed that a kind of hierarchical Design model based on Gabor filtering is used to carry out feature extraction.Gabor filtering is that a kind of pair of texture is sensitive
Method, be able to reflect out textural characteristics of the image on different scale and direction, the edge in image procossing be widely applied
Test problems.
Summary of the invention
The embodiment provides a kind of recognition methods of character picture, are effectively known from character picture with realizing
It Chu not character.
To achieve the goals above, this invention takes following technical solutions.
A kind of recognition methods of character picture, comprising:
Original character image Image is normalized to obtain floating type image array ImageSquare；
Convolution algorithm is carried out to the floating type image array ImageSquare using multiple filtering cores, is obtained comprising not
With one group of image array ImageDst of textural characteristics；
The corresponding feature vector Features of the original character image Image is obtained, from one group of image array
Multilayer feature is extracted in ImageDst, and the multilayer feature is arranged into described eigenvector Features, data are obtained
Sample；
Layer-by-layer classification processing is carried out to the data sample using the multi-level classifier with cascade connection, is identified
Character in the original character image Image.
Further, described that original character image Image is normalized to obtain floating type image array
ImageSquare, comprising:
If original character image is Image, image Image gray processing treated image is ImageGray, takes image
The longest edge of ImageGray is denoted as LenSide, calculates gray average of the image ImageGray at four angular vertexs
MeanVal, the outside of two long sides of image ImageGray fill identical quantity, gray value be meanVald pixel,
Image ImageGray is set to become the image array of a size LenSide × LenSide, then by the matrix image normalizing
Turn to 32 × 32 floating type image array ImageSquare.
Further, described that convolution is carried out to the floating type image array ImageSquare using multiple filtering cores
Operation obtains one group of image array ImageDst comprising different texture feature, comprising:
12 Gabor filtering cores are designed, the kernel function of the Gabor filtering is as follows:
X '=x cos θ+y sin θ
Y '=- x sin θ+y cos θ
Wherein θ indicates the direction of filtering core,Indicate the phase of cosine part, γ indicates the space aspect ratio of filtering core, λ
Indicate that the wavelength of cosine part, σ indicate the standard deviation in Gaussian function；
Convolution fortune is carried out to the floating type image array ImageSquare using 12 Gabor filtering cores respectively
It calculates, extracts 12 character feature images in total of 3 kinds of different scales on 4 directions, kinds of characters characteristic image includes not
All character feature images are formed one group of image array ImageDst by same textural characteristics.
Further, the corresponding feature vector Features of the acquisition original character image Image, from described
Multilayer feature is extracted in one group of image matrix group ImageDst, by multilayer feature setting to described eigenvector
In Features, data sample is obtained, comprising:
Remember that the corresponding feature vector of image Image is Features, calculates separately out one group of image array
In ImageDst in every characteristic image all pixels point mean value and standard deviation and the feature as first layer, by described first
The feature of layer is arranged into feature vector Features；
Every characteristic image in one group of image array ImageDst is divided into the image moment that 4 sizes are 16 × 16
Battle array calculates separately out the mean value and standard deviation and the feature as the second layer of each 16 × 16 image array, by described second
The feature of layer is arranged into feature vector Features；
Every characteristic image in one group of image array ImageDst is divided into the image array that 16 sizes are 8 × 8,
The mean value and standard deviation and the feature as third layer for calculating separately out each 8 × 8 image array, by the spy of the third layer
Sign is arranged into feature vector Features；
The feature vector Features of three layers of characteristic information will be contained as data sample.
Further, the use there is the multi-level classifier of cascade connection to carry out to the data sample layer-by-layer
Classification processing identifies the character in the original character image Image, comprising:
There is the multi-level classifier of cascade connection according to sample attribute setting, each layer includes multiple classifiers, often
The classifier on a upper layer cascades the classifier of one or more lower layer, and sub-classification device is to its cascade upper layer classifier
Refinement, each classifier respectively correspond a kind of sample attribute of classification, choose the other training data of every type respectively to corresponding
Classifier is trained, and obtains trained cascade classifier；
Using the trained cascade classifier according to top-down sequence, from different perspectives to the data sample
Successively classify, when the data sample corresponds to some classifier in the classification results of a upper level, then choose it is described some
Cascade next layer of the classifier of classifier carries out further classification to the data sample, until the classification of the bottom
Device is completed to the classification processing of the data sample, by the classifier on all levels to the classification results of the sample data into
Row synthesis, obtains the character identification result of the original character image Image.
As can be seen from the technical scheme provided by the above-mentioned embodiment of the present invention, the embodiment of the present invention is based on image classification
Thought devises a kind of multi-level cascade classifier model using SVM classifier, realize it is a kind of first sort out identify afterwards by
Grade recognition methods, reduces the class number of single layer classifier, to improve recognition accuracy to a certain extent.The present invention is real
The method for applying example is succinct, strong operability.
The additional aspect of the present invention and advantage will be set forth in part in the description, these will become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, required use in being described below to embodiment
Attached drawing be briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for this
For the those of ordinary skill of field, without any creative labor, it can also be obtained according to these attached drawings other
Attached drawing.
Fig. 1 is a kind of level characteristics model and cascade classifier for character picture identification provided in an embodiment of the present invention
Design flow diagram.
Fig. 2 is a kind of expression original character image provided in an embodiment of the present invention, and carries out gray scale to original character image
The schematic diagram of filling.
The visualization signal that Fig. 3 is a kind of kernelGroup being made of 12 filtering cores provided in an embodiment of the present invention
Figure.
Fig. 4 is a kind of hierarchical Design model schematic provided in an embodiment of the present invention.
Fig. 5 is a kind of cascade classifier design diagram provided in an embodiment of the present invention.
Specific embodiment
Embodiments of the present invention are described below in detail, the example of the embodiment is shown in the accompanying drawings, wherein from beginning
Same or similar element or element with the same or similar functions are indicated to same or similar label eventually.Below by ginseng
The embodiment for examining attached drawing description is exemplary, and for explaining only the invention, and is not construed as limiting the claims.
Those skilled in the art of the present technique are appreciated that unless expressly stated, singular " one " used herein, " one
It is a ", " described " and "the" may also comprise plural form.It is to be further understood that being arranged used in specification of the invention
Diction " comprising " refer to that there are the feature, integer, step, operation, element and/or component, but it is not excluded that in the presence of or addition
Other one or more features, integer, step, operation, element, component and/or their group.It should be understood that when we claim member
Part is " connected " or when " coupled " to another element, it can be directly connected or coupled to other elements, or there may also be
Intermediary element.In addition, " connection " used herein or " coupling " may include being wirelessly connected or coupling.Wording used herein
"and/or" includes one or more associated any cells for listing item and all combinations.
Those skilled in the art of the present technique are appreciated that unless otherwise defined, all terms used herein (including technology art
Language and scientific term) there is meaning identical with the general understanding of those of ordinary skill in fields of the present invention.Should also
Understand, those terms such as defined in the general dictionary, which should be understood that, to be had and the meaning in the context of the prior art
The consistent meaning of justice, and unless defined as here, it will not be explained in an idealized or overly formal meaning.
In order to facilitate understanding of embodiments of the present invention, it is done by taking several specific embodiments as an example below in conjunction with attached drawing further
Explanation, and each embodiment does not constitute the restriction to the embodiment of the present invention.
The embodiment of the present invention is set mainly for the character feature extracting method and recognition classifier in character recognition process
Meter.The embodiment of the present invention is based on Gabor filter design level characteristics models, and the overall situation that not only can reflect character picture is special
Sign, moreover it is possible to the local feature of character stroke is shown, to achieve the purpose that extract character picture validity feature.Classifier is set
Meter is also the important ring of OCR, and good classifier design can be reasonably using the character picture feature being drawn into, and it is accurate to obtain
Reliable character identification result.With the increase of classification, the identification error rate of single layer classifier can also be greatly increased.Base of the present invention
In the thought of image classification, a kind of grade of SVM (support vector machine, support vector machines) classifier design is utilized
The model for joining classifier, realizes a kind of recognition methods step by step first sorted out and identified afterwards, reduces single layer classification to a certain extent
The class number of device, to improve recognition accuracy.
The embodiment of the present invention proposes the level characteristics model and cascade classifier model for character picture identification, the party
Pretreatment is normalized to input picture first in method, and different directions and different scale feature can be extracted by then devising
Gabor filtering core, and image is filtered.It is designed followed by hierarchical model, and extracts character picture feature.
Cascade classifier is finally designed, classified according to cascade classifier to sample data and is trained step by step.
A kind of process flow of the recognition methods for character picture that the embodiment of the present invention proposes is as shown in Figure 1, include as follows
Processing step:
Step S110, the pretreatment of character picture.
Original character image first pair is normalized.Original character image is generally length and width not equal rectangle,
Gray scale filling is carried out to original character image first, is made using the length of original character image longest edge as the pros of side length
Shape image block is finally normalized to 32 × 32 floating type image array.
Note original character image is Image, and the image after gray processing is ImageGray, and the longest edge of ImageGray is taken to remember
For LenSide, calculates the gray average that ImageGray is located at four angular vertexs and be denoted as meanVal, at ImageGray two
Identical quantity is filled in the outside of long side, and gray value is the pixel of meanVald, and ImageGray is made to become a size
The image array of LenSide × LenSide, is denoted as ImageSquare.A kind of expression original character provided in an embodiment of the present invention
Image, and the schematic diagram of gray scale filling is carried out as shown in Fig. 2, the outside of arrow meaning black line is to original character image
The region being filled.Finally, ImageSquare to be normalized to 32 × 32 floating type image array.
Based on the size of different original images, above-mentioned size 32 × 32 be can be varied (it is appropriate increase or
Reduce), but as obtained by experiment, the picture on A4 paper is shot for most of mobile phone or camera, each word in figure
Size is close to 32 × 32, and being normalized to such a size is to be conducive to the reservation of original image information, so the size
Setting be relatively reasonable.
Step S120, the design of Gabor filtering core and the filtering processing of character picture.
Gabor (adding primary) filtering is the important filter function of detection image textural characteristics, and the stroke feature of character can also be with
Regard a kind of textural characteristics as, so, this method devises 12 Gabor filtering cores, uses this 12 Gabor filtering cores respectively
Convolution algorithm is carried out to the floating type image array ImageSquare, 3 kinds of different scales on 4 directions can be extracted
12 character feature images in total, kinds of characters characteristic image include different textural characteristics, by all character feature figures
As composition image array ImageDst.
The kernel function of Gabor filtering is as follows:
X '=x cos θ+y sin θ
Y '=- x sin θ+y cos θ
Wherein θ indicates the direction of filtering core,Indicate the phase of cosine part, γ indicates the space aspect ratio of filtering core, λ
Indicate that the wavelength of cosine part, σ indicate the standard deviation in Gaussian function.Take σ=π/3, γ=0.5,θ=0, π/4,
Pi/2,3 π/4, λ=2,4,6, it is available after combination to represent 4 directions, 12 filtering kernel functions of 3 scales, setting
The template size of kernel function is 11 × 11, and constructs corresponding one group of kernel function template kernelGroup.
The present invention filters core by the different parameter combinations of the setting Gabor kernel function filtered to obtain one group of Gabor, into
And extraction characteristics of image effectively and reasonably.
As shown in Figure 3.Convolution algorithm is carried out to ImageSquare (Size32 × 32) respectively using this 12 filtering cores,
12 one group of image arrays comprising different texture feature have been obtained, ImageDst is denoted as.
Step S130, the design of hierarchical model and the feature extraction of character picture.
In order to preferably extract the global characteristics of character and the local feature of stroke, this method gradually will include textural characteristics
Image be subdivided into several pieces, acquire every piece of mean value and standard deviation every time, and by the mean value of the block of all same sizes and side
Differential pressure enters feature vector, the characteristic value as this layer.
In order to extract the global characteristics and local feature of character simultaneously, Fig. 4 is one kind provided in an embodiment of the present invention
Hierarchical Design model schematic.
Remember that the corresponding feature vector of Image is Features, seeks every characteristic image in one group of image array ImageDst
Feature of the mean value and standard deviation of all pixels point as first layer, and the feature of the first layer is arranged to feature vector
In Features.
Then, every characteristic image in one group of image array ImageDst is divided into the image moment that 4 sizes are 16 × 16
Battle array, acquires the mean value and standard deviation of each image array (Size16 × 16), by each image array (Size16 × 16) respectively
Feature as the second layer of mean value and standard deviation, and the feature of the second layer is arranged into feature vector Features.
Finally, every characteristic image in one group of image array ImageDst is divided into the image moment that 16 sizes are 8 × 8
Battle array, acquires the mean value and standard deviation of each image array (Size8 × 8), by the equal of each image array (Size8 × 8) respectively
Value and feature of the standard deviation as third layer, and the feature of the third layer is arranged into feature vector Features.
In practical applications, it can also continue to be subdivided into more layers, at least two layers of this method, but be not limited only to two layers, than
Such as, can there are the 4th layer, layer 5 feature.
In this way, we just obtained not only comprising global characteristics but also include local feature feature vector Features, will wrap
The feature vector Features of multilayer feature information is contained as the data sample for having extracted feature.
Step S140, the design and training of cascade classifier.
In order to solve the problems, such as single layer classifier output classification it is excessive, this method first by sample according to natural quality step by step
Classify, and the label on appropriate level is set for sample.Then top-down, training point is gone according to the classification of current level
Class device finally obtains the multi-level classifier with cascade connection.When using cascade classifier, equally using top-down
Process select the classifier of next stage, and finally obtain desired result every time according to the classification results of sample.
In order to reduce the output classification of single layer classifier, recognition accuracy is improved, the present invention is arranged according to sample attribute to be had
There is the multi-level cascade classifier of cascade connection.Each layer includes multiple classifiers, and the classifier on each upper layer cascades one
Or the classifier of multiple lower layers, sub-classification device are the refinements to its cascade upper layer classifier, each classifier is right respectively
A kind of sample attribute of classification is answered, the corresponding classifier of the other training data of every type is chosen respectively and is trained, trained
Good cascade classifier.The cascade classifier is based on SVM classifier and constructs.
It is layer-by-layer from different perspectives to the data sample using trained cascade classifier according to top-down sequence
Classify, when the data sample corresponds to some classifier in the classification results of a upper level, then some is classified described in selection
Cascade next layer of the classifier of device carries out further classification to the data sample, until the classifier of the bottom is complete
The classification processing of the pairs of data sample carries out classification results of the classifier on all levels to the sample data comprehensive
It closes, obtains the character identification result of the original character image Image.
Firstly, being classified step by step from different perspectives according to the natural quality of sample to sample, each layer of classification can be regarded as
Level-one, every level-one are all the subdivisions to upper level.For the data sample of all extraction features, i.e. feature vector
Features marks i label { y for it1,y2j,…,yij, yijBelonging to indicating the sample under j-th of classifier of i-stage
Classification.We are basic model with SVM classifier, have label y using allijTraining data training j-th of i-stage classification
Device.During identification, equally use top-down process, by classifier i-stage classification results, to determine to survey
The classifier that sample sheet should be selected in i+1 grade finally completes the process of identification character recognition on afterbody classifier.
Fig. 5 is a kind of cascade classifier design diagram provided in an embodiment of the present invention, and first order classifier is by feature sample
Originally be divided into four classes --- Chinese character, English, number and other, if input is that " Chinese " image block by first order classifier judges it
The first order is classified as " Chinese character ", and then is entered second level classifier and judged that its specific category is by second level classifier
" Chinese ".
In conclusion thought of the embodiment of the present invention based on image classification, is devised a kind of multi-level using SVM classifier
Cascade classifier model, realize a kind of recognition methods step by step first sorted out and identified afterwards, reduce single layer point to a certain extent
The class number of class device, to improve recognition accuracy.The method of the embodiment of the present invention is succinct, strong operability.
The embodiment of the present invention is by the design to Gabor filtering core and hierarchical model, so that the character picture extracted is special
Sign is more significant, is more advantageous to the identification of image.The design of cascade classifier is substantially reduced using the thought divided and rule
The complicated classification degree of single classifier compared with single classifier can greatly improve the accuracy rate of image recognition.
Those of ordinary skill in the art will appreciate that: attached drawing is the schematic diagram of one embodiment, module in attached drawing or
Process is not necessarily implemented necessary to the present invention.
As seen through the above description of the embodiments, those skilled in the art can be understood that the present invention can
It realizes by means of software and necessary general hardware platform.Based on this understanding, technical solution of the present invention essence
On in other words the part that contributes to existing technology can be embodied in the form of software products, the computer software product
It can store in storage medium, such as ROM/RAM, magnetic disk, CD, including some instructions are used so that a computer equipment
(can be personal computer, server or the network equipment etc.) executes the certain of each embodiment or embodiment of the invention
Method described in part.
All the embodiments in this specification are described in a progressive manner, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for device or
For system embodiment, since it is substantially similar to the method embodiment, so describing fairly simple, related place is referring to method
The part of embodiment illustrates.Apparatus and system embodiment described above is only schematical, wherein the conduct
The unit of separate part description may or may not be physically separated, component shown as a unit can be or
Person may not be physical unit, it can and it is in one place, or may be distributed over multiple network units.It can root
According to actual need that some or all of the modules therein is selected to achieve the purpose of the solution of this embodiment.Ordinary skill
Personnel can understand and implement without creative efforts.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
In the technical scope disclosed by the present invention, any changes or substitutions that can be easily thought of by anyone skilled in the art,
It should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with scope of protection of the claims
Subject to.
Claims (4)
1. a kind of recognition methods of character picture characterized by comprising
Original character image Image is normalized to obtain floating type image array ImageSquare；
Convolution algorithm is carried out to the floating type image array ImageSquare using multiple filtering cores, is obtained comprising different lines
Manage one group of image array ImageDst of feature；
The corresponding feature vector Features of the original character image Image is obtained, from one group of image array
Multilayer feature is extracted in ImageDst, and the multilayer feature is arranged into described eigenvector Features, data are obtained
Sample；
Layer-by-layer classification processing is carried out to the data sample using the multi-level classifier with cascade connection, is identified described
Character in original character image Image；
The corresponding feature vector Features of the acquisition original character image Image, from one group of image array
Multilayer feature is extracted in group ImageDst, the multilayer feature is arranged into described eigenvector Features, is counted
According to sample, comprising: the corresponding feature vector of note image Image is Features, calculates separately out one group of image array
In ImageDst in every characteristic image all pixels point mean value and standard deviation and the feature as first layer, by described first
The feature of layer is arranged into feature vector Features；Every characteristic image in one group of image array ImageDst is divided into
The image array that 4 sizes are 16 × 16 calculates separately out the mean value and standard deviation and conduct of each 16 × 16 image array
The feature of the second layer is arranged into feature vector Features the feature of the second layer；By one group of image array
Every characteristic image is divided into the image array that 16 sizes are 8 × 8 in ImageDst, calculates separately out each 8 × 8 image moment
The mean value and standard deviation and the feature as third layer of battle array, by the feature setting of the third layer to feature vector Features
In；The feature vector Features of three layers of characteristic information will be contained as data sample.
2. according to the method described in claim 1, described be normalized to obtain floating-point to original character image Image
Type image array ImageSquare, comprising:
If original character image is Image, image Image gray processing treated image is ImageGray, takes image
The longest edge of ImageGray is denoted as LenSide, calculates gray average of the image ImageGray at four angular vertexs
MeanVal, the outside of two long sides of image ImageGray fill identical quantity, gray value be meanVald pixel,
Image ImageGray is set to become the image array of a size LenSide × LenSide, then by described image matrix normalizing
Turn to 32 × 32 floating type image array ImageSquare.
3. the recognition methods of character picture according to claim 2, which is characterized in that described to use multiple filtering cores pair
The floating type image array ImageSquare carries out convolution algorithm, obtains one group of image array comprising different texture feature
ImageDst, comprising:
12 Gabor filtering cores are designed, the kernel function of the Gabor filtering is as follows:
X '=x cos θ+y sin θ
Y '=- x sin θ+y cos θ
Wherein θ indicates the direction of filtering core,Indicate the phase of cosine part, γ indicates the space aspect ratio of filtering core, and λ is indicated
The wavelength of cosine part, σ indicate the standard deviation in Gaussian function；
Convolution algorithm is carried out to the floating type image array ImageSquare using 12 Gabor filtering cores respectively, is mentioned
12 character feature images in total of 3 kinds of different scales on 4 directions are got, kinds of characters characteristic image includes different line
Feature is managed, all character feature images are formed into one group of image array ImageDst.
4. the recognition methods of character picture according to claim 1, which is characterized in that the use has cascade connection
Multi-level classifier layer-by-layer classification processing is carried out to the data sample, identify in the original character image Image
Character, comprising:
There is the multi-level classifier of cascade connection according to sample attribute setting, each layer includes multiple classifiers, Mei Geshang
The classifier of layer cascades the classifier of one or more lower layer, and sub-classification device is to the thin of its cascade upper layer classifier
Change, each classifier respectively corresponds a kind of sample attribute of classification, chooses the other training data of every type respectively to corresponding point
Class device is trained, and obtains trained cascade classifier；
It is layer-by-layer from different perspectives to the data sample using the trained cascade classifier according to top-down sequence
Classify, when the data sample corresponds to some classifier in the classification results of a upper level, then some is classified described in selection
Cascade next layer of the classifier of device carries out further classification to the data sample, until the classifier of the bottom is complete
The classification processing of the pairs of data sample carries out classification results of the classifier on all levels to the sample data comprehensive
It closes, obtains the character identification result of the original character image Image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610603330.9A CN106228166B (en) | 2016-07-27 | 2016-07-27 | The recognition methods of character picture |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610603330.9A CN106228166B (en) | 2016-07-27 | 2016-07-27 | The recognition methods of character picture |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106228166A CN106228166A (en) | 2016-12-14 |
| CN106228166B true CN106228166B (en) | 2019-05-21 |
Family
ID=57534087
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610603330.9A Active CN106228166B (en) | 2016-07-27 | 2016-07-27 | The recognition methods of character picture |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106228166B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106909894B (en) * | 2017-02-14 | 2018-08-14 | 北京深瞐科技有限公司 | Vehicle brand type identifier method and system |
| CN108734159B (en) * | 2017-04-18 | 2022-06-03 | 苏宁易购集团股份有限公司 | Method and system for detecting sensitive information in image |
| CN109426831B (en) * | 2017-08-30 | 2022-12-13 | 腾讯科技（深圳）有限公司 | Image similarity matching and model training method and device and computer equipment |
| CN110503090B (en) * | 2019-07-09 | 2021-11-09 | 中国科学院信息工程研究所 | Character detection network training method based on limited attention model, character detection method and character detector |
| CN111488843A (en) * | 2020-04-16 | 2020-08-04 | 贵州安防工程技术研究中心有限公司 | Face sunglasses distinguishing method based on step-by-step inhibition of missing report and false report rate |
| CN111783787B (en) * | 2020-07-10 | 2023-08-25 | 上海淇毓信息科技有限公司 | Method and device for recognizing image characters and electronic equipment |
| CN112560893B (en) * | 2020-11-13 | 2022-11-04 | 贝壳技术有限公司 | Picture texture matching method and device, electronic medium and storage medium |
| CN113781607B (en) * | 2021-09-17 | 2023-09-19 | 平安科技（深圳）有限公司 | Processing method, device, equipment and storage medium for labeling data of OCR (optical character recognition) image |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101697228A (en) * | 2009-10-15 | 2010-04-21 | 东莞市步步高教育电子产品有限公司 | Method for processing text images |
| CN102509112A (en) * | 2011-11-02 | 2012-06-20 | 珠海逸迩科技有限公司 | Number plate identification method and identification system thereof |
| CN103996057A (en) * | 2014-06-12 | 2014-08-20 | 武汉科技大学 | Real-time handwritten digital recognition method based on multi-feature fusion |
| US9014481B1 (en) * | 2014-04-22 | 2015-04-21 | King Fahd University Of Petroleum And Minerals | Method and apparatus for Arabic and Farsi font recognition |
| CN105760891A (en) * | 2016-03-02 | 2016-07-13 | 上海源庐加佳信息科技有限公司 | Chinese character verification code recognition method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7724957B2 (en) * | 2006-07-31 | 2010-05-25 | Microsoft Corporation | Two tiered text recognition |
-
2016
- 2016-07-27 CN CN201610603330.9A patent/CN106228166B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101697228A (en) * | 2009-10-15 | 2010-04-21 | 东莞市步步高教育电子产品有限公司 | Method for processing text images |
| CN102509112A (en) * | 2011-11-02 | 2012-06-20 | 珠海逸迩科技有限公司 | Number plate identification method and identification system thereof |
| US9014481B1 (en) * | 2014-04-22 | 2015-04-21 | King Fahd University Of Petroleum And Minerals | Method and apparatus for Arabic and Farsi font recognition |
| CN103996057A (en) * | 2014-06-12 | 2014-08-20 | 武汉科技大学 | Real-time handwritten digital recognition method based on multi-feature fusion |
| CN105760891A (en) * | 2016-03-02 | 2016-07-13 | 上海源庐加佳信息科技有限公司 | Chinese character verification code recognition method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106228166A (en) | 2016-12-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106228166B (en) | The recognition methods of character picture | |
| CN110033000B (en) | Text detection and identification method for bill image | |
| Zalpour et al. | A new approach for oil tank detection using deep learning features with control false alarm rate in high-resolution satellite imagery | |
| CN108830209B (en) | Remote sensing image road extraction method based on generation countermeasure network | |
| Epshtein et al. | Detecting text in natural scenes with stroke width transform | |
| CN110298338A (en) | A kind of file and picture classification method and device | |
| CN106446954A (en) | Character recognition method based on depth learning | |
| CN109784197B (en) | Pedestrian re-identification method based on hole convolution and attention mechanics learning mechanism | |
| CN107844740A (en) | A kind of offline handwriting, printing Chinese character recognition methods and system | |
| CN103544504B (en) | Scene character recognition method based on multi-scale map matching core | |
| CN108681735A (en) | Optical character recognition method based on convolutional neural networks deep learning model | |
| CN109993201A (en) | A kind of image processing method, device and readable storage medium storing program for executing | |
| CN103136504A (en) | Face recognition method and device | |
| CN106548169A (en) | Fuzzy literal Enhancement Method and device based on deep neural network | |
| CN103810274A (en) | Multi-feature image tag sorting method based on WordNet semantic similarity | |
| CN111914838A (en) | License plate recognition method based on text line recognition | |
| CN111753873A (en) | Image detection method and device | |
| CN107085731A (en) | A kind of image classification method based on RGB D fusion features and sparse coding | |
| CN107220598A (en) | Iris Texture Classification based on deep learning feature and Fisher Vector encoding models | |
| CN109614866A (en) | Method for detecting human face based on cascade deep convolutional neural networks | |
| CN110110610A (en) | A kind of event detecting method for short-sighted frequency | |
| Pirrone et al. | Papy-s-net: A siamese network to match papyrus fragments | |
| CN110188750A (en) | A kind of natural scene picture character recognition method based on deep learning | |
| Kobchaisawat et al. | Thai text localization in natural scene images using convolutional neural network | |
| CN106650696A (en) | Handwritten electrical element identification method based on singular value decomposition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |