CN106228166A - The recognition methods of character picture - Google Patents
The recognition methods of character picture Download PDFInfo
- Publication number
- CN106228166A CN106228166A CN201610603330.9A CN201610603330A CN106228166A CN 106228166 A CN106228166 A CN 106228166A CN 201610603330 A CN201610603330 A CN 201610603330A CN 106228166 A CN106228166 A CN 106228166A
- Authority
- CN
- China
- Prior art keywords
- image
- feature
- character
- grader
- image array
- 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.)
- Granted
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)
- Character Input (AREA)
- Image Analysis (AREA)
Abstract
The invention provides the recognition methods of a kind of character picture.The method specifically includes that to be normalized original character image Image and obtains floating type image array ImageSquare;Use multiple filtering core that floating type image array ImageSquare is carried out convolution algorithm, obtain the one group of image array ImageDst comprising different texture feature;Obtain original character image Image characteristic of correspondence vector Features, from one group of image array ImageDst, extract multilayer feature, multilayer feature is arranged in characteristic vector Features, obtain data sample;Use the multi-level grader with cascade connection that data sample carries out successively classification to process, identify the character in original character image Image.The thought that the present invention classifies based on image, SVM classifier is utilized to devise a kind of multi-level cascade classifier model, achieve a kind of recognition methods step by step first sorted out and identify afterwards, decrease the class number of monolayer grader to a certain extent, thus improve recognition accuracy.The method of the present invention is succinct, workable.
Description
Technical field
The present invention relates to character recognition technologies field, particularly relate to the recognition methods of a kind of character picture.
Background technology
Optical character recognition (Optical Character Recognition, OCR) is pattern recognition (Pattern
Recognition) an important research direction in field, its main purpose be by hand-written, print, scan or text in image
Be converted to machine code, be a kind of digitized process of text message.In recent years, along with the development of science and technology, OCR technique not only existed
The Digitization of literature of ancient book plays great function, is also applied in many intelligent scene, as normal in life
The charge station's Car license recognition seen, and Google's glasses provide wearer according to the word seen and advise accordingly.Character recognition
Process is generally divided into several steps such as Image semantic classification, Character segmentation, character recognition and post processing, and each step is required for multiple
Technology participates in jointly.
Abroad the research work of OCR technique is started to walk relatively early, and English character has total amount character stroke little, single to connect
Deng distinct characteristic, these features are all prone to from image extract.Chinese character recognition is then the most challenging in OCR technique
One content, one of them main difficulty is that the feature extraction of Chinese character.Chinese character is developed by pictograph, character
Quantity is many, font type is many, structure is complicated, and has a number of nearly word form, and these features both increase extraction Chinese character to be had
The difficulty of effect feature, thus have influence on recognition result.
The conventional method extracting character feature has statistical nature to extract, architectural feature is extracted.Statistical nature extracts main
It is, by methods such as projection, statistics with histogram, each pixel in image carried out half-tone information statistics, thus forms character
Characteristic vector.The method is poor to distinguishing property of details, it is impossible to effectively distinguish nearly word form.Feature extraction side based on structure
Method is mainly extracted the stroke information of character, combines the feature such as direction, cross point, can preferably identify detailed information, but
The most also can be the most sensitive to noise.In recent years due to the burning hot development of degree of depth study, occur in that and utilize neutral net to carry out
The method of feature extraction, although there is the accuracy of identification that comparison is high, but technology realizes cost height, and Chinese character quantity is various, god
Becoming sufficiently complex through the design of network, these are all unfavorable for the exploitation of middle-size and small-size OCR system.Based on the problems referred to above, the present invention
Propose a kind of hierarchical Design model based on Gabor filtering to be used for carrying out feature extraction.Gabor filtering is a kind of sensitive to texture
Method, it is possible to reflect image textural characteristics on different scale and direction, its extensive application image process in edge
Test problems.
Summary of the invention
The embodiment provides the recognition methods of a kind of character picture, to realize effectively knowing from character picture
Do not go out character.
To achieve these goals, this invention takes following technical scheme.
A kind of recognition methods of character picture, including:
Original character image Image is normalized and obtains floating type image array ImageSquare;
Use multiple filtering core that described floating type image array ImageSquare is carried out convolution algorithm, obtain comprising not
One group of image array ImageDst with textural characteristics;
Obtain described original character image Image characteristic of correspondence vector Features, from described one group of image array
ImageDst extracts multilayer feature, described multilayer feature is arranged in described characteristic vector Features, obtains data
Sample;
Use the multi-level grader with cascade connection that described data sample carries out successively classification to process, identify
Character in described original character image Image.
Further, described be normalized original character image Image obtains floating type image array
ImageSquare, including:
If original character image is Image, the image after image Image gray processing processes is ImageGray, takes image
The longest edge of ImageGray is designated as LenSide, calculates image ImageGray and is positioned at the gray average at four angular vertexs
MeanVal, two long limits of image ImageGray outside fill equal number, gray value be the pixel of meanVald,
Making image ImageGray become a size is the image array of LenSide × LenSide, then by described matrix image normalizing
Turn to the floating type image array ImageSquare of 32 × 32.
Further, the multiple filtering core of described use carries out convolution to described floating type image array ImageSquare
Computing, obtains the one group of image array ImageDst comprising different texture feature, including:
Designing 12 Gabor filtering cores, the kernel function of described Gabor filtering is as follows:
X '=x cos θ+y sin θ
Y '=-x sin θ+y cos θ
Wherein θ represents the direction of filtering core,Representing the phase place of cosine part, γ represents the space aspect ratio of filtering core, λ
Representing the wavelength of cosine part, σ represents the standard deviation in Gaussian function;
Use described 12 Gabor filtering cores that described floating type image array ImageSquare is carried out convolution fortune respectively
Calculating, extract 12 character feature images altogether of 3 kinds of different scales on 4 directions, kinds of characters characteristic image comprises not
Same textural characteristics, forms one group of image array ImageDst by all of character feature image.
Further, described acquisition described original character image Image characteristic of correspondence vector Features, from described
One group of image array group ImageDst extracts multilayer feature, described multilayer feature is arranged described characteristic vector
In Features, obtain data sample, including:
Note image Image characteristic of correspondence vector is Features, calculates described one group of image array respectively
The average of all pixels and standard deviation the feature as ground floor in every characteristic image in ImageDst, by described first
The feature of layer is arranged in characteristic vector Features;
It is divided into 4 sizes to be the image moment of 16 × 16 every characteristic image in described one group of image array ImageDst
Battle array, calculates the average of the image array of each 16 × 16 and standard deviation the feature as the second layer, respectively by described second
The feature of layer is arranged in characteristic vector Features;
It is divided into 16 sizes to be the image array of 8 × 8 every characteristic image in described one group of image array ImageDst,
Calculate the average of the image array of each 8 × 8 and standard deviation the feature as third layer respectively, by the spy of described third layer
Levy and arrange in characteristic vector Features;
Using contain three layers of characteristic information characteristic vector Features as data sample.
Further, described use has the multi-level grader of cascade connection and carries out described data sample successively
Classification processes, and identifies the character in described original character image Image, including:
Arrange the multi-level grader with cascade connection according to sample attribute, each layer includes multiple grader, often
The grader on individual upper strata cascades the grader of one or more lower floor, and sub classification device is the upper strata grader cascading it
Refinement, the sample attribute of each grader corresponding kind respectively, choose the training data of every kind respectively to correspondence
Grader is trained, and obtains the cascade classifier trained;
The cascade classifier trained described in using is according to top-down order, to described data sample from different perspectives
Successively classify, when described data sample is at classification results certain grader corresponding of last layer time, then choose described certain
The grader of next layer of grader cascade carries out classification further to described data sample, until the classification of the bottom
Device completes the classification to described data sample and processes, and the grader on all levels is entered the classification results of described sample data
Row is comprehensive, obtains the character identification result of described original character image Image.
The technical scheme that thered is provided by embodiments of the invention described above is it can be seen that what the embodiment of the present invention was classified based on image
Thought, utilizes SVM classifier to devise a kind of multi-level cascade classifier model, it is achieved that a kind of first sort out identify afterwards by
Level recognition methods, decreases the class number of monolayer grader to a certain extent, thus improves recognition accuracy.The present invention is real
The method executing example is succinct, workable.
Aspect and advantage that the present invention adds will part be given in the following description, and these will become from the following description
Obtain substantially, or recognized by the practice of the present invention.
Accompanying drawing explanation
In order to be illustrated more clearly that the technical scheme of the embodiment of the present invention, required use in embodiment being described below
Accompanying drawing be briefly described, it should be apparent that, below describe in accompanying drawing be only some embodiments of the present invention, for this
From the point of view of the those of ordinary skill of field, on the premise of not paying creative work, it is also possible to obtain other according to these accompanying drawings
Accompanying drawing.
A kind of level characteristics model for character picture identification that Fig. 1 provides for the embodiment of the present invention and cascade classifier
Design flow diagram.
A kind of expression original character image that Fig. 2 provides for the embodiment of the present invention, and original character image is carried out gray scale
The schematic diagram filled.
The visualization signal of a kind of kernelGroup being made up of 12 filtering cores that Fig. 3 provides for the embodiment of the present invention
Figure.
A kind of hierarchical Design model schematic that Fig. 4 provides for the embodiment of the present invention.
A kind of cascade classifier design diagram that Fig. 5 provides for the embodiment of the present invention.
Detailed description of the invention
Embodiments of the present invention are described below in detail, and the example of described embodiment is shown in the drawings, the most ad initio
Represent same or similar element to same or similar label eventually or there is the element of same or like function.Below by ginseng
The embodiment examining accompanying drawing description is exemplary, is only used for explaining the present invention, and is not construed as limiting the claims.
Those skilled in the art of the present technique are appreciated that unless expressly stated, singulative used herein " ", "
Individual ", " described " and " being somebody's turn to do " may also comprise plural form.It is to be further understood that use in the description of the present invention arranges
Diction " including " refers to there is described feature, integer, step, operation, element and/or assembly, but it is not excluded that existence or adds
Other features one or more, integer, step, operation, element, assembly and/or their group.It should be understood that when we claim unit
Part is " connected " or during " coupled " to another element, and it can be directly connected or coupled to other elements, or can also exist
Intermediary element.Additionally, " connection " used herein or " coupling " can include wireless connections or couple.Wording used herein
"and/or" includes one or more any cell listing item being associated and all combinations.
Those skilled in the art of the present technique are appreciated that unless otherwise defined, and all terms used herein (include technology art
Language and scientific terminology) have with the those of ordinary skill in art of the present invention be commonly understood by identical meaning.Also should
Being understood by, those terms defined in such as general dictionary should be understood that the meaning having with the context of prior art
The meaning that justice is consistent, and unless defined as here, will not explain by idealization or the most formal implication.
For ease of the understanding to the embodiment of the present invention, do as a example by several specific embodiments further below in conjunction with accompanying drawing
Explanation, and each embodiment is not intended that the restriction to the embodiment of the present invention.
The embodiment of the present invention sets mainly for the character feature extracting method in character recognition process and recognition classifier
Meter.The embodiment of the present invention is level characteristics model based on Gabor filter design, can not only reflect that the overall situation of character picture is special
Levy, moreover it is possible to show the local feature of character stroke, thus reach to extract the purpose of character picture validity feature.Setting of grader
Meter is also a ring important for OCR, and good classifier design can reasonably utilize the character picture feature being drawn into, and draws accurately
Character identification result reliably.Along with the increase of classification, the identification error rate of monolayer grader also can be greatly increased.Base of the present invention
In the thought of image classification, utilize a kind of level of SVM (support vector machine, support vector machine) classifier design
The model of connection grader, it is achieved that a kind of first sorts out the recognition methods step by step identified afterwards, decreases monolayer to a certain extent and classifies
The class number of device, thus improve recognition accuracy.
The embodiment of the present invention proposes the level characteristics model for character picture identification and cascade classifier model, the party
First method is normalized pretreatment to input picture, then devises and can extract different directions and different scale feature
Gabor filtering core, and be filtered image processing.It is followed by hierarchical model is designed, and extracts character picture feature.
Finally design cascade classifier, according to cascade classifier sample data classified and train step by step.
The handling process of the recognition methods of a kind of character picture that the embodiment of the present invention proposes is as it is shown in figure 1, include as follows
Process step:
Step S110, the pretreatment of character picture.
First to original character image be normalized.Original character image is generally the rectangle that length and width do not wait,
First original character image is carried out gray scale filling, make the pros of a length of length of side with original character image longest edge
Shape image block, is finally normalized to the floating type image array of 32 × 32.
Note original character image is Image, and the image after gray processing is ImageGray, takes the longest edge note of ImageGray
For LenSide, the gray average that calculating ImageGray is positioned at four angular vertexs is designated as meanVal, ImageGray two
Equal number is filled in the outside on long limit, and gray value is the pixel of meanVald, makes ImageGray become a size and is
The image array of LenSide × LenSide, is designated as ImageSquare.A kind of expression original character that the embodiment of the present invention provides
Image, and original character image is carried out the schematic diagram of gray scale filling as in figure 2 it is shown, the outside of arrow indication black line is
The region being filled.Finally, ImageSquare is normalized to the floating type image array of 32 × 32.
Size based on different original images, above-mentioned size 32 × 32 be can be varied from (suitable increase or
Reduce), but by experiment gained, for the picture on major part mobile phone or camera shooting A4 paper, each word in figure
Size, close to 32 × 32, is normalized to such a size and is advantageous for the reservation of original image information, so this size
Setting be relatively reasonable.
Step S120, the design of Gabor filtering core and the Filtering Processing of character picture.
Gabor (adding the primary) filtering is the important filter function of detection image texture characteristic, and the stroke feature of character can also
Regard a kind of textural characteristics as, so, this method devises 12 Gabor filtering cores, uses this 12 Gabor filtering cores respectively
Described floating type image array ImageSquare is carried out convolution algorithm, 3 kinds of different scales on 4 directions can be extracted
12 character feature images altogether, kinds of characters characteristic image comprises different textural characteristics, by all of character feature figure
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 represents the direction of filtering core,Representing the phase place of cosine part, γ represents the space aspect ratio of filtering core, λ
Representing the wavelength of cosine part, σ represents 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, can obtain after combination representing 4 directions, 12 filtering core functions of 3 yardsticks, set
The template size of kernel function is 11 × 11, and constructs corresponding one group of kernel function template kernelGroup.
The present invention obtains one group of Gabor filter core by the different parameter group arranging the kernel function of Gabor filtering is incompatible, enters
And extraction characteristics of image effectively and reasonably.
As shown in Figure 3.These 12 filtering cores are used respectively ImageSquare (Size32 × 32) to be carried out convolution algorithm,
Obtain 12 one group of image arrays comprising different texture feature, be denoted as ImageDst.
Step S130, the design of hierarchical model and the feature extraction of character picture.
The local feature of global characteristics Yu stroke in order to preferably extract character, this method progressively will comprise textural characteristics
Image be subdivided into some pieces, try to achieve average and the standard deviation of every piece, and by the average of the block of all formed objects and side every time
Differential pressure enters characteristic vector, as the eigenvalue of this layer.
In order to extract global characteristics and the local feature of character, the one that Fig. 4 provides for the embodiment of the present invention simultaneously
Hierarchical Design model schematic.
Note Image characteristic of correspondence vector is Features, seeks every characteristic image in one group of image array ImageDst
The average of all pixels and standard deviation are as the feature of ground floor, and the feature of this ground floor is arranged characteristic vector
In Features.
Then, it is divided into 4 sizes to be the image moment of 16 × 16 every characteristic image in one group of image array ImageDst
Battle array, tries to achieve average and the standard deviation of each image array (Size16 × 16), respectively by each image array (Size16 × 16)
Average and standard deviation as the feature of the second layer, and the feature of this second layer is arranged in characteristic vector Features.
Finally, it is divided into 16 sizes to be the image moment of 8 × 8 every characteristic image in one group of image array ImageDst
Battle array, tries to achieve average and the standard deviation of each image array (Size8 × 8) respectively, equal by each image array (Size8 × 8)
Value and standard deviation are as the feature of third layer, and the feature of this third layer are arranged in characteristic vector Features.
In actual applications, it is also possible to continue to be subdivided into more layers, the method at least two-layer, but it is not limited only to two-layer, than
As, can there be the 4th layer, layer 5 feature.
So, we have just obtained not only comprising global characteristics but also comprise characteristic vector Features of local feature, will bag
Contain characteristic vector Features of multilayer feature information as the data sample extracting feature.
Step S140, the design of cascade classifier and training.
The problem too much in order to solve monolayer grader output classification, this method is first by sample according to natural quality step by step
Classify, and the label on appropriate level is set for sample.The most top-down, go training point according to the classification of current level
Class device, finally obtains the multi-level grader with cascade connection.When using cascade classifier, same employing is top-down
Process, every time according to the classification results to sample, select the grader of next stage, and finally give desired result.
In order to reduce the output classification of monolayer grader, improving recognition accuracy, the present invention arranges tool according to sample attribute
There is the multi-level cascade classifier of cascade connection.Each layer includes multiple grader, and the grader on each upper strata cascades one
Or the grader of multiple lower floors, sub classification device is the refinement to the upper strata grader that it cascades, and each grader is the most right
Answering the sample attribute of a kind, the grader choosing the training data of every kind respectively corresponding is trained, and is trained
Good cascade classifier.This cascade classifier builds based on SVM classifier.
The cascade classifier that use trains is according to top-down order, to described data sample the most successively
Classify, when described data sample is at classification results certain grader corresponding of last layer time, then choose certain classification described
The grader of next layer of device cascade carries out classification further to described data sample, until the grader of the bottom is complete
The classification of paired described data sample processes, and the grader on all levels is combined the classification results of described sample data
Close, obtain the character identification result of described original character image Image.
First, according to the natural quality of sample, classifying sample the most step by step, each layer of classification can be regarded as
One-level, every one-level is all the segmentation to upper level.For the data sample of all extraction features, i.e. characteristic vector
Features, marks i label { y for it1,y2j,…,yij, yijRepresent belonging to this sample is under i-stage jth grader
Classification.We are model based on SVM classifier, uses and all has label yijTraining data training i-stage jth classification
Device.During identifying, the top-down process of same employing, by grader at the classification results of i-stage, determine to survey
This grader that should select in i+1 level of sample, finally completes to identify the process of character recognition on afterbody grader.
A kind of cascade classifier design diagram that Fig. 5 provides for the embodiment of the present invention, first order grader is by feature sample
Originally be divided into four class Chinese characters, English, numeral and other, if input is for " Chinese " image block, by first order grader, it is judged that its
The first order is categorized as " Chinese character ", and then enters second level grader, by second level grader, it is judged that its specific category is
" Chinese ".
In sum, the thought that the embodiment of the present invention is classified based on image, utilize SVM classifier to devise a kind of multi-level
Cascade classifier model, it is achieved that a kind of first sort out the recognition methods step by step identified afterwards, decrease monolayer to a certain extent and divide
The class number of class device, thus improve recognition accuracy.The method of the embodiment of the present invention is succinct, workable.
The embodiment of the present invention is by the design to Gabor filtering core and hierarchical model so that the character picture of extraction is special
Levy more significantly, be more beneficial for the identification of image.The design of cascade classifier, substantially uses the thought divided and rule to reduce
The complicated classification degree of single grader, compared with single grader, can be greatly enhanced the accuracy rate of image recognition.
One of ordinary skill in the art will appreciate that: accompanying drawing is the schematic diagram of an embodiment, module in accompanying drawing or
Flow process is not necessarily implemented necessary to the present invention.
As seen through the above description of the embodiments, those skilled in the art it can be understood that to the present invention can
The mode adding required general hardware platform by software realizes.Based on such understanding, technical scheme essence
On the part that in other words prior art contributed can embody with the form of software product, this computer software product
Can be stored in storage medium, such as ROM/RAM, magnetic disc, CD etc., including some instructions with so that a computer equipment
(can be personal computer, server, or the network equipment etc.) performs some of each embodiment of the present invention or embodiment
Method described in part.
Each embodiment in this specification all uses the mode gone forward one by one to describe, identical similar portion between each embodiment
Dividing and see mutually, what each embodiment stressed is the difference with other embodiments.Especially for device or
For system embodiment, owing to it is substantially similar to embodiment of the method, so describing fairly simple, relevant part sees method
The part of embodiment illustrates.Apparatus and system embodiment described above is only schematically, wherein said conduct
The unit of separating component explanation can be or may not be physically separate, the parts shown as unit can be or
Person may not be physical location, i.e. may be located at a place, or can also be distributed on multiple NE.Can root
Factually border need select some or all of module therein to realize the purpose of the present embodiment scheme.Ordinary skill
Personnel, in the case of not paying creative work, are i.e. appreciated that and implement.
The above, the only present invention preferably detailed description of the invention, but protection scope of the present invention is not limited thereto,
Any those familiar with the art in the technical scope that the invention discloses, the change that can readily occur in or replacement,
All should contain within protection scope of the present invention.Therefore, protection scope of the present invention should be with scope of the claims
It is as the criterion.
Claims (5)
1. the recognition methods of a character picture, it is characterised in that including:
Original character image Image is normalized and obtains floating type image array ImageSquare;
Use multiple filtering core that described floating type image array ImageSquare is carried out convolution algorithm, obtain comprising different stricture of vagina
One group of image array ImageDst of reason feature;
Obtain described original character image Image characteristic of correspondence vector Features, from described one group of image array
ImageDst extracts multilayer feature, described multilayer feature is arranged in described characteristic vector Features, obtains data
Sample;
Use the multi-level grader with cascade connection that described data sample carries out successively classification to process, identify described
Character in original character image Image.
Method the most according to claim 1, described be normalized original character image Image obtains floating-point
Type image array ImageSquare, including:
If original character image is Image, the image after image Image gray processing processes is ImageGray, takes image
The longest edge of ImageGray is designated as LenSide, calculates image ImageGray and is positioned at the gray average at four angular vertexs
MeanVal, two long limits of image ImageGray outside fill equal number, gray value be the pixel of meanVald,
Making image ImageGray become a size is the image array of LenSide × LenSide, then by described matrix image normalizing
Turn to the floating type image array ImageSquare of 32 × 32.
The recognition methods of character picture the most according to claim 2, it is characterised in that the multiple filtering core pair of described use
Described floating type image array ImageSquare carries out convolution algorithm, obtains the one group of image array comprising different texture feature
ImageDst, including:
Designing 12 Gabor filtering cores, the kernel function of described Gabor filtering is as follows:
X '=x cos θ+y sin θ
Y '=-x sin θ+y cos θ
Wherein θ represents the direction of filtering core,Representing the phase place of cosine part, γ represents the space aspect ratio of filtering core, and λ represents
The wavelength of cosine part, σ represents the standard deviation in Gaussian function;
Use described 12 Gabor filtering cores that described floating type image array ImageSquare is carried out convolution algorithm respectively, carry
Getting 12 character feature images altogether of 3 kinds of different scales on 4 directions, kinds of characters characteristic image comprises different stricture of vaginas
Reason feature, forms one group of image array ImageDst by all of character feature image.
The recognition methods of character picture the most according to claim 3, it is characterised in that the described original character of described acquisition
Image Image characteristic of correspondence vector Features, extracts multilayer feature from described one group of image array group ImageDst,
Described multilayer feature is arranged in described characteristic vector Features, obtains data sample, including:
Note image Image characteristic of correspondence vector is Features, calculates respectively in described one group of image array ImageDst
In every characteristic image, the average of all pixels and standard deviation the feature as ground floor, set the feature of described ground floor
Put in characteristic vector Features;
It is divided into 4 sizes to be the image array of 16 × 16 every characteristic image in described one group of image array ImageDst, point
Do not calculate the average of the image array of each 16 × 16 and standard deviation the feature as the second layer, by the spy of the described second layer
Levy and arrange in characteristic vector Features;
It is divided into 16 sizes to be the image array of 8 × 8 every characteristic image in described one group of image array ImageDst, respectively
Calculate the average of the image array of each 8 × 8 and standard deviation the feature as third layer, the feature of described third layer is set
Put in characteristic vector Features;
Using contain three layers of characteristic information characteristic vector Features as data sample.
The recognition methods of character picture the most according to claim 4, it is characterised in that described use has cascade connection
Multi-level grader described data sample is carried out successively classification process, identify in described original character image Image
Character, including:
Arrange the multi-level grader with cascade connection according to sample attribute, each layer includes multiple grader, Mei Geshang
The grader of layer cascades the grader of one or more lower floor, and sub classification device is thin to the upper strata grader that it cascades
Changing, the sample attribute of each grader corresponding kind respectively, correspondence is divided by the training data choosing every kind respectively
Class device is trained, and obtains the cascade classifier trained;
The cascade classifier trained described in using is according to top-down order, to described data sample the most successively
Classify, when described data sample is at classification results certain grader corresponding of last layer time, then choose certain classification described
The grader of next layer of device cascade carries out classification further to described data sample, until the grader of the bottom is complete
The classification of paired described data sample processes, and the grader on all levels is combined the classification results of described sample data
Close, obtain the character identification result of described 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 true CN106228166A (en) | 2016-12-14 |
CN106228166B 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) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106909894A (en) * | 2017-02-14 | 2017-06-30 | 北京深瞐科技有限公司 | Vehicle brand type identifier method and system |
CN108734159A (en) * | 2017-04-18 | 2018-11-02 | 苏宁云商集团股份有限公司 | The detection method and system of sensitive information in a kind of image |
CN109426831A (en) * | 2017-08-30 | 2019-03-05 | 腾讯科技(深圳)有限公司 | The method, apparatus and computer equipment of picture Similarity matching and model training |
CN110503090A (en) * | 2019-07-09 | 2019-11-26 | 中国科学院信息工程研究所 | Character machining network training method, character detection method and character machining device based on limited attention model |
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 |
CN111783787A (en) * | 2020-07-10 | 2020-10-16 | 上海淇毓信息科技有限公司 | Method and device for identifying image characters and electronic equipment |
CN112560893A (en) * | 2020-11-13 | 2021-03-26 | 贝壳技术有限公司 | Picture texture matching method and device, electronic medium and storage medium |
CN113781607A (en) * | 2021-09-17 | 2021-12-10 | 平安科技(深圳)有限公司 | Method, device and equipment for processing annotation data of OCR (optical character recognition) image and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080025610A1 (en) * | 2006-07-31 | 2008-01-31 | Microsoft Corporation | Two tiered text recognition |
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 |
-
2016
- 2016-07-27 CN CN201610603330.9A patent/CN106228166B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080025610A1 (en) * | 2006-07-31 | 2008-01-31 | Microsoft Corporation | Two tiered text recognition |
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 |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106909894A (en) * | 2017-02-14 | 2017-06-30 | 北京深瞐科技有限公司 | Vehicle brand type identifier method and system |
CN108734159A (en) * | 2017-04-18 | 2018-11-02 | 苏宁云商集团股份有限公司 | The detection method and system of sensitive information in a kind of image |
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 |
CN109426831A (en) * | 2017-08-30 | 2019-03-05 | 腾讯科技(深圳)有限公司 | The method, apparatus and computer equipment of picture Similarity matching and model training |
CN110503090A (en) * | 2019-07-09 | 2019-11-26 | 中国科学院信息工程研究所 | Character machining network training method, character detection method and character machining device based on limited attention model |
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 |
CN111783787A (en) * | 2020-07-10 | 2020-10-16 | 上海淇毓信息科技有限公司 | Method and device for identifying image characters and electronic equipment |
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 |
CN112560893A (en) * | 2020-11-13 | 2021-03-26 | 贝壳技术有限公司 | Picture texture matching method and device, electronic medium and storage medium |
CN113781607A (en) * | 2021-09-17 | 2021-12-10 | 平安科技(深圳)有限公司 | Method, device and equipment for processing annotation data of OCR (optical character recognition) image 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 |
Also Published As
Publication number | Publication date |
---|---|
CN106228166B (en) | 2019-05-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106228166A (en) | The recognition methods of character picture | |
CN103514456B (en) | Image classification method and device based on compressed sensing multi-core learning | |
Singh et al. | From strings to things: Knowledge-enabled vqa model that can read and reason | |
CN107424159A (en) | Image, semantic dividing method based on super-pixel edge and full convolutional network | |
CN111091105A (en) | Remote sensing image target detection method based on new frame regression loss function | |
CN101520894B (en) | Method for extracting significant object based on region significance | |
CN110175613A (en) | Street view image semantic segmentation method based on Analysis On Multi-scale Features and codec models | |
Ahranjany et al. | A very high accuracy handwritten character recognition system for Farsi/Arabic digits using convolutional neural networks | |
CN103544504B (en) | Scene character recognition method based on multi-scale map matching core | |
CN108345850A (en) | The scene text detection method of the territorial classification of stroke feature transformation and deep learning based on super-pixel | |
CN109409384A (en) | Image-recognizing method, device, medium and equipment based on fine granularity image | |
CN106548169A (en) | Fuzzy literal Enhancement Method and device based on deep neural network | |
CN111652273B (en) | Deep learning-based RGB-D image classification method | |
CN108681735A (en) | Optical character recognition method based on convolutional neural networks deep learning model | |
CN103810274A (en) | Multi-feature image tag sorting method based on WordNet semantic similarity | |
CN113256649B (en) | Remote sensing image station selection and line selection semantic segmentation method based on deep learning | |
CN109886161A (en) | A kind of road traffic index identification method based on possibility cluster and convolutional neural networks | |
CN105989336A (en) | Scene recognition method based on deconvolution deep network learning with weight | |
CN115527095A (en) | Multi-scale target detection method based on combined recursive feature pyramid | |
CN115131797A (en) | Scene text detection method based on feature enhancement pyramid network | |
Xu et al. | Feature-based constraint deep CNN method for mapping rainfall-induced landslides in remote regions with mountainous terrain: An application to Brazil | |
CN117372898A (en) | Unmanned aerial vehicle aerial image target detection method based on improved yolov8 | |
CN109034213A (en) | Hyperspectral image classification method and system based on joint entropy principle | |
CN106227836A (en) | Associating visual concept learning system and method is supervised with the nothing of word based on image | |
CN117853955A (en) | Unmanned aerial vehicle small target detection method based on improved YOLOv5 |
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 |