CN107256378A - Language Identification and device - Google Patents
Language Identification and device Download PDFInfo
- Publication number
- CN107256378A CN107256378A CN201710272779.6A CN201710272779A CN107256378A CN 107256378 A CN107256378 A CN 107256378A CN 201710272779 A CN201710272779 A CN 201710272779A CN 107256378 A CN107256378 A CN 107256378A
- Authority
- CN
- China
- Prior art keywords
- images
- recognized
- pixel
- energy
- textural characteristics
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/263—Language identification
-
- 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
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/40—Document-oriented image-based pattern recognition
- G06V30/41—Analysis of document content
- G06V30/413—Classification of content, e.g. text, photographs or tables
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the present invention provides a kind of Language Identification and device.This method includes：The textural characteristics vector of images to be recognized is obtained using multi-wavelet transformation method, the textural characteristics vector of images to be recognized is inputted in predetermined SVM, obtain the languages of images to be recognized Chinese version, realize and textural characteristics vector is obtained using multi-wavelet transformation method, because the process that multi-wavelet transformation method obtains textural characteristics vector is very fast, therefore, the recognition efficiency of Language Identification provided in an embodiment of the present invention is higher.Simultaneously as the characteristic of multi-wavelet transformation method, the recognition accuracy of the Language Identification of the embodiment of the present invention is also higher.Therefore, the efficiency and accuracy rate of languages identification are improved using Language Identification provided in an embodiment of the present invention.
Description
Technical field
The present invention relates to machine learning techniques, more particularly to a kind of Language Identification and device.
Background technology
The demand interacted with people and computer increases, in order that man-machine interaction is more efficient, it would be desirable to have one
Individual system can handle the input of a variety of such as block letter and handwritten form document form.Although computer is widely used in almost
All spectra, but paper document still possesses consequence within the ensuing some time.In addition, the business of physical form
Industry document is widely present to meet a variety of demands in reality.For example, fax paper, show file in court etc..Cause
This, we in the urgent need to can automatically extract, analyze and store the information from existing paper form, and, when needed
The method and system safeguarded and accessed.All these problems are attributed to text image analysis.Languages identification is text image
An important component in process field, its application scenarios have text image sequence, search online comprising language-specific
Text document, design multilingual optical character identification (Optical Character Recognition, OCR) system and
Languages autotext searching system based on basic file etc..
At present, can based on image textural characteristics carry out languages identification.The existing languages identification based on textural characteristics
Method mainly has：Based on gray level co-occurrence matrix method, based on fractal dimension method and several based on Gabor filter method etc..This
A little methods are required to extract first the textural characteristics of image.It is that figure is extracted according to co-occurrence matrix based on gray level co-occurrence matrix method
The characteristic value of picture knows method for distinguishing so as to carry out languages.In practical application, because the calculating of gray level co-occurrence matrixes is more complicated, one
As the complexity of computing can be reduced by the pretreatment of image binaryzation.The gray level co-occurrence matrix of image after so handling
Three values can be produced, languages identification is can be used in using these three values as textural characteristics.Method based on fractal dimension it is basic
It by image understanding is multiple Fractal Set that theory, which is, and the textural characteristics of image pass through the calculating to corresponding Generalized Dimension Spectrum
.Method based on fractal dimension is handled text image as a fractal model, and the textural characteristics of image pass through
Fractal dimension that difference box-counting dimension is calculated is expressed, so as to complete the task of languages identification.Based on Gabor filter method
The multi-channel model based on Gabor filter is constructed, so as to carry out texture feature extraction to text image, it is possible to achieve languages are known
Other purpose.
But, there is the defect of computationally intensive, texture feature extraction time length, therefore, mesh in these above-mentioned methods
Preceding Language Identification is less efficient.
The content of the invention
To solve problems of the prior art, the present invention provides a kind of Language Identification and device, to improve language
Plant the efficiency of identification.
In a first aspect, the embodiment of the present invention provides a kind of Language Identification, including：
The textural characteristics vector of images to be recognized is obtained using multi-wavelet transformation method；
The textural characteristics vector of the images to be recognized is inputted in predetermined SVM, the images to be recognized is obtained
The languages of Chinese version.
In method as implied above, the use multi-wavelet transformation method obtains the textural characteristics vector of images to be recognized,
Including：
The images to be recognized is handled using filtering interpolation method, the coefficient matrix of the images to be recognized is obtained；
Using the first half row element of all rows in the coefficient matrix as the first scaling function of second order m ultiwavelet parameter,
Using the later half row element of all rows in the coefficient matrix as the second scaling function of second order m ultiwavelet parameter；
The second order m ultiwavelet is determined according to the parameter of the parameter of first scaling function and second scaling function；
The coefficient matrix is decomposed according to the second order m ultiwavelet, obtain the images to be recognized first is preset
The details subgraph of quantity and the ll channel of the second predetermined number；
The average energy value and energy variance of each details subgraph are determined respectively；
The textural characteristics of the average energy value and energy variance of each details subgraph as the images to be recognized are sweared
Amount.
In method as implied above, before the use filtering interpolation method is pre-processed to the images to be recognized, institute
Stating method also includes：
The energy of each pixel of the images to be recognized is normalized, obtains to be identified after normalization
Image；
The use filtering interpolation method is handled the images to be recognized, including：
The images to be recognized after the normalization is handled using filtering interpolation method.
In method as implied above, place is normalized in the energy of each pixel to the images to be recognized
Reason, obtains the images to be recognized after normalization, including：
According to formulaObtain each pixel normalization in the images to be recognized
Energy afterwards；Wherein, m represents the row sequence number of pixel in the images to be recognized, and n represents pixel in the images to be recognized
Row sequence number, W (m, n) represents the energy of pixel (m, n), and Q represents total per a line pixel in the images to be recognized
The total number of number and each row pixel；
Energy after being normalized according to all pixels point determines the images to be recognized after the normalization.
In method as implied above, the average energy value and energy variance that each details subgraph is determined respectively, bag
Include：
According to formulaIt is determined that the energy value of each details subgraph；Wherein：A represents the details
The row sequence number of pixel in subgraph, b represents the row sequence number of pixel in the details subgraph, and l=(1,2,3,4) represents every grade
Decompose four details subgraphs on same direction；J=(1,2), represents decomposed class；K=(1,2,3), represents level, vertical
With diagonal three directions；N represents the total number of every a line pixel of the details subgraph and the total number of each row pixel,
W (a, b) represents the energy of pixel (a, b)；
According to formulaIt is determined that the average energy value of each details subgraph；
According to formulaIt is determined that the energy variance of each details subgraph.
In method as implied above, by the input of the textural characteristics vector of the images to be recognized it is predetermined support to
In amount machine SVM, before the languages for obtaining the images to be recognized Chinese version, methods described also includes：
The textural characteristics vector that training image concentrates each training image is obtained using multi-wavelet transformation method；
According to the textural characteristics vector of each training image and the languages calibration result of each training image,
Determine the classifying face of the SVM；
The textural characteristics vector by the images to be recognized is inputted in predetermined SVM, is obtained described to be identified
The languages of image Chinese version, including：
According to the textural characteristics vector of the images to be recognized and the classifying face of the SVM, the figure to be identified is determined
As the languages of Chinese version.
Second aspect, the embodiment of the present invention provides a kind of languages identifying device, including：
First acquisition module, the textural characteristics vector for obtaining images to be recognized using multi-wavelet transformation method；
Second acquisition module, for the textural characteristics vector of the images to be recognized to be inputted into predetermined supporting vector
In machine SVM, the languages of the images to be recognized Chinese version are obtained.
In device as implied above, first acquisition module specifically for：
The images to be recognized is handled using filtering interpolation method, the coefficient matrix of the images to be recognized is obtained；
Using the first half row element of all rows in the coefficient matrix as the first scaling function of second order m ultiwavelet parameter,
Using the later half row element of all rows in the coefficient matrix as the second scaling function of second order m ultiwavelet parameter；
The second order m ultiwavelet is determined according to the parameter of the parameter of first scaling function and second scaling function；
The coefficient matrix is decomposed according to the second order m ultiwavelet, obtain the images to be recognized first is preset
The details subgraph of quantity and the ll channel of the second predetermined number；
The average energy value and energy variance of each details subgraph are determined respectively；
The textural characteristics of the average energy value and energy variance of each details subgraph as the images to be recognized are sweared
Amount.
In device as implied above, described device also includes：
Normalized module, the energy for each pixel to the images to be recognized is normalized,
Obtain the images to be recognized after normalization；
First acquisition module is handled the images to be recognized using filtering interpolation method, is specifically included：
The images to be recognized after the normalization is handled using filtering interpolation method.
In device as implied above, the normalized module is used for：
According to formulaObtain each pixel normalization in the images to be recognized
Energy afterwards；Wherein, m represents the row sequence number of pixel in the images to be recognized, and n represents pixel in the images to be recognized
Row sequence number, W (m, n) represents the energy of pixel (m, n), and Q represents total per a line pixel in the images to be recognized
The total number of number and each row pixel；
Energy after being normalized according to all pixels point determines the images to be recognized after the normalization.
Language Identification and device provided in an embodiment of the present invention, obtain to be identified by using multi-wavelet transformation method
The textural characteristics vector of image, the textural characteristics vector of images to be recognized is inputted in predetermined SVM, obtains figure to be identified
As the languages of Chinese version, realize and textural characteristics vector is obtained using multi-wavelet transformation method, because multi-wavelet transformation method is obtained
Take that the process of textural characteristics vector is very fast, therefore, the recognition efficiency of Language Identification provided in an embodiment of the present invention is higher.Together
When, due to the characteristic of multi-wavelet transformation method, the recognition accuracy of the Language Identification of the embodiment of the present invention is also higher.Cause
This, the efficiency and accuracy rate of languages identification are improved using Language Identification provided in an embodiment of the present invention.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the accompanying drawing used required in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
Some embodiments of invention, for those of ordinary skill in the art, without having to pay creative labor, may be used also
To obtain other accompanying drawings according to these accompanying drawings.
Fig. 1 is the schematic flow sheet of Language Identification embodiment provided in an embodiment of the present invention；
A kind of schematic flow sheet for concrete implementation mode that Fig. 2 is S101 in embodiment illustrated in fig. 1；
Fig. 3 is the schematic diagram of progress multi-wavelet transformation in Fig. 2；
Fig. 4 is the experimental result schematic diagram in Fig. 3；
Fig. 5 illustrates for a kind of flow of concrete methods of realizing of Language Identification embodiment provided in an embodiment of the present invention
Figure；
Fig. 6 A are the schematic diagram for the first image library for verifying Language Identification embodiment provided in an embodiment of the present invention；
Fig. 6 B are the schematic diagram for the second image library for verifying Language Identification embodiment provided in an embodiment of the present invention；
Fig. 7 is the structural representation of languages identifying device embodiment provided in an embodiment of the present invention.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation is described, it is clear that described embodiment is only a part of embodiment of the invention, rather than whole embodiments.It is based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of creative work is not made
Embodiment, belongs to the scope of protection of the invention.
Term " first ", " second ", " the 3rd " in description and claims of this specification and above-mentioned accompanying drawing and "
The (if present)s such as four " are for distinguishing similar object, without for describing specific order or precedence.It should manage
The data that solution is so used can be exchanged in the appropriate case, so as to embodiments of the invention described herein can with except
Here the order beyond those for illustrating or describing is implemented.In addition, term " comprising " and " having " and their any deformation,
It is intended to that covering is non-exclusive to be included, for example, containing the process of series of steps or unit, method, system, product or setting
Standby those steps or the unit for being not necessarily limited to clearly list, but may include not list clearly or for these mistakes
The intrinsic other steps of journey, method, product or equipment or unit.
Technical scheme is described in detail with specific embodiment below.These specific implementations below
Example can be combined with each other, and may be repeated no more for same or analogous concept or process in some embodiments.
Fig. 1 is the schematic flow sheet of Language Identification embodiment provided in an embodiment of the present invention.As shown in figure 1, this hair
The Language Identification that bright embodiment is provided comprises the following steps：
S101：The textural characteristics vector of images to be recognized is obtained using multi-wavelet transformation method.
Specifically, Language Identification provided in an embodiment of the present invention can be performed by computer.Determine image Chinese version
Languages can apply to the online text document of sequence to text image or search comprising language-specific etc..Based on languages
Computer in the different application scenarios of identification, the embodiment of the present invention can be the server of search engine, or, data in literature
Server in storehouse.
The texture of image is a kind of visual signature for reflecting homogeneity phenomenon in image, refers to the image pixel intensities of spatial variations
Function.Therefore, the textural characteristics vector of images to be recognized can reflect the image pixel intensities with spatial variations of images to be recognized.
Images to be recognized in the embodiment of the present invention can scan obtained text image by paper material, or, it is to be compiled by word
Collect the converted text image of the document of Software Create.The content of word in images to be recognized can be the small of various language
Say and document etc..
A kind of schematic flow sheet for concrete implementation mode that Fig. 2 is S101 in embodiment illustrated in fig. 1.As shown in Fig. 2
In a kind of possible implementation, the detailed process of the textural characteristics vector of images to be recognized is obtained using m ultiwavelet changing method
It is as follows：
S1011：Images to be recognized is handled using filtering interpolation method, the coefficient matrix of images to be recognized is obtained.
Because multi-wavelet transformation is only applicable to vector signal, therefore, multi-wavelet transformation is carried out to picture signal, it is necessary to
Pre-filtering first is carried out to the row and column of image, then by the row and column of the image Jing Guo pre-filtering, constituted according to certain rule
Vector signal, then carry out multi-wavelet transformation.
Represent a width N × N images to be recognized.Wherein, N represents this image
The total number of the pixel of every a line and each row, it can be seen that this image is a square image, ai,jRepresent picture
The pixel value of vegetarian refreshments (i, j), 1≤i≤N, 0≤j≤N, what pixel (i, j) was represented is the pixel of the i-th row jth row.
The detailed process of filtering interpolation method is as follows：
Images to be recognized A every a line is constituted into row vector signal in the following way first：
Then to Airow(n) preposition pre-filtering is carried out：
Wherein,It represents the impulse response of prefilter, then Iirow(n) I of composition1For 2N rows, N/2
The matrix of row.
Enter the preposition pre-filtering of ranks again：
By I1Each row constitute column vector signal in the following way：
Then to Bicol(n) preposition pre-filtering is carried out：
The I finally givenicol(n) I of composition2It is exactly the coefficient matrix of the images to be recognized.
S1012：Using the first half row element of all rows in coefficient matrix as the first scaling function of second order m ultiwavelet ginseng
Number, using the later half row element of all rows in coefficient matrix as the second scaling function of second order m ultiwavelet parameter.
Specifically, optionally, Biorthogonal Multiwavelets " ghmap2 " are chosen in the embodiment of the present invention to divide coefficient matrix
Solution.Ghmap2 is the multiple wavelet function of biorthogonal, is constructed by two scaling functions.The second order m ultiwavelet has two parameters：The
One scaling function and the second scaling function.The parameter of the first scaling function is determined and the parameter of the second scaling function is determined that
Second order m ultiwavelet.The parameter of first scaling function and the second scaling function refers to the first scaling function and the second scaling function
Scale factor.The parameter of the first scaling function and the parameter of the second scaling function can be determined according to coefficient matrix.
S1013：Second order m ultiwavelet is determined according to the parameter of the parameter of the first scaling function and the second scaling function.
S1014：Coefficient matrix is decomposed according to second order m ultiwavelet, the first predetermined number of images to be recognized is obtained
The ll channel of details subgraph and the second predetermined number.
Specifically, in the embodiment of the present invention, two-stage decomposition has been carried out to coefficient matrix, 4 approximators may finally be obtained
Figure and 24 details subgraphs.
Fig. 3 is the schematic diagram of progress multi-wavelet transformation in Fig. 2.As shown in figure 3, first being carried out to images to be recognized at pre-filtering
Reason, obtains the coefficient matrix of images to be recognized.Level of decomposition and two grades of decomposition are carried out to coefficient matrix again.Have four after level of decomposition
Class figure：LL, HL, LH and HH.The four classes figure is according to multiresolution Analysis Theory that two-dimensional discrete wavelet conversion is equivalent into by two
Group mirror filter is decomposed, the parameter that the process of reconstruction signal is obtained.It is low in LL correspondence row, column difference convolution resolution filters
Bandpass filter, HL correspondence row convolution decomposes high-pass filter, row convolution and decomposes low pass filter, and LH and HH are analogized with secondary.Wherein
HH, LH and HL maintain the edge detail information of original image, feature the minutia of image, referred to as details subgraph.LL is protected
The more energy information of artwork, referred to as ll channel are stayed.During two grades are decomposed, LL figures are decomposed again, 16 subgraphs are obtained.
In this 16 subgraphs, there are 4 LL figures, 4 LL Figure 31 are ll channel, and other figures after two grades of decomposition are details subgraph, because
This, 12+12=24 of the quantity of details subgraph.Fig. 4 is the experimental result schematic diagram in Fig. 3, and Figure 41 is approximator in Fig. 4
Figure, other figures are details subgraph.Details subgraph and ll channel in the embodiment of the present invention refer to the figure after two grades of decomposition.
S1015：The average energy value and energy variance of each details subgraph are determined respectively.
Specifically, can be determined according to the energy of each pixel in details subgraph each details subgraph average energy value and
Energy variance.Details subgraph is stored with a matrix type, and what the value of each element was represented in matrix is what this element was represented
The energy of pixel.
According to formulaIt is determined that the energy value of each details subgraph.Wherein：A represents details subgraph
The row sequence number of middle pixel, b represents the row sequence number of pixel in details subgraph, and l=(1,2,3,4) represents that every grade of decomposition is same
Four details subgraphs on individual direction；J=(1,2), represents decomposed class；K=(1,2,3), represents level, vertical and diagonal three
Individual direction；N represents the total number of every a line pixel of details subgraph and the total number of each row pixel.It can be seen that, carefully
Knot figure is square.According to formulaIt is determined that the average energy value of each details subgraph.According to formulaIt is determined that the energy variance of each details subgraph.
S1016：Using the average energy value and energy variance of each details subgraph as images to be recognized textural characteristics vector.
In view of for text image, the information distribution on different directions and frequency is different, and details subgraph pin
Ability to express to marginal information is strong, therefore, the average energy value using full details subgraph and energy side in the embodiment of the present invention
Difference constructs the textural characteristics of images to be recognized.After the average energy value and energy variance of each details subgraph is defined,
Using these average energy values and energy variance as images to be recognized textural characteristics vector.Because the quantity of details subgraph is 24
Individual, each details subgraph has the measurement of two dimensions of average energy value and energy variance, therefore, and textural characteristics vector is 48 dimensions.
Textural characteristics vector representation is as follows：
S102：The textural characteristics vector of images to be recognized is inputted in predetermined SVM, images to be recognized Chinese is obtained
This languages.
Specifically, SVMs (Support Vector Machine, SVM) is determined according to training image collection.
Before S102, it is thus necessary to determine that the classifying face in SVM, the process is：Training image is obtained using multi-wavelet transformation method to concentrate
The textural characteristics vector of each training image；According to the textural characteristics vector of each training image and the language of each training image
Calibration result is planted, SVM classifying face is determined.The languages calibration result of each training image is referred to according to artificial or its other party
The correct languages result for the training image Chinese version that formula is determined.Optionally, by the textural characteristics vector of images to be recognized
Input in predetermined SVM, obtaining the languages of images to be recognized Chinese version can be：According to the textural characteristics of images to be recognized
The classifying face of vector and SVM, determines the languages of images to be recognized Chinese version.
SVM can be determined using LIBSVM kits in the embodiment of the present invention.This is one simple and effective on SVM
Kit facilitates us to be adjusted for concrete application there is provided multiple interfaces.Simultaneously there is provided many default parameters,
Regulation parameter is simplified, while there is provided the function of cross-verification.Classifying step is as follows：
1) data set is prepared according to following form：
<label><index1>:<value1><index2>:<value2>...
Wherein<label>For mark one classification.<index>It, with 1 integer started, can be discontinuous to be；<
value>For real number.
2) operational efficiency is improved using scale methods scaled data；
3) selection Radial basis kernel function is used as kernel function；
4) training regulation parameter C and g；
5) training of model is completed according to parameter C and g；
6) tested using the SVM classifier trained on test set.
Optionally, energy normalized can be carried out to images to be recognized.For different images to be recognized, have different
Energy, so that the calculating of subgraph energy is influenceed, therefore, before multi-wavelet transformation is carried out, to each pixel of images to be recognized
Energy be normalized, obtain normalization after images to be recognized.Then filtering interpolation method is being used to images to be recognized
When being handled, to be handled using filtering interpolation method the images to be recognized after normalization.
Can be according to formulaObtain each pixel normalization in images to be recognized
Energy afterwards.Wherein, m represents the row sequence number of pixel in images to be recognized, and n represents the row sequence of pixel in images to be recognized
Number, W (m, n) represents the energy of pixel (m, n), and Q represents the total number and each row per a line pixel in images to be recognized
The total number of pixel.It is determined that all pixels point normalization after energy after, by the energy of these pixels with the shape of matrix
Formula is arranged, then what the matrix was represented is the images to be recognized after normalization.That is, after being normalized according to all pixels point
Energy determines the images to be recognized after normalization.
Fig. 5 illustrates for a kind of flow of concrete methods of realizing of Language Identification embodiment provided in an embodiment of the present invention
Figure.As shown in figure 5, there is an images to be recognized 51, energy normalized is carried out to the images to be recognized 51.Treating after normalization is known
Other image carries out m ultiwavelet decomposition, obtains exploded view, the textural characteristics of the images to be recognized can be obtained from these exploded views
Vector.Textural characteristics vector is inputted in SVM, you can with the languages for the text for obtaining the images to be recognized.Here languages can
To be English, Chinese, Russian, Japanese or Arabic etc..
It should be noted that a SVM is the grader of one two classification.In embodiments of the present invention, can be true in advance
Fixed multiple SVM, to realize multilingual identification.For example, graders of the SVM1 for Chinese and English, SVM2 is Russian and Japanese
Grader, the then SVM systems that SVM1 and SVM2 is constituted can realize the identification of 4 languages.
The process of Language Identification provided in an embodiment of the present invention is verified in description below.Fig. 6 A are implemented for the checking present invention
The schematic diagram of first image library of the Language Identification embodiment that example is provided.Fig. 6 B are checking language provided in an embodiment of the present invention
Plant the schematic diagram of the second image library of recognition methods embodiment.
Because languages recognize the not authoritative text image storehouse that can be obtained of research field of problem, therefore, this hair
It is bright to gather and establish two text image storehouses the method for examining proposition.In order to test multi-wavelet transformation texture feature extraction side
Method specially establishes the second image library for the robustness of picture quality.
First image library：The document that copy editor's software is formed is converted into rest image, and word content mostlys come from
The multi-national linguistic data of Chinese and English novel and spoken and written languages of ethnic minorities Translation Services Center, size is 128x128, is stored as 8 gray level images,
Comprising China and British, Tibetan, each 300 width of text image for tieing up four kinds of words, almost without quality degradation phenomenon.Contain 8 per piece image
Line of text.As shown in Figure 6A.
Second image library：Text image is obtained by paper material scanning, image sources are collection and low-quality on internet
The conversion of the portable document format (Portable Document Format, PDF) of amount.Size is 128x 128, is stored as 8
Position gray level image, comprising English, in two kinds of words each 200 width of text image.Such image sources are more complicated, generally with matter
Measure the phenomenon degenerated, such as fracture, noise, inclination etc..As shown in Figure 6B.
Hardware and software configuration is as follows used in the verification process：1st, hardware environment：Central processing unit (Central
Processing Unit, CPU)：Intel (R) Core (TM) Quad [email protected], internal memory：4G；2nd, software environment：Operation
System：64 operating systems of Windows 8.Experiment porch：MATLAB R2014a
In the verification process, the evaluation index of algorithm performance is divided into two parts：The efficiency and languages of languages identification are known
Other effect.The extraction time of efficiency Primary Reference textural characteristics vector, the accuracy rate of effect Primary Reference cross validation and call together
The rate of returning.The definition of accuracy rate is：1. accuracy rate=the correct sample number of classification/is divided into the total number of samples of the category；Recall rate is determined
Justice is：2. such sample number in the correct sample number/sample set of recall rate=classification.
Tested respectively in the first image library and the second image library, and using the method for ten ten folding cross validations
Algorithm is verified, accuracy rate and recall rate, and the signature analysis time of statistical average is obtained.By calculating, the present invention is single
The extraction time of width image is in 0.4639s or so.Table 1 is the experimental result for the first image library.Table 2 is for the second image
The experimental result in storehouse.
Table 1 is directed to the experimental result of the first image library
Table 2 is directed to the experimental result of the second image library
As can be seen from Table 1 and Table 2, using the standard of the English image classification of Language Identification provided in an embodiment of the present invention
True rate is 90% or so, and the accuracy rate of identification is higher.
Language Identification provided in an embodiment of the present invention, images to be recognized is obtained by using multi-wavelet transformation method
Textural characteristics vector, the textural characteristics vector of images to be recognized is inputted in predetermined SVM, obtains images to be recognized Chinese
This languages, realize and obtain textural characteristics vector using multi-wavelet transformation method, because multi-wavelet transformation method obtains texture
The process of characteristic vector is very fast, and therefore, the recognition efficiency of Language Identification provided in an embodiment of the present invention is higher.Meanwhile, by
In the characteristic of multi-wavelet transformation method, the recognition accuracy of the Language Identification of the embodiment of the present invention is also higher.Therefore, use
Language Identification provided in an embodiment of the present invention improves the efficiency and accuracy rate of languages identification.
Fig. 7 is the structural representation of languages identifying device embodiment provided in an embodiment of the present invention.As shown in fig. 7, this hair
The languages identifying device that bright embodiment is provided includes following module：
First acquisition module 71, the textural characteristics vector for obtaining images to be recognized using multi-wavelet transformation method.
Optionally, the first acquisition module 71 specifically for：Images to be recognized is handled using filtering interpolation method, obtained
The coefficient matrix of images to be recognized；Using the first half row element of all rows in coefficient matrix as second order m ultiwavelet the first yardstick letter
Several parameters, using the later half row element of all rows in coefficient matrix as the second scaling function of second order m ultiwavelet parameter；Root
Second order m ultiwavelet is determined according to the parameter of the first scaling function and the parameter of the second scaling function；According to second order m ultiwavelet to coefficient square
Battle array is decomposed, and obtains the details subgraph and the ll channel of the second predetermined number of the first predetermined number of images to be recognized；Point
The average energy value and energy variance of each details subgraph are not determined；Using the average energy value and energy variance of each details subgraph as
The textural characteristics vector of images to be recognized.
Second acquisition module 72, for the textural characteristics vector of images to be recognized to be inputted in predetermined SVM, is obtained
The languages of images to be recognized Chinese version.
Optionally, the device also includes：Normalized module, the energy for each pixel to images to be recognized
It is normalized, obtains the images to be recognized after normalization.Then the first acquisition module 71 treats knowledge using filtering interpolation method
Other image is handled, and is specifically included：The images to be recognized after normalization is handled using filtering interpolation method.
Normalized module specifically for：According to formulaObtain images to be recognized
In energy after each pixel normalization；Wherein, m represents the row sequence number of pixel in images to be recognized, and n represents figure to be identified
The row sequence number of pixel as in, W (m, n) represents the energy of pixel (m, n), and Q is represented in images to be recognized per a line pixel
Total number and each row pixel total number；Energy after being normalized according to all pixels point determines that treating after normalization is known
Other image.
The device also includes：3rd acquisition module, concentrates each for obtaining training image using multi-wavelet transformation method
The textural characteristics vector of training image；Determining module, for the textural characteristics vector according to each training image and each instruction
Practice the languages calibration result of image, determine SVM classifying face.Then the second acquisition module 72 is specifically for according to images to be recognized
The classifying face of textural characteristics vector and SVM, determines the languages of images to be recognized Chinese version.
Languages identifying device provided in an embodiment of the present invention is particularly used in the languages identification side for performing embodiment illustrated in fig. 1
Method, its implementation process is similar with technical principle, and here is omitted.
Languages identifying device provided in an embodiment of the present invention, by setting the first acquisition module, for being become using m ultiwavelet
Change the textural characteristics vector that method obtains images to be recognized, the second acquisition module, for the textural characteristics of images to be recognized to be sweared
In the predetermined SVM of amount input, the languages of images to be recognized Chinese version are obtained, realizes and is obtained using multi-wavelet transformation method
Textural characteristics vector, because the process that multi-wavelet transformation method obtains textural characteristics vector is very fast, therefore, the embodiment of the present invention is carried
The recognition efficiency of the languages identifying device of confession is higher.Simultaneously as the characteristic of multi-wavelet transformation method, the language of the embodiment of the present invention
The recognition accuracy for planting identifying device is also higher.Therefore, languages identifying device provided in an embodiment of the present invention improves languages knowledge
Other efficiency and accuracy rate.
One of ordinary skill in the art will appreciate that：Realizing all or part of step of above-mentioned each method embodiment can lead to
The related hardware of programmed instruction is crossed to complete.Foregoing program can be stored in a computer read/write memory medium.The journey
Sequence upon execution, performs the step of including above-mentioned each method embodiment；And foregoing storage medium includes：ROM, RAM, magnetic disc or
Person's CD etc. is various can be with the medium of store program codes.
Finally it should be noted that：Various embodiments above is merely illustrative of the technical solution of the present invention, rather than its limitations；To the greatest extent
The present invention is described in detail with reference to foregoing embodiments for pipe, it will be understood by those within the art that：Its according to
The technical scheme described in foregoing embodiments can so be modified, or which part or all technical characteristic are entered
Row equivalent substitution；And these modifications or replacement, the essence of appropriate technical solution is departed from various embodiments of the present invention technology
The scope of scheme.
Claims (10)
1. a kind of Language Identification, it is characterised in that including：
The textural characteristics vector of images to be recognized is obtained using multi-wavelet transformation method；
The textural characteristics vector of the images to be recognized is inputted in predetermined support vector machines, waits to know described in acquisition
The languages of other image Chinese version.
2. according to the method described in claim 1, it is characterised in that the use multi-wavelet transformation method obtains images to be recognized
Textural characteristics vector, including：
The images to be recognized is handled using filtering interpolation method, the coefficient matrix of the images to be recognized is obtained；
Using the first half row element of all rows in the coefficient matrix as the first scaling function of second order m ultiwavelet parameter, by institute
The later half row element for stating all rows in coefficient matrix is used as the parameter of the second scaling function of second order m ultiwavelet；
The second order m ultiwavelet is determined according to the parameter of the parameter of first scaling function and second scaling function；
The coefficient matrix is decomposed according to the second order m ultiwavelet, the first predetermined number of the images to be recognized is obtained
Details subgraph and the second predetermined number ll channel；
The average energy value and energy variance of each details subgraph are determined respectively；
Using the average energy value and energy variance of each details subgraph as the images to be recognized textural characteristics vector.
3. method according to claim 2, it is characterised in that the use filtering interpolation method is entered to the images to be recognized
Before row pretreatment, methods described also includes：
The energy of each pixel of the images to be recognized is normalized, the figure to be identified after normalization is obtained
Picture；
The use filtering interpolation method is handled the images to be recognized, including：
The images to be recognized after the normalization is handled using filtering interpolation method.
4. method according to claim 3, it is characterised in that the energy of each pixel to the images to be recognized
Amount is normalized, and obtains the images to be recognized after normalization, including：
According to formulaObtain in the images to be recognized after each pixel normalization
Energy；Wherein, m represents the row sequence number of pixel in the images to be recognized, and n represents the row of pixel in the images to be recognized
Sequence number, W (m, n) represents the energy of pixel (m, n), Q represent total number in the images to be recognized per a line pixel and
The total number of each row pixel；
Energy after being normalized according to all pixels point determines the images to be recognized after the normalization.
5. the method according to claim 2-4, it is characterised in that the energy for determining each details subgraph respectively
Average and energy variance, including：
According to formulaIt is determined that the energy value of each details subgraph；Wherein：A represents the details subgraph
The row sequence number of middle pixel, b represents the row sequence number of pixel in the details subgraph, and l=(1,2,3,4) represents every grade of decomposition
Four details subgraphs on same direction；J=(1,2), represents decomposed class；K=(1,2,3), represents level, vertical and right
Three, angle direction；N represents the total number of every a line pixel of the details subgraph and the total number of each row pixel, W (a,
B) energy of pixel (a, b) is represented；
According to formulaIt is determined that the average energy value of each details subgraph；
According to formulaIt is determined that the energy variance of each details subgraph.
6. according to the method described in claim 1, it is characterised in that inputted by the textural characteristics vector of the images to be recognized
In predetermined support vector machines, before the languages for obtaining the images to be recognized Chinese version, methods described also includes：
The textural characteristics vector that training image concentrates each training image is obtained using multi-wavelet transformation method；
According to the textural characteristics vector of each training image and the languages calibration result of each training image, it is determined that
The classifying face of the SVM；
The textural characteristics vector by the images to be recognized is inputted in predetermined SVM, obtains the images to be recognized
The languages of Chinese version, including：
According to the textural characteristics vector of the images to be recognized and the classifying face of the SVM, determine in the images to be recognized
The languages of text.
7. a kind of languages identifying device, it is characterised in that including：
First acquisition module, the textural characteristics vector for obtaining images to be recognized using multi-wavelet transformation method；
Second acquisition module, for the textural characteristics vector of the images to be recognized to be inputted into predetermined SVMs
In SVM, the languages of the images to be recognized Chinese version are obtained.
8. device according to claim 7, it is characterised in that first acquisition module specifically for：
The images to be recognized is handled using filtering interpolation method, the coefficient matrix of the images to be recognized is obtained；
Using the first half row element of all rows in the coefficient matrix as the first scaling function of second order m ultiwavelet parameter, by institute
The later half row element for stating all rows in coefficient matrix is used as the parameter of the second scaling function of second order m ultiwavelet；
The second order m ultiwavelet is determined according to the parameter of the parameter of first scaling function and second scaling function；
The coefficient matrix is decomposed according to the second order m ultiwavelet, the first predetermined number of the images to be recognized is obtained
Details subgraph and the second predetermined number ll channel；
The average energy value and energy variance of each details subgraph are determined respectively；
Using the average energy value and energy variance of each details subgraph as the images to be recognized textural characteristics vector.
9. device according to claim 8, it is characterised in that described device also includes：
Normalized module, the energy for each pixel to the images to be recognized is normalized, and obtains
Images to be recognized after normalization；
First acquisition module is handled the images to be recognized using filtering interpolation method, is specifically included：
The images to be recognized after the normalization is handled using filtering interpolation method.
10. device according to claim 9, it is characterised in that the normalized module is used for：
According to formulaObtain the energy after each pixel normalization in the images to be recognized
Amount；Wherein, m represents the row sequence number of pixel in the images to be recognized, and n represents the row sequence of pixel in the images to be recognized
Number, W (m, n) represents the energy of pixel (m, n), and Q represents total number in the images to be recognized per a line pixel and every
The total number of one row pixel；
Energy after being normalized according to all pixels point determines the images to be recognized after the normalization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710272779.6A CN107256378A (en) | 2017-04-24 | 2017-04-24 | Language Identification and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710272779.6A CN107256378A (en) | 2017-04-24 | 2017-04-24 | Language Identification and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107256378A true CN107256378A (en) | 2017-10-17 |
Family
ID=60027222
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710272779.6A Pending CN107256378A (en) | 2017-04-24 | 2017-04-24 | Language Identification and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107256378A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108172212A (en) * | 2017-12-25 | 2018-06-15 | 横琴国际知识产权交易中心有限公司 | A kind of voice Language Identification and system based on confidence level |
| CN110334705A (en) * | 2019-06-25 | 2019-10-15 | 华中科技大学 | A kind of Language Identification of the scene text image of the global and local information of combination |
| CN111027528A (en) * | 2019-11-22 | 2020-04-17 | 华为技术有限公司 | Language identification method and device, terminal equipment and computer readable storage medium |
| CN111339787A (en) * | 2018-12-17 | 2020-06-26 | 北京嘀嘀无限科技发展有限公司 | Language identification method and device, electronic equipment and storage medium |
| CN114462397A (en) * | 2022-01-20 | 2022-05-10 | 连连（杭州）信息技术有限公司 | Language identification model training method, language identification method and device and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103295001A (en) * | 2013-06-03 | 2013-09-11 | 哈尔滨理工大学 | Invoice information recognition equipment and invoice information recognition method both applied to handhold equipment |
| CN105139041A (en) * | 2015-08-21 | 2015-12-09 | 北京旷视科技有限公司 | Method and device for recognizing languages based on image |
| CN105760901A (en) * | 2016-01-27 | 2016-07-13 | 南开大学 | Automatic language identification method for multilingual skew document image |
-
2017
- 2017-04-24 CN CN201710272779.6A patent/CN107256378A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103295001A (en) * | 2013-06-03 | 2013-09-11 | 哈尔滨理工大学 | Invoice information recognition equipment and invoice information recognition method both applied to handhold equipment |
| CN105139041A (en) * | 2015-08-21 | 2015-12-09 | 北京旷视科技有限公司 | Method and device for recognizing languages based on image |
| CN105760901A (en) * | 2016-01-27 | 2016-07-13 | 南开大学 | Automatic language identification method for multilingual skew document image |
Non-Patent Citations (2)
| Title |
|---|
| 顾立娟 等: "基于多小波变换的文本图像文种识别", 《电子设计工程》 * |
| 顾立娟: "基于文本图像纹理特征的文种识别技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108172212A (en) * | 2017-12-25 | 2018-06-15 | 横琴国际知识产权交易中心有限公司 | A kind of voice Language Identification and system based on confidence level |
| CN108172212B (en) * | 2017-12-25 | 2020-09-11 | 横琴国际知识产权交易中心有限公司 | Confidence-based speech language identification method and system |
| CN111339787A (en) * | 2018-12-17 | 2020-06-26 | 北京嘀嘀无限科技发展有限公司 | Language identification method and device, electronic equipment and storage medium |
| CN111339787B (en) * | 2018-12-17 | 2023-09-19 | 北京嘀嘀无限科技发展有限公司 | Language identification method and device, electronic equipment and storage medium |
| CN110334705A (en) * | 2019-06-25 | 2019-10-15 | 华中科技大学 | A kind of Language Identification of the scene text image of the global and local information of combination |
| CN110334705B (en) * | 2019-06-25 | 2021-08-03 | 华中科技大学 | Language identification method of scene text image combining global and local information |
| CN111027528A (en) * | 2019-11-22 | 2020-04-17 | 华为技术有限公司 | Language identification method and device, terminal equipment and computer readable storage medium |
| CN111027528B (en) * | 2019-11-22 | 2023-10-03 | 华为技术有限公司 | Language identification method, device, terminal equipment and computer readable storage medium |
| CN114462397A (en) * | 2022-01-20 | 2022-05-10 | 连连（杭州）信息技术有限公司 | Language identification model training method, language identification method and device and electronic equipment |
| CN114462397B (en) * | 2022-01-20 | 2023-09-22 | 连连（杭州）信息技术有限公司 | Language identification model training method, language identification method, device and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| RU2661750C1 (en) | Symbols recognition with the use of artificial intelligence | |
| CN107256378A (en) | Language Identification and device | |
| Sermanet et al. | Convolutional neural networks applied to house numbers digit classification | |
| Agrawal et al. | Automated bank cheque verification using image processing and deep learning methods | |
| CN113591546B (en) | Semantic enhancement type scene text recognition method and device | |
| CN108446621A (en) | Bank slip recognition method, server and computer readable storage medium | |
| CN103886342B (en) | Hyperspectral image classification method based on spectrums and neighbourhood information dictionary learning | |
| CN105512657B (en) | Character identifying method and equipment | |
| CN109446345A (en) | Nuclear power file verification processing method and system | |
| CN111339935B (en) | Optical remote sensing picture classification method based on interpretable CNN image classification model | |
| CN109492625A (en) | A kind of human face identification work-attendance checking method based on width study | |
| CN114170608A (en) | Super-resolution text image recognition method, device, equipment and storage medium | |
| CN109740515A (en) | One kind reading and appraising method and device | |
| CN109255289A (en) | A kind of across aging face identification method generating model based on unified formula | |
| CN110781672A (en) | Question bank production method and system based on machine intelligence | |
| CN112069900A (en) | Bill character recognition method and system based on convolutional neural network | |
| CN108509833A (en) | A kind of face identification method, device and equipment based on structured analysis dictionary | |
| Khuong et al. | Generating synthetic handwritten mathematical expressions from a LaTeX sequence or a MathML script | |
| CN106650696A (en) | Handwritten electrical element identification method based on singular value decomposition | |
| CN109034213A (en) | Hyperspectral image classification method and system based on joint entropy principle | |
| Mariyathas et al. | Sinhala handwritten character recognition using convolutional neural network | |
| Li et al. | Manchu word recognition based on convolutional neural network with spatial pyramid pooling | |
| Thuon et al. | Improving isolated glyph classification task for palm leaf manuscripts | |
| CN106503706B (en) | The method of discrimination of Chinese character pattern cutting result correctness | |
| CN112200789A (en) | Image identification method and device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171017 |