CN106408002A - Hand-written manchu alphabet identification system - Google Patents
Hand-written manchu alphabet identification system Download PDFInfo
- Publication number
- CN106408002A CN106408002A CN201610752646.4A CN201610752646A CN106408002A CN 106408002 A CN106408002 A CN 106408002A CN 201610752646 A CN201610752646 A CN 201610752646A CN 106408002 A CN106408002 A CN 106408002A
- Authority
- CN
- China
- Prior art keywords
- manchu alphabet
- tuple
- manchu
- alphabet
- eigenvalue
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- 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
- G06V30/22—Character recognition characterised by the type of writing
- G06V30/226—Character recognition characterised by the type of writing of cursive writing
-
- 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)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a hand-written manchu alphabet identification system, belongs to the mode identification technology field and solves a problem that manchu alphabets only having one extracted single characteristic because of letter particularities have no noise interference prevention capability. The system is characterized by comprising a processing assembly (1202), wherein the processing assembly (1202) comprises one or more processors (1220) for instruction performing to accomplish steps of a manchu identification method. The method comprises steps that a K-neighbour method is employed to carry out classification processing on manchu alphabet characteristic values after dimension reduction, and manchu alphabets corresponding to to-be-identified manchu alphabet images are acquired. The system is advantaged in that on one hand, dimension reduction and classification calculation efficiency improvement are realized, on the other hand, the manchu alphabet characteristic values after dimension reduction are easier to distinguish, relatively good noise interference prevention capability is realized, and thereby accuracy of the manchu alphabets identified through employing the K-neighbour method according to the manchu alphabet characteristic values after dimension reduction is relatively high.
Description
Technical field
The present invention relates to mode identification technology, the system of more particularly, to a kind of hand-written Manchu alphabet identification.
Background technology
Full language, as a kind of one of ancient language, is a kind of minority language of great representative, studies ethnic groups
Character recognition method is protection and the important technical of succession ethnic mountainous regions legacy, has important social value and goes through
History meaning.The handwriting recognition of Chinese character has met practical demand at present, and the off-line handwritten recognition product of English and numeral is also
Very ripe, but the identification of the handwriting of the language of the Manchus of punctuating is still one piece of blank field, waits to study.
The utility model patent of Authorization Notice No. CN 205451106 U discloses a kind of hand-written Manchu alphabet acquisition system,
It is based on this patent, it is possible to achieve to hand-written Manchu alphabet collecting work, sets up Manchu alphabet storehouse according to this, obtain Manchu alphabet number
According to sample, obtain training sample.
The feature that traditional character recognition method extracts is usually single features, but for Manchu alphabet, because of its word
Particularity, only extracts single features and does not possess certain noise antijamming capability.Therefore, when noise jamming is larger, can lead to
The discrimination of Manchu alphabet is relatively low.
Content of the invention
For overcoming problems of the prior art, the present invention discloses a kind of system of hand-written Manchu alphabet identification.
A kind of system of hand-written Manchu alphabet identification, including process assembly, process assembly includes one or more processors
Carry out execute instruction, to complete the step that method for distinguishing known in the language of the Manchus；
Described language of the Manchus recognition methodss comprise the steps：
It is respectively adopted Directional feature extraction method and thick meshed feature extracting method Manchu alphabet image to be identified is carried
Take feature, obtain Manchu alphabet eigenvalue；
Method using the linear judgment analysis of LDA carries out dimensionality reduction to described Manchu alphabet eigenvalue, obtains expiring after dimensionality reduction
Word mother's eigenvalue；
Classification process is carried out to the Manchu alphabet eigenvalue after described dimensionality reduction using K- near neighbor method, obtains described to be identified
The corresponding Manchu alphabet of Manchu alphabet image.
Further, described system also includes following one or more assemblies：Memorizer (1204), power supply module
(1206), multimedia groupware (1208), audio-frequency assembly (1210), the interface (1212) of input/output and communication component
(1214)；
Described memorizer (1204) is configured to store various types of data to support the operation in equipment, memorizer
(1204) realized by any kind of volatibility or non-volatile memory device or combinations thereof；
Described electric power assembly (1206) provides electric power for various assemblies, and electric power assembly (1206) includes power-supply management system,
One or more power supplys, and other generate with for identifying system, management and the distribution assembly that is associated of electric power；
Described multimedia groupware (1208) includes one output interface of offer between described identifying system and user
Screen, this screen includes liquid crystal display and touch panel, to receive the input signal from user；
Described audio-frequency assembly (1210) is configured to output and/or input audio signal；
The interface (1212) of described input/output provides interface for process assembly (1202) and peripheral interface module between；
Described communication component (1214) is configured to facilitate wired or wireless way between identifying system and other equipment
Communication.
Further, described it is respectively adopted Directional feature extraction method and thick meshed feature extracting method is expired to be identified
Civilian letter image extraction feature, the method obtaining Manchu alphabet eigenvalue, including：
Manchu alphabet image to be identified carries out 8 Directional feature extraction, obtains the direction character value of Manchu alphabet；
Thick meshed feature extraction is carried out to Manchu alphabet image to be identified, obtains the grid search-engine value of Manchu alphabet；
The direction character value of Manchu alphabet and grid search-engine value are combined as string, obtain Manchu alphabet eigenvalue.
Further, described classification process is carried out to the Manchu alphabet eigenvalue after described dimensionality reduction using K- near neighbor method,
The method obtaining the described corresponding Manchu alphabet of Manchu alphabet image to be identified, including：
Safeguard size be k by apart from descending priority query, for storing tuple to be identified.Random from
Choose k tuple in tuple to be identified as initial arest neighbors tuple, calculate respectively test tuple to this k tuple away from
From, by be identified unit deck label and distance be stored in priority query；
Traversal training tuple set, calculates the distance of currently tuple to be identified and test tuple, compare gained apart from L with preferential
Ultimate range Lmax in level queue, obtains final priority query；
Traversal finishes, and calculates many several classes ofs of k tuple in priority query, and the classification as test tuple.
Further, the described method using the linear judgment analysis of LDA carries out dimensionality reduction to described Manchu alphabet eigenvalue, obtains
In the step of Manchu alphabet eigenvalue to after dimensionality reduction, the acquisition methods of dimensionality reduction matrix, including：
Set up the hand-written language of the Manchus storehouse of storage language of the Manchus data sample；
It is respectively adopted Directional feature extraction method and thick meshed feature extracting method to the Manchu alphabet in hand-written language of the Manchus storehouse
Image zooming-out feature, obtains Manchu alphabet eigenvalue；Distribution data space storage Manchu alphabet eigenvalue and label；
Calculate Different categories of samples expectation and this expectation of gross sample；
Calculate covariance matrix Sw in covariance matrix Sb and class between class；
Seek the characteristic vector of matrix Sw-1Sb, obtain projection vector.
Further, described classification process is carried out to the Manchu alphabet eigenvalue after described dimensionality reduction using K- near neighbor method,
Obtain the described corresponding Manchu alphabet of Manchu alphabet image to be identified, the k value in k- near neighbor method determines method, including：
Set up the hand-written language of the Manchus storehouse of storage language of the Manchus data sample；
It is respectively adopted Directional feature extraction method and thick meshed feature extracting method to the Manchu alphabet in hand-written language of the Manchus storehouse
Image zooming-out feature, obtains Manchu alphabet eigenvalue；
Method using the linear judgment analysis of LDA carries out dimensionality reduction to described Manchu alphabet eigenvalue, obtains expiring after dimensionality reduction
Word mother's eigenvalue；
Distribution data space stores training data and test tuple, parameter preset K respectively；
Safeguard size be k by apart from descending priority query, for storing tuple to be identified, at random from
Choose k tuple in tuple to be identified as initial arest neighbors tuple, calculate respectively test tuple to this k tuple away from
From, by be identified unit deck label and distance be stored in priority query；
Traversal training tuple set, calculates the distance of currently tuple to be identified and test tuple, compare gained apart from L with preferential
Ultimate range Lmax in level queue, obtains final priority query；
Traversal finishes, and calculates many several classes ofs of k tuple in priority query, and the classification as test tuple.
Test tuple set is completed rear calculation error rate, continues to set different K value re -training, finally takes error rate
Minimum K value.
Further, the k value in k- near neighbor method determines in method, described described Manchu alphabet eigenvalue is dropped
In the step of dimension, the acquisition methods of the dimensionality reduction matrix of use, including：
Distribution data space storage Manchu alphabet eigenvalue and label；
Calculate Different categories of samples expectation and this expectation of gross sample；
Calculate covariance matrix Sw in covariance matrix Sb and class between class；
Seek the characteristic vector of matrix Sw-1Sb, obtain projection vector.
Further, described compare ultimate range Lmax in L and priority query for the gained, obtain final preferential
The method of level queue, including：
When described gained is more than ultimate range Lmax in priority query apart from L, then gives up this tuple, travel through next
Individual tuple.
When described gained is less than ultimate range Lmax in priority query apart from L, then delete in priority query
The tuple of big distance, current training tuple is stored in priority query.
Further, described to Manchu alphabet image zooming-out feature to be identified before, have to Manchu alphabet to be identified
The step that image pattern carries out pretreatment.
Further, described pretreatment, including carrying out the linear normalization of character boundary, plus virtual to Manchu alphabet image
Point in the resampling of point in stroke, the non-linear normalizing of character, stroke, stroke one or more of smooth.
The technical scheme that the present invention provides can include following beneficial effect：
Extract, using Directional feature extraction and thick meshed feature, the method combining Manchu alphabet image to be identified is carried
Take feature so that the Manchu alphabet eigenvalue obtaining is more accurate, there is preferable noise antijamming capability, so that follow-up drop
Identify that Manchu alphabet accuracy rate is higher using KNN grader according to this Manchu alphabet eigenvalue after dimension.
By LDA dimension-reduction treatment is carried out to Manchu alphabet eigenvalue, thus decreasing the calculation times of processor, improve
Treatment effeciency.In addition Manchu alphabet characteristic to be identified has scattered in the class scatter of maximum and the class of minimum in the projected
Degree, is the later stage to carry out classification using K- near neighbor method to provide condition.
Because KNN method is mainly by neighbouring sample limited around, rather than by differentiating that the method for class field to determine institute
Belong to classification, therefore for the intersection as this class field of hand-written Manchu alphabet or overlapping more sample set to be divided, KNN side
Method is more suitable for compared with additive method, is obtained in that higher discrimination.Additionally, being easily programmed using K- near neighbor method, and it is not required to
Optimize, improve the efficiency of identification.
By being trained to K- Nearest Neighbor Classifier, obtain suitable model parameter so as to be identified after dimensionality reduction
After Manchu alphabet image feature value enters grader, there is accurate classifying quality.
By Manchu alphabet image to be identified is carried out with the operation of pretreatment, can reduce well in class and make a variation to identification
During interference, be conducive to improve Manchu alphabet image recognition efficiency.
The embodiment of the present invention, the method being combined using direction character and thick meshed feature is to Manchu alphabet figure to be identified
As extracting feature so that the Manchu alphabet eigenvalue obtaining is more accurate；Further, using LDA linear judgment analysis method pair
Manchu alphabet eigenvalue carries out dimensionality reduction, on the one hand decreases dimension and improves classified counting efficiency, after on the other hand making dimensionality reduction
Manchu alphabet eigenvalue be more prone to distinguish, there is preferable noise antijamming capability so that adopt K- near neighbor method
Higher according to the Manchu alphabet accuracy rate that the Manchu alphabet eigenvalue after this dimensionality reduction identifies.
It should be appreciated that above general description and detailed description hereinafter are only exemplary and explanatory, not
The disclosure can be limited.
Brief description
Accompanying drawing herein is merged in description and constitutes the part of this specification, shows the enforcement meeting the present invention
Example, and be used for explaining the principle of the present invention together with description.
Figure 1A is the flow chart of a kind of hand-written Manchu alphabet recognition methodss according to an exemplary embodiment.
Figure 1B is a kind of method flow diagram of the acquisition Manchu alphabet eigenvalue according to an exemplary embodiment.
Fig. 1 C is a kind of method flow diagram of the Manchu alphabet eigenvalue classification according to an exemplary embodiment.
Fig. 2 is the flow chart of the another kind hand-written Manchu alphabet recognition methodss according to an exemplary embodiment.
Fig. 2A is a kind of method stream that Manchu alphabet sample is carried out with model training according to an exemplary embodiment
Cheng Tu.
Fig. 3 is a kind of method flow diagram of the acquisition LDA dimensionality reduction matrix according to an exemplary embodiment.
Fig. 4 is a kind of method flow diagram of the adjustment K- neighbour's parameter according to an exemplary embodiment.
Fig. 5 is the block diagram of a kind of hand-written Manchu alphabet identifying device according to an exemplary embodiment.
Fig. 6 is the block diagram of the another kind hand-written Manchu alphabet identifying device according to an exemplary embodiment.
Fig. 7 is the block diagram of the another kind hand-written Manchu alphabet identifying device according to an exemplary embodiment.
Fig. 8 A is the block diagram of the another kind hand-written Manchu alphabet identifying device according to an exemplary embodiment.
Fig. 8 B is the block diagram of the another kind hand-written Manchu alphabet identifying device according to an exemplary embodiment.
Fig. 9 is the block diagram of the another kind hand-written Manchu alphabet identifying device according to an exemplary embodiment.
Figure 10 is the block diagram of another kind of formality Ei Manchu alphabet identifying device according to an exemplary embodiment.
Figure 11 is the block diagram of the another kind hand-written Manchu alphabet identifying device according to an exemplary embodiment.
Figure 12 is a kind of block diagram being applied to hand-written Manchu alphabet identifying device according to an exemplary embodiment.
Specific embodiment
Here will in detail exemplary embodiment be illustrated, its example is illustrated in the accompanying drawings.Explained below is related to
During accompanying drawing, unless otherwise indicated, the same numbers in different accompanying drawings represent same or analogous key element.Following exemplary embodiment
Described in embodiment do not represent all embodiments consistent with the present invention.On the contrary, they be only with such as appended
The example of the consistent apparatus and method of some aspects being described in detail in claims, the present invention.
Figure 1A is the flow chart of a kind of hand-written Manchu alphabet recognition methodss according to an exemplary embodiment, such as Figure 1A
Shown, this hand-written Manchu alphabet recognition methods can be used in various terminal equipment, and the method comprises the following steps：
In step S101, it is respectively adopted Directional feature extraction method and thick meshed feature extracting method is expired to be identified
Civilian letter image extracts feature, obtains Manchu alphabet eigenvalue.
Directional feature extraction refers to each language of the Manchus character picture is divided into the grid of M × M, then in each grid extract with
Related N (the 4,8) direction character in locus, this N (4,8) individual direction can be defined as respectively upper and lower, left and right or upper,
Under, left and right, upper left, lower-left, upper right, bottom right.This feature extracting method makes the direction character of point and sky in handwritten stroke
Between related, rather than and time correlation.Can be good at solving changing, due to order of strokes and number of strokes, the discrimination bringing
Decline problem.The fixing dimension that this method can obtain having single one physical meaning from each Manchu alphabet sample simultaneously
Characteristic vector.
Thick meshed feature is belonging to one kind of local feature in statistical nature, embody Manchu alphabet global shape point
Cloth.The method is divided into M × M grid sample letter, then counts the pixel quantity in each grid successively, in cognitive phase
The statistical nature of character can be used as and realize the identification to character successively by the combinations of features of each lattice statistical is risen.
Pixel in each grid of this algorithm can reflect a part of feature of letter, and algorithm is simple, it is easy to accomplish.
Need when actual characteristic extracts that travel direction feature extraction and thick meshed feature carry respectively to Manchu alphabet image
Take, therefore as shown in Figure 1B, step S101 comprises the steps：
In step S1011,8 Directional feature extraction are carried out to Manchu alphabet image to be identified, obtains Manchu alphabet
Direction character value.
It is in this embodiment, described that to carry out 8 Directional feature extraction methods to Manchu alphabet image to be identified as follows：Right
After Manchu alphabet image carries out pretreatment, new hand-written Manchu alphabet and Manchu alphabet tracing point are obtained before nonlinear change
The mapping relations of point afterwards, are divided into letter 8 × 8 grid afterwards, and the center taking each lattice is sampling center, obtains 8
× 8 sampled point, then to these sampling center Gabor envelope extraction 8- direction characters.
In step S1012, thick meshed feature extraction is carried out to Manchu alphabet image to be identified, obtains Manchu alphabet
Grid search-engine value.
It is in this embodiment, described that to carry out thick meshed feature extracting method to Manchu alphabet image to be identified as follows：Ask
Go out the outer rim of the final Manchu alphabet image obtaining after pretreatment；Then become adding the Manchu alphabet image segmentation after frame
8 × 8 grids；Count the black pixel ratio in each grid successively；Finally 8 × 8 black pixels after statistics are formed a line
Form the grid search-engine vector of 64 dimensions.
In step S1013, the direction character value of Manchu alphabet and grid search-engine value are combined as string, obtain language of the Manchus word
Female eigenvalue.
Directional feature extraction method makes the direction character of point and space correlation in handwritten stroke, and embodiment is in grid
Difference on direction for the stroke, the method strong antijamming capability, but it is higher to realize process complexity, and complex in the application；
Pixel in each grid of thick meshed feature extraction and application reflecting a part of feature of letter, realize simple, distinguishes by the method
The ability of similar letter is strong, but grid search-engine anti-stroke Position disturbance ability is poor.Therefore, because the spy that Manchu alphabet is constituted
Different property is it is impossible to be extracted using traditional single features, and the feature extracting method being combined using described two methods.
Extract, using Directional feature extraction and thick meshed feature, the method combining Manchu alphabet image to be identified is carried
Take feature so that the Manchu alphabet eigenvalue obtaining is more accurate, there is preferable noise antijamming capability, so that follow-up drop
Identify that Manchu alphabet accuracy rate is higher using KNN grader according to this Manchu alphabet eigenvalue after dimension.
In step s 102, dimensionality reduction is carried out to described Manchu alphabet eigenvalue using the method for the linear judgment analysis of LDA, obtain
Manchu alphabet eigenvalue to after dimensionality reduction.
Linear discriminent analyzes (LinearDiscriminantAnalysis, LDA), also referred to as Fisher linear discriminant
(FisherLinearDiscrininant, FLD), is the classic algorithm of pattern recognition.The basic thought of linear discriminent analysis
It is that the pattern sample of higher-dimension is projected to best discriminant technique vector space, extract classification information and compressive features space dimensionality to reach
Effect, after projection, Assured Mode sample is between class distance and minimum class that new subspace (after projecting) has maximum
Distance, that is, sample have optimal separability on this space.LDA is a kind of linear dimension-reduction algorithm having supervision, keeps with PCA
Data message is different, and LDA is so that the data point after dimensionality reduction is easily distinguished as much as possible.So adopting LDA side here
Method carries out dimensionality reduction operation to Manchu alphabet eigenvalue.
In this embodiment, dimensionality reduction is carried out to described Manchu alphabet eigenvalue using the method for the linear judgment analysis of LDA, first
First need to calculate dimensionality reduction matrix in the training process, obtain best projection vector, the i.e. group moment of new subspace in linear space
Battle array.Then Manchu alphabet eigenvector projection to be identified to be realized special to Manchu alphabet in identification process to this subspace
The dimensionality reduction of value indicative.
In this embodiment, the purpose of LDA linear discriminent analysis is to carry out dimension-reduction treatment to Manchu alphabet eigenvalue, with
Improve follow-up treatment effeciency；In addition, after by LDA, within class scatter matrix converges Manchu alphabet characteristic to be identified
Total numerical value is less, the numerical value in inter _ class relationship matrix larger so that between the class got the degree of coupling relatively low, polymerization in class
Degree is higher, is the later stage to carry out classification using K- near neighbor method to provide condition.
In step s 103, classification process is carried out to the Manchu alphabet eigenvalue after described dimensionality reduction using K- near neighbor method,
Obtain the described corresponding Manchu alphabet of Manchu alphabet image to be identified.
K nearest neighbor (k-NearestNeighbor, KNN) sorting algorithm, is the method for a comparative maturity in theory, is also
One of simplest machine learning algorithm.The thinking of the method is：If the k in feature space, sample is most like
Great majority in the sample of (i.e. closest in feature space) belong to some classification, then this sample falls within this classification.KNN
In algorithm, selected neighbours are the objects of correctly classification.The method is on categorised decision only according to closest one
The classification of individual or several sample is determining the classification belonging to sample to be divided.
KNN algorithm itself is simply effective, and it is a kind of lazy-learning algorithm.In this embodiment, algorithm uses
Model actually corresponds to the division to Manchu alphabet feature space.The selection of K value, distance metric and categorised decision rule are institutes
State three fundamentals of KNN algorithm in this embodiment：
Selecting of 1.K value can be to the result generation significant impact of algorithm.K value is less mean only nearer with input example
Training example just can work to predicting the outcome, but be susceptible to over-fitting；If K value is larger, advantage is to reduce
The estimation difference practised, but the error of approximation that shortcoming is study increases, at this moment also can be to pre- with input example training example farther out
Survey is worked, and leads to prediction to make a mistake.In the application of this embodiment, it is trained by using training set, adjusting parameter K is
Obtain the most suitable K value of classifying quality eventually, parameter training process describes in detail later.
2. the categorised decision rule in this algorithm is majority voting, that is, by input example K closest to training example
In many several classes ofs determine the classification of input examples.
3. distance metric adopts Lp distance (p=2), i.e. Euclidean distance.Due to having carried out pretreatment behaviour in this embodiment
Make so that after before tolerance, the value of each attribute is standardization, it is to avoid have larger initial codomain attribute and
There is the weighted problem of the attribute of less initial codomain.
When being classified using K- nearest neighbor algorithm, needs are calculated and are compared, therefore as shown in Figure 1 C, step S103 bag
Include following steps：
In step S1031, safeguard size be k by apart from descending priority query, treat for storage
Identification tuple.Choose k tuple at random as initial arest neighbors tuple from tuple to be identified, calculate test tuple respectively and arrive
The distance of this k tuple, unit's deck label to be identified and distance are stored in priority query.
In step S1032, traversal training tuple set, calculate currently tuple to be identified and the distance of test tuple, compare
Ultimate range Lmax in L with priority query for the gained, obtains final priority query.
In this embodiment, compare ultimate range Lmax in L and priority query for the gained, obtain final preferential
Level queue, including：
When described gained is more than ultimate range Lmax in priority query apart from L, then gives up this tuple, travel through next
Individual tuple.
When described gained is less than ultimate range Lmax in priority query apart from L, then delete in priority query
The tuple of big distance, current training tuple is stored in priority query.
In step S1033, traversal finishes, and calculates many several classes ofs of k tuple in priority query, and as test
The classification of tuple.
Although KNN method also relies on limit theorem from principle, in classification decision-making, only with minimal amount of adjacent sample
This is relevant.Because KNN method is mainly by neighbouring sample limited around rather than affiliated by differentiating that the method for class field to determine
Classification, therefore for the intersection as this class field of hand-written Manchu alphabet or overlapping more sample set to be divided, KNN method
It is more suitable for compared with additive method.
In this embodiment, it is easily programmed using K- near neighbor method, and do not need to optimize；
In this embodiment, hand-written Manchu alphabet sample size to be identified is big, K value size to fit, and K- nearest neighbour classification is by mistake
Difference is little, and discrimination is higher.
In addition, as shown in Figure 1A, before step S101, also include：
In the step s 100, pretreatment operation is carried out to Manchu alphabet image to be identified.
The main purpose of pretreatment is to reduce the variation between different samples in same Manchu alphabet class, that is, strengthen in class
The degree of polymerization.
In this embodiment, Manchu alphabet image to be identified is carried out with pretreatment operation mainly to include to be identified is expired
Civilian letter image carries out the linear normalization of character boundary, plus virtual stroke, the non-linear normalizing of character, point in stroke
The operation such as smooth of the point in resampling, stroke.
By Manchu alphabet image to be identified is carried out with the operation of pretreatment, can reduce well in class and make a variation to identification
During interference, be conducive to improve Manchu alphabet image recognition efficiency.
Fig. 2 is the flow chart of the another kind hand-written Manchu alphabet recognition methodss according to an exemplary embodiment, such as Fig. 2
Show, before step S101, can also comprise the steps：
In step S104, carry out model training using Manchu alphabet data sample.
In this embodiment, as shown in Figure 2 A, step S104 comprises the steps：
In step S1041, carry out LDA dimensionality reduction matrix calculus, obtain best projection vector in linear space.
In this embodiment, the purpose carrying out LDA dimensionality reduction matrix calculus is：By being calculated best projection vector, and
So that the sample after projection has divergence in the class scatter of maximum and the class of minimum.
Fig. 3 is a kind of method flow diagram carrying out LDA dimensionality reduction matrix calculus according to an exemplary embodiment, including
Following steps：
In step s 201, pretreatment is carried out to Manchu alphabet image pattern to be identified；
In step S202, it is respectively adopted Directional feature extraction method and thick meshed feature extracting method to Manchu alphabet sample
This extraction feature, obtains Manchu alphabet eigenvalue；
In step S203, distribution data space stores Manchu alphabet eigenvalue and label；
In step S204, calculate Different categories of samples expectation and this expectation of gross sample；
In step S205, calculate covariance matrix Sw in covariance matrix Sb and class between class；
In this embodiment, step S205 be by each sample in all Manchu alphabet eigenvalue samples according to belonging to oneself
Manchu alphabet class calculate the summation of sample and overall covariance matrix, this from macroscopically describe all classes and overall it
Between discrete degree of redundancy.Then calculate in each Manchu alphabet class the covariance matrix between each sample and affiliated class it
With what this matrix was portrayed is each sample and the dispersion between such, institute in this embodiment in class as a whole
The class feature portrayed is to be made up of the average value matrix of each sample in class.
In step S206, seek matrix Sw-1The characteristic vector of Sb, obtains projection vector.
In this embodiment, step S206 mainly uses Fisher criterion and calculates eigenvalue and characteristic vector, Fisher
Differentiate that divergence in the class scatter on projection vector and class is combined together criterion function by sample, for determining optimum projection side
To providing a criterion.When criterion function reach maximum vector when, determined by projecting direction be exactly best projection vector.
Concrete grammar is：After obtaining eigenvalue and characteristic vector, retain d maximum eigenvalue, and corresponding to this d eigenvalue
Characteristic vector, and by orthogonal for these characteristic vectors standardization, thus constituting the basic matrix of new subspace, that is, project to
Amount.
In step S1042, by setting different K value repetition trainings, the parameter value of adjustment K- neighbour.
In this embodiment, purpose K- Nearest Neighbor Classifier being trained is parameter K of adjustment grader, so that fall
After Manchu alphabet image feature value to be identified after dimension enters grader, there is accurate classifying quality.
Fig. 4 is a kind of method flow diagram that K- Nearest Neighbor Classifier is trained according to an exemplary embodiment,
Comprise the steps：
In step S301, pretreatment is carried out to Manchu alphabet image pattern to be identified；
In step s 302, it is respectively adopted Directional feature extraction method and thick meshed feature extracting method to described to be identified
Manchu alphabet image pattern extract feature, obtain Manchu alphabet eigenvalue；
In step S303, the method using the linear judgment analysis of LDA carries out dimensionality reduction to described Manchu alphabet eigenvalue, obtains
Manchu alphabet eigenvalue to after dimensionality reduction；
In step s 304, distribution data space stores training data and test tuple, parameter preset K respectively；
In step S305, safeguard size be k by apart from descending priority query, wait to know for storing
Other tuple.Choose k tuple at random as initial arest neighbors tuple from tuple to be identified, calculate test tuple respectively and arrive this
The distance of k tuple, unit's deck label to be identified and distance are stored in priority query；
In step S306, traversal training tuple set, calculate currently tuple to be identified and the distance of test tuple, compare institute
Obtain ultimate range Lmax in L with priority query, obtain final priority query；
In one embodiment, compare ultimate range Lmax in L and priority query for the gained, obtain final preferential
Level queue, including：
When described gained is more than ultimate range Lmax in priority query apart from L, then gives up this tuple, travel through next
Individual tuple.
When described gained is less than ultimate range Lmax in priority query apart from L, then delete in priority query
The tuple of big distance, current training tuple is stored in priority query.
In step S307, traversal finishes, and calculates many several classes ofs of k tuple in priority query, and as test
The classification of tuple.
In step S308, test tuple set is completed rear calculation error rate, continues the different K values of setting and again instructs
Practice, finally take the minimum K value of error rate.
As can be seen here, the above-mentioned mode being trained using hand-written Manchu alphabet data sample simple, effectively, be easy to real
Existing.
Above-described embodiment, by carrying out model training to sample, it is possible to obtain preferably model parameter, thus carrying for identification
For condition.
Corresponding with aforementioned hand-written Manchu alphabet recognition methodss embodiment, the disclosure additionally provides hand-written Manchu alphabet identification
Device embodiment.
Fig. 6 is the block diagram of a kind of hand-written Manchu alphabet identifying device according to an exemplary embodiment, as Fig. 6 institute
Show, hand-written Manchu alphabet identifying device includes：Extraction module 51, dimensionality reduction module 52 and sort module 53.
Extraction module 51, is configured to using Directional feature extraction method and thick meshed feature extracting method to be identified
Manchu alphabet image zooming-out feature, obtains Manchu alphabet eigenvalue；
Dimensionality reduction module 52, is configured to using the method for the linear judgment analysis of LDA, described Manchu alphabet eigenvalue be carried out
Dimensionality reduction, obtains the Manchu alphabet eigenvalue after dimensionality reduction；
Sort module 53, is configured to using K- near neighbor method, the Manchu alphabet eigenvalue after described dimensionality reduction be classified
Process, obtain the described corresponding Manchu alphabet of Manchu alphabet image to be identified.
In one embodiment, described device also includes：
Pretreatment module 54, is configured to Manchu alphabet image to be identified and carries out pretreatment.
The main purpose of pretreatment is to reduce the variation between different samples in same Manchu alphabet class, that is, strengthen in class
The degree of polymerization.
In this embodiment, Manchu alphabet image to be identified is carried out with pretreatment operation mainly to include to be identified is expired
Civilian letter image carries out the linear normalization of character boundary, plus virtual stroke, the non-linear normalizing of character, point in stroke
The operation such as smooth of the point in resampling, stroke.
In one embodiment, described device also includes：
Training module 50, is configured to, with Manchu alphabet data sample and carries out model training.
In one embodiment, described training module includes：
First process submodule 501, is configured to LDA dimensionality reduction matrix calculus, obtains best projection vector in linear space；
Second processing submodule 502, is configured to set different K value repetition trainings, the parameter value of adjustment K- neighbour.
In one embodiment, described first process submodule 501 includes：
Pretreatment module 5011, is configured to carry out pretreatment to described Manchu alphabet image pattern to be identified.
Extraction module 5012, is configured to using Directional feature extraction method and thick meshed feature extracting method to be identified
Manchu alphabet image pattern extract feature, obtain Manchu alphabet eigenvalue；
LDA dimensionality reduction matrix calculus module 5013, is configured to LDA dimensionality reduction matrix is calculated, obtains in linear space
Best projection vector.
In this embodiment, the purpose carrying out LDA dimensionality reduction matrix calculus is：By being calculated best projection vector, and
So that the sample after projection has divergence in the class scatter of maximum and the class of minimum.
In one embodiment, described second processing submodule 502 includes：
Pretreatment module 5021, is configured to carry out pretreatment to described Manchu alphabet image pattern to be identified.
Extraction module 5022, is configured to using Directional feature extraction method and thick meshed feature extracting method to be identified
Manchu alphabet image pattern extract feature, obtain Manchu alphabet eigenvalue；
Dimensionality reduction module 5023, is configured to using the method for the linear judgment analysis of LDA, described Manchu alphabet eigenvalue be entered
Row dimensionality reduction, obtains the Manchu alphabet eigenvalue after dimensionality reduction；
KNN training module 5024, is configured to set different K value repetition trainings, the parameter value of adjustment K- neighbour.
In this embodiment, purpose K- Nearest Neighbor Classifier being trained is parameter K of adjustment grader, so that fall
After Manchu alphabet image feature value after dimension enters grader, there is accurate classifying quality.
In one embodiment, described extraction module 51 is configured to carry using Directional feature extraction method and thick meshed feature
Take method to Manchu alphabet image zooming-out feature to be identified, obtain Manchu alphabet eigenvalue.Extraction module 51 includes：
Directional feature extraction submodule 511, is configured to carry out 8 direction characters to Manchu alphabet image to be identified carry
Take, or 8 Directional feature extraction are carried out to Manchu alphabet image pattern to be identified, obtain the direction character value of Manchu alphabet.
In this embodiment, described 8 Directional feature extraction are carried out to Manchu alphabet image to be identified, or to be identified
Manchu alphabet image pattern to carry out 8 Directional feature extraction methods as follows：After pretreatment is carried out to Manchu alphabet image, obtain
The mapping relations of the new hand-written Manchu alphabet and Manchu alphabet tracing point point before and after nonlinear change, afterwards letter point
Become 8 × 8 grid, the center taking each lattice is sampling center, obtains 8 × 8 sampled point, then to these centers of sampling
With Gabor envelope extraction 8- direction character.
Thick meshed feature extracting sub-module 512, is configured to carry out thick meshed feature to Manchu alphabet image to be identified
Extract, or thick meshed feature extraction is carried out to Manchu alphabet image pattern to be identified, obtain the grid search-engine of Manchu alphabet
Value.
In this embodiment, described thick meshed feature extraction is carried out to Manchu alphabet image to be identified, or treat knowledge
It is as follows that other Manchu alphabet image pattern carries out thick meshed feature extracting method：Obtain the final language of the Manchus obtaining after pretreatment
The outer rim of letter image；Then the Manchu alphabet image segmentation adding after frame is become 8 × 8 grids；Count each net successively
Black pixel ratio in lattice；Finally 8 × 8 black pixels after statistics are formed a line and form the grid search-engine vector of 64 dimensions.
In one embodiment, described sort module 53 is configured to using K- near neighbor method to the language of the Manchus word after described dimensionality reduction
Female eigenvalue carries out classification process, obtains the described corresponding Manchu alphabet of Manchu alphabet image to be identified.K- near neighbor method
Thinking is：If a sample is most in the sample of the k in feature space most like (i.e. closest in feature space)
Number belongs to some classification, then this sample falls within this classification.In KNN algorithm, selected neighbours are correctly to divide
The object of class.The method only to determine sample to be divided according to the classification of one or several closest samples on categorised decision
Affiliated classification.
In one embodiment, described sort module 53 includes：
Metrics calculation unit 531, is configured to calculate currently tuple to be identified and the distance testing tuple.
Distance metric adopts Lp distance (p=2), i.e. Euclidean distance.Due to having carried out pretreatment operation in this embodiment,
So that after before tolerance, the value of each attribute is standardization, it is to avoid there is the attribute of larger initial codomain and have
The weighted problem of the attribute of less initial codomain.
Relatively select unit 532, is configured to compare the ultimate range in gained distance and priority query, obtains final
Priority query.Traversal is waited after terminating, to select many several classes ofs of K tuple in priority query, as test tuple
Classification.
In this embodiment, compare ultimate range Lmax in L and priority query for the gained, obtain final preferential
Level queue, including：
When described gained is more than ultimate range Lmax in priority query apart from L, then gives up this tuple, travel through next
Individual tuple.
When described gained is less than ultimate range Lmax in priority query apart from L, then delete in priority query
The tuple of big distance, current training tuple is stored in priority query.
With regard to the device in above-described embodiment, wherein modules, the concrete mode of submodule execution operation is having
It has been described in detail in the embodiment closing the method, explanation will be not set forth in detail herein.
Figure 12 is a kind of block diagram being applied to Manchu alphabet identifying device according to an exemplary embodiment.For example,
Device 1200 can be ARM, mobile phone, messaging devices, the embedded device such as tablet device.
With reference to Figure 12, device 1200 can include following one or more assemblies：Process assembly 1202, memorizer 1204,
Power supply module 1206, multimedia groupware 1208, audio-frequency assembly 1210, the interface 1212 of input/output (I/O), and communication set
Part 1214.
The integrated operation of the usual control device 1200 of process assembly 1202, such as with display, data calculate, data communication and
The associated operation of record operation.Process assembly 1202 can include one or more processors 1220 and carry out execute instruction, with complete
Become all or part of step of above-mentioned method.Additionally, process assembly 1202 can include one or more modules, it is easy to process
Interaction between assembly 1202 and other assemblies.For example, process assembly 1202 can include multi-media module, to facilitate multimedia
Interaction between assembly 1208 and process assembly 1202.
Memorizer 1204 is configured to store various types of data to support the operation in equipment 1200.These data
Example include on device 1200 operation any application program or method instruction, text data, message, picture, depending on
Frequency etc..Memorizer 1204 can be realized by any kind of volatibility or non-volatile memory device or combinations thereof, such as
Static RAM (SRAM), Electrically Erasable Read Only Memory (EEPROM), erasable programmable is read-only to be deposited
Reservoir (EPROM), programmable read only memory (PROM), read only memory (ROM), magnetic memory, flash memory etc..
Electric power assembly 1206 provides electric power for the various assemblies of device 1200.Electric power assembly 1206 can include power management
System, one or more power supplys, and other generate, manage and distribute, with for device 1200, the assembly that electric power is associated.
Multimedia groupware 1208 includes the screen of one output interface of offer between described device 1200 and user.Should
Screen includes liquid crystal display (LCD) and touch panel (TP), to receive the input signal from user.In some embodiments
In, multimedia groupware 1208 includes one or more photographic head.When equipment 1200 is in operator scheme, such as exposal model or video recording
During pattern, photographic head can receive the multi-medium data of outside.
Audio-frequency assembly 1210 is configured to output and/or input audio signal.For example, audio-frequency assembly 1210 includes a wheat
Gram wind (MIC), when device 1200 is in operator scheme, such as recording mode when, mike is configured to receive external audio signal.
The audio signal being received can be further stored in memorizer 1204 or send via communication component 1214.In some enforcements
In example, audio-frequency assembly 1210 also includes a speaker, for exports audio signal.
I/O interface 1212 is for providing interface, above-mentioned peripheral interface module between process assembly 1202 and peripheral interface module
Can be button, touch key-press etc..These buttons can include but is not limited to：Volume button, SR.
Communication component 1214 is configured to facilitate the communication of wired or wireless way between device 1200 and other equipment.Dress
Put 1200 and can access wireless network based on communication standard, such as WiFi, 2G or 3G, or combinations thereof.Exemplary at one
In embodiment, described communication component 1214 also includes serial communication module, to complete the communication with host computer.
In the exemplary embodiment, device 1200 can be by one or more application specific integrated circuits (ASIC), numeral
Signal processor (DSP), digital signal processing appts (DSPD), PLD (PLD), ready-made programmable gate array
(FPGA) controller, microcontroller, microprocessor or other electronic components are realized, for executing said method.
Those skilled in the art, after considering description and putting into practice disclosure disclosed herein, will readily occur to its of the disclosure
His embodiment.The application is intended to any modification, purposes or the adaptations of the disclosure, these modifications, purposes or
Person's adaptations are followed the general principle of the disclosure and are included the undocumented common knowledge in the art of the disclosure
Or conventional techniques.Description and embodiments be considered only as exemplary, the true scope of the disclosure and spirit by following
Claim is pointed out.
It should be appreciated that the disclosure is not limited to be described above and precision architecture illustrated in the accompanying drawings, and
And various modifications and changes can carried out without departing from the scope.The scope of the present disclosure only to be limited by appended claim.
Claims (10)
1. a kind of system of hand-written Manchu alphabet identification, includes process assembly (1202), process assembly (1202) inclusion one or
Multiple processors (1220) carry out execute instruction, to complete the step that method for distinguishing known in the language of the Manchus；
Described language of the Manchus recognition methodss comprise the steps：
It is respectively adopted Directional feature extraction method and thick meshed feature extracting method special to Manchu alphabet image zooming-out to be identified
Levy, obtain Manchu alphabet eigenvalue；
Method using the linear judgment analysis of LDA carries out dimensionality reduction to described Manchu alphabet eigenvalue, obtains the language of the Manchus word after dimensionality reduction
Female eigenvalue；
Classification process is carried out to the Manchu alphabet eigenvalue after described dimensionality reduction using K- near neighbor method, obtains described to be identified expiring
The corresponding Manchu alphabet of civilian letter image.
2. the system of hand-written Manchu alphabet identification as claimed in claim 1 is it is characterised in that also include following one or more
Assembly：Memorizer (1204), power supply module (1206), multimedia groupware (1208), audio-frequency assembly (1210), input/output
Interface (1212) and communication component (1214)；
Described memorizer (1204) is configured to store various types of data to support the operation in equipment, memorizer (1204)
Realized by any kind of volatibility or non-volatile memory device or combinations thereof；
Described electric power assembly (1206) is various assemblies offer electric power, electric power assembly (1206) inclusion power-supply management system, one
Or multiple power supply, and the assembly that other generate with for identifying system, management and distribution electric power are associated；
Described multimedia groupware (1208) includes the screen of one output interface of offer between described identifying system and user,
This screen includes liquid crystal display and touch panel, to receive the input signal from user；
Described audio-frequency assembly (1210) is configured to output and/or input audio signal；
The interface (1212) of described input/output provides interface for process assembly (1202) and peripheral interface module between；
Described communication component (1214) is configured to facilitate the communication of wired or wireless way between identifying system and other equipment.
3. the system of hand-written Manchu alphabet identification as claimed in claim 1 is it is characterised in that described be respectively adopted direction character
Extracting method and thick meshed feature extracting method, to Manchu alphabet image zooming-out feature to be identified, obtain Manchu alphabet eigenvalue
Method, including：
Manchu alphabet image to be identified carries out 8 Directional feature extraction, obtains the direction character value of Manchu alphabet；
Thick meshed feature extraction is carried out to Manchu alphabet image to be identified, obtains the grid search-engine value of Manchu alphabet；
The direction character value of Manchu alphabet and grid search-engine value are combined as string, obtain Manchu alphabet eigenvalue.
4. the system of hand-written Manchu alphabet identification as claimed in claim 1 is it is characterised in that described employing K near neighbor method pair
Manchu alphabet eigenvalue after described dimensionality reduction carries out classification process, obtains the described corresponding language of the Manchus of Manchu alphabet image to be identified
The method of letter, including：
Safeguard size be k by apart from descending priority query, for storing tuple to be identified.Random from waiting to know
Choose k tuple in other tuple as initial arest neighbors tuple, calculate test tuple respectively to the distance of this k tuple, will
Unit's deck label to be identified and distance are stored in priority query；
Traversal training tuple set, calculates currently tuple to be identified and the distance of test tuple, compares gained apart from L and priority team
Ultimate range Lmax in row, obtains final priority query；
Traversal finishes, and calculates many several classes ofs of k tuple in priority query, and the classification as test tuple.
5. the system of hand-written Manchu alphabet identification as claimed in claim 1 is it is characterised in that described employing LDA linearly adjudicates
The method of analysis carries out dimensionality reduction to described Manchu alphabet eigenvalue, obtains in the step of the Manchu alphabet eigenvalue after dimensionality reduction, fall
The acquisition methods of dimension matrix, including：
Set up the hand-written language of the Manchus storehouse of storage language of the Manchus data sample；
It is respectively adopted Directional feature extraction method and thick meshed feature extracting method to the Manchu alphabet image in hand-written language of the Manchus storehouse
Extract feature, obtain Manchu alphabet eigenvalue；Distribution data space storage Manchu alphabet eigenvalue and label；
Calculate Different categories of samples expectation and this expectation of gross sample；
Calculate covariance matrix Sw in covariance matrix Sb and class between class；
Seek matrix Sw-1The characteristic vector of Sb, obtains projection vector.
6. the system of hand-written Manchu alphabet identification as claimed in claim 1 is it is characterised in that described employing K near neighbor method pair
Manchu alphabet eigenvalue after described dimensionality reduction carries out classification process, obtains the described corresponding language of the Manchus of Manchu alphabet image to be identified
Letter, the k value in k near neighbor method determines method, including：
Set up the hand-written language of the Manchus storehouse of storage language of the Manchus data sample；
It is respectively adopted Directional feature extraction method and thick meshed feature extracting method to the Manchu alphabet image in hand-written language of the Manchus storehouse
Extract feature, obtain Manchu alphabet eigenvalue；
Method using the linear judgment analysis of LDA carries out dimensionality reduction to described Manchu alphabet eigenvalue, obtains the language of the Manchus word after dimensionality reduction
Female eigenvalue；
Distribution data space stores training data and test tuple, parameter preset K respectively；
Safeguard size be k by apart from descending priority query, for storing tuple to be identified, at random from waiting to know
Choose k tuple in other tuple as initial arest neighbors tuple, calculate test tuple respectively to the distance of this k tuple, will
Unit's deck label to be identified and distance are stored in priority query；
Traversal training tuple set, calculates currently tuple to be identified and the distance of test tuple, compares gained apart from L and priority team
Ultimate range Lmax in row, obtains final priority query；
Traversal finishes, and calculates many several classes ofs of k tuple in priority query, and the classification as test tuple.
Test tuple set is completed rear calculation error rate, continues to set different K value re -training, finally takes error rate minimum
K value.
7. the system of hand-written Manchu alphabet identification as claimed in claim 6 is it is characterised in that the k value in k near neighbor method is true
Determine in method, described described Manchu alphabet eigenvalue is carried out in the step of dimensionality reduction, the acquisition methods of the dimensionality reduction matrix of use,
Including：
Distribution data space storage Manchu alphabet eigenvalue and label；
Calculate Different categories of samples expectation and this expectation of gross sample；
Calculate covariance matrix Sw in covariance matrix Sb and class between class；
Seek matrix Sw-1The characteristic vector of Sb, obtains projection vector.
8. the system of the hand-written Manchu alphabet identification as described in claim 4 or 6 is it is characterised in that the described gained that compares is apart from L
With ultimate range Lmax in priority query, the method obtaining final priority query, including：
When described gained is more than ultimate range Lmax in priority query apart from L, then give up this tuple, the next unit of traversal
Group.
When described gained is less than ultimate range Lmax in priority query apart from L, then delete in priority query maximum away from
From tuple, current training tuple is stored in priority query.
9. the hand-written Manchu alphabet identification as described in any one of claim 1-8 system it is characterised in that described to be identified
Manchu alphabet image zooming-out feature before, there is the step carrying out pretreatment to Manchu alphabet image pattern to be identified.
10. the system of hand-written Manchu alphabet identification as described in any of claims 9 is it is characterised in that described pretreatment,
Including Manchu alphabet image is carried out with character boundary linear normalization, plus virtual stroke, the non-linear normalizing of character, stroke
On the resampling of point, point in stroke one or more of smooth.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610752646.4A CN106408002A (en) | 2016-08-29 | 2016-08-29 | Hand-written manchu alphabet identification system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610752646.4A CN106408002A (en) | 2016-08-29 | 2016-08-29 | Hand-written manchu alphabet identification system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106408002A true CN106408002A (en) | 2017-02-15 |
Family
ID=58003772
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610752646.4A Pending CN106408002A (en) | 2016-08-29 | 2016-08-29 | Hand-written manchu alphabet identification system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106408002A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108596183A (en) * | 2018-04-24 | 2018-09-28 | 大连民族大学 | The overdivided region merging method of language of the Manchus component cutting |
| CN109063670A (en) * | 2018-08-16 | 2018-12-21 | 大连民族大学 | Block letter language of the Manchus word recognition methods based on prefix grouping |
| CN109558830A (en) * | 2018-11-27 | 2019-04-02 | 钟祥博谦信息科技有限公司 | Hand-written recognition method, device and equipment |
| CN110210476A (en) * | 2019-05-24 | 2019-09-06 | 北大方正集团有限公司 | Basic character component clustering method, device, equipment and computer readable storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1570958A (en) * | 2004-04-23 | 2005-01-26 | 清华大学 | Method for identifying multi-font multi-character size print form Tibetan character |
| CN101510259A (en) * | 2009-03-18 | 2009-08-19 | 西北民族大学 | On-line identification method and recognition system for 'ding' of handwriting Tibet character |
| CN102622610A (en) * | 2012-03-05 | 2012-08-01 | 西安电子科技大学 | Handwritten Uyghur character recognition method based on classifier integration |
-
2016
- 2016-08-29 CN CN201610752646.4A patent/CN106408002A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1570958A (en) * | 2004-04-23 | 2005-01-26 | 清华大学 | Method for identifying multi-font multi-character size print form Tibetan character |
| CN101510259A (en) * | 2009-03-18 | 2009-08-19 | 西北民族大学 | On-line identification method and recognition system for 'ding' of handwriting Tibet character |
| CN102622610A (en) * | 2012-03-05 | 2012-08-01 | 西安电子科技大学 | Handwritten Uyghur character recognition method based on classifier integration |
Non-Patent Citations (4)
| Title |
|---|
| 玛依热•依布拉音 等: "联机手写维吾尔文字母识别方法", 《模式识别与人工智能》 * |
| 百度百科: "LDA", 《HTTPS://BAIKE.BAIDU.COM/HISTORY/LDA/106311832》 * |
| 百度百科: "临近算法", 《HTTPS://BAIKE.BAIDU.COM/HISTORY/%E9%82%BB%E8%BF%91%E7%AE%97%E6%B3%95/73677073》 * |
| 许爽 等: "满文识别技术研究与分析", 《大连民族学院学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108596183A (en) * | 2018-04-24 | 2018-09-28 | 大连民族大学 | The overdivided region merging method of language of the Manchus component cutting |
| CN109063670A (en) * | 2018-08-16 | 2018-12-21 | 大连民族大学 | Block letter language of the Manchus word recognition methods based on prefix grouping |
| CN109558830A (en) * | 2018-11-27 | 2019-04-02 | 钟祥博谦信息科技有限公司 | Hand-written recognition method, device and equipment |
| CN110210476A (en) * | 2019-05-24 | 2019-09-06 | 北大方正集团有限公司 | Basic character component clustering method, device, equipment and computer readable storage medium |
| CN110210476B (en) * | 2019-05-24 | 2021-04-09 | 北大方正集团有限公司 | Character component clustering method, device, equipment and computer readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109697416B (en) | Video data processing method and related device | |
| CN105242779B (en) | A kind of method and mobile intelligent terminal of identification user action | |
| US9104242B2 (en) | Palm gesture recognition method and device as well as human-machine interaction method and apparatus | |
| CN102687140B (en) | For contributing to the method and apparatus of CBIR | |
| US7447338B2 (en) | Method and system for face detection using pattern classifier | |
| CN110210535A (en) | Neural network training method and device and image processing method and device | |
| CN101331476B (en) | System and method for handwritten character recognition | |
| CN109977262A (en) | The method, apparatus and processing equipment of candidate segment are obtained from video | |
| CN109948447B (en) | Character network relation discovery and evolution presentation method based on video image recognition | |
| CN109952614A (en) | The categorizing system and method for biomone | |
| Li et al. | Df 2 net: Discriminative feature learning and fusion network for rgb-d indoor scene classification | |
| CN106028134A (en) | Detect sports video highlights for mobile computing devices | |
| CN106408002A (en) | Hand-written manchu alphabet identification system | |
| CN110020592A (en) | Object detection model training method, device, computer equipment and storage medium | |
| CN104504362A (en) | Face detection method based on convolutional neural network | |
| CN112819686B (en) | Image style processing method and device based on artificial intelligence and electronic equipment | |
| CN110781829A (en) | Light-weight deep learning intelligent business hall face recognition method | |
| CN104331498A (en) | Method for automatically classifying webpage content visited by Internet users | |
| CN101930549B (en) | Second generation curvelet transform-based static human detection method | |
| CN113239914B (en) | Classroom student expression recognition and classroom state evaluation method and device | |
| EP2535787B1 (en) | 3D free-form gesture recognition system and method for character input | |
| CN107590427A (en) | Monitor video accident detection method based on space-time interest points noise reduction | |
| CN113076903A (en) | Target behavior detection method and system, computer equipment and machine readable medium | |
| CN115545103A (en) | Abnormal data identification method, label identification method and abnormal data identification device | |
| CN110175500B (en) | Finger vein comparison method, device, computer equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into 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: 20170215 |