CN109034021A - A kind of recognition methods again for easily obscuring digital handwriting body - Google Patents
A kind of recognition methods again for easily obscuring digital handwriting body Download PDFInfo
- Publication number
- CN109034021A CN109034021A CN201810767088.8A CN201810767088A CN109034021A CN 109034021 A CN109034021 A CN 109034021A CN 201810767088 A CN201810767088 A CN 201810767088A CN 109034021 A CN109034021 A CN 109034021A
- Authority
- CN
- China
- Prior art keywords
- digital
- dimensional space
- digital handwriting
- handwriting body
- handwriting
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 25
- 239000011159 matrix material Substances 0.000 claims description 27
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 238000002474 experimental method Methods 0.000 claims description 6
- 238000013461 design Methods 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 229910017435 S2 In Inorganic materials 0.000 claims 1
- 238000010801 machine learning Methods 0.000 abstract description 2
- 238000003909 pattern recognition Methods 0.000 abstract description 2
- 238000012800 visualization Methods 0.000 description 23
- 230000006870 function Effects 0.000 description 18
- 238000011156 evaluation Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 4
- 238000013507 mapping Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
Classifications
-
- 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
- G06V30/32—Digital ink
- G06V30/333—Preprocessing; Feature extraction
-
- 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
- G06V30/24—Character recognition characterised by the processing or recognition method
- G06V30/242—Division of the character sequences into groups prior to recognition; Selection of dictionaries
- G06V30/244—Division of the character sequences into groups prior to recognition; Selection of dictionaries using graphical properties, e.g. alphabet type or font
-
- 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
- G06V30/32—Digital ink
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Character Discrimination (AREA)
Abstract
The invention discloses a kind of recognition methods again for easily obscuring digital handwriting body, belong to pattern-recognition and machine learning techniques field.The present invention first identifies that easily obscuring hand number writes body 4,9 and easily obscure hand number and write body 3,5,8, again identifies that it, can effectively identify the similar handwriting digital of shape according to handwriting digital 0-9;It is better than existing t-SNE simultaneously, avoids the identified mistake of a large amount of handwriting digitals compared to existing method and reduce working efficiency.
Description
Technical field
The present invention relates to a kind of recognition methods again for easily obscuring digital handwriting body, belong to pattern-recognition and machine learning techniques
Field.
Background technique
With the extensive use of human-computer interaction and image recognition technology in life, the identification problem of digital handwriting body is obtained
Increasingly in-depth study.Since Different Culture, Different Individual have different writing styles, and the even same person,
Due to writing the difference of the extraneous factors such as environment, ways of writing environment, it all may cause and write the inconsistent of result, this makes often
A handwriting digital presented be characterized in it is diversified.Existing number 4 is similar with the shape of number 9 in digital handwriting body, number
Word 3,5,8 is also easily obscured because shape is similar, this makes troubles to the identification of handwriting digital.It in practical applications, if will
Number identification mistake in phone, it will cause number errors, to delay important thing;If by the postcode number in address
Identify mistake, it will cause the miscarryings of mail, express delivery to pass.
Summary of the invention
The present invention provides a kind of recognition methods again for easily obscuring digital handwriting body, to obscure digital handwriting body for realizing easy
Identification.
The technical scheme is that a kind of recognition methods again for easily obscuring digital handwriting body, the method comprises the following steps:
Step1: it is pre-processed to easily digital handwriting body is obscured: n easily being obscured the composition experiment of digital handwriting body picture
Sample database, then convert each picture to the vector of 1*m, then the vector of n 1*m is subjected to binary conversion treatment, most end form respectively
Digital handwriting body observing matrix is easily obscured at the n*m being made of 0 and 1;
Step2: it to the gaussian kernel function of n*m easily obscured digital handwriting body observing matrix and calculate its higher dimensional space, goes forward side by side
Row normalized is grouped the gaussian kernel function of the higher dimensional space after normalization, adds respectively to each grouping
Power, obtains the Weighted Gauss kernel function of higher dimensional space;
Step3: digital handwriting body observing matrix design conditions are obscured to easy using the Weighted Gauss kernel function of step Step2
Probability, then higher dimensional space joint probability is calculated with obtained conditional probability;
Step4: out of normal distribution, the Y matrix of one n*2 of random initializtion calculates low-dimensional to the Y matrix of n*2
Space gaussian kernel function calculates lower dimensional space joint probability using lower dimensional space gaussian kernel function;It is general according to higher dimensional space joint
Rate and lower dimensional space joint probability use KL divergence calculating target function, finally judge objective function and the last time of current iteration
Whether the difference of the objective function of iteration restrains: if difference restrains, exporting the dimensionality reduction result Y matrix of lower dimensional space;Otherwise,
Then continue iteration and update objective function, until difference restrains.
It is described easily to obscure digital handwriting body and obtain as follows:
Digital handwritten form 0-9 is pre-processed: by r digital handwritten form picture composition experiment sample library, then will be each
Picture is converted into the vector of 1*s, then the vector of r 1*s is carried out binary conversion treatment respectively, ultimately forms the r* being made of 0 and 1
The digital handwriting body observing matrix of s;
T-SNE algorithm is carried out to the digital handwriting body observing matrix of r*s, dimensionality reduction is calculated as a result, obtaining according to dimensionality reduction result
4,9 easily obscure digital handwriting body easily to obscure digital handwriting body, 3,5,8 out.
In the step S2, three groups are grouped into, three groups of weights being corresponding in turn to are (0,1), [1], (1 ,+∞).
The beneficial effects of the present invention are: the present invention first identifies that easily obscuring hand number writes body 4,9 and from digital handwriting body 0-9
Easily obscure hand number and write body 3,5,8, then easily obscure hand number to every kind and write body and again identify that, can effectively identify shape
Similar handwriting digital;It is better than existing t-SNE simultaneously, it is identified to avoid a large amount of handwriting digitals compared to existing method
Mistake and reduce working efficiency.
Detailed description of the invention
Fig. 1 is a kind of experimentation figure of recognition methods again for easily obscuring digital handwriting body of the present invention;
Fig. 2 is a kind of specific experiment flow chart of recognition methods again for easily obscuring digital handwriting body of the present invention;
Fig. 3 is that the present invention easily obscures handwriting digital 0-9 progress t-SNE algorithm dimensionality reduction result figure one;
Fig. 4 is that the present invention easily obscures handwriting digital 0-9 progress t-SNE algorithm dimensionality reduction result figure two;
Fig. 5 is that the present invention easily obscures handwriting digital 0-9 progress t-SNE algorithm dimensionality reduction result figure three;
Fig. 6 is that the present invention easily obscures handwriting digital 4,9, and the dimensionality reduction that t-SNE algorithm is carried out in the case where 2000 groups can
Depending on changing result figure;
Fig. 7 is that the present invention easily obscures handwriting digital 4,9, and the dimensionality reduction that t-SNE algorithm is carried out in the case where 4000 groups can
Depending on changing result figure;
Fig. 8 is that the present invention easily obscures handwriting digital 3,5,8, and the dimensionality reduction of t-SNE algorithm is carried out in the case where 2000 groups
Visualization result figure;
Fig. 9 is that the present invention easily obscures handwriting digital 3,5,8, and the dimensionality reduction of t-SNE algorithm is carried out in the case where 4000 groups
Visualization result figure;
Figure 10 present invention easily obscures handwriting digital 4,9, and weighting t-SNE algorithm is grouped in the case where 2000 groups
Dimension reduction and visualization result figure;
Figure 11 is that the present invention easily obscures handwriting digital 4,9, and weighting t-SNE algorithm is grouped in the case where 4000 groups
Dimension reduction and visualization result figure;
Figure 12 is that the present invention easily obscures handwriting digital 3,5,8, and weighting t-SNE is grouped in the case where 2000 groups and is calculated
The dimension reduction and visualization result figure of method;
Figure 13 is that the present invention easily obscures handwriting digital 3,5,8, and weighting t-SNE is grouped in the case where 4000 groups and is calculated
The dimension reduction and visualization result figure of method;
Figure 14 is that the present invention easily obscures handwriting digital in the case where 2000 groups, then the recall ratio of identification and evaluation index
Comparison diagram.
Figure 15 is that the present invention easily obscures handwriting digital in the case where 4000 groups, then the recall ratio of identification and evaluation index
Comparison diagram.
Figure 16 is that the present invention easily obscures handwriting digital in the case where 2000 groups, then the precision ratio of identification and evaluation index
Comparison diagram.
Figure 17 is that the present invention easily obscures handwriting digital in the case where 4000 groups, then the precision ratio of identification and evaluation index
Comparison diagram.
Specific embodiment
Embodiment 1: it is a kind of easily to obscure the recognition methods again of digital handwriting body as shown in Fig. 1-17, the method it is specific
Steps are as follows:
First, handwriting digital 0-9 picture is pre-processed, i.e., it is real easily to obscure r digital handwriting body picture composition
Sample database is tested, then converts each picture to the vector of 1*s, then the vector of r 1*s is subjected to binary conversion treatment respectively, finally
Form the r*s being made of 0 and 1 easily obscures digital handwriting body observing matrix;S=row * column;
Second, the digital handwriting body observing matrix that handwriting digital 0-9 is formed calculates the Gaussian kernel letter in higher dimensional space
Number:
The digital handwriting body observing matrix of formation is calculated to the conditional probability p of higher dimensional space againi|jAnd pj|i:
Wherein, δ is the vector standard deviation of Gaussian function.
Third, then calculate the joint probability p of higher dimensional spaceijAre as follows:
With lesser standard deviation, the Y matrix of random initializtion one n*2 for meeting normal distribution, to the Y matrix meter of n*2
Lower dimensional space gaussian kernel function is calculated, lower dimensional space joint probability q is calculated according to the gaussian kernel function of lower dimensional spaceijAre as follows:
This method final optimization pass object definition are as follows: combined by the joint probability and lower dimensional space that minimize higher dimensional space general
KL divergence between rate obtains optimal result.Shown in the objective function of KL divergence such as formula (6):
The gradient of objective function is formula (7):
The initial value y of gradient descent method mirror point(0)It is generated at random from Gaussian Profile, in order to accelerate the speed of optimizing
And avoid falling into local optimum, then momentum term u is added, then the newer of available low-dimensional mirror point y:Wherein, u indicates current iteration number.Fig. 2 is complete flow chart,
U indicates iteration time in figure, and L indicates maximum number of iterations, and ε indicates convergency value.
Fig. 3,4,5 are respectively that the digital handwriting body observing matrix of 2500 handwriting digital 0-9 is carried out by t-SNE algorithm
Dimensionality reduction and the dimensionality reduction result for being mapped to two-dimensional coordinate;Fig. 3,4,5 are identical sample, are repeatedly identified, are all 4,9 easily to obscure, 3,5,8
Easily obscure.Wherein 0-9 digital handwriting body be have a bad handwriting easily obscure digital handwriting body.
4th, it can be seen that from Fig. 3-Fig. 5 relative to other numbers, since the shape of number 4 and digital 9 is similar, can
Occurs more coincidence in result depending on changing, the distribution of number 3,5,8 is also more close, and easily obscuring for being difficult to differentiate between is hand-written
Body number 4,9 easily obscures handwriting digital 3,5,8;(easily obscures handwriting digital 4,9 to be tested as a kind of, easily obscure
Handwriting digital 3,5,8 is tested as a kind of)
5th, first step operation, image are carried out to easily obscuring digital handwriting body 4,9 and easily obscuring digital handwriting body 3,5,8
Pretreatment, then carries out second step operation to it, carries out dimensionality reduction using t-SNE algorithm, obtain visualization dimensionality reduction as a result, as Fig. 6,7,
8, shown in 9.
6th, to easily obscuring digital handwriting body 4,9 and easily obscure digital handwriting body 3,5,8 and carry out first step operation respectively,
Image preprocessing, obtained digital handwriting body observing matrix of easily obscuring are grouped weighting t-SNE algorithm (i.e. GWt-SNE) meter
It calculates, specific as follows:
It is pre-processed to easily digital handwriting body is obscured: n easily being obscured digital handwriting body picture composition experiment sample library,
It again converts each picture to the vector of 1*m, then the vector of n 1*m is subjected to binary conversion treatment respectively, ultimately form by 0 and 1
The n*m's of composition easily obscures digital handwriting body observing matrix;
The gaussian kernel function that digital handwriting body observing matrix calculates its higher dimensional space is easily obscured to n*m, such as formula (1), and
It is normalized, obtains matrix M:
Work as dij=dminWhen, it is 0 that M, which is minimized,;Work as dij=dmaxWhen, it is 1 that M, which is maximized,.By formula (1) it is recognised that plus
The gaussian kernel function of power can indicate are as follows:
Wherein, α indicates the weight of each distance classification, the corresponding different grouping of the selection of weight, in the range of (0,1),
[1], (1 ,+∞).This use-case is easily obscured digital handwriting body 4,9 and easily obscure digital handwriting body 3,5,8 use with [0,
0.25), [0.25,0.625], (0.625,1.0] section grouping.I.e. [0,0.25) respective weights (0,1), [0.25,0.625]
Respective weights [1], (0.625,1.0] respective weights (1 ,+∞), when specifically calculating, the value in (0,1), the section (1 ,+∞) is random
It chooses.
The present invention obtains Weighted distance d*=α d=α | | xi-xj||2.The conditional probability p of sample higher dimensional space at this timei|j
And pj|iBecome:
Third step operation is carried out to it again, calculates higher dimensional space joint probability p according to formula (4), (5), (6), (7)ij, it is low
Dimension space joint probability is qijAnd final objective function.Packet-weighted t-SNE Algorithm Demo as a result, as Figure 10,11,
12, shown in 13, Figure 10 is the visualization easily obscured digital handwriting body 4,9 and carry out GW t-SNE algorithm dimensionality reduction in 2000 samples
Result figure, Figure 11 are the visualization result easily obscured digital handwriting body 4,9 and carry out GWt-SNE algorithm dimensionality reduction in 4000 samples
Figure, Figure 12 is the visualization result easily obscured digital handwriting body 3,5,8 and carry out GW t-SNE algorithm dimensionality reduction in 2000 samples
Figure, Figure 13 is the visualization result easily obscured digital handwriting body 3,5,8 and carry out GW t-SNE algorithm dimensionality reduction in 4000 samples
Figure;
Finally, carrying out metrics evaluation to it with recall ratio and precision ratio.
1 classification results confusion matrix of table
The real example of recall ratio P=/(real example+vacation positive example), is shown in formula (12):
The real example of precision ratio R=/(real example+vacation counter-example), such as formula (13):
Evaluation index, which is taken, calculates its center position, and determines range radius of circle, and it includes opposite in region for calculating
The number for the mapping point answered is calculated, and evaluation index is compared as shown in Figure 14,15,16,17, is compared with the prior art:
Fig. 6 is the visualization result easily obscured digital handwriting body 4,9 and carry out t-SNE algorithm dimensionality reduction in 2000 samples
Figure, can be clearly seen that, spacing is big between same numbers mapping point by Fig. 6;Figure 10 is easily to obscure digital handwriting body 4,9 to exist
The visualization result figure that GW t-SNE algorithm dimensionality reduction is carried out when 2000 samples finds phase after GW t-SNE algorithm dimensionality reduction
It is even closer with being contacted between data point, and relative distribution between different data points.
Fig. 7 is the visualization result easily obscured digital handwriting body 4,9 and carry out t-SNE algorithm dimensionality reduction in 4000 samples
Figure, number 4,9 have intersection in visualization figure, illustrate that number 4,9 can not be distinctly divided into two classes under t-SNE method.Figure 11
The visualization result figure of GW t-SNE algorithm dimensionality reduction, handwritten form are carried out easily to obscure digital handwriting body 4,9 in 4000 samples
Though 4,9 still have cross section, two classes can be distinctly divided into, the application algorithm is substantially better than t- in two-dimensional visualization figure
SNE algorithm.
Fig. 8 is the visualization result easily obscured digital handwriting body 3,5,8 and carry out t-SNE algorithm dimensionality reduction in 2000 samples
Figure, can be clearly seen that by Fig. 8, have apparent intersection between number, and each point distribution is sparse.Figure 12 is easily to obscure number
Word handwritten form 3,5,8 carries out the visualization result figure of GW t-SNE algorithm dimensionality reduction in 2000 samples, calculates by GW t-SNE
After method dimensionality reduction, the phenomenon that finding to contact very closely between identical data point, and do not obstructed completely by other numbers.
Fig. 9 is the visualization result easily obscured digital handwriting body 3,5,8 and carry out t-SNE algorithm dimensionality reduction in 4000 samples
Figure, the same numbers distribution in visualization figure of number 3,5,8 is no very close, intersects and the point of cluster mistake is more.Figure 13 is
Easily obscure the visualization result figure that digital handwriting body 3,5,8 carries out GW t-SNE algorithm dimensionality reduction in 4000 samples, handwritten form
3,5,8 same sections connection obvious closer and not excessive intersection and overlapping, thus after dimensionality reduction two-dimensional visualization it is poly-
In class Comparative result, it can be deduced that GW t-SNE algorithm of the present invention is better than t-SNE algorithm.
To sum up: the present invention can write body 4,9 and easily obscure hand number according to hand number is easily obscured writes the progress of body 3,5,8 again
Identification, can effectively identify the similar handwriting digital of shape, have raising work simultaneously for easily obscuring hand number and writing body also
Make efficiency, avoids the identified mistake of a large amount of handwriting digitals and reduce working efficiency.
Above in conjunction with attached drawing, the embodiment of the present invention is explained in detail, but the present invention is not limited to above-mentioned
Embodiment within the knowledge of a person skilled in the art can also be before not departing from present inventive concept
Put that various changes can be made.
Claims (3)
1. a kind of recognition methods again for easily obscuring digital handwriting body, it is characterised in that: the method comprises the following steps:
Step1: it is pre-processed to easily digital handwriting body is obscured: n easily being obscured digital handwriting body picture composition experiment sample
Library, then convert each picture to the vector of 1*m, then the vector of n 1*m is subjected to binary conversion treatment respectively, it ultimately forms by 0
The n*m's constituted with 1 easily obscures digital handwriting body observing matrix;
Step2: easily obscuring digital handwriting body observing matrix and calculate the gaussian kernel function of its higher dimensional space to n*m, and returned
One change processing, is grouped the gaussian kernel function of the higher dimensional space after normalization, is weighted, obtains to each grouping respectively
To the Weighted Gauss kernel function of higher dimensional space;
Step3: general to easily digital handwriting body observing matrix design conditions are obscured using the Weighted Gauss kernel function of step Step2
Rate, then higher dimensional space joint probability is calculated with obtained conditional probability;
Step4: out of normal distribution, the Y matrix of one n*2 of random initializtion calculates lower dimensional space to the Y matrix of n*2
Gaussian kernel function calculates lower dimensional space joint probability using lower dimensional space gaussian kernel function;According to higher dimensional space joint probability and
Lower dimensional space joint probability uses KL divergence calculating target function, finally judges the objective function and last time iteration of current iteration
The difference of objective function whether restrain: if difference restrains, export the dimensionality reduction result Y matrix of lower dimensional space;Otherwise, then after
Continuous iteration updates objective function, until difference restrains.
2. the recognition methods according to claim 1 again for easily obscuring digital handwriting body, it is characterised in that: described easily to obscure number
Word handwritten form obtains as follows:
Digital handwritten form 0-9 is pre-processed: by r digital handwritten form picture composition experiment sample library, then by each picture
It is converted into the vector of 1*s, then the vector of r 1*s is subjected to binary conversion treatment respectively, ultimately forms the r*s's being made of 0 and 1
Digital handwriting body observing matrix;
To the digital handwriting body observing matrix of r*s carry out t-SNE algorithm dimensionality reduction is calculated as a result, obtain 4 according to dimensionality reduction result,
9 easily obscure digital handwriting body easily to obscure digital handwriting body, 3,5,8.
3. the recognition methods according to claim 1 again for easily obscuring digital handwriting body, it is characterised in that: the step S2
In, three groups are grouped into, three groups of weights being corresponding in turn to are (0,1), [1], (1 ,+∞).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810767088.8A CN109034021B (en) | 2018-07-13 | 2018-07-13 | Re-identification method for confusable digital handwriting |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810767088.8A CN109034021B (en) | 2018-07-13 | 2018-07-13 | Re-identification method for confusable digital handwriting |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109034021A true CN109034021A (en) | 2018-12-18 |
CN109034021B CN109034021B (en) | 2022-05-20 |
Family
ID=64642064
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810767088.8A Active CN109034021B (en) | 2018-07-13 | 2018-07-13 | Re-identification method for confusable digital handwriting |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109034021B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112183335A (en) * | 2020-09-28 | 2021-01-05 | 中国人民大学 | Handwritten image recognition method and system based on unsupervised learning |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2004059569A1 (en) * | 2002-12-26 | 2004-07-15 | Koninklijke Philips Electronics N.V. | Method and system for three-dimentional handwriting recognition |
CN106096066A (en) * | 2016-08-17 | 2016-11-09 | 盐城工学院 | The Text Clustering Method embedded based on random neighbor |
-
2018
- 2018-07-13 CN CN201810767088.8A patent/CN109034021B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2004059569A1 (en) * | 2002-12-26 | 2004-07-15 | Koninklijke Philips Electronics N.V. | Method and system for three-dimentional handwriting recognition |
CN106096066A (en) * | 2016-08-17 | 2016-11-09 | 盐城工学院 | The Text Clustering Method embedded based on random neighbor |
Non-Patent Citations (2)
Title |
---|
LAURENS VAN DER MAATEN ET AL.: "Visualizing Data using t-SNE", 《JOURNAL OF MACHINE LEARNING RESEARCH 9 》 * |
詹威威 等: "自适应加权t-SNE算法及其在脑网络状态观测矩阵降维中的应用研究", 《计算机应用研究 优先出版》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112183335A (en) * | 2020-09-28 | 2021-01-05 | 中国人民大学 | Handwritten image recognition method and system based on unsupervised learning |
Also Published As
Publication number | Publication date |
---|---|
CN109034021B (en) | 2022-05-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CA2954516C (en) | Touch classification | |
CN111931710B (en) | Online handwritten character recognition method and device, electronic equipment and storage medium | |
CN113344019A (en) | K-means algorithm for improving decision value selection initial clustering center | |
Zhang et al. | Fast covariance matching with fuzzy genetic algorithm | |
Wang et al. | Human peripheral blood leukocyte classification method based on convolutional neural network and data augmentation | |
CN113032613B (en) | Three-dimensional model retrieval method based on interactive attention convolution neural network | |
Wang et al. | A residual-attention offline handwritten Chinese text recognition based on fully convolutional neural networks | |
CN106650820A (en) | Matching recognition method of handwritten electrical component symbols and standard electrical component symbols | |
CN105843925A (en) | Similar image searching method based on improvement of BOW algorithm | |
CN108459993B (en) | Complex high-dimensional system optimization method based on rapid peak-tracking sampling | |
CN110472651B (en) | Target matching and positioning method based on edge point local characteristic value | |
CN109034021A (en) | A kind of recognition methods again for easily obscuring digital handwriting body | |
CN114358166A (en) | Multi-target positioning method based on self-adaptive k-means clustering | |
CN114492566A (en) | Weight-adjustable high-dimensional data dimension reduction method and system | |
CN106203456A (en) | Coal dust Algorithm for Overlapping Granule separation method based on improved differential evolution particle cluster algorithm | |
CN104965928A (en) | Chinese character image retrieval method based on shape matching | |
Tang et al. | Online chemical symbol recognition for handwritten chemical expression recognition | |
CN110222645B (en) | Gesture misidentification feature discovery method | |
Samanta et al. | Script independent online handwriting recognition | |
Chen et al. | A handwritten character recognition algorithm based on artificial immune | |
Li et al. | Rotation-free online handwritten Chinese character recognition using two-stage convolutional neural network | |
CN108960347A (en) | A kind of recruitment evaluation system and method for convolutional neural networks handwriting recongnition Ranking Stability | |
CN115393876A (en) | Online signature identification method based on neural network | |
Le et al. | Multiple distribution data description learning algorithm for novelty detection | |
Chen et al. | Wafer maps defect recognition based on transfer learning of handwritten pre-training network |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |