CN110232337A - Chinese character image stroke extraction based on full convolutional neural networks, system - Google Patents
Chinese character image stroke extraction based on full convolutional neural networks, system Download PDFInfo
- Publication number
- CN110232337A CN110232337A CN201910454930.7A CN201910454930A CN110232337A CN 110232337 A CN110232337 A CN 110232337A CN 201910454930 A CN201910454930 A CN 201910454930A CN 110232337 A CN110232337 A CN 110232337A
- Authority
- CN
- China
- Prior art keywords
- stroke
- overlapping region
- chinese character
- character image
- extraction
- 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
- 238000000605 extraction Methods 0.000 title claims abstract description 81
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 56
- 238000000034 method Methods 0.000 claims abstract description 73
- 239000011159 matrix material Substances 0.000 claims abstract description 60
- 238000012549 training Methods 0.000 claims abstract description 24
- 230000001427 coherent effect Effects 0.000 claims abstract description 15
- 239000002131 composite material Substances 0.000 claims description 28
- 239000000284 extract Substances 0.000 claims description 24
- 230000005484 gravity Effects 0.000 claims description 10
- 238000003860 storage Methods 0.000 claims description 10
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 210000000988 bone and bone Anatomy 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000003909 pattern recognition Methods 0.000 abstract description 2
- 238000010189 synthetic method Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 9
- 108091006146 Channels Proteins 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- 238000013459 approach Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012805 post-processing Methods 0.000 description 3
- 238000007639 printing Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 239000011248 coating agent Substances 0.000 description 2
- 238000000576 coating method Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000012916 structural analysis Methods 0.000 description 2
- 230000003796 beauty Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000013441 quality evaluation Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000026676 system process Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06V30/2268—Character recognition characterised by the type of writing of cursive writing using stroke segmentation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Character Discrimination (AREA)
Abstract
The invention belongs to computer vision and area of pattern recognition, and in particular to a kind of Chinese character image stroke extraction based on full convolutional neural networks, system, it is intended to it is difficult to solve the problems, such as that the handwritten character strokes of Free Writing are extracted.The method of the present invention includes: to carry out extracted region to the Chinese character image of acquisition；Skeletonizing operation is carried out to overlapping region, non-overlapping region；The coherent degree between any stroke section of overlapping region after calculating skeletonizing；The stroke section for belonging to same stroke in overlapping region is connected, is merged into complete matrix morphology stroke with the stroke section being connected directly in non-overlapping region.One aspect of the present invention is in the case where the overlapping of the handwritten Chinese character stroke of Free Writing, the Strokes extraction of handwritten Chinese character still may be implemented, on the other hand it uses character synthetic method and obtains training sample, and its subsidiary different labeled information in different task, greatly save human cost.
Description
Technical field
The invention belongs to computer vision and area of pattern recognition, and in particular to a kind of based on full convolutional neural networks
Chinese character image stroke extraction, system.
Background technique
The Strokes extraction of Chinese character image has weight in the Text region research and related application based on structural analysis
Want status.Chinese handwritten/print character individual character classification based on depth learning technology has been achieved for quite high accuracy,
But in many applications, people are not only concerned about the classification of character, also concern stroke explanation, writing quality evaluation, shape beauty
The problems such as change, font design, and this just needs that the stroke in character image is split and is extracted.
For the Strokes extraction problem of off line Chinese character, the past has algorithm, and there are two main classes: direct extraction method and base
In the extracting method of character skeleton.Wherein, the method directly extracted is mainly used for printed character, has smoothly in character picture
Edge, simple stroke shape, the stroke width of fixation and clearly between stroke when relationship, the effects of this kind of methods compared with
Well, such as the researchers such as Tseng and Chuang [1] sum up the rule of some versatilities from the charcter topology of a variety of printing type faces
Rule, carries out Strokes extraction by heuristic rule；Print character is pen according to similar regular cutting by [2] such as Cao and Tan
Section (total 3 seed types) are drawn, these stroke sections are screened again later, are reassembled as independent stroke；Lee and Wu [3] is by print character
Image is expressed as the form of figure (Graph), and the connection between stroke section is inferred according to contour feature in the overlapping region of stroke
Relationship；Chen et al. [4] then learns two-dimensional manifold out from standard letter, then with template character (wherein pen corresponding with manifold
Draw to have extracted and finish) instruct the Strokes extraction of true printing specimen.When handling off-line handwritten character image, due to free hand
Write characters diversity and complexity with higher in relationship between stroke shape and stroke, are directly extracted using heuristic rule
Stroke is extremely difficult to ideal effect.Therefore, existing major part carries out the work of Strokes extraction all for off-line handwritten character
It is to be operated on character skeleton, the Strokes extraction task in this operation handlebar connected region grade is reduced in lines rank
Extraction [5].When carrying out Strokes extraction on character skeleton, rule and direct extraction method used by existing major part method
In relevant portion be similar.Strokes extraction based on skeletonizing faces backbone distortion (especially stroke overlapping region), pen
The problem of drawing section connection ambiguity, does not well solve method still so far.
Generally speaking, although researchers propose much about Strokes extraction in Chinese printing/hand-written character image
Method, but the character for still comparing specification being primarily upon.For the hand-written character of Free Writing, due to stroke form and position
It sets changeable, in addition the case where stroke overlapping region is extremely complex, brings huge challenge to Strokes extraction, existing method is not yet
Provide satisfactory result.
Following documents is technical background data related to the present invention:
[1]Lin Yu Tseng and Chen-Tsun Chuang."An efficient knowledge-based
stroke extraction method for multi-font Chinese characters."Pattern
Recognition,25(12):1445-1458,1992.
[2]Ruini Cao and Chew Lim Tan."A model of stroke extraction from
Chinese character images."In.Proceedings of the 15th International Conference
on Pattern Recognition,2000.
[3]Chungnan Lee and Bohom Wu."A Chinese-character-stroke-extraction
algorithm based on contour information."Pattern Recognition,31(6):651-663,
1998.
[4]Xudong Chen,Zhouhui Lian,Yingmin Tang,and Jianguo Xiao."An
automatic stroke extraction method using manifold learning."In.Proceedings of
Eurographics,2017.
[5]Cheng-Lin Liu,In-Jung Kim,and Jin H.Kim."Model-based stroke
extraction and matching for handwritten Chinese character recognition."
Pattern Recognition,34(12):2339-2352,2001.
[6]Tie-Qiang Wang and Cheng-Lin Liu,"Fully convolutional network
based skeletonization for handwritten Chinese characters."AAAI Conference on
Artificial Intelligence,2018.
[7]Byungsoo Kim,Oliver Wang,A.Cengizand Markus Gross."
Semantic segmentation for line drawing vectorization using neural networks."
In Proceedings of Eurographics,2017.
Summary of the invention
In order to solve the above problem in the prior art, i.e. the handwritten character strokes of Free Writing extract difficult problem,
The present invention provides a kind of Chinese character image stroke extraction based on full convolutional neural networks, changing extracting method includes:
Step S10 obtains Chinese character image as input picture；
Step S20 extracts the overlapping region figure of stroke in the input picture；The input picture removes the friendship
Folded region part is non-overlapping administrative division map；
Step S30 carries out skeletonizing operation to the overlapping region figure, non-overlapping administrative division map respectively, obtains overlapping region
Matrix morphology stroke section set, non-overlapping region framework form stroke section set；
Step S40 is based on the overlapping region matrix morphology stroke section set, calculates and links up between any two stroke section
Spend matrix；All elements are all larger than or belong to same stroke equal to two stroke sections of preset threshold in the coherent degree matrix；
The stroke section for belonging to same stroke in the overlapping region is connected by step S50, and by the stroke section with it is described
The stroke section being connected directly in non-overlapping region is merged into complete matrix morphology stroke.
In some preferred embodiments, " Chinese character image is obtained as input picture " in step S10, method
Are as follows:
Collected Chinese character image is obtained, global Binarization methods or local auto-adaptive based on OTSU method are passed through
Binarization methods remove the background of the Chinese character image of acquisition, obtain the foreground image of Chinese character image, and by the prospect
Image is as input picture.
In some preferred embodiments, " overlapping region of stroke in the input picture is extracted in step S20
Figure；It is non-overlapping administrative division map that the input picture, which removes the overlapping region part, ", method are as follows:
Step S201 is based on the input picture, extracts network constricted path by overlapping region and extracts the input figure
The feature of picture；
Step S202, it is symmetrical by extracting network constricted path with overlapping region based on the feature of the input picture
Path expander is inversely generated, and stroke overlapping region figure is obtained；The input picture removes the overlapping region part
For non-overlapping administrative division map；
Wherein, it is to be constructed based on full convolutional neural networks for extracting the input figure that the overlapping region, which extracts network,
The network of the overlapping region figure of stroke as in.
In some preferred embodiments, " the degree matrix that links up between any two stroke section is calculated " in step S40, side
Method are as follows:
Step S401 chooses any two stroke section in the overlapping region matrix morphology stroke section set, is denoted as respectively
S1、S2；
Step S402, in the stroke section S1、S2On uniformly choose N number of point, be denoted as set respectively
Step S403 calculates the set using the full convolutional network of conditionalWithMiddle any two point belongs to the probability of same stroke, obtains N × N number of probability, constitutes stroke section S1、S2
Between coherent degree matrix.
In some preferred embodiments, the training sample of the skeletal extraction network, acquisition methods are as follows:
Step B10 using the stroke coordinate point sequence of hand script Chinese input equipment character as the skeleton of composite characters image, and sets pen
Draw width；
Step B20, based on the skeleton of the composite characters image, according to the stroke width of setting, by the composite characters
The skeleton expansion of image is to have the stroke of width, obtains composite characters image；The corresponding skeleton of the composite characters image
For the training sample of the skeletal extraction network.
In some preferred embodiments, the overlapping region extracts the training sample of network, acquisition methods are as follows:
Step G10, the step of using the above-mentioned Chinese character image stroke extraction based on full convolutional neural networks
The method of B10- step B20 obtains composite characters image；
It is corresponding to calculate the composite characters image for step G20, the stroke coordinate point sequence information based on composite characters image
Stroke overlapping region；The composite characters image extracts the instruction of network with corresponding stroke overlapping region for the overlapping region
Practice sample.
In some preferred embodiments, " stroke overlapping region matrix morphology stroke section collection is obtained in step S30
Close, the non-overlapping region framework form stroke section set of stroke " it is additionally provided with the Optimization Steps of matrix morphology stroke section later,
Its method are as follows:
The center of gravity for calculating each overlapping region in the overlapping region figure obtains the adjacent institute in the center of gravity corresponding region
There is skeletal point, overlapping region center of gravity is connect in skeleton drawing one by one with adjacent skeletal point, the overlapping region bone after being optimized
Frame form stroke section set；
Based on the non-overlapping region of the stroke, Skeleton pixel point is recalled by the method for cluster, after being optimized
Non-overlapping region framework form stroke section set.
In some preferred embodiments, in step S50 " the stroke section for belonging to same stroke in overlapping region is connected,
And the stroke section being connected directly in the stroke section and non-overlapping region is merged into complete matrix morphology stroke " also set later
It is equipped with the step of raw stroke form is restored, method are as follows:
Pixel in the input picture is associated with the complete matrix morphology stroke, obtains Chinese character image pen
It draws, is the raw stroke form of input picture.
Another aspect of the present invention proposes a kind of Chinese character image Strokes extraction system based on full convolutional neural networks
System, including input module, region extraction module, skeletonizing module, stroke judgment module, skeleton link block, output module；
The input module is configured to obtain Chinese character image as input picture and input；
The region extraction module is configured to extract the overlapping region figure of stroke in the input picture；It is described defeated
Entering image to remove the overlapping region part is non-overlapping administrative division map；
The skeletonizing module is configured to carry out skeletonizing operation to the overlapping region figure, non-overlapping administrative division map, obtain
Overlapping region matrix morphology stroke section set, non-overlapping region framework form stroke section set；
The stroke judgment module is configured to the overlapping region matrix morphology stroke section set, calculates any two
Link up degree matrix between a stroke section；It is described it is coherent degree matrix in all elements be all larger than or equal to preset threshold two strokes
Section belongs to same stroke；
The skeleton link block is configured to for the stroke section for belonging to same stroke in the overlapping region being connected, and will
The stroke section being connected directly in the stroke section and the non-overlapping region is merged into complete matrix morphology stroke；
The output module is configured to the complete matrix morphology stroke that will acquire output.
The third aspect of the present invention proposes a kind of storage device, wherein be stored with a plurality of program, described program be suitable for by
Processor is loaded and is executed to realize the above-mentioned Chinese character image stroke extraction based on full convolutional neural networks.
The fourth aspect of the present invention proposes a kind of processing unit, including processor, storage device；The processor is fitted
In each program of execution；The storage device is suitable for storing a plurality of program；Described program be suitable for loaded by processor and executed with
Realize the above-mentioned Chinese character image stroke extraction based on full convolutional neural networks.
Beneficial effects of the present invention:
(1) the present invention is based on the Chinese character image stroke extraction of full convolutional neural networks, the conditional of use is complete
Convolutional neural networks can fully describe the spatial relationship between stroke form, position and stroke in Chinese print character, nothing
Need additional post-processing approach.
(2) the present invention is based on the Chinese character image stroke extractions of full convolutional neural networks, for Chinese handwritten word
The changeable problem of the writing style and stroke width of symbol carries out stroke on single pixel width skeleton using skeleton extraction module
It extracts, this operation had not only maintained the structure of hand-written character, but also can significantly save calculation amount.
(3) the present invention is based on the Chinese character image stroke extractions of full convolutional neural networks, by Chinese character image
Middle stroke overlapping region detected, and individually be handled, and all stroke section set for converging at same overlapping region are obtained, right
Stroke section in this set is analyzed two-by-two, describes the two stroke sections with the coherent degree matrix of the stroke between two stroke sections
Between relationship, judge that the two stroke sections connect into a stroke, avoid handwritten character strokes extract omit or redundancy.
(4) the present invention is based on the Chinese character image stroke extraction of full convolutional neural networks, a kind of training is provided
With character picture synthetic method, millions of off line Chinese handwritten characters images can be automatically generated, and is attached to it not
With the different labeled information in task, human cost has greatly been saved.
Detailed description of the invention
By reading a detailed description of non-restrictive embodiments in the light of the attached drawings below, the application's is other
Feature, objects and advantages will become more apparent upon:
Fig. 1 is the flow diagram of the Chinese character image stroke extraction the present invention is based on full convolutional neural networks；
Fig. 2 is a kind of friendship of embodiment of Chinese character image stroke extraction the present invention is based on full convolutional neural networks
The full convolutional neural networks structural schematic diagram of folded extracted region network and skeletonizing Web vector graphic；
Fig. 3 is a kind of area of embodiment of Chinese character image stroke extraction the present invention is based on full convolutional neural networks
Extract network and skeletal extraction network training data training process schematic diagram in domain；
Fig. 4 is a kind of hand of embodiment of Chinese character image stroke extraction the present invention is based on full convolutional neural networks
The post-processing approach schematic diagram of write characters skeletonizing；
Fig. 5 is a kind of item of embodiment of Chinese character image stroke extraction the present invention is based on full convolutional neural networks
The full convolutional network training method schematic diagram of part formula；
Fig. 6 is a kind of base of embodiment of Chinese character image stroke extraction the present invention is based on full convolutional neural networks
In the Strokes extraction exemplary diagram of the coherent degree of stroke.
Specific embodiment
The application is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is only used for explaining related invention, rather than the restriction to the invention.It also should be noted that in order to just
Part relevant to related invention is illustrated only in description, attached drawing.
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.The application is described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.
A kind of Chinese character image stroke extraction based on full convolutional neural networks of the invention, comprising:
Step S10 obtains Chinese character image as input picture；
Step S20 extracts the overlapping region figure of stroke in the input picture；The input picture removes the friendship
Folded region part is non-overlapping administrative division map；
Step S30 carries out skeletonizing operation to the overlapping region figure, non-overlapping administrative division map, obtains overlapping region skeleton
Form stroke section set, non-overlapping region framework form stroke section set；
Step S40 is based on the overlapping region matrix morphology stroke section set, calculates and links up between any two stroke section
Spend matrix；All elements are all larger than or belong to same stroke equal to two stroke sections of preset threshold in the coherent degree matrix；
The stroke section for belonging to same stroke in the overlapping region is connected by step S50, and by the stroke section with it is described
The stroke section being connected directly in non-overlapping region is merged into complete matrix morphology stroke.
In order to more clearly to the present invention is based on the progress of the Chinese character image stroke extraction of full convolutional neural networks
Illustrate, step each in embodiment of the present invention method is unfolded to be described in detail below with reference to Fig. 1.
The Chinese character image stroke extraction based on full convolutional neural networks of an embodiment of the present invention, including step
Rapid S10- step S50, each step are described in detail as follows:
Step S10 obtains Chinese character image as input picture.
Text region is an important branch of CRT technology, and the most difficult problem in identification field it
One.The Strokes extraction of Chinese character has critical role in the Text region research and related application based on structural analysis.Text
Word identification is divided into Machine printed character recognition and handwritten text identification two major classes again, and Machine printed character recognition is advised due to text formatting
Model, strokes sharp, current research have had a great development, and the variation of handwriting font is more, and person's handwriting is connected more, and because
Cause the word that similarity is high in handwritten text more for handwriting deformation, therefore there are many identification relative difficulty of handwriting.
" Chinese character image is obtained as input picture " in step S10, method are as follows:
Collected Chinese character image is obtained, global Binarization methods or local auto-adaptive based on OTSU method are passed through
Binarization methods remove the background of the Chinese character image of acquisition, obtain the foreground image of Chinese character image, and by the prospect
Image is as input picture.The present invention is based on the Chinese character image stroke extractions of full convolutional neural networks, not only can be with
The Strokes extraction for carrying out handwritten Chinese character image, equally has well the Strokes extraction of block letter Chinese character image
Effect.
The purpose of image binaryzation is the influence eliminating the gray value of stroke pixel and extracting to subsequent stroke, is based on OTSU
The global Binarization methods of method are generally used for the more uniform Chinese character image of light application ratio, and local auto-adaptive binaryzation is calculated
Method is generally used for the even image of uneven illumination.The method of image binaryzation can be selected in conjunction with the characteristics of image there are also very much
Suitable image binaryzation method, this is no longer going to repeat them.
Step S20 extracts the overlapping region figure of stroke in the input picture；The input picture removes the friendship
Folded region part is non-overlapping administrative division map.
In the embodiment of the present invention, network is extracted using the overlapping region constructed based on full convolutional neural networks, is extracted
The region removed outside overlapping region after the overlapping region of input picture, in input picture is non-overlapping region.Overlapping region mentions
Network is taken symmetrically partially to be formed by two: constricted path and path expander.Constricted path is used to extract characteristics of image, is characterized in
A certain class object is different from corresponding (essence) feature of other class objects or the set of characteristic or these features and characteristic, special
Sign is by measuring or handling the data that can be extracted.Feature of the path expander based on acquisition is inversely generated, and character is obtained
Stroke overlapping region figure.
Step S201 is based on the input picture, extracts the input by the constricted path that overlapping region extracts network
The feature of image.
Step S202, it is symmetrical by extracting network constricted path with overlapping region based on the feature of the input picture
Path expander is inversely generated, and stroke overlapping region figure is obtained；The input picture removes the overlapping region part
For non-overlapping administrative division map.
Wherein, it is to be constructed based on full convolutional neural networks for extracting the input figure that the overlapping region, which extracts network,
The network of the overlapping region figure of stroke as in.
Step S30 carries out skeletonizing operation to the overlapping region figure, non-overlapping administrative division map, obtains overlapping region skeleton
Form stroke section set, non-overlapping region framework form stroke section set.
In the embodiment of the present invention, stroke overlapping region figure, non-overlapping area are carried out using based on full convolutional neural networks
The skeletonizing of domain figure operates.
Fig. 2 is that the present invention is based on a kind of friendships of embodiment of Chinese character image stroke extraction of full convolutional neural networks
The full convolutional neural networks structural schematic diagram of extracted region Web vector graphic is folded, picture of the upper left corner with text is the defeated of network in figure
Entering picture, picture of the upper right corner with two o'clock is the overlapping region image of network output, remaining block represents convolutional calculation unit, in order to
It is easy to understand and describes, in Fig. 2 of the present invention and example, each computing unit is made of three-layer coil lamination, the downward arrow in left side
Head represents the down-sampling process in constricted path, and the upward arrow in right side represents the upper sampling process in path expander, and two right
Claim the even depth position in path, i.e., arrow to the right represents the connection of residual error formula.
Overlapping region extracts the training of the full convolutional neural networks of network and skeletal extraction Web vector graphic, it is desirable that millions of
The hand-written character sample data of meter, each sample labeling goes out stroke overlapping region, and carries out people by obtaining hand-written character image
Work label is difficult to obtain the training sample of such quantity.The present invention is by with hand script Chinese input equipment character composite characters image, because of connection
Machine hand-written character has stroke track information (stroke coordinate point sequence), it is easy to calculate stroke overlapping region, avoid manually marking
The burden of note, so as to obtain a large amount of training sample in a short time.By the stroke point sequence of hand script Chinese input equipment character as
The skeleton of character picture, and to each stroke assign a stroke width, skeleton is extended to the stroke of width, just obtain with
The consistent synthesis character image of true character image, wherein the intersection of the pixel of different strokes is exactly stroke overlapping region.
The training sample of the skeletal extraction network, acquisition methods are as follows:
Step B10 using the stroke coordinate point sequence of hand script Chinese input equipment character as the skeleton of composite characters image, and sets pen
Draw width.
Step B20, based on the skeleton of the composite characters image, according to the stroke width of setting, by the composite characters
The skeleton expansion of image is to have the stroke of width, obtains composite characters image；The corresponding skeleton of the composite characters image
For the training sample of the skeletal extraction network.
The overlapping region extracts the training sample of network, acquisition methods are as follows:
Step G10, the step of using the above-mentioned Chinese character image stroke extraction based on full convolutional neural networks
The method of B10- step B20 obtains composite characters image；
It is corresponding to calculate the composite characters image for step G20, the stroke coordinate point sequence information based on composite characters image
Stroke overlapping region；The composite characters image extracts the instruction of network with corresponding stroke overlapping region for the overlapping region
Practice sample.
Fig. 3 is that the present invention is based on a kind of areas of embodiment of Chinese character image stroke extraction of full convolutional neural networks
Network and skeletal extraction network training data training process schematic diagram are extracted in domain, and the module of first row is that training sample is defeated
Enter, secondary series represents four groups of convolution units, and third column, the 4th column, the 5th column, the 6th each convolution unit arranged are different scales
Under conventional convolution branch, the 7th be classified as network training output result.When being played a role jointly due to convolution and pondization operation, meeting
Reduce picture size at double, so when need to up-sample layer to restore picture size, using can learn in the embodiment of the present invention
It up-samples layer and up-samples operation instead of bilinear interpolation, model is enabled to recover more image details.Final
Multi-scale feature fusion stage, the present invention are directly merged last forecast image using convolution operation and obtain final output figure
Picture, this operation, which can make full use of in bigger receptive field more local messages come the central point for inferring current receptive field, is
It is no to be judged as skeletal point.
" overlapping region matrix morphology stroke section set, non-overlapping region framework form stroke section collection are obtained in step S30
Close " it is additionally provided with the Optimization Steps of matrix morphology stroke, method later are as follows:
The center of gravity for calculating each overlapping region in the overlapping region figure obtains the adjacent institute in the center of gravity corresponding region
There is skeletal point, overlapping region center of gravity is connect in skeleton drawing one by one with adjacent skeletal point, the overlapping region bone after being optimized
Frame form stroke section set；
Based on the non-overlapping region, Skeleton pixel point is recalled by the method for cluster, the non-overlapping area after being optimized
Domain matrix morphology stroke section set.
As shown in figure 4, for the present invention is based on a kind of realities of Chinese character image stroke extraction of full convolutional neural networks
The post-processing approach schematic diagram of the hand-written character skeletonizing of example is applied, K-means is K mean cluster method, bwmorph_thin@
Matlab is traditional thinning algorithm, and sigmoid represents sigmoid function, also referred to as S sigmoid growth curve.It is poly- by K-means
Class, by pixel in image be generally divided into the transition point between skeletal point, non-skeleton point, skeletal point and non-skeleton point these three
Classification.By this operation, most skeletal points can be called back, but inevitably bring the superfluous of a small amount of non-skeleton
Yu Dian, in this regard, it is superfluous to use the simple rule in tradition refinement (Thinning) algorithm to leave out these in an example of the invention
Yu Dian guarantees that the lines in skeleton drawing are single pixel width.Next, for the overlapping region outlined in figure with rectangle,
The region (the as starting point of four arrows in figure) is indicated using the center of gravity in the region, this region produces four with skeletal point
Abutment points (terminal of four arrows in figure), then focus point and four abutment points are connected in skeleton drawing, it can be to this region
Obtain an ideal skeletonizing result.
Step S40 is based on the overlapping region matrix morphology stroke section set, calculates and links up between any two stroke section
Spend matrix；All elements are all larger than or belong to same stroke equal to two stroke sections of preset threshold in the coherent degree matrix.
In the embodiment of the present invention, the degree square that links up between any two stroke section is carried out using the full convolutional neural networks of conditional
Battle array calculates.The full convolutional network of conditional is constructed based on full convolutional neural networks, using the conditional method training of basic point guidance:
The input of network is binary channels form, is exported as single channel form, and first channel of input is hand-written character figure
Picture, second channel are a mask of basic point, this mask keeps identical with character picture size, only in retrieval pixel
Numerical value is 1 at the coordinate of basic point, and other positions numerical value is 0.Mask serve as full convolutional network conditional input, network it is defeated
It is out the complete stroke comprising basic point in mask.Wherein, VDSR unit is Standard convolution unit, and the structure of itself is very simple,
It is only made of convolutional layer, active coating and batch normalization layer, in this department pattern of the invention, the quantity of VDSR unit is depended on
Convolution kernel size and the size of image the two variables, i.e. VDSR unit need successively to stack until the receptive field of model covers completely
Until covering whole picture.
As shown in figure 5, for the present invention is based on a kind of realities of Chinese character image stroke extraction of full convolutional neural networks
The full convolutional network training method schematic diagram of conditional of example is applied, left side is that the binary channels of network inputs, hand-written character image, basic point
A mask, right side be network single channel export, for the complete stroke comprising basic point in mask, intermediate VDSR UNITS
For Standard convolution unit, conv represents convolutional layer, and ReLU represents active coating, and BatchNorm represents batch normalization layer.
Step S401 chooses any two stroke section in the overlapping region matrix morphology stroke section set, is denoted as respectively
S1、S2。
Step S402, in the stroke section S1、S2On uniformly choose N number of point, be denoted as set respectively
Step S403 calculates the set using the full convolutional network of conditionalWithMiddle any two point belongs to the probability of same stroke, obtains N × N number of probability, constitutes stroke section S1、S2
Between coherent degree matrix.
The probability that any two point in two set belongs to same stroke is calculated, shown in method such as formula (1):
Wherein, pnnForWithBelong to the probability of same stroke, fvdsr() represents the full convolutional network of conditional.
S1、S2Between coherent degree matrix, as shown in formula (2):
In one embodiment of the invention, set the value of matrix all greater than or be equal to preset threshold value 0.5, then stroke
Section S1And S2Belong to the same stroke.
As shown in fig. 6, for the present invention is based on a kind of realities of Chinese character image stroke extraction of full convolutional neural networks
The Strokes extraction exemplary diagram based on the coherent degree of stroke of example is applied, the upper left corner is handwritten Chinese character sample instance, and the upper right corner is hand
The skeleton drawing of Chinese character sample instance is write, the point in the figure of the lower left corner is reconnaissance example in overlapping region stroke section, and the lower right corner is
The result of handwritten Chinese character sample extraction stroke exports.
The stroke section for belonging to same stroke in the overlapping region is connected by step S50, and by the stroke section with it is described
The stroke section being connected directly in non-overlapping region is merged into complete matrix morphology stroke.
In step S50 " the stroke section for belonging to same stroke in overlapping region is connected, and by the stroke section with it is non-overlapping
The stroke section being connected directly in region is merged into complete matrix morphology stroke " it is additionally provided with the recovery of raw stroke form later
Step, method are as follows:
Pixel in the input picture is associated with the complete matrix morphology stroke, obtains Chinese character image pen
It draws, is the raw stroke form of input picture.
The Chinese character image Strokes extraction system based on full convolutional neural networks of second embodiment of the invention, including it is defeated
Enter module, region extraction module, skeletonizing module, stroke judgment module, skeleton link block, output module；
The input module is configured to obtain Chinese character image as input picture and input；
The region extraction module is configured to extract the overlapping region figure of stroke in the input picture；It is described defeated
Entering image to remove the overlapping region part is non-overlapping administrative division map；
The skeletonizing module is configured to carry out skeletonizing operation to the overlapping region figure, non-overlapping administrative division map, obtain
Overlapping region matrix morphology stroke section set, non-overlapping region framework form stroke section set；
The stroke judgment module is configured to the overlapping region matrix morphology stroke section set, calculates any two
Link up degree matrix between a stroke section；It is described it is coherent degree matrix in all elements be all larger than or equal to preset threshold two strokes
Section belongs to same stroke；
The skeleton link block is configured to for the stroke section for belonging to same stroke in the overlapping region being connected, and will
The stroke section being connected directly in the stroke section and the non-overlapping region is merged into complete matrix morphology stroke；
The output module is configured to the complete matrix morphology stroke that will acquire output.
Person of ordinary skill in the field can be understood that, for convenience and simplicity of description, foregoing description
The specific work process of system and related explanation, can refer to corresponding processes in the foregoing method embodiment, details are not described herein.
It should be noted that the Chinese character image Strokes extraction provided by the above embodiment based on full convolutional neural networks
System only the example of the division of the above functional modules in practical applications, can according to need and by above-mentioned function
Can distribution completed by different functional modules, i.e., by the embodiment of the present invention module or step decompose or combine again,
For example, the module of above-described embodiment can be merged into a module, multiple submodule can also be further split into, with complete with
The all or part of function of upper description.For module involved in the embodiment of the present invention, the title of step, it is only for area
Divide modules or step, is not intended as inappropriate limitation of the present invention.
A kind of storage device of third embodiment of the invention, wherein being stored with a plurality of program, described program is suitable for by handling
Device is loaded and is executed to realize the above-mentioned Chinese character image stroke extraction based on full convolutional neural networks.
A kind of processing unit of fourth embodiment of the invention, including processor, storage device；Processor is adapted for carrying out each
Program；Storage device is suitable for storing a plurality of program；Described program is suitable for being loaded by processor and being executed to realize above-mentioned base
In the Chinese character image stroke extraction of full convolutional neural networks.
Person of ordinary skill in the field can be understood that, for convenience and simplicity of description, foregoing description
The specific work process and related explanation of storage device, processing unit, can refer to corresponding processes in the foregoing method embodiment,
Details are not described herein.
Those skilled in the art should be able to recognize that, mould described in conjunction with the examples disclosed in the embodiments of the present disclosure
Block, method and step, can be realized with electronic hardware, computer software, or a combination of the two, software module, method and step pair
The program answered can be placed in random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electric erasable and can compile
Any other form of storage well known in journey ROM, register, hard disk, moveable magnetic disc, CD-ROM or technical field is situated between
In matter.In order to clearly demonstrate the interchangeability of electronic hardware and software, in the above description according to function generally
Describe each exemplary composition and step.These functions are executed actually with electronic hardware or software mode, depend on technology
The specific application and design constraint of scheme.Those skilled in the art can carry out using distinct methods each specific application
Realize described function, but such implementation should not be considered as beyond the scope of the present invention.
Term " includes " or any other like term are intended to cover non-exclusive inclusion, so that including a system
Process, method, article or equipment/device of column element not only includes those elements, but also including being not explicitly listed
Other elements, or further include the intrinsic element of these process, method, article or equipment/devices.
So far, it has been combined preferred embodiment shown in the drawings and describes technical solution of the present invention, still, this field
Technical staff is it is easily understood that protection scope of the present invention is expressly not limited to these specific embodiments.Without departing from this
Under the premise of the principle of invention, those skilled in the art can make equivalent change or replacement to the relevant technologies feature, these
Technical solution after change or replacement will fall within the scope of protection of the present invention.
Claims (11)
1. a kind of Chinese character image stroke extraction based on full convolutional neural networks, which is characterized in that the extracting method
Include:
Step S10 obtains Chinese character image as input picture；
Step S20 extracts the overlapping region figure of stroke in the input picture；The input picture removes the crossover region
Domain part is non-overlapping administrative division map；
Step S30 carries out skeletonizing operation to the overlapping region figure, non-overlapping administrative division map, obtains overlapping region matrix morphology
Stroke section set, non-overlapping region framework form stroke section set；
Step S40 is based on the overlapping region matrix morphology stroke section set, calculates the degree square that links up between any two stroke section
Battle array；All elements are all larger than or belong to same stroke equal to two stroke sections of preset threshold in the coherent degree matrix；
The stroke section for belonging to same stroke in the overlapping region is connected by step S50, and by the stroke section and the non-friendship
The stroke section being connected directly in folded region is merged into complete matrix morphology stroke.
2. the Chinese character image stroke extraction according to claim 1 based on full convolutional neural networks, feature
It is, " obtains Chinese character image as input picture " in step S10, method are as follows:
Collected Chinese character image is obtained, global Binarization methods or local auto-adaptive two-value based on OTSU method are passed through
Change algorithm, remove the background of the Chinese character image of acquisition, obtain the foreground image of Chinese character image, and by the foreground image
As input picture.
3. the Chinese character image stroke extraction according to claim 1 based on full convolutional neural networks, feature
It is, " extracts the overlapping region figure of stroke in the input picture in step S20；The input picture removes the friendship
Folded region part is non-overlapping administrative division map ", method are as follows:
Step S201 is based on the input picture, extracts network constricted path by overlapping region and extracts the input picture
Feature；
Step S202 is symmetrically expanded based on the feature of the input picture by extracting network constricted path with overlapping region
Path is inversely generated, and overlapping region figure is obtained；It is non-overlapping region that the input picture, which removes the overlapping region part,
Figure；
Wherein, it is to be constructed based on full convolutional neural networks for extracting in the input picture that the overlapping region, which extracts network,
The network of the overlapping region figure of stroke.
4. the Chinese character image stroke extraction according to claim 1 based on full convolutional neural networks, feature
It is, " calculates the degree matrix that links up between any two stroke section " in step S40, method are as follows:
Step S401 chooses any two stroke section in the overlapping region matrix morphology stroke section set, is denoted as S respectively1、S2；
Step S402, in the stroke section S1、S2On uniformly choose N number of point, be denoted as set respectively
Step S403 calculates the set using the full convolutional network of conditionalWithMiddle any two point belongs to the probability of same stroke, obtains N × N number of probability, constitutes stroke section S1、S2
Between coherent degree matrix.
5. the Chinese character image stroke extraction according to claim 1 based on full convolutional neural networks, feature
It is, the training sample of the skeletal extraction network, acquisition methods are as follows:
Step B10 using the stroke coordinate point sequence of hand script Chinese input equipment character as the skeleton of composite characters image, and sets stroke width
Degree；
Step B20, based on the skeleton of the composite characters image, according to the stroke width of setting, by the composite characters image
Skeleton expansion be to have the stroke of width, obtain composite characters image；The corresponding skeleton of the composite characters image is institute
State the training sample of skeletal extraction network.
6. the Chinese character image stroke extraction according to claim 5 based on full convolutional neural networks, the friendship
The training sample of folded extracted region network, acquisition methods are as follows:
Step G10, using the Chinese character image stroke extraction based on full convolutional neural networks described in claim 5
The method of step B10- step B20 obtains composite characters image；
Step G20, the stroke coordinate point sequence information based on composite characters image calculate the corresponding pen of the composite characters image
Draw overlapping region；The composite characters image extracts the training sample of network with corresponding stroke overlapping region for the overlapping region
This.
7. the Chinese character image stroke extraction according to claim 1 based on full convolutional neural networks, step S30
In " obtain overlapping region matrix morphology stroke section set, non-overlapping region framework form stroke section set " after be additionally provided with bone
The Optimization Steps of frame form stroke section, method are as follows:
The center of gravity for calculating each overlapping region in the stroke overlapping region obtains the center of gravity corresponding region adjoining
Overlapping region center of gravity is connect in skeleton drawing, the overlapping region after being optimized by all skeletal points one by one with adjacent skeletal point
Matrix morphology stroke section set；
Based on the non-overlapping region of the stroke, Skeleton pixel point is recalled by the method for cluster, the non-friendship after being optimized
Folded region framework form stroke.
8. the Chinese character image Strokes extraction side according to claim 1-7 based on full convolutional neural networks
Method, in step S50 " the stroke section for belonging to same stroke in the overlapping region is connected, and by the stroke section with it is described non-
The stroke section being connected directly in overlapping region is merged into complete matrix morphology stroke " after to be additionally provided with raw stroke form extensive
Multiple step, method are as follows:
Pixel in the character picture of the acquisition is associated with the complete matrix morphology stroke, obtains Chinese character image pen
It draws, is the raw stroke form of input picture.
9. a kind of Chinese character image Strokes extraction system based on full convolutional neural networks, which is characterized in that including inputting mould
Block, region extraction module, skeletonizing module, stroke judgment module, skeleton link block, output module；
The input module is configured to obtain Chinese character image as input picture and input；
The region extraction module is configured to extract the overlapping region figure of stroke in the input picture；The input figure
It is non-overlapping administrative division map as removing the overlapping region part；
The skeletonizing module is configured to carry out skeletonizing operation to the stroke overlapping region figure, non-overlapping administrative division map,
Obtain overlapping region matrix morphology stroke section set, non-overlapping region framework form stroke section set；
The stroke judgment module is configured to the overlapping region matrix morphology stroke section set, calculates any two pen
Draw the degree matrix that links up between section；It is described it is coherent degree matrix in all elements be all larger than or equal to preset threshold value two stroke sections
Belong to same stroke；
The skeleton link block is configured to for the stroke section for belonging to same stroke in the overlapping region being connected, and will be described
The stroke section being connected directly in stroke section and the non-overlapping region is merged into complete matrix morphology stroke；
The output module is configured to the complete matrix morphology stroke that will acquire output.
10. a kind of storage device, wherein being stored with a plurality of program, which is characterized in that described program is suitable for by processor load simultaneously
It executes to realize the described in any item Chinese character image Strokes extraction sides based on full convolutional neural networks claim 1-8
Method.
11. a kind of processing unit, including
Processor is adapted for carrying out each program；And
Storage device is suitable for storing a plurality of program；
It is characterized in that, described program is suitable for being loaded by processor and being executed to realize:
The described in any item Chinese character image stroke extractions based on full convolutional neural networks of claim 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910454930.7A CN110232337B (en) | 2019-05-29 | 2019-05-29 | Chinese character image stroke extraction method and system based on full convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910454930.7A CN110232337B (en) | 2019-05-29 | 2019-05-29 | Chinese character image stroke extraction method and system based on full convolution neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110232337A true CN110232337A (en) | 2019-09-13 |
| CN110232337B CN110232337B (en) | 2021-02-02 |
Family
ID=67858809
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910454930.7A Active CN110232337B (en) | 2019-05-29 | 2019-05-29 | Chinese character image stroke extraction method and system based on full convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110232337B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110660019A (en) * | 2019-09-29 | 2020-01-07 | 华北电力大学 | Small data set simplified stroke generation method based on BPL |
| CN110969681A (en) * | 2019-11-29 | 2020-04-07 | 山东浪潮人工智能研究院有限公司 | Method for generating handwriting characters based on GAN network |
| CN112862025A (en) * | 2021-03-08 | 2021-05-28 | 成都字嗅科技有限公司 | Chinese character stroke filling method, system, terminal and medium based on computer |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1271140A (en) * | 1999-04-21 | 2000-10-25 | 中国科学院自动化研究所 | Handwriting identifying method based on grain analysis |
| CN101853126A (en) * | 2010-05-12 | 2010-10-06 | 中国科学院自动化研究所 | Real-time identification method for on-line handwriting sentences |
| CN105205448A (en) * | 2015-08-11 | 2015-12-30 | 中国科学院自动化研究所 | Character recognition model training method based on deep learning and recognition method thereof |
| US20170017835A1 (en) * | 2013-06-09 | 2017-01-19 | Apple Inc. | Multi-script handwriting recognition using a universal recognizer |
| CN108229397A (en) * | 2018-01-04 | 2018-06-29 | 华南理工大学 | Method for text detection in image based on Faster R-CNN |
| CN108345850A (en) * | 2018-01-23 | 2018-07-31 | 哈尔滨工业大学 | The scene text detection method of the territorial classification of stroke feature transformation and deep learning based on super-pixel |
| CN108345833A (en) * | 2018-01-11 | 2018-07-31 | 深圳中兴网信科技有限公司 | The recognition methods of mathematical formulae and system and computer equipment |
| CN106446896B (en) * | 2015-08-04 | 2020-02-18 | 阿里巴巴集团控股有限公司 | Character segmentation method and device and electronic equipment |
-
2019
- 2019-05-29 CN CN201910454930.7A patent/CN110232337B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1271140A (en) * | 1999-04-21 | 2000-10-25 | 中国科学院自动化研究所 | Handwriting identifying method based on grain analysis |
| CN101853126A (en) * | 2010-05-12 | 2010-10-06 | 中国科学院自动化研究所 | Real-time identification method for on-line handwriting sentences |
| US20170017835A1 (en) * | 2013-06-09 | 2017-01-19 | Apple Inc. | Multi-script handwriting recognition using a universal recognizer |
| CN106446896B (en) * | 2015-08-04 | 2020-02-18 | 阿里巴巴集团控股有限公司 | Character segmentation method and device and electronic equipment |
| CN105205448A (en) * | 2015-08-11 | 2015-12-30 | 中国科学院自动化研究所 | Character recognition model training method based on deep learning and recognition method thereof |
| CN108229397A (en) * | 2018-01-04 | 2018-06-29 | 华南理工大学 | Method for text detection in image based on Faster R-CNN |
| CN108345833A (en) * | 2018-01-11 | 2018-07-31 | 深圳中兴网信科技有限公司 | The recognition methods of mathematical formulae and system and computer equipment |
| CN108345850A (en) * | 2018-01-23 | 2018-07-31 | 哈尔滨工业大学 | The scene text detection method of the territorial classification of stroke feature transformation and deep learning based on super-pixel |
Non-Patent Citations (2)
| Title |
|---|
| BAOTIAN HU 等: "Stroke Sequence-Dependent Deep Convolutional Neural", 《ARXIV:1610.04057V1》 * |
| TIE-QIANG WANG 等: "DeepAD: A Deep Learning Based Approach to Stroke-Level Abnormality Detection in Handwritten Chinese Character Recognition", 《2018 IEEE INTERNATIONAL CONFERENCE ON DATA MINING (ICDM)》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110660019A (en) * | 2019-09-29 | 2020-01-07 | 华北电力大学 | Small data set simplified stroke generation method based on BPL |
| CN110969681A (en) * | 2019-11-29 | 2020-04-07 | 山东浪潮人工智能研究院有限公司 | Method for generating handwriting characters based on GAN network |
| CN110969681B (en) * | 2019-11-29 | 2023-08-29 | 山东浪潮科学研究院有限公司 | Handwriting word generation method based on GAN network |
| CN112862025A (en) * | 2021-03-08 | 2021-05-28 | 成都字嗅科技有限公司 | Chinese character stroke filling method, system, terminal and medium based on computer |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110232337B (en) | 2021-02-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chaudhry et al. | Leaf-qa: Locate, encode & attend for figure question answering | |
| Naz et al. | Urdu Nastaliq recognition using convolutional–recursive deep learning | |
| Alwzwazy et al. | Handwritten digit recognition using convolutional neural networks | |
| Rahman et al. | Bangla handwritten character recognition using convolutional neural network | |
| CN105426919B (en) | The image classification method of non-supervisory feature learning is instructed based on conspicuousness | |
| CN101377854B (en) | Method for simulating Chinese characters hand-written handwriting by a computer | |
| CN110232337A (en) | Chinese character image stroke extraction based on full convolutional neural networks, system | |
| CN104834941A (en) | Offline handwriting recognition method of sparse autoencoder based on computer input | |
| Abandah et al. | Novel moment features extraction for recognizing handwritten Arabic letters | |
| CN110334724B (en) | Remote sensing object natural language description and multi-scale correction method based on LSTM | |
| Asthana et al. | Handwritten multiscript numeral recognition using artificial neural networks | |
| CN109685065A (en) | Printed page analysis method, the system of paper automatic content classification | |
| CN101599180B (en) | Automatic generation method of imitative computer calligraphy based on handwriting style | |
| CN108805223A (en) | A kind of recognition methods of seal character text and system based on Incep-CapsNet networks | |
| Islam et al. | A CNN based approach for garments texture design classification | |
| Chen et al. | Page segmentation for historical handwritten document images using conditional random fields | |
| Kim et al. | The structure of deep neural network for interpretable transfer learning | |
| Indira et al. | Classification and Recognition of Printed Hindi Characters Using Artificial Neural Networks | |
| Dokare et al. | Recognition of handwritten devanagari character using convolutional neural network | |
| CN112837332A (en) | Creative design generation method, device, terminal, storage medium and processor | |
| Liu et al. | Online handwritten Mongolian word recognition using MWRCNN and position maps | |
| Jung et al. | On-line recognition of cursive Korean characters using graph representation | |
| Khan et al. | Isolated handwritten pashto characters recognition using KNN classifier | |
| Ghaleb et al. | Graph modeling based segmentation of handwritten Arabic text into constituent sub-words | |
| Hasan et al. | A new state of art deep learning approach for Bangla handwritten digit recognition using SVM classifier |
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 |