CN109241974A - A kind of recognition methods and system of text image - Google Patents
A kind of recognition methods and system of text image Download PDFInfo
- Publication number
- CN109241974A CN109241974A CN201810965342.5A CN201810965342A CN109241974A CN 109241974 A CN109241974 A CN 109241974A CN 201810965342 A CN201810965342 A CN 201810965342A CN 109241974 A CN109241974 A CN 109241974A
- Authority
- CN
- China
- Prior art keywords
- text
- image
- network
- normotopia
- compression
- 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 52
- 230000006835 compression Effects 0.000 claims abstract description 59
- 238000007906 compression Methods 0.000 claims abstract description 59
- 238000012549 training Methods 0.000 claims abstract description 51
- 238000013528 artificial neural network Methods 0.000 claims abstract description 20
- 230000008569 process Effects 0.000 claims abstract description 16
- 230000006870 function Effects 0.000 claims abstract description 12
- 238000010801 machine learning Methods 0.000 claims abstract description 8
- 238000013527 convolutional neural network Methods 0.000 claims description 13
- 238000003776 cleavage reaction Methods 0.000 claims description 9
- 239000003550 marker Substances 0.000 claims description 9
- 230000007017 scission Effects 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 5
- 230000001537 neural effect Effects 0.000 claims 2
- 238000012217 deletion Methods 0.000 claims 1
- 230000037430 deletion Effects 0.000 claims 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 10
- 238000005516 engineering process Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 238000012512 characterization method Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000001771 impaired effect Effects 0.000 description 2
- 210000004218 nerve net Anatomy 0.000 description 2
- 238000012015 optical character recognition Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000010408 sweeping Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000017105 transposition Effects 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/60—Rotation of whole images or parts thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
- G06T9/002—Image coding using neural 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
- G06V30/14—Image acquisition
- G06V30/148—Segmentation of character regions
- G06V30/153—Segmentation of character regions using recognition of characters or words
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Character Discrimination (AREA)
Abstract
The present invention relates to a kind of recognition methods of text image, it include: to rotate images to be recognized input picture compression normotopia network, so that the text in images to be recognized is in horizontal position, compression of images normotopia network is obtained by the method training of machine learning, has the function of image rotation;The image recognition text exported from compression of images normotopia network.The invention has the benefit that by carrying out compression and normotopia automatically to images to be recognized with convolution self-encoding encoder, and identified using text identification neural network, it ensure that the accuracy of Text region, eliminate artificial pretreated process, hand labor is saved, is provided convenience for user.
Description
Technical field
The invention belongs to technical field of character recognition, and in particular to a kind of recognition methods and system of text image.
Background technique
OCR software for discerning characters refers to and utilizes OCR (Optical Character Recognition, optical character identification)
Word content on picture, photo is converted directly into the software of editable text by technology.
Existing Text region process includes: that paper document is converted into electronic document by electronic equipment, such as by sweeping
It retouches instrument or digital camera obtains the image file of paper document;OCR software for discerning characters is analyzed and processed image file, obtains
Take text and layout information.
The above method in the actual operation process, due to electronic equipment obtain image file be difficult to ensure it is horizontally arranged,
Therefore operator is needed to rotate manually to image text, adjustment character arranging direction to level, when papery text to be identified
When gear number amount is more, operator's larger workload causes recognition efficiency low, and manual operation is easy error, it is also difficult to protect
Demonstrate,prove recognition accuracy.
Can it be those skilled in the art's urgent need to resolve that a kind of more convenient text image recognition method therefore be provided
Problem.
Summary of the invention
In order to solve the problems, such as that text identification low efficiency of the existing technology, accuracy rate are low, the present invention provides one kind
The recognition methods and system of text image have the characteristics that recognition efficiency is high, accuracy rate is high.
It is convenient for people to use the object of the present invention is to provide one kind and saves manual labor and the higher text diagram of recognition efficiency
As recognition methods and identifying system.
The recognition methods of the text image of specific embodiment according to the present invention comprising: images to be recognized is inputted
Compression of images normotopia network is rotated, so that the text in the images to be recognized is in horizontal position, described image pressure
Contracting normotopia network is obtained by the method training of machine learning, has the function of image rotation;
To the image recognition text of described image compression normotopia network output.
Preferably, while described image compression normotopia network rotates the images to be recognized, also in text diagram
As edge addition marker site, the marker site is used to distinguish the text and white space in text image;
To the process of the image recognition text of described image compression normotopia network output are as follows: compress normotopia net from described image
The image recognition text that network is exported according to the marker site.
Preferably, described image compression normotopia network is while rotate the images to be recognized, also to it is described to
Identification image is compressed.
Preferably, it line by line, is word for word cut to described through compression and the progress of postrotational images to be recognized according to the mark point
It cuts;
Images to be recognized input text identification neural network after cutting is subjected to text identification, the text identification nerve
Network is obtained by the method training of machine learning, has text recognition function.
Preferably, the acquisition process of the text identification neural network includes:
Establish character library;
Build convolutional neural networks of classifying more;
It chooses the text in character library and is spliced into complete image, input described image compression normotopia network is compressed
Character library training convolutional neural networks after compressing normotopia Web compression using described image;
Obtain the text identification neural network.
Preferably, the text identification neural network by the convolutional neural networks convolutional layer, pond layer, full articulamentum
It is constituted with corresponding network weight.
Preferably, described image compression normotopia network is made of the convolutional layer and pond layer of the convolutional neural networks.
Preferably, the acquisition methods of described image compression normotopia network include:
Obtain training image text;
Rotation normotopia is carried out to training image, as training target, and marks original image as training set;
According to the sample of training target, text is word for word cut line by line, cleavage site is added in text interval;
Training sample and training target input convolution self-encoding encoder are trained, the convolution after the completion of training is encoded certainly
The full articulamentum in decoder is deleted, obtains to have to automatically correct and compresses normotopia network with the described image of compressed capability.
Preferably, to the process of text identification use distributed processing mode, multiple groups text identification neural network simultaneously into
Row work;And the result of distributed text identification is integrated in order, obtain final text identification result.
The identifying system of the text image of specific embodiment according to the present invention, comprising:
Text image obtains module, and the text image obtains module for obtaining user's images to be recognized;
Compression of images normotopia network, described image compress normotopia network be used for user's images to be recognized of the acquisition into
Row rotation and compression;
Text cutting module, the text cutting module are word for word cut the image after rotary compression line by line;With
And
Text identification module, the text identification module carry out identification to the image after cutting and export corresponding text.
The recognition methods and system of a kind of text image provided in an embodiment of the present invention, beneficial effect include: by self-editing
The mode that code device and convolutional neural networks combine, makes the user do not need again to pre-process original image, provide for user
It is convenient, while maintaining higher Text region precision;The cumbersome step for simplifying existing Text region, makes Text region exist
It can be completed in consolidated network system.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of flow diagram one of the recognition methods of the text image provided according to an exemplary embodiment;
Fig. 2 is a kind of flow diagram two of the recognition methods of the text image provided according to an exemplary embodiment;
Fig. 3 is the flow diagram of the text identification neural network composition provided according to an exemplary embodiment;
Fig. 4 is the flow diagram one of the composition of the compression of images normotopia network provided according to an exemplary embodiment;
Fig. 5 is the flow diagram two of the composition of the compression of images normotopia network provided according to an exemplary embodiment;
Fig. 6 is the structural schematic diagram of the whole identification network provided according to an exemplary embodiment;
Fig. 7 is the structural schematic diagram of the character identification system provided according to an exemplary embodiment;
Fig. 8 is provided according to an exemplary embodiment through rotating the schematic diagram with the text image after compression processing.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, technical solution of the present invention will be carried out below
Detailed description.Obviously, described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Base
Embodiment in the present invention, those of ordinary skill in the art are obtained all without making creative work
Other embodiment belongs to the range that the present invention is protected.
Shown in referring to Fig.1, the embodiment provides a kind of recognition methods of text image, comprising:
101, images to be recognized is obtained.
102, images to be recognized input picture compression normotopia network is rotated, so that at the text in images to be recognized
In horizontal position;Wherein compression of images normotopia network is obtained by the method training of machine learning, has image rotation
Function;
103, the image recognition text exported from compression of images normotopia Network device.
A kind of recognition methods of text image provided in this embodiment, is combined by self-encoding encoder and convolutional neural networks
Mode, make the user do not need again to pre-process original image, provide convenience for user, while maintaining higher text
Accuracy of identification;The cumbersome step for simplifying existing Text region, can be completed Text region in consolidated network system.
As a kind of feasible implementation of above-described embodiment, convolution self-encoding encoder includes by multiple convolutional layers, Chi Hua
The encoder that layer is constituted and the decoder being made of anti-pond layer and warp lamination.Convolutional layer includes multiple convolution kernels, to input
Image carries out feature extraction and obtains characteristic pattern;The activation primitive of convolutional layer can be with are as follows: hk=σ (x*Wk+βk).Pond layer is to feature
Figure carries out noise reduction sampling operation, to reduce the calculation amount of convolution operation.Deconvolution is operated to its corresponding volume of every characteristic pattern
The transposition of product core carries out convolution operation and simultaneously sums, and activation primitive can be with are as follows: y=σ (∑ hk*(WT)k+c)。
It carries out rotating positive bit manipulation being because the orientation of the most text of image of shooting is not in page top edge
Horizontal position causes the segmentation of text and identification difficulty to increase, and accuracy reduces, using convolution self-encoding encoder to the text of input
Image carries out rotation transformation, keeps the image text orientation of output horizontal.
Referring to shown in Fig. 2, in a specific embodiment of the invention, compression of images normotopia network handles identify that image carries out
While rotation, marker site also is added at text image edge, marker site is used to distinguish the text and sky in text image
White region;The image recognition text exported from compression of images normotopia network are as follows: according to marker site from compression of images normotopia network
The image recognition text of output;
While compression of images normotopia network handles identification image is rotated, also images to be recognized is compressed;
It line by line, is word for word cut according to mark point to through compression and the progress of postrotational images to be recognized;
Images to be recognized input text identification neural network after cutting is subjected to text identification, text identification neural network
It is to be obtained by the method training of machine learning, there is text recognition function.
Carrying out compression to image while using compression of images normotopia network is because having from code book body good
Compression of images ability, if being only used for image rotation excessively waste of resource.Therefore after being rotated to image to image into
Row compression will avoid the wasting of resources.Autocoder is that a kind of have three layers of a neural network: input layer, hidden layer (coding layer) and
Decoding layer.The purpose of the network is its input of reconstruct, and the study of its hidden layer is made to arrive the well-characterized of the input.Autocoder mind
It is a kind of unsupervised machine learning algorithm through network, applies backpropagation, target value can be arranged to equal with input value.
The training objective of autocoder is will to input to copy to output.In inside, it has a description for characterizing the code of its input
Hidden layer.The convolution that the present invention uses has been compatible with denoising autocoder from coding, and randomly part is using impaired input
Identity function risk is solved, so that autocoder must be restored or be denoised.What this technology can be used for being inputted
Well-characterized.Good characterization refers to that the characterization that can steadily obtain from impaired input, the characterization can be used for restoring it
Corresponding noiseless input.
After carrying out rotary compression to image, because treated, image includes cleavage site, can be easily by image
It is word for word cut line by line.Referring to shown in Fig. 8, as a specific embodiment of the invention, to text image after rotary compression
Cleavage site information include three kinds of data, wherein first data indicate that line number, second data indicate that x-axis is sat from left to right
Mark, third data indicate y-axis coordinate, have all carried out the label of label and position to every row of text in this way, so that it may real easily
Now to the cutting of text.
Referring to shown in Fig. 3, as a kind of feasible implementation of above-described embodiment, text can be obtained by following procedure
This identification neural network:
301, character library is established;
302, convolutional neural networks of classifying are built more;
303, it chooses the text in character library and is spliced into complete image, input picture compression normotopia network is compressed, utilized
Character library training convolutional neural networks after compression of images normotopia Web compression;
304, the convolutional neural networks after interception training, obtain text identification neural network.
In a specific embodiment of the invention, the training process of a convolution Text region network includes:
The scanned picture for first collecting different literals, establishes a complete character library, is carried out using compression of images normotopia network
Compression processing, as sample set;
Convolutional neural networks are initialized, using random parameter assignment network weight, are in network wait instruct
Practice state, network convolutional layer uses Relu activation primitive, i.e. f (x)=max (0, x);
The character library being collected into is upset into sequence and is grouped, then by batch initialized convolutional neural networks of input, to net
Network is trained;
Training progress is observed, cross validation is carried out to training result, until network performance tends to restrain, completes training.
Text image after compression of images network processes includes the high-level information of original image, filters and has cut and is non-
Necessary information.
Referring to shown in Fig. 4, as a kind of feasible implementation of above-described embodiment, the acquisition of compression of images normotopia network
Process may include:
401, training image text is obtained;
402, rotation normotopia is carried out to training image, as training target, and marks original image as training set;
403, according to the sample of training target, text is word for word cut line by line, cleavage site is added in text interval;
404, training sample and training target input convolution self-encoding encoder are trained, certainly by the convolution after the completion of training
Coding deletes the full articulamentum in decoder, and obtaining has the compression of images normotopia network automatically corrected with compressed capability.
Referring to Figure 5, in a specific embodiment of the invention, one convolution self-encoding encoder of training includes following mistake
Journey:
501, training sample is collected;
502, training sample is slightly rotated, training set is added, and mark original image;
503, the picture after rotation normotopia in training sample is word for word cut;
504, cleavage site is added in cut place, is spliced into full picture as training target;
505, convolution self-encoding encoder is initialized, random value assignment network is used;
506, training sample random ordering is arranged, inputs convolution self-encoding encoder in batches and is trained until convergence.
Training method is the reconstructed error between the reconstructed image and training target for minimizing convolution self-encoding encoder.Training loss
Function uses lowest mean square difference function, i.e.,
Wherein yiFor training target target value,For the value of reconstructed image.The more new formula of convolutional network parameter are as follows:
It is in a specific embodiment of the invention, whole to identify that network includes: compression of images normotopia net referring to shown in Fig. 6
Network, program and text identification neural network that the images to be recognized after rotary compression is word for word cut.Due to figure
Output result as compressing normotopia network has apparent cleavage site, therefore text cutting is not necessarily in conventional text cutting side
Formula.Can dynamic scan input picture, using cleavage site as boundary, image between cleavage site connects a Text region nerve net
Network.It is whole to identify that network be split as two subsystems, Text region nerve net if server-side processes ability is enough
Network can be directly connected to convolution self-encoding encoder end, form a complete neural network.The design avoid because text by
A large amount of communication congestions between GPU and CPU produced by word is divided, the significant increase utilization efficiency and calculating speed of GPU.
In some embodiments of the invention, distributed processing mode can be used in the process of Text region, multiple whole
Body identification network works at the same time, and recognition speed can be substantially improved.
The embodiments of the present invention also provide a kind of systems of text image identification, comprising:
Text image obtains module, for obtaining user's images to be recognized;
Compression of images normotopia network, for user's images to be recognized of acquisition to be rotated and compressed;
Text cutting module, for word for word being cut the image after rotary compression line by line;And
Text identification module exports corresponding text for carrying out identification to the image after cutting.
In some embodiments of the invention, the use environment that text image integrally identifies includes multiple terminals and one
A server end, the system that server end is equipped with the identification of above-mentioned text image.Terminal can be but not limited to various energy operation figures
As in the personal computer of mathematical formulae detection method, laptop, personal digital assistant, smart phone, tablet computer and
Portable wearable device etc..Server can be the server for realizing simple function, be also possible to realize the clothes of multiple functions
Business device, specifically can be independent physical server, is also possible to physical server cluster.It is identified needed for client terminal shooting
Text, such as a examination paper, are sent to server end by network;Server end by picture to be measured using convolution from code machine into
Row automation pretreatment, reuses Text region network and is identified, obtain final result.Identify content of text by network again
Communication return is carried out to client, user obtains recognition result.
In a specific embodiment of the invention, the computer equipment of server end includes the place connected by system bus
Manage device, memory, network interface, display screen and input unit.Wherein, the processor is for providing calculating and control ability, branch
Support the operation of entire terminal.The memory of computer equipment includes non-volatile memory medium and built-in storage, non-volatile to deposit
Storage media is stored with operating system and computer program, when which is executed by processor, so that processor is realized
Mathematical formulae detection method in a kind of image.Built-in storage in computer equipment can also store computer program, the calculating
When machine program is executed by processor, processor may make to execute a kind of recognition methods of whole text image.Computer equipment
Network interface with terminal for communicating.The input unit of computer equipment can be the touch layer covered on display screen, can also be with
It is external keyboard, Trackpad or mouse etc., input unit can obtain the operation interface that user uses finger to show display screen
The instruction of generation, such as obtain user and input image to be detected etc. by clicking the particular options in terminal.Display screen can be used for
Show the text filed of input interface or output.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
The technical staff for being familiar with this art field in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all cover
Within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (10)
1. a kind of recognition methods of text image characterized by comprising
Images to be recognized input picture compression normotopia network is rotated, so that the text in the images to be recognized is in water
Flat position, described image compression normotopia network are obtained by the method training of machine learning, have the function of image rotation;
To the image recognition text of described image compression normotopia network output.
2. the method according to claim 1, wherein described image compresses normotopia network to the images to be recognized
While rotation, marker site also is added at text image edge, the marker site is for distinguishing in text image
Text and white space;
To the process of the image recognition text of described image compression normotopia network output are as follows: compress normotopia network root from described image
The image recognition text exported according to the marker site.
3. according to the method described in claim 2, it is characterized in that, described image compresses normotopia network to the images to be recognized
While rotation, also the images to be recognized is compressed.
4. according to the method described in claim 3, it is characterized in that,
It line by line, is word for word cut to described through compression and the progress of postrotational images to be recognized according to the mark point;
Images to be recognized input text identification neural network after cutting is subjected to text identification, the text identification neural network
It is to be obtained by the method training of machine learning, there is text recognition function.
5. according to the method described in claim 4, it is characterized in that, the acquisition process of the text identification neural network includes:
Establish character library;
Build convolutional neural networks of classifying more;
It chooses the text in character library and is spliced into complete image, input described image compression normotopia network carries out compression and utilizes the figure
As the character library training convolutional neural networks after compression normotopia Web compression;
Obtain the text identification neural network.
6. according to the method described in claim 5, it is characterized in that, the text identification neural network is by the convolutional Neural net
Convolutional layer, pond layer, full articulamentum and the corresponding network weight of network are constituted.
7. according to the method described in claim 5, it is characterized in that, described image compresses normotopia network by the convolutional Neural net
The convolutional layer and pond layer of network form.
8. the method according to the description of claim 7 is characterized in that the acquisition methods of described image compression normotopia network include:
Obtain training image text;
Rotation normotopia is carried out to training image, as training target, and marks original image as training set;
According to the sample of training target, text is word for word cut line by line, cleavage site is added in text interval;
Training sample and training target input convolution self-encoding encoder are trained, the convolution after the completion of training is encoded into deletion certainly
Full articulamentum in decoder obtains to have to automatically correct and compresses normotopia network with the described image of compressed capability.
9. method according to any one of claims 1 to 8, which is characterized in that used to the process of text identification distributed
Processing mode, text identification neural network described in multiple groups work simultaneously;And in order to the result of distributed text identification
It is integrated, obtains final text identification result.
10. a kind of identifying system of text image characterized by comprising
Text image obtains module, and the text image obtains module for obtaining user's images to be recognized;
Compression of images normotopia network, described image compression normotopia network is for revolving user's images to be recognized of the acquisition
Turn and compresses;
Text cutting module, the text cutting module are word for word cut the image after rotary compression line by line;And text
This identification module, the text identification module carry out identification to the image after cutting and export corresponding text.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810965342.5A CN109241974B (en) | 2018-08-23 | 2018-08-23 | Text image identification method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810965342.5A CN109241974B (en) | 2018-08-23 | 2018-08-23 | Text image identification method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109241974A true CN109241974A (en) | 2019-01-18 |
CN109241974B CN109241974B (en) | 2020-12-01 |
Family
ID=65069329
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810965342.5A Active CN109241974B (en) | 2018-08-23 | 2018-08-23 | Text image identification method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109241974B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110674811A (en) * | 2019-09-04 | 2020-01-10 | 广东浪潮大数据研究有限公司 | Image recognition method and device |
CN111242024A (en) * | 2020-01-11 | 2020-06-05 | 北京中科辅龙科技股份有限公司 | Method and system for recognizing legends and characters in drawings based on machine learning |
CN111444908A (en) * | 2020-03-25 | 2020-07-24 | 腾讯科技(深圳)有限公司 | Image recognition method, device, terminal and storage medium |
CN111695385A (en) * | 2019-03-15 | 2020-09-22 | 杭州海康威视数字技术股份有限公司 | Text recognition method, device and equipment |
US11216960B1 (en) | 2020-07-01 | 2022-01-04 | Alipay Labs (singapore) Pte. Ltd. | Image processing method and system |
CN117496531A (en) * | 2023-11-02 | 2024-02-02 | 四川轻化工大学 | Construction method of convolution self-encoder capable of reducing Chinese character recognition resource overhead |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104915386A (en) * | 2015-05-25 | 2015-09-16 | 中国科学院自动化研究所 | Short text clustering method based on deep semantic feature learning |
CN105469047A (en) * | 2015-11-23 | 2016-04-06 | 上海交通大学 | Chinese detection method based on unsupervised learning and deep learning network and system thereof |
US20160283814A1 (en) * | 2015-03-25 | 2016-09-29 | Alibaba Group Holding Limited | Method and apparatus for generating text line classifier |
CN107247950A (en) * | 2017-06-06 | 2017-10-13 | 电子科技大学 | A kind of ID Card Image text recognition method based on machine learning |
CN107403130A (en) * | 2017-04-19 | 2017-11-28 | 北京粉笔未来科技有限公司 | A kind of character identifying method and character recognition device |
-
2018
- 2018-08-23 CN CN201810965342.5A patent/CN109241974B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160283814A1 (en) * | 2015-03-25 | 2016-09-29 | Alibaba Group Holding Limited | Method and apparatus for generating text line classifier |
CN104915386A (en) * | 2015-05-25 | 2015-09-16 | 中国科学院自动化研究所 | Short text clustering method based on deep semantic feature learning |
CN105469047A (en) * | 2015-11-23 | 2016-04-06 | 上海交通大学 | Chinese detection method based on unsupervised learning and deep learning network and system thereof |
CN107403130A (en) * | 2017-04-19 | 2017-11-28 | 北京粉笔未来科技有限公司 | A kind of character identifying method and character recognition device |
CN107247950A (en) * | 2017-06-06 | 2017-10-13 | 电子科技大学 | A kind of ID Card Image text recognition method based on machine learning |
Non-Patent Citations (2)
Title |
---|
BAOGUANG SHI,XINGGANG WANG,PENGYUAN LYU,CONG YAO,XIANG BAI: "Robust Scene Text Recognition with Automatic Rectification", 《 2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
白翔,杨明锟,石葆光,廖明辉: "基于深度学习的场景文字检测与识别", 《中国科学:信息科学》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111695385A (en) * | 2019-03-15 | 2020-09-22 | 杭州海康威视数字技术股份有限公司 | Text recognition method, device and equipment |
CN111695385B (en) * | 2019-03-15 | 2023-09-26 | 杭州海康威视数字技术股份有限公司 | Text recognition method, device and equipment |
CN110674811A (en) * | 2019-09-04 | 2020-01-10 | 广东浪潮大数据研究有限公司 | Image recognition method and device |
CN111242024A (en) * | 2020-01-11 | 2020-06-05 | 北京中科辅龙科技股份有限公司 | Method and system for recognizing legends and characters in drawings based on machine learning |
CN111444908A (en) * | 2020-03-25 | 2020-07-24 | 腾讯科技(深圳)有限公司 | Image recognition method, device, terminal and storage medium |
CN111444908B (en) * | 2020-03-25 | 2024-02-02 | 腾讯科技(深圳)有限公司 | Image recognition method, device, terminal and storage medium |
US11216960B1 (en) | 2020-07-01 | 2022-01-04 | Alipay Labs (singapore) Pte. Ltd. | Image processing method and system |
CN117496531A (en) * | 2023-11-02 | 2024-02-02 | 四川轻化工大学 | Construction method of convolution self-encoder capable of reducing Chinese character recognition resource overhead |
CN117496531B (en) * | 2023-11-02 | 2024-05-24 | 四川轻化工大学 | Construction method of convolution self-encoder capable of reducing Chinese character recognition resource overhead |
Also Published As
Publication number | Publication date |
---|---|
CN109241974B (en) | 2020-12-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109241974A (en) | A kind of recognition methods and system of text image | |
Zhu et al. | Image inpainting by end-to-end cascaded refinement with mask awareness | |
CN114066902A (en) | Medical image segmentation method, system and device based on convolution and transformer fusion | |
Couturier et al. | Image denoising using a deep encoder-decoder network with skip connections | |
CN105493095A (en) | Adaptive and recursive filtering for sample submission | |
CN113657404B (en) | Image processing method of Dongba pictograph | |
CN110610180A (en) | Method, device and equipment for generating recognition set of wrongly-recognized words and storage medium | |
CN105095939A (en) | Two-dimensional code vision optimization method | |
CN108898562A (en) | A kind of mobile device image defogging method based on deep learning | |
JP7389824B2 (en) | Object identification method and device, electronic equipment and storage medium | |
CN112380566A (en) | Method, apparatus, electronic device, and medium for desensitizing document image | |
Sasaki et al. | Learning to restore deteriorated line drawing | |
CN112017162B (en) | Pathological image processing method, pathological image processing device, storage medium and processor | |
CN115358917B (en) | Method, equipment, medium and system for migrating non-aligned faces of hand-painted styles | |
CN113553797A (en) | End-to-end consistent-style character and image generation method | |
CN111768466A (en) | Image filling method, device, equipment and storage medium | |
US20200320291A1 (en) | Techniques to determine document recognition errors | |
CN111833413B (en) | Image processing method, image processing device, electronic equipment and computer readable storage medium | |
CN111862343B (en) | Three-dimensional reconstruction method, device, equipment and computer readable storage medium | |
CN116484878B (en) | Semantic association method, device, equipment and storage medium of power heterogeneous data | |
CN116863419A (en) | Method and device for lightening target detection model, electronic equipment and medium | |
CN114299204B (en) | Three-dimensional cartoon character model generation method and device | |
CN113610856A (en) | Method and device for training image segmentation model and image segmentation | |
Wang et al. | ThemeStation: Generating Theme-Aware 3D Assets from Few Exemplars | |
CN117523593B (en) | Patient medical record data processing method and system |
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 |