CN110363190A - A kind of character recognition method, device and equipment - Google Patents
A kind of character recognition method, device and equipment Download PDFInfo
- Publication number
- CN110363190A CN110363190A CN201910681467.XA CN201910681467A CN110363190A CN 110363190 A CN110363190 A CN 110363190A CN 201910681467 A CN201910681467 A CN 201910681467A CN 110363190 A CN110363190 A CN 110363190A
- Authority
- CN
- China
- Prior art keywords
- text
- subgraph
- pixel
- image
- identification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Character Discrimination (AREA)
Abstract
This specification embodiment provides a kind of character recognition method, device and equipment.The described method includes: receiving images to be recognized；According to the corresponding characteristic value of pixel in the images to be recognized, text filed subgraph is obtained；The text filed subgraph is inputted into Text region model, obtains the identification lteral data corresponding to the text filed subgraph；The Text region model, including the machine learning model gone out based on endorsed image data sample training；It integrates the identification lteral data and feeds back.By the above method, the step of improving the accuracy of Text region in image, simplify Text region, so as to fast be accurately realized Text region.
Description
Technical field
This specification embodiment is related to field of computer technology, in particular to a kind of character recognition method, device and equipment.
Background technique
In the past, for the text in image, although the mankind can easily identify text therein, for computer
For, text can not be directly identified from image.As the image comprising text that people are contacted is more and more, entirely
Portion by manpower to the text in image carry out identification and it is unrealistic.Character recognition technology also just comes into being.Existing text
Identification technology be mostly according to the training of a large amount of data sample, obtain can to the training pattern that text is identified and then
It is trained using the training pattern.
But it when carrying out Text region using the training pattern in the prior art, generally requires to be utilized respectively multiple
Stage is executed, and realizes that process is relatively complicated, when identifying text, needs to be in advance single word by character string cutting
Symbol, is identified respectively with this, these single characters are carried out splicing again after the completion of identification and are integrated into normal text, so that in text
Efficiency when word identifies is also more low.In addition, it is directed to scalloping, fuzzy situation, it often just can not be compared in the initial stage
The region where text is judged well, brings no small difficulty to Text region.Therefore, needing one kind at present can facilitate
It is accurately realized the technology of Text region.
Summary of the invention
The purpose of this specification embodiment is to provide a kind of character recognition method, device and equipment, to solve the prior art
In in identification text be to need individually to annotate character, and identify the not high problem of accuracy, so that realization is quick and precisely
Ground carries out Text region.
To solve the above-mentioned problems, this specification embodiment proposes a kind of character recognition method, device and equipment, specifically
Implement as follows:
A kind of character recognition method, comprising:
Receive images to be recognized；
According to the corresponding characteristic value of pixel in the images to be recognized, text filed subgraph is obtained；
The text filed subgraph is inputted into Text region model as a whole, obtains corresponding to the text filed son
The identification lteral data of image；The Text region model, including the machine gone out based on endorsed image data sample training
Learning model；
It integrates the identification lteral data and feeds back.
A kind of character recognition device, comprising:
Image receiver module, for receiving images to be recognized；
Text filed subgraph obtains module, for obtaining according to the corresponding characteristic value of pixel in the images to be recognized
Take text filed subgraph；
Identify that lteral data obtains module, for the text filed subgraph to be inputted Text region mould as a whole
Type obtains the identification lteral data corresponding to the text filed subgraph；The Text region model, including based on endorsed
Image data sample training go out machine learning model；
Feedback module, for integrating the identification lteral data and feeding back.
A kind of Text region equipment, comprising:
Memory, for storing computer instruction；
Processor performs the steps of reception images to be recognized for executing the computer instruction；According to it is described to
It identifies the corresponding characteristic value of pixel in image, obtains text filed subgraph；As a whole by the text filed subgraph
Text region model is inputted, the identification lteral data corresponding to the text filed subgraph is obtained；The Text region model,
Including the machine learning model gone out based on endorsed image data sample training；It integrates the identification lteral data and feeds back.
The technical solution provided by above this specification embodiment is as it can be seen that this specification embodiment is receiving figure to be identified
As after, first according to the corresponding characteristic value of pixel in images to be recognized, text filed subgraph is obtained.Utilize pixel spy
Value indicative is to the text filed region for being judged to specify that needs are analyzed first, thus fuzzy in face of image or the feelings such as distort
When condition, the region that can be accurately distributed to text is identified, improves the robustness of recognition methods.For text filed
Subgraph is directly inputted and is identified in Text region model, without carrying out cutting according to character in text filed subgraph,
Simplify execution step.The Text region model, including the machine learning gone out based on endorsed image data sample training
Model enables the Text region model to identify text included in text filed subgraph or character.Most
The text that will identify that afterwards is integrated and is fed back, so as to complete this Text region process.Pass through above-mentioned Text region side
The introduction of the embodiment of method is as can be seen that this method can efficiently identify out the region where text, and simplify text
The operating procedure of identification easily and accurately identifies text so as to realize.
Detailed description of the invention
In order to illustrate more clearly of this specification embodiment or technical solution in the prior art, below will to embodiment or
Attached drawing needed to be used in the description of the prior art is briefly described, it should be apparent that, the accompanying drawings in the following description is only
The some embodiments recorded in this specification, for those of ordinary skill in the art, before not making the creative labor
It puts, is also possible to obtain other drawings based on these drawings.
Fig. 1 is a kind of flow chart of character recognition method of this specification embodiment；
Fig. 2 is a kind of module map of character recognition device of this specification embodiment；
Fig. 3 is a kind of module map of Text region equipment of this specification embodiment；
Fig. 4 A is a kind of schematic diagram of the document of text filed subgraph to be determined of this specification embodiment；
Fig. 4 B is a kind of schematic diagram for the document that text filed subgraph has been determined of this specification embodiment；
Fig. 5 is a kind of structure chart of CRNN convolution loop neural network model of this specification embodiment.
Specific embodiment
Below in conjunction with the attached drawing in this specification embodiment, the technical solution in this specification embodiment is carried out clear
Chu is fully described by, it is clear that described embodiment is only this specification a part of the embodiment, rather than whole implementation
Example.The embodiment of base in this manual, those of ordinary skill in the art are obtained without making creative work
Every other embodiment, all should belong to this specification protection range.
A kind of character recognition method of this specification embodiment is introduced below in conjunction with attached drawing 1.The execution of the method
Main body is computer equipment, and the computer equipment includes but is not limited to server, industrial personal computer (industrial control computer), PC machine
Deng.The method is shown in the specific implementation steps are as follows:
S100: images to be recognized is received.
Images to be recognized is the targeted image for needing to identify text therein of this specification embodiment.To
Identify that in image include character to be identified and/or character string.The character includes English alphabet, number, symbol, Chinese character
Deng.
The images to be recognized, which can be, is submitted directly to the computer equipment for identification by user, is also possible to it
When his computer equipment has the image for needing to be identified to wherein text in task implementation procedure, described image is sent
To the computer equipment for executing the character recognition method.It may include that other come that the computer, which receives images to be recognized also,
Source, with no restriction to this.
The images to be recognized may include the images such as document image, paper image, handwritten text image.It is described to be identified
Image for itself format and related particular content with no restriction.
S200: according to the corresponding characteristic value of pixel in the images to be recognized, text filed subgraph is obtained.
Image is made of numerous pixels.Pixel is the smallest partition unit of image, is corresponding with location information in the picture
And characteristic value.Generally in the image comprising text, the color of text and the color different from of image background color.Pass through analysis chart
As in the case where all pixels corresponding characteristic value, in the difference according to corresponding to the text filed characteristic value with image background color,
It can determine the region that text is distributed.
But the image that the region that is distributed of text is constituted is often irregular, in later use Text region model
When identifying to text, if numerical value corresponding to text is directly inputted Text region model, phase may can not be obtained
The recognition effect answered.Therefore, in order to which subsequent process can better implement, finally obtained text filed subgraph can be
Preferably identification is realized to meet the format for the training sample that text identification model is directed in rectangular area comprising text
Effect.
It, can be first according to the corresponding characteristic value of pixel in the images to be recognized, meter in order to realize above-mentioned technical effect
The corresponding activation score value of pixel is calculated, then obtains the pixel that activation score value is greater than point threshold, as activation pixel, finally again
Using image outline detection algorithm, the acquisition of text filed subgraph is realized.
Characteristic value can be used to indicate that specific value corresponding to the feature of pixel.For example, under normal circumstances, text pair
The color of the pixel answered pixel corresponding with image background is not identical.It can be using the color value of pixel as corresponding feature
Value, the color value may include R numerical value, G numerical value and B numerical value when indicating the color of pixel using RGB；In benefit
When indicating the color of pixel with CMY, the color value may include C numerical value, M numerical value and Y numerical value.Utilize HSV, HIS
When indicating the color of pixel Deng other color spaces, the color value can also be set according to specific circumstances, herein no longer
It repeats.
In some cases, text is typically distributed across specific position, such as the center of image, can also be corresponding by pixel
Location information be set as characteristic value, corresponding activation pixel is obtained according to corresponding location information, constructs text filed subgraph
Picture.After the characteristic value for calculating all pixels, the activation pixel for meeting text filed feature can be filtered out accordingly, construct text
Region subgraph.
Activation score value refer to the pixel with it is text filed between correlation score.It activates score value higher, more may indicate
In text filed.Obtain the corresponding activation score value of each pixel and then by the corresponding activation score value of pixel with set in advance
Fixed point threshold is compared.If the activation score value of pixel is greater than point threshold, the pixel is obtained as activation pixel.Swash
Look exactly like element is the pixel being distributed in text filed substantially.Pixel is activated according to these, recycles image outline detection algorithm,
Text filed subgraph can be obtained.Described image contour detecting algorithm, may include Sobel operator, Laplacian operator,
The edge detection algorithms such as Canny operator, determine text distribution image edge after, that is, can determine text distribution image.
The above process is illustrated using a specific example.It as shown in Figure 4 A, is a list figure to be identified
Picture.The activation score value of the pixel in image is calculated first, further according to activation score value, is compared with activation point threshold
Compared with determining activation pixel therein.In Fig. 4 A, the position of pixel is activated to be determined substantially.Further according to determining activation pixel,
According to image outline detection algorithm, text filed subgraph is determined, as shown in Figure 4 B, text filed subgraph is by white box table
Show.Using text filed subgraph as the image identified in subsequent process by Text region model.
It, when identifying the text in image, needs for character string to be split as single word different from the prior art
Symbol, then each character is known otherwise respectively.It can in images to be recognized in this method embodiment for being identified
With the character string for including a character or being made of multiple characters.It in the follow-up process, can directly will include multiple characters
Input text image identification model in.So, operation is simplified, the efficiency of Text region is improved.
Preferably, when the images to be recognized is document image, numerous tables is usually contained in the document image
Lattice item.Subsequent Text region is accurately executed in order to more convenient, avoids mixing in the corresponding text of different zones in document
The problem of to cause identification to malfunction, the table line in document can be continued to identify first with table line detecting algorithm.Root
According to the table line in the obtained document image of identification, document image is cut, obtain at least one cutting image and then
Successively according to the corresponding characteristic value of pixel in each cutting image, corresponding text filed subgraph is obtained respectively.
Above-mentioned steps in order to facilitate understanding are illustrated using a specific example.It as shown in Figure 4 A, is a Zhang Dan
According to, in document have different regions, respectively indicate different meanings.First choice identifies the table line in document, then
According to table line, after carrying out cutting to document, multiple regions as illustrated in the drawing are formed, corresponding to different cutting images,
These cutting images are identified respectively again, to reduce in identification process because obscuring confusion caused by region, are improved
The accuracy rate of Text region.
S300: inputting Text region model for the text filed subgraph as a whole, obtains corresponding to the text
The identification lteral data of region subgraph.
The Text region model can be the machine learning model gone out based on endorsed image data sample training.
By being annotated in advance to the corresponding text of image each in image data sample, recycle machine learning to the model into
Row training, can directly identify the text in image.
Preferably, the machine learning model gone out based on endorsed image data sample training, can be CRNN volumes
Product Recognition with Recurrent Neural Network model.
In one embodiment, the CRNN convolution loop neural network model, including convolutional layer, batch standardization layer,
Pond layer, dropout layers, circulation layer, transcription layer.Specifically, the CRNN convolution loop neural network model can be such as Fig. 5 institute
Show, the CRNN convolution loop neural network model may include convolutional layer, circulation layer and transcription layer three parts.Convolutional layer master
It is used to from the image of input extract characteristic sequence, circulation layer is used to predict the label distribution of the characteristic sequence proposed, and turns
Record layer is mainly by converting final recognition result by operations such as duplicate removal integration for label distribution.
Specifically, in above-mentioned convolutional layer, and can wrap containing 5 modules, it may include again smaller in each module
Convolutional layer.Module 1, module 2 are made of convolutional layer, batch standardization layer, pond layer, and module 3 is by two layers of convolutional layer and batch rule
Generalized layer, pond layer form, and module 4 is made of three-layer coil lamination and batch standardization layer, pond layer, and module 5 is by convolutional layer and criticizes
Standardize layer composition.Convolutional layer in module carries out convolution operation mainly for image and is convenient for extract the feature in image
Subsequent step is identified.
Batch standardization layer is mainly used for the data for transmitting convolutional layer and standardizes again, so that the mean value of output data is close
0, standard deviation is close to 1, so that the phenomenon that alleviating the gradient disappearance or gradient explosion in training process, improves the convergence of machine learning
Speed accelerates training progress.
Pond layer is mainly used for reducing received data, to improve calculating speed, increases the Shandong of extracted feature
Stick
There are problems that over-fitting in order to avoid generating model, dropout layers can be increased after module 5, it is main to use
In temporarily deleting at random the neuron in neural network, to increase the generalization ability of model.
Since the Text region model is based on directly on what character string was trained, when obtaining training data without
It needs to divide, is marked without to each character, character string is labeled respectively directly.So, not only greatly
Reduce the workload that the training stage is labeled data sample greatly, model service stage also can directly to character string into
Row identification, simplifies identification process, improves the efficiency of Text region.
It,, can be with for front and back there are the character string of semantic relation under specifically identification scene meanwhile for character string
It is further determined using the relationship, to guarantee that recognition result is more accurate.
During practical application, due to may be more similar between different types of character, such as letter o and number 0
May be more similar, carrying out identification according to image may bad identification.In response to this, it is respectively trained out using the above method
English identification model, digital identification model and Chinese Character Recognition model etc., concrete application are directed to different application scenarios benefits in the process
It is identified with corresponding model, so as to further increase the accuracy rate of identification.
In the training stage, by utilizing the data sample being labeled based on character string to the Text region model
It is trained, so that the Text region model can not only directly identify character, also has for character string preferable
Recognition effect simplifies training step, also improves text so as to be identified directly against the character string in input picture
The efficiency of word identification.
S400: the integration identification lteral data is simultaneously fed back.
After identifying above-mentioned text filed subgraph, obtain corresponding respectively to each text filed subgraph
Identify lteral data.In order to enable the lteral data showed still retains the bandwagon effect in original image, by the text
Data are integrated according to position of the corresponding text filed subgraph in images to be recognized, so that it is guaranteed that the text identified
Bandwagon effect it is normal.
In the case where the images to be recognized is document image, if before for the table line pair in the document image
Document image is cut, then the identification lteral data is set to the corresponding region of cutting image first, then will include
There is the cutting image of identification lteral data to be restored according to the corresponding position in images to be recognized, is preferably read to realize
Read effect.
After by identification lteral data integration, by the identification lteral data after the integration according to reception data
It is fed back in source.For example, the lteral data after identification is fed back to user for the images to be recognized that user directly provides；
For the images to be recognized that other servers provide, the lteral data after identification is fed back into corresponding server.
Above method step is illustrated below with a specific Sample Scenario.User needs to count at year end
It is used for the expense of office appliance in 1 year, but since document quantity is excessive, it is more to check that each document and counting can waste one by one
Time, and it is easy error.The problem can be better solved by carrying out identification for the corresponding data in document.It collects first all
Document, the corresponding scan image of these documents is committed to the server for being used for carrying out Text region.Server is in reception
After stating image, according to the corresponding characteristic value of pixel in image, it is first determined the text filed subgraph for dividing this to be distributed.It will
Text filed subgraph sequentially inputs Text region model, after identifying the wherein text datas such as corresponding character or character string.
The text data that these are identified is integrated and fed back further according to position corresponding to text filed subgraph.User receives
To after all text datas, directly wherein keyword relevant to office appliance is scanned for, so that it is determined that office appliance
Corresponding price, and the expense of office appliance is finally counted, save corresponding resource and time.
Can be seen that text recognition method that this specification embodiment is introduced by the above method and Sample Scenario can
To using the corresponding characteristic value of pixel in image, accurately text filed subgraph therein is determined, before recycling
Text therein is identified using the machine learning model that the image data sample training of annotation goes out, so as to quickly quasi-
It really realizes text identification, improves the efficiency of text identification, and ensure and accurately improve the Shandong of character recognition method
Stick.
A kind of embodiment of character recognition device of this specification introduced below, described device are integrated in the computer and set
It is standby.The specific module of described device is as shown in Figure 2:
Image receiver module 210, for receiving images to be recognized；
Text filed subgraph obtains module 220, is used for according to the corresponding characteristic value of pixel in the images to be recognized,
Obtain text filed subgraph；
Identify that lteral data obtains module 230, for the text filed subgraph to be inputted Text region as a whole
Model obtains the identification lteral data corresponding to the text filed subgraph；The Text region model, including be based on being criticized
The machine learning model that the image data sample training of note goes out；
Feedback module 240, for integrating the identification lteral data and feeding back.
As shown in figure 3, this specification embodiment provides a kind of computer equipment.The computer equipment may include storage
Device and processor.
In the present embodiment, the memory can be implemented in any suitable manner.For example, the memory can be
Read-only memory, mechanical hard disk, solid state hard disk or USB flash disk etc..The memory can be used for storing computer instruction.
In the present embodiment, the processor can be implemented in any suitable manner.For example, processor can take example
Such as microprocessor or processor and storage can by (micro-) processor execute computer readable program code (such as software or
Firmware) computer-readable medium, logic gate, switch, specific integrated circuit (Application Specific
Integrated Circuit, ASIC), programmable logic controller (PLC) and the form etc. for being embedded in microcontroller.
The processor can execute the computer instruction and perform the steps of reception images to be recognized；According to described
The corresponding characteristic value of pixel in images to be recognized, obtains text filed subgraph；Using the text filed subgraph as whole
Body inputs Text region model, obtains the identification lteral data corresponding to the text filed subgraph；The Text region mould
Type, including the machine learning model gone out based on endorsed image data sample training；Integrate the identification lteral data and anti-
Feedback.
In the 1990s, the improvement of a technology can be distinguished clearly be on hardware improvement (for example,
Improvement to circuit structures such as diode, transistor, switches) or software on improvement (improvement for method flow).So
And with the development of technology, the improvement of current many method flows can be considered as directly improving for hardware circuit.
Designer nearly all obtains corresponding hardware circuit by the way that improved method flow to be programmed into hardware circuit.Cause
This, it cannot be said that the improvement of a method flow cannot be realized with hardware entities module.For example, programmable logic device
(Programmable Logic Device, PLD) (such as field programmable gate array (Field Programmable Gate
Array, FPGA)) it is exactly such a integrated circuit, logic function determines device programming by user.By designer
Voluntarily programming comes a digital display circuit " integrated " on a piece of PLD, designs and makes without asking chip maker
Dedicated IC chip.Moreover, nowadays, substitution manually makes IC chip, this programming is also used instead mostly " is patrolled
Volume compiler (logic compiler) " software realizes that software compiler used is similar when it writes with program development,
And the source code before compiling also write by handy specific programming language, this is referred to as hardware description language
(Hardware Description Language, HDL), and HDL is also not only a kind of, but there are many kind, such as ABEL
(Advanced Boolean Expression Language)、AHDL(Altera Hardware Description
Language)、Confluence、CUPL(Cornell University Programming Language)、HDCal、JHDL
(Java Hardware Description Language)、Lava、Lola、MyHDL、PALASM、RHDL(Ruby
Hardware Description Language) etc., VHDL (Very-High-Speed is most generally used at present
Integrated Circuit Hardware Description Language) and Verilog2.Those skilled in the art
It will be apparent to the skilled artisan that only needing method flow slightly programming in logic and being programmed into integrated circuit with above-mentioned several hardware description languages
In, so that it may it is readily available the hardware circuit for realizing the logical method process.
System, device, module or the unit that above-described embodiment illustrates can specifically realize by computer chip or entity,
Or it is realized by the product with certain function.It is a kind of typically to realize that equipment is computer.Specifically, computer for example may be used
Think personal computer, laptop computer, cellular phone, camera phone, smart phone, personal digital assistant, media play
It is any in device, navigation equipment, electronic mail equipment, game console, tablet computer, wearable device or these equipment
The combination of equipment.
As seen through the above description of the embodiments, those skilled in the art can be understood that this specification
It can realize by means of software and necessary general hardware platform.Based on this understanding, the technical solution of this specification
Substantially the part that contributes to existing technology can be embodied in the form of software products in other words, the computer software
Product can store in storage medium, such as ROM/RAM, magnetic disk, CD, including some instructions are used so that a computer
Equipment (can be personal computer, server or the network equipment etc.) executes each embodiment of this specification or embodiment
Certain parts described in method.
All the embodiments in this specification are described in a progressive manner, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for system reality
For applying example, since it is substantially similar to the method embodiment, so being described relatively simple, related place is referring to embodiment of the method
Part explanation.
This specification can be used in numerous general or special purpose computing system environments or configuration.Such as: personal computer,
Server computer, handheld device or portable device, laptop device, multicomputer system, microprocessor-based system,
Set top box, programmable consumer-elcetronics devices, network PC, minicomputer, mainframe computer including any of the above system are set
Standby distributed computing environment etc..
This specification can describe in the general context of computer-executable instructions executed by a computer, such as journey
Sequence module.Generally, program module include routines performing specific tasks or implementing specific abstract data types, programs, objects,
Component, data structure etc..This specification can also be practiced in a distributed computing environment, in these distributed computing environment
In, by executing task by the connected remote processing devices of communication network.In a distributed computing environment, program module
It can be located in the local and remote computer storage media including storage equipment.
Although depicting this specification by embodiment, it will be appreciated by the skilled addressee that there are many become for this specification
Shape and the spirit changed without departing from this specification, it is desirable to which the attached claims include these deformations and change without departing from this
The spirit of specification.
Claims (10)
1. a kind of character recognition method characterized by comprising
Receive images to be recognized；
According to the corresponding characteristic value of pixel in the images to be recognized, text filed subgraph is obtained；
The text filed subgraph is inputted into Text region model as a whole, obtains corresponding to the text filed subgraph
Identification lteral data；The Text region model includes: the machine learning gone out based on endorsed image data sample training
Model；
It integrates the identification lteral data and feeds back.
2. the method as described in claim 1, which is characterized in that the images to be recognized, including document image.
3. method according to claim 2, which is characterized in that after the reception images to be recognized, further includes:
According to the table line in document image, document image is cut, obtains at least one cutting image；
Correspondingly, described obtain text filed subgraph according to the corresponding characteristic value of pixel in the images to be recognized, packet
It includes:
According to the corresponding characteristic value of pixel at least one described cutting image, text filed subgraph is obtained.
4. the method as described in claim 1, which is characterized in that include at least one word in the text filed subgraph
Symbol or character string.
5. the method as described in claim 1, which is characterized in that described according to the corresponding spy of pixel in the images to be recognized
Value indicative obtains text filed subgraph, comprising:
According to the corresponding characteristic value of pixel in the images to be recognized, the corresponding activation score value of pixel is calculated；
The pixel that activation score value is greater than point threshold is obtained, as activation pixel；
Based on the activation pixel, text filed subgraph is obtained using image outline detection algorithm.
6. the method as described in claim 1, which is characterized in that the machine gone out based on endorsed image data sample training
Device learning model, including CRNN convolution loop neural network model.
7. the method as described in claim 1, which is characterized in that the endorsed image data sample includes: for character
The image data sample that string is annotated.
8. the method as described in claim 1, which is characterized in that described to integrate the identification lteral data and feed back, comprising:
The position that the identification lteral data corresponds in images to be recognized is integrated；
Identification lteral data after feedback integration.
9. a kind of character recognition device characterized by comprising
Image receiver module, for receiving images to be recognized；
Text filed subgraph obtains module, for obtaining text according to the corresponding characteristic value of pixel in the images to be recognized
Region subgraph；
It identifies that lteral data obtains module, for the text filed subgraph to be inputted Text region model, is corresponded to
The identification lteral data of the text filed subgraph；The Text region model, including based on endorsed image data sample
Originally the machine learning model trained；
Feedback module, for integrating the identification lteral data and feeding back.
10. a kind of Text region equipment characterized by comprising
Memory, for storing computer instruction；
Processor performs the steps of reception images to be recognized for executing the computer instruction；According to described to be identified
The corresponding characteristic value of pixel in image, obtains text filed subgraph；The text filed subgraph is inputted into Text region mould
Type obtains the identification lteral data corresponding to the text filed subgraph；The Text region model, including based on endorsed
Image data sample training go out machine learning model；It integrates the identification lteral data and feeds back.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910681467.XA CN110363190A (en) | 2019-07-26 | 2019-07-26 | A kind of character recognition method, device and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910681467.XA CN110363190A (en) | 2019-07-26 | 2019-07-26 | A kind of character recognition method, device and equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110363190A true CN110363190A (en) | 2019-10-22 |
Family
ID=68222330
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910681467.XA Pending CN110363190A (en) | 2019-07-26 | 2019-07-26 | A kind of character recognition method, device and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110363190A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110929664A (en) * | 2019-11-29 | 2020-03-27 | 汉海信息技术(上海)有限公司 | Image recognition method and device |
| CN111353502A (en) * | 2020-02-25 | 2020-06-30 | 北京眸视科技有限公司 | Digital table identification method and device and electronic equipment |
| CN111612157A (en) * | 2020-05-22 | 2020-09-01 | 四川无声信息技术有限公司 | Training method, character recognition method, device, storage medium and electronic equipment |
| CN113343997A (en) * | 2021-05-19 | 2021-09-03 | 北京百度网讯科技有限公司 | Optical character recognition method, device, electronic equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101593276A (en) * | 2008-05-29 | 2009-12-02 | 汉王科技股份有限公司 | A kind of video OCR image-text separation method and system |
| JP2014096154A (en) * | 2013-12-03 | 2014-05-22 | Canon Software Inc | Image processing device and method, and program |
| CN107016387A (en) * | 2016-01-28 | 2017-08-04 | 苏宁云商集团股份有限公司 | A kind of method and device for recognizing label |
| CN109086756A (en) * | 2018-06-15 | 2018-12-25 | 众安信息技术服务有限公司 | A kind of text detection analysis method, device and equipment based on deep neural network |
| CN109241894A (en) * | 2018-08-28 | 2019-01-18 | 南京安链数据科技有限公司 | A kind of specific aim ticket contents identifying system and method based on form locating and deep learning |
| CN109284750A (en) * | 2018-08-14 | 2019-01-29 | 北京市商汤科技开发有限公司 | Bank slip recognition method and device, electronic equipment and storage medium |
-
2019
- 2019-07-26 CN CN201910681467.XA patent/CN110363190A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101593276A (en) * | 2008-05-29 | 2009-12-02 | 汉王科技股份有限公司 | A kind of video OCR image-text separation method and system |
| JP2014096154A (en) * | 2013-12-03 | 2014-05-22 | Canon Software Inc | Image processing device and method, and program |
| CN107016387A (en) * | 2016-01-28 | 2017-08-04 | 苏宁云商集团股份有限公司 | A kind of method and device for recognizing label |
| CN109086756A (en) * | 2018-06-15 | 2018-12-25 | 众安信息技术服务有限公司 | A kind of text detection analysis method, device and equipment based on deep neural network |
| CN109284750A (en) * | 2018-08-14 | 2019-01-29 | 北京市商汤科技开发有限公司 | Bank slip recognition method and device, electronic equipment and storage medium |
| CN109241894A (en) * | 2018-08-28 | 2019-01-18 | 南京安链数据科技有限公司 | A kind of specific aim ticket contents identifying system and method based on form locating and deep learning |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110929664A (en) * | 2019-11-29 | 2020-03-27 | 汉海信息技术(上海)有限公司 | Image recognition method and device |
| CN110929664B (en) * | 2019-11-29 | 2023-05-30 | 汉海信息技术(上海)有限公司 | Image recognition method and device |
| CN111353502A (en) * | 2020-02-25 | 2020-06-30 | 北京眸视科技有限公司 | Digital table identification method and device and electronic equipment |
| CN111353502B (en) * | 2020-02-25 | 2023-06-30 | 北京眸视科技有限公司 | Digital table identification method and device and electronic equipment |
| CN111612157A (en) * | 2020-05-22 | 2020-09-01 | 四川无声信息技术有限公司 | Training method, character recognition method, device, storage medium and electronic equipment |
| CN111612157B (en) * | 2020-05-22 | 2023-06-30 | 四川无声信息技术有限公司 | Training method, character recognition device, storage medium and electronic equipment |
| CN113343997A (en) * | 2021-05-19 | 2021-09-03 | 北京百度网讯科技有限公司 | Optical character recognition method, device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110363190A (en) | A kind of character recognition method, device and equipment | |
| US20230177821A1 (en) | Document image understanding | |
| Rong et al. | Recognizing text-based traffic guide panels with cascaded localization network | |
| CN110033018B (en) | Graph similarity judging method and device and computer readable storage medium | |
| CN110348294A (en) | The localization method of chart, device and computer equipment in PDF document | |
| CN111061898A (en) | Image processing method, image processing device, computer equipment and storage medium | |
| CN112699775A (en) | Certificate identification method, device and equipment based on deep learning and storage medium | |
| EP3058513B1 (en) | Multi-color channel detection for note recognition and management | |
| CN111724336B (en) | Image processing apparatus, image processing method, and image processing system | |
| US20230137337A1 (en) | Enhanced machine learning model for joint detection and multi person pose estimation | |
| CN110889437B (en) | Image processing method and device, electronic equipment and storage medium | |
| WO2021237227A1 (en) | Method and system for multi-language text recognition model with autonomous language classification | |
| KR102440198B1 (en) | VIDEO SEARCH METHOD AND APPARATUS, COMPUTER DEVICE, AND STORAGE MEDIUM | |
| CN115937887A (en) | Method and device for extracting document structured information, electronic equipment and storage medium | |
| WO2023038722A1 (en) | Entry detection and recognition for custom forms | |
| CN113780116A (en) | Invoice classification method and device, computer equipment and storage medium | |
| CN110750501A (en) | File retrieval method and device, storage medium and related equipment | |
| CN116225956A (en) | Automated testing method, apparatus, computer device and storage medium | |
| CN113486171B (en) | Image processing method and device and electronic equipment | |
| CN114387600A (en) | Text feature recognition method and device, computer equipment and storage medium | |
| CN113220125A (en) | Finger interaction method and device, electronic equipment and computer storage medium | |
| CN113822143A (en) | Text image processing method, device, equipment and storage medium | |
| KR102026475B1 (en) | Processing visual input | |
| CN113128496A (en) | Method, device and equipment for extracting structured data from image | |
| CN110543238A (en) | Desktop interaction method based on artificial intelligence |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191022 |