CN107679533A - Character recognition method and device - Google Patents
Character recognition method and device Download PDFInfo
- Publication number
- CN107679533A CN107679533A CN201710890221.4A CN201710890221A CN107679533A CN 107679533 A CN107679533 A CN 107679533A CN 201710890221 A CN201710890221 A CN 201710890221A CN 107679533 A CN107679533 A CN 107679533A
- Authority
- CN
- China
- Prior art keywords
- character image
- network
- distortion
- layer
- image
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/243—Aligning, centring, orientation detection or correction of the image by compensating for image skew or non-uniform image deformations
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Character Discrimination (AREA)
Abstract
The disclosure is directed to a kind of character recognition method and device.This method includes：By character image to be identified input character image generation network, the first reparation character image is obtained；Wherein, the character image generation network trains to obtain by not distorting character image and distortion character image；Character image, which is repaired, according to described first carries out Text region.The character recognition method and device of the disclosure, character image to be identified can be repaired, obtain repairing character image, and be identified to repairing the text information in character image, the accuracy rate thus, it is possible to greatly improve Text region.
Description
Technical field
This disclosure relates to image identification technical field, more particularly to a kind of character recognition method and device.
Background technology
In correlation technique, Text region refers to word in image is identified, verified and recorded using computer etc.
The technology of reason.People will handle substantial amounts of word, form and text in production and life, and character recognition technology can be significantly
Mitigate the work of people.At present, it is relatively low for the Text region accuracy rate in distortion character image, improve in distortion character image
Text region accuracy rate it is significant.
The content of the invention
To overcome problem present in correlation technique, the disclosure provides a kind of character recognition method and device.
According to the first aspect of the embodiment of the present disclosure, there is provided a kind of character recognition method, including：
By character image to be identified input character image generation network, the first reparation character image is obtained；Wherein, the text
Word image generates network and trains to obtain by not distorting character image and distortion character image；
Character image, which is repaired, according to described first carries out Text region.
In a kind of possible implementation, methods described also includes：
Distortion processing is carried out to the character image that do not distort, obtains the distortion character image；
Character image and the distortion character image are not distorted according to described, and training differentiates network and generation network, described
Differentiate that network is used to differentiate the reparation character image and the uniformity for not distorting character image；
Network and the differentiation network are generated described in repetition training, reaches predetermined threshold value or the differentiation net in frequency of training
, will when the differentiation result of network shows the consistent sexual satisfaction preparatory condition that the reparation character image does not distort character image with described
Current generation network is defined as the character image generation network.
In a kind of possible implementation, character image and the distortion character image, training are not distorted according to described
Differentiate network and generation network, including：
The distortion character image is inputted into the generation network, obtains the second reparation character image；
The character image and described second that do not distort is repaired into the character image input differentiation network, obtains being used for table
Show that the second reparation character image does not distort the whether consistent differentiation result of character image with described；
According to the differentiation result, the value for differentiating parameter in network or the generation network is adjusted.
In a kind of possible implementation, the generation network includes the multiple coding modules connected by residual error mode
With multiple decoder modules, the coding module includes convolutional layer, line rectification function layer and maximum pond layer, the decoder module
Including convolutional layer, line rectification function layer and maximum pond layer.
It is described to differentiate that network includes multiple coding modules, the Duo Gequan being sequentially connected in a kind of possible implementation
Articulamentum and threshold function table layer, the coding module include convolutional layer, line rectification function layer and maximum pond layer.
According to the second aspect of the embodiment of the present disclosure, there is provided a kind of character recognition device, including：
Repair module, for character image to be identified input character image to be generated into network, obtain the first reparation word graph
Picture；Wherein, the character image generation network trains to obtain by not distorting character image and distortion character image；
Identification module, Text region is carried out for repairing character image according to described first.
In a kind of possible implementation, described device also includes：
Processing module, for carrying out distortion processing to the character image that do not distort, obtain the distortion character image；
Training module, for not distorting character image and the distortion character image according to described, training differentiate network with
Network is generated, the differentiation network is used to differentiate the reparation character image and the uniformity for not distorting character image；
Determining module, for generating network and the differentiation network described in repetition training, reach default threshold in frequency of training
Value or the differentiation result for differentiating network show that the reparation character image and the uniformity for not distorting character image expire
During sufficient preparatory condition, current generation network is defined as the character image and generates network.
In a kind of possible implementation, the training module is used for：
The distortion character image is inputted into the generation network, obtains the second reparation character image；
The character image and described second that do not distort is repaired into the character image input differentiation network, obtains being used for table
Show that the second reparation character image does not distort the whether consistent differentiation result of character image with described；
According to the differentiation result, the value for differentiating parameter in network or the generation network is adjusted.
In a kind of possible implementation, the generation network includes the multiple coding modules connected by residual error mode
With multiple decoder modules, the coding module includes convolutional layer, line rectification function layer and maximum pond layer, the decoder module
Including convolutional layer, line rectification function layer and maximum pond layer.
It is described to differentiate that network includes multiple coding modules, the Duo Gequan being sequentially connected in a kind of possible implementation
Articulamentum and threshold function table layer, the coding module include convolutional layer, line rectification function layer and maximum pond layer.
According to the third aspect of the embodiment of the present disclosure, there is provided a kind of character recognition device, including：Processor；For storing
The memory of processor-executable instruction；Wherein, the processor is configured as performing above-mentioned method.
According to the fourth aspect of the embodiment of the present disclosure, there is provided a kind of non-volatile computer readable storage medium storing program for executing, deposit thereon
Computer program instructions are contained, the computer program instructions realize above-mentioned method when being executed by processor.
The technical scheme provided by this disclosed embodiment can include the following benefits：The character recognition method of the disclosure
And device, network is generated by the way that character image to be identified is inputted into character image, the first reparation character image is obtained, according to first
Repair character image and carry out Text region, wherein, character image generates network by not distorting character image and distortion word graph
Obtained as training, thus, it is possible to repair character image to be identified, obtain repairing character image, and to repairing character image
In text information be identified, the accuracy rate thus, it is possible to greatly improve Text region.
It should be appreciated that the general description and following detailed description of the above are only exemplary and explanatory, not
The disclosure can be limited.
Brief description of the drawings
Accompanying drawing herein is merged in specification and forms the part of this specification, shows the implementation for meeting the disclosure
Example, and be used to together with specification to explain the principle of the disclosure.
Fig. 1 is a kind of flow chart of character recognition method according to an exemplary embodiment.
Fig. 2 is a kind of flow chart of character recognition method according to an exemplary embodiment.
Fig. 3 is the schematic block diagram of the generation network according to an exemplary embodiment.
Fig. 4 is the schematic block diagram of the differentiation network according to an exemplary embodiment.
Fig. 5 is a kind of block diagram of character recognition device according to an exemplary embodiment.
Fig. 6 is an a kind of schematical block diagram of character recognition device according to an exemplary embodiment.
Fig. 7 is a kind of block diagram of device 800 for Text region according to an exemplary embodiment.
Embodiment
Here exemplary embodiment will be illustrated in detail, its example is illustrated in the accompanying drawings.Following description is related to
During accompanying drawing, unless otherwise indicated, the same numbers in different accompanying drawings represent same or analogous key element.Following exemplary embodiment
Described in embodiment do not represent all embodiments consistent with the disclosure.On the contrary, they be only with it is such as appended
The example of the consistent apparatus and method of some aspects be described in detail in claims, the disclosure.
Fig. 1 is a kind of flow chart of character recognition method according to an exemplary embodiment.This method is used for word
Identification equipment, the disclosure are without limitation.As shown in figure 1, the method comprising the steps of S11 and step S12.
In step s 11, by character image to be identified input character image generation network, the first reparation word graph is obtained
Picture；Wherein, character image generation network trains to obtain by not distorting character image and distortion character image.
Wherein, the input of character image generation network can be character image to be identified, and output can be the first reparation text
Word image.First repairs character image repairs character image corresponding to character image to be identified.First repair character image with
Character image to be identified is of the same size and resolution ratio.Character image to be identified can be the character image that word is twisted
Or the character image that word is not twisted or is not almost twisted, the disclosure are without limitation.
In step s 12, repair character image according to first and carry out Text region.
Wherein, the word in the first reparation character image can include Chinese character, English character, numerical character and symbol
One or more in character, the first word for repairing in character image can be one in handwritten text and print hand writing
Item is multinomial, and the disclosure is without limitation.
It should be noted that it will be appreciated by those skilled in the art that there are various ways to realize basis in correlation technique
First, which repairs character image, carries out Text region, such as template matching method, structured analysis method or feature extraction etc., the disclosure pair
This is not limited.
The character recognition method of the disclosure, character image to be identified can be repaired, obtain repairing character image, and
It is identified to repairing the text information in character image, the accuracy rate thus, it is possible to greatly improve Text region.
Fig. 2 is a kind of flow chart of character recognition method according to an exemplary embodiment.As shown in Fig. 2 the party
Method includes step S21 to step S25.
In the step s 21, distortion processing is carried out to not distorting character image, obtains distorting character image.
Wherein, the character image that word is not twisted or is not almost twisted, distortion text can be referred to by not distorting character image
Word image can refer to the character image that word is twisted.
In a kind of possible implementation, selection does not distort character image, and what random distortion was chosen does not distort word graph
Picture, obtain distorting character image.
In step S22, according to character image and distortion character image is not distorted, training differentiates network and generation network,
Differentiate that network is used to differentiate the uniformity repaired character image and do not distort character image.
It should be noted that it will be appreciated by those skilled in the art that character image and distortion text are not distorted in step S21
Word image, which is used to train, differentiates network and generation network.During hands-on, character image and torsion are not distorted for every group
Bent character image, alternately training differentiate network and generation network.Furthermore, it is necessary to obtain it is multigroup it is different do not distort character image and
Character image is distorted, repetition training differentiates network and generation network, and the steady of network is generated with the character image that enhancing training obtains
Qualitative and adaptability.
Wherein, alternately training differentiates that network and generation network can not distort character image and distortion word with pointer to every group
Image, in the case where keeping the parameter constant of generation network, training differentiates network, then differentiates the parameter of network not in holding
In the case of change, training generation network.Alternately training differentiates network and generation network, until generation network is according to distortion word graph
As generation second repair character image, differentiate network can not differentiate do not distort character image and second repair character image whether one
Cause, such as differentiate network output 0.5, i.e., the second reparation character image has 50% probability consistent with not distorting character image, has
50% probability is inconsistent with not distorting character image.
In step S23, repetition training generation network and differentiation network, reach predetermined threshold value in frequency of training or differentiate net
When the differentiation result of network shows to repair character image with the consistent sexual satisfaction preparatory condition for not distorting character image, by current life
It is defined as character image generation network into network.
In a kind of possible implementation, generation network includes the multiple coding modules connected by residual error mode
(Encode) and multiple decoder modules (Decode), coding module include convolutional layer, line rectification function (ReLu, Rectified
Linear Unit) layer and maximum pond layer (Max Pooling), decoder module includes convolutional layer, line rectification function layer and most
Great Chiization layer.
Wherein, coding module is used to encode image.Decoder module is used for encoding what is obtained by coding module
Image is decoded.Coding module and decoder module can change image resolution ratio and image channel number, such as increase image
Resolution ratio simultaneously reduces image channel number, or reduces image resolution ratio and increase image channel number.Convolutional layer, line rectification function layer
It is a basic processing unit in coding module and decoder module with maximum pond layer.
Fig. 3 is the schematic block diagram of the generation network according to an exemplary embodiment.As shown in figure 3, generation network is
One ten layers of coding module-decoder module structure, including 5 coding modules connected by residual error mode and 5 decoding moulds
Block.Each coding module and each decoder module include 1 convolutional layer, 1 line rectification function layer and 1 maximum pond
Layer.Wherein, coding module be respectively Encode1 (n*32*3), Encode2 (n/2*16*64), Encode3 (n/4*8*128),
Encode4 (n/8*4*256) and Encode5 (16/n*2*512).Decoder module be respectively Decode1 (16/n*2*512),
Decode2 (n/8*4*256), Decode3 (n/4*8*128), Decode4 (n/2*16*64) and Decode5 (n*32*3).Can
With understanding, the n*32 in n*32*3 can represent image resolution ratio, and 32 can represent short side resolution ratio in image, and 3 can be with
Represent image channel number.
It is 32 by short side resolution adjustment in character image to be identified, long side is pressed as an example of the implementation
According to proportional zoom, the first character image is obtained.First character image is inputted into generation network.First character image is by generation net
Coding module in network is encoded, then is decoded by the decoder module in generation network, obtains the first reparation word graph
Picture.
In a kind of possible implementation, multiple coding modules, multiple full connections that network includes being sequentially connected are differentiated
(FC, Fully Connected Layers) layer and threshold function table (Sigmoid) layer, coding module include convolutional layer, linear whole
Stream function layer and maximum pond layer.
Wherein, coding module is used to encode image.Full articulamentum is used to represent the distributed nature learnt
It is mapped to sample labeling space.Threshold function table layer is used for variable mappings between [0,1].Coding module can change image point
Resolution and image channel number, such as reduce image resolution ratio and increase image channel number.Convolutional layer, line rectification function layer and most
Great Chiization layer is a basic processing unit in coding module.
Fig. 4 is the schematic block diagram of the differentiation network according to an exemplary embodiment.As shown in figure 4, differentiate network bag
Include 5 coding modules being sequentially connected, 2 full articulamentums and 1 threshold function table layer.Each coding module include 1 convolutional layer,
1 line rectification function layer and 1 maximum pond layer.Wherein, coding module is respectively Encode1 ' (n*32*6), Encode2 '
(n/2*16*64), Encode3 ' (n/4*8*128), Encode4 ' (n/8*4*256) and Encode5 ' (16/n*2*512).Can
With understanding, the n*32 in n*32*3 can represent image resolution ratio, and 32 can represent the resolution ratio of short side in image, and 3 can
To represent image channel number.
In a kind of possible implementation, according to character image and distortion character image is not distorted, training differentiates network
It can include with generation network (step S22)：Distortion character image input is generated into network, obtains the second reparation character image；
To not distort character image and second repair character image input differentiate network, obtain be used for represent second repair character image with
The whether consistent differentiation result of character image is not distorted；According to result is differentiated, adjustment differentiates network or generates parameter in network
Value.
Wherein, the value for generating parameter in network can refer to the convolution that each coding module and decoder module include in Fig. 3
The value of parameter in layer, line rectification function layer and maximum pond layer.The value of parameter can refer to each in Fig. 4 in differentiation network
Convolutional layer, line rectification function layer and the maximum pond layer and each full articulamentum and threshold function table layer that coding module includes
The value of middle parameter.
In a kind of possible implementation, the input for differentiating network can be not distort character image and second to repair text
The fused images that word image obtains after being merged, output can be for representing that second repairs whether character image is not distort
The differentiation result of character image.For example, image channel number corresponding to not distorting character image and the second reparation character image is
3, it will not distort character image and the second reparation character image is merged, and obtain the fused images that image channel number is 6, will melt
Image is closed as the input for differentiating network.
In a kind of possible implementation, alternating training differentiates network and generation network, and is used according to differentiation result
Back-propagation algorithm adjustment differentiates network and generates the value of parameter in network, until differentiating that network and generation network are all restrained.
Wherein, differentiate that network and generation network are all restrained to refer to and differentiate that result is in stable state or frequency of training reaches default threshold
Value.
In a kind of possible implementation, generation network G is determined using formula 1；
Wherein, G represents generation network, and D represents to differentiate network,Represent loss result corresponding to generation network, x
Expression does not distort character image, and D (x) represents the differentiation result that x obtains as input,Represent to differentiate and damaged corresponding to network
Result is lost, z represents distortion character image, and G (z) represents the generation result that z obtains as input, i.e. G (z) represents that second repairs text
Word image, the differentiation result that D (G (z)) expression G (z) obtain as input, E [| | x-G (z) | |1Expression does not distort character image
With the smooth loss (Smooth L1Loss) of the second difference for repairing character image.
It is understood that generation network G is the network for generating image, it receives random noise z, passes through noise
Z generation image G (z).It is the network for differentiation to differentiate network D, and output differentiates result.Differentiate that result represents that input differentiates network
Image whether be true picture probability.Differentiate that result is 1 and represents that input differentiates that the image 100% of network is truly to scheme
Picture, differentiate that result is 0 and represents that input differentiates that the image of network is unlikely to be true picture.
During hands-on, the target for generating network G is just to try to generate true picture and go to cheat to differentiate network D.
And differentiate network D target and be just to try to the image and true picture of the generation of generation network G to be distinguished from.Thus network is generated
G and differentiation network D constitute a dynamic gambling process.Under optimal state, the result of last game is made a living networking
Network G can generate the image G (z) for being enough to mix the spurious with the genuine, and differentiate that network D is difficult to differentiate that the image of generation network G generation is actually
It is not true, therefore D (G (z))=0.5.
It should be noted that those skilled in the art, it should be appreciated that x represents true picture, z represents input generation net
Network G noise, and G (z) represents the image of generation network G generation.D (x) represents to differentiate that network D judges whether true picture x is true
Real probability.D (G (z)) represents to differentiate that network D judges to generate the whether real probability of image of network G generation.Because x is exactly
True picture, so for differentiating network D, D (x) is better closer to 1.And G (z) is to generate the image that network G generates, institute
So that for differentiating network D, D (x) is better closer to 0.Generate the purpose of network G：D (G (z)) is to differentiate network D
Judge to generate the whether real probability of image that network G generates, generation network G should wish the image of oneself generation
Closer to true better.That is, generation network G wish D (G (z)) as far as possible greatly, at this momentValue can diminish.Therefore the foremost of formula 1
Mark beDifferentiate network D purpose：Differentiate that network D ability is stronger, D (x) should be bigger, and D (G (z)) should be got over
It is small, at this momentValue can become big.Therefore formula 1
The mark of foremost be
In step s 24, by character image to be identified input character image generation network, the first reparation word graph is obtained
Picture.
Description for the step may refer to step S11.
In step s 25, repair character image according to first and carry out Text region.
Description for the step may refer to step S12.
It should be noted that it will be appreciated by those skilled in the art that step S21 to step S23, which is training, obtains word graph
As the process of generation network, step S24 to step S25 is the process that actual use character image generates network.Training process is
Abnormal process, the use of process is normality process.
In a kind of possible implementation, the character image that encapsulation training obtains in Text region equipment generates net
Network, to cause character image generation network handles identification character image may be reused in Text region equipment to be repaired,
Obtain repairing character image, and be identified to repairing the character information in character image, known thus, it is possible to greatly improve word
Other accuracy rate.
The character recognition method of the disclosure, using production confrontation network generate the training of network so that generation net
Network has preferable repair ability to distortion character image, can repair to obtain and not distort that character image is same or analogous to be repaiied
Multiple character image, the accuracy rate thus, it is possible to greatly improve Text region.
Fig. 5 is a kind of block diagram of character recognition device according to an exemplary embodiment.Reference picture 5, the device bag
Include：Repair module 51, for character image to be identified input character image to be generated into network, obtain the first reparation character image；
Wherein, the character image generation network trains to obtain by not distorting character image and distortion character image；Identification module 52,
Text region is carried out for repairing character image according to described first.
Fig. 6 is an a kind of schematical block diagram of character recognition device according to an exemplary embodiment.Reference picture
6：
In a kind of possible implementation, described device also includes：Processing module 53, for not distorting word to described
Image carries out distortion processing, obtains the distortion character image；Training module 54, for according to it is described do not distort character image with
The distortion character image, training differentiate network and generation network, and the differentiation network is used to differentiate the reparation character image
With the uniformity for not distorting character image；Determining module 55, for generating network and the differentiation net described in repetition training
Network, frequency of training reach predetermined threshold value or it is described differentiate network differentiation result show it is described reparation character image and it is described not
When distorting the consistent sexual satisfaction preparatory condition of character image, current generation network is defined as the character image and generates net
Network.
In a kind of possible implementation, the training module 54 is used for：By described in the distortion character image input
Network is generated, obtains the second reparation character image；The character image and described second that do not distort is repaired into character image input
The differentiation network, obtain for representing that the second reparation character image does not distort whether character image is consistent to be sentenced with described
Other result；According to the differentiation result, the value for differentiating parameter in network or the generation network is adjusted.
In a kind of possible implementation, the generation network includes the multiple coding modules connected by residual error mode
With multiple decoder modules, the coding module includes convolutional layer, line rectification function layer and maximum pond layer, the decoder module
Including convolutional layer, line rectification function layer and maximum pond layer.
It is described to differentiate that network includes multiple coding modules, the Duo Gequan being sequentially connected in a kind of possible implementation
Articulamentum and threshold function table layer, the coding module include convolutional layer, line rectification function layer and maximum pond layer.
On the device in above-described embodiment, wherein modules perform the concrete mode of operation in relevant this method
Embodiment in be described in detail, explanation will be not set forth in detail herein.
The character recognition device of the disclosure, character image to be identified can be repaired, obtain repairing character image, and
It is identified to repairing the text information in character image, the accuracy rate thus, it is possible to greatly improve Text region.
Fig. 7 is a kind of block diagram of device 800 for Text region according to an exemplary embodiment.For example, dress
It can be mobile phone to put 800, computer, digital broadcast terminal, messaging devices, game console, tablet device, medical treatment
The equipment that equipment, body-building equipment, personal digital assistant etc. have character identification function.
Reference picture 7, device 800 can include following one or more assemblies：Processing component 802, memory 804, power supply
Component 806, multimedia groupware 808, audio-frequency assembly 810, the interface 812 of input/output (I/O), sensor cluster 814, and
Communication component 816.
The integrated operation of the usual control device 800 of processing component 802, such as communicated with display, call, data, phase
The operation that machine operates and record operation is associated.Processing component 802 can refer to including one or more processors 820 to perform
Order, to complete all or part of step of above-mentioned method.In addition, processing component 802 can include one or more modules, just
Interaction between processing component 802 and other assemblies.For example, processing component 802 can include multi-media module, it is more to facilitate
Interaction between media component 808 and processing component 802.
Memory 804 is configured as storing various types of data to support the operation in device 800.These data are shown
Example includes the instruction of any application program or method for being operated on device 800, contact data, telephone book data, disappears
Breath, picture, video etc..Memory 804 can be by any kind of volatibility or non-volatile memory device or their group
Close and realize, as static RAM (SRAM), Electrically Erasable Read Only Memory (EEPROM) are erasable to compile
Journey read-only storage (EPROM), programmable read only memory (PROM), read-only storage (ROM), magnetic memory, flash
Device, disk or CD.
Power supply module 806 provides electric power for the various assemblies of device 800.Power supply module 806 can include power management system
System, one or more power supplys, and other components associated with generating, managing and distributing electric power for device 800.
Multimedia groupware 808 is included in the screen of one output interface of offer between described device 800 and user.One
In a little embodiments, screen can include liquid crystal display (LCD) and touch panel (TP).If screen includes touch panel, screen
Curtain may be implemented as touch-screen, to receive the input signal from user.Touch panel includes one or more touch sensings
Device is with the gesture on sensing touch, slip and touch panel.The touch sensor can not only sensing touch or sliding action
Border, but also detect and touched or the related duration and pressure of slide with described.In certain embodiments, more matchmakers
Body component 808 includes a front camera and/or rear camera.When device 800 is in operator scheme, such as screening-mode or
During video mode, front camera and/or rear camera can receive outside multi-medium data.Each front camera and
Rear camera can be a fixed optical lens system or have focusing and optical zoom capabilities.
Audio-frequency assembly 810 is configured as output and/or input audio signal.For example, audio-frequency assembly 810 includes a Mike
Wind (MIC), when device 800 is in operator scheme, during such as call model, logging mode and speech recognition mode, microphone by with
It is set to reception external audio signal.The audio signal received can be further stored in memory 804 or via communication set
Part 816 is sent.In certain embodiments, audio-frequency assembly 810 also includes a loudspeaker, for exports audio signal.
I/O interfaces 812 provide interface between processing component 802 and peripheral interface module, and above-mentioned peripheral interface module can
To be keyboard, click wheel, button etc..These buttons may include but be not limited to：Home button, volume button, start button and lock
Determine button.
Sensor cluster 814 includes one or more sensors, and the state for providing various aspects for device 800 is commented
Estimate.For example, sensor cluster 814 can detect opening/closed mode of device 800, and the relative positioning of component, for example, it is described
Component is the display and keypad of device 800, and sensor cluster 814 can be with 800 1 components of detection means 800 or device
Position change, the existence or non-existence that user contacts with device 800, the orientation of device 800 or acceleration/deceleration and device 800
Temperature change.Sensor cluster 814 can include proximity transducer, be configured to detect in no any physical contact
The presence of neighbouring object.Sensor cluster 814 can also include optical sensor, such as CMOS or ccd image sensor, for into
As being used in application.In certain embodiments, the sensor cluster 814 can also include acceleration transducer, gyro sensors
Device, Magnetic Sensor, pressure sensor or temperature sensor.
Communication component 816 is configured to facilitate the communication of wired or wireless way between device 800 and other equipment.Device
800 can access the wireless network based on communication standard, such as WiFi, 2G or 3G, or combinations thereof.In an exemplary implementation
In example, communication component 816 receives broadcast singal or broadcast related information from external broadcasting management system via broadcast channel.
In one exemplary embodiment, the communication component 816 also includes near-field communication (NFC) module, to promote junction service.Example
Such as, in NFC module radio frequency identification (RFID) technology can be based on, Infrared Data Association (IrDA) technology, ultra wide band (UWB) technology,
Bluetooth (BT) technology and other technologies are realized.
In the exemplary embodiment, device 800 can be believed by one or more application specific integrated circuits (ASIC), numeral
Number processor (DSP), digital signal processing appts (DSPD), PLD (PLD), field programmable gate array
(FPGA), controller, microcontroller, microprocessor or other electronic components are realized, for performing the above method.
In the exemplary embodiment, a kind of non-transitorycomputer readable storage medium including instructing, example are additionally provided
Such as include the memory 804 of instruction, above-mentioned instruction can be performed to complete the above method by the processor 820 of device 800.For example,
The non-transitorycomputer readable storage medium can be ROM, random access memory (RAM), CD-ROM, tape, floppy disk
With optical data storage devices etc..
Those skilled in the art will readily occur to the disclosure its after considering specification and putting into practice invention disclosed herein
Its embodiment.The application is intended to any modification, purposes or the adaptations of the disclosure, these modifications, purposes or
Person's adaptations follow the general principle of the disclosure and including the undocumented common knowledges in the art of the disclosure
Or conventional techniques.Description and embodiments are considered only as exemplary, and the true scope of the disclosure and spirit are by following
Claim is pointed out.
It should be appreciated that the precision architecture that the disclosure is not limited to be described above and is shown in the drawings, and
And various modifications and changes can be being carried out without departing from the scope.The scope of the present disclosure is only limited by appended claim.
Claims (12)
- A kind of 1. character recognition method, it is characterised in that including：By character image to be identified input character image generation network, the first reparation character image is obtained；Wherein, the word graph Train to obtain by not distorting character image and distortion character image as generating network；Character image, which is repaired, according to described first carries out Text region.
- 2. according to the method for claim 1, it is characterised in that methods described also includes：Distortion processing is carried out to the character image that do not distort, obtains the distortion character image；Character image and the distortion character image are not distorted according to described, and training differentiates network and generation network, the differentiation Network is used to differentiate the reparation character image and the uniformity for not distorting character image；Network and the differentiation network are generated described in repetition training, reaches predetermined threshold value in frequency of training or described differentiates network , will be current when differentiating that result shows the consistent sexual satisfaction preparatory condition that the reparation character image does not distort character image with described Generation network be defined as character image generation network.
- 3. according to the method for claim 2, it is characterised in that do not distort character image and the distortion word according to described Image, training differentiate network and generation network, including：The distortion character image is inputted into the generation network, obtains the second reparation character image；The character image and described second that do not distort is repaired into the character image input differentiation network, obtains being used to represent institute State the second reparation character image and do not distort the whether consistent differentiation result of character image with described；According to the differentiation result, the value for differentiating parameter in network or the generation network is adjusted.
- 4. according to the method for claim 1, it is characterised in that the generation network is more including being connected by residual error mode Individual coding module and multiple decoder modules, the coding module include convolutional layer, line rectification function layer and maximum pond layer, institute Stating decoder module includes convolutional layer, line rectification function layer and maximum pond layer.
- 5. according to the method for claim 2, it is characterised in that described to differentiate that network includes the multiple coding moulds being sequentially connected Block, multiple full articulamentums and threshold function table layer, the coding module include convolutional layer, line rectification function layer and maximum pond Layer.
- A kind of 6. character recognition device, it is characterised in that including：Repair module, for character image to be identified input character image to be generated into network, obtain the first reparation character image；Its In, the character image generation network trains to obtain by not distorting character image and distortion character image；Identification module, Text region is carried out for repairing character image according to described first.
- 7. device according to claim 6, it is characterised in that described device also includes：Processing module, for carrying out distortion processing to the character image that do not distort, obtain the distortion character image；Training module, for not distorting character image and the distortion character image according to, training differentiates network and generation Network, the differentiation network are used to differentiate the reparation character image and the uniformity for not distorting character image；Determining module, for generating network and the differentiation network described in repetition training, frequency of training reach predetermined threshold value or The differentiation result for differentiating network shows that the reparation character image and the consistent sexual satisfaction for not distorting character image are pre- If during condition, current generation network is defined as the character image and generates network.
- 8. device according to claim 7, it is characterised in that the training module is used for：The distortion character image is inputted into the generation network, obtains the second reparation character image；The character image and described second that do not distort is repaired into the character image input differentiation network, obtains being used to represent institute State the second reparation character image and do not distort the whether consistent differentiation result of character image with described；According to the differentiation result, the value for differentiating parameter in network or the generation network is adjusted.
- 9. device according to claim 6, it is characterised in that the generation network is more including being connected by residual error mode Individual coding module and multiple decoder modules, the coding module include convolutional layer, line rectification function layer and maximum pond layer, institute Stating decoder module includes convolutional layer, line rectification function layer and maximum pond layer.
- 10. device according to claim 7, it is characterised in that described to differentiate that network includes the multiple codings being sequentially connected Module, multiple full articulamentums and threshold function table layer, the coding module include convolutional layer, line rectification function layer and maximum pond Layer.
- A kind of 11. character recognition device, it is characterised in that including：Processor；For storing the memory of processor-executable instruction；Wherein, the processor is configured as the method described in any one in perform claim requirement 1 to 5.
- 12. a kind of non-volatile computer readable storage medium storing program for executing, is stored thereon with computer program instructions, it is characterised in that institute State and method in claim 1 to 5 described in any one is realized when computer program instructions are executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710890221.4A CN107679533A (en) | 2017-09-27 | 2017-09-27 | Character recognition method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710890221.4A CN107679533A (en) | 2017-09-27 | 2017-09-27 | Character recognition method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107679533A true CN107679533A (en) | 2018-02-09 |
Family
ID=61137533
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710890221.4A Pending CN107679533A (en) | 2017-09-27 | 2017-09-27 | Character recognition method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107679533A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109871841A (en) * | 2019-02-14 | 2019-06-11 | 腾讯科技（深圳）有限公司 | Image processing method, device, terminal and storage medium |
| CN110084172A (en) * | 2019-04-23 | 2019-08-02 | 北京字节跳动网络技术有限公司 | Character recognition method, device and electronic equipment |
| CN110363189A (en) * | 2018-04-09 | 2019-10-22 | 珠海金山办公软件有限公司 | A kind of document content restorative procedure, device, electronic equipment and readable storage medium storing program for executing |
| CN110889469A (en) * | 2019-09-19 | 2020-03-17 | 北京市商汤科技开发有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN111402156A (en) * | 2020-03-11 | 2020-07-10 | 腾讯科技（深圳）有限公司 | Restoration method and device for smear image, storage medium and terminal equipment |
| CN111723788A (en) * | 2019-03-19 | 2020-09-29 | 杭州海康威视数字技术股份有限公司 | Character recognition method and device |
| CN111783756A (en) * | 2019-04-03 | 2020-10-16 | 北京市商汤科技开发有限公司 | Text recognition method and device, electronic equipment and storage medium |
| WO2022047662A1 (en) * | 2020-09-02 | 2022-03-10 | Intel Corporation | Method and system of neural network object recognition for warpable jerseys with multiple attributes |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101661613A (en) * | 2009-08-27 | 2010-03-03 | 北京交通大学 | Image restoration method based on image segmentation, and system therefor |
| CN102063621A (en) * | 2010-11-30 | 2011-05-18 | 汉王科技股份有限公司 | Method and device for correcting geometric distortion of character lines |
| CN103413271A (en) * | 2013-07-18 | 2013-11-27 | 西安交通大学 | Document image rectifying method based on local information |
| CN104809733A (en) * | 2015-05-08 | 2015-07-29 | 中北大学 | Ancient building wall polluted inscription character image edge extraction method |
| CN104899571A (en) * | 2015-06-12 | 2015-09-09 | 成都数联铭品科技有限公司 | Random sample generation method for recognition of complex character |
| CN106648418A (en) * | 2016-11-14 | 2017-05-10 | 北京爱知之星科技股份有限公司 | Handwriting correction method and device |
| CN107133934A (en) * | 2017-05-18 | 2017-09-05 | 北京小米移动软件有限公司 | Image completion method and device |
-
2017
- 2017-09-27 CN CN201710890221.4A patent/CN107679533A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101661613A (en) * | 2009-08-27 | 2010-03-03 | 北京交通大学 | Image restoration method based on image segmentation, and system therefor |
| CN102063621A (en) * | 2010-11-30 | 2011-05-18 | 汉王科技股份有限公司 | Method and device for correcting geometric distortion of character lines |
| CN103413271A (en) * | 2013-07-18 | 2013-11-27 | 西安交通大学 | Document image rectifying method based on local information |
| CN104809733A (en) * | 2015-05-08 | 2015-07-29 | 中北大学 | Ancient building wall polluted inscription character image edge extraction method |
| CN104899571A (en) * | 2015-06-12 | 2015-09-09 | 成都数联铭品科技有限公司 | Random sample generation method for recognition of complex character |
| CN106648418A (en) * | 2016-11-14 | 2017-05-10 | 北京爱知之星科技股份有限公司 | Handwriting correction method and device |
| CN107133934A (en) * | 2017-05-18 | 2017-09-05 | 北京小米移动软件有限公司 | Image completion method and device |
Non-Patent Citations (1)
| Title |
|---|
| RAYMOND A. YEH: "Semantic Image Inpainting with Deep Generative Models", 《CVPR2017》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110363189A (en) * | 2018-04-09 | 2019-10-22 | 珠海金山办公软件有限公司 | A kind of document content restorative procedure, device, electronic equipment and readable storage medium storing program for executing |
| CN110363189B (en) * | 2018-04-09 | 2021-09-24 | 珠海金山办公软件有限公司 | Document content restoration method and device, electronic equipment and readable storage medium |
| CN109871841B (en) * | 2019-02-14 | 2023-04-21 | 腾讯科技（深圳）有限公司 | Image processing method, device, terminal and storage medium |
| CN109871841A (en) * | 2019-02-14 | 2019-06-11 | 腾讯科技（深圳）有限公司 | Image processing method, device, terminal and storage medium |
| CN111723788A (en) * | 2019-03-19 | 2020-09-29 | 杭州海康威视数字技术股份有限公司 | Character recognition method and device |
| CN111723788B (en) * | 2019-03-19 | 2023-12-29 | 杭州海康威视数字技术股份有限公司 | Character recognition method and device |
| CN111783756B (en) * | 2019-04-03 | 2024-04-16 | 北京市商汤科技开发有限公司 | Text recognition method and device, electronic equipment and storage medium |
| CN111783756A (en) * | 2019-04-03 | 2020-10-16 | 北京市商汤科技开发有限公司 | Text recognition method and device, electronic equipment and storage medium |
| CN110084172A (en) * | 2019-04-23 | 2019-08-02 | 北京字节跳动网络技术有限公司 | Character recognition method, device and electronic equipment |
| CN110889469A (en) * | 2019-09-19 | 2020-03-17 | 北京市商汤科技开发有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN110889469B (en) * | 2019-09-19 | 2023-07-21 | 北京市商汤科技开发有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN111402156B (en) * | 2020-03-11 | 2021-08-03 | 腾讯科技（深圳）有限公司 | Restoration method and device for smear image, storage medium and terminal equipment |
| CN111402156A (en) * | 2020-03-11 | 2020-07-10 | 腾讯科技（深圳）有限公司 | Restoration method and device for smear image, storage medium and terminal equipment |
| WO2022047662A1 (en) * | 2020-09-02 | 2022-03-10 | Intel Corporation | Method and system of neural network object recognition for warpable jerseys with multiple attributes |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107679533A (en) | Character recognition method and device | |
| CN107609560A (en) | Character recognition method and device | |
| CN107679483A (en) | Number plate recognition methods and device | |
| CN107527059B (en) | Character recognition method and device and terminal | |
| CN107798327A (en) | Character identifying method and device | |
| CN106530227A (en) | Image restoration method and device | |
| CN107845062A (en) | image generating method and device | |
| CN106651955A (en) | Method and device for positioning object in picture | |
| CN107993210A (en) | Image repair method, device and computer-readable recording medium | |
| CN106792170A (en) | Method for processing video frequency and device | |
| CN107944447A (en) | Image classification method and device | |
| CN105426857A (en) | Training method and device of face recognition model | |
| CN105335754A (en) | Character recognition method and device | |
| CN105335713A (en) | Fingerprint identification method and device | |
| CN110096209A (en) | Handwriting trace display methods and device | |
| TW201234213A (en) | Multimedia input method | |
| CN106682736A (en) | Image identification method and apparatus | |
| CN108737891A (en) | Video material processing method and processing device | |
| CN106778531A (en) | Face detection method and device | |
| CN107230137A (en) | Merchandise news acquisition methods and device | |
| CN105354560A (en) | Fingerprint identification method and device | |
| CN109819288A (en) | Determination method, apparatus, electronic equipment and the storage medium of advertisement dispensing video | |
| CN108345581A (en) | A kind of information identifying method, device and terminal device | |
| CN107038428A (en) | Vivo identification method and device | |
| CN110717399A (en) | Face recognition method and electronic terminal equipment |
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 |