CN106488313A - A kind of TV station symbol recognition method and system - Google Patents
A kind of TV station symbol recognition method and system Download PDFInfo
- Publication number
- CN106488313A CN106488313A CN201610931021.4A CN201610931021A CN106488313A CN 106488313 A CN106488313 A CN 106488313A CN 201610931021 A CN201610931021 A CN 201610931021A CN 106488313 A CN106488313 A CN 106488313A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- station symbol
- video
- array
- 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
- 238000000034 method Methods 0.000 title claims abstract description 47
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 204
- 238000012549 training Methods 0.000 claims abstract description 19
- 230000007935 neutral effect Effects 0.000 claims description 9
- 238000004891 communication Methods 0.000 claims description 6
- 230000001537 neural effect Effects 0.000 claims description 5
- 238000012937 correction Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000010606 normalization Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 2
- WSNMPAVSZJSIMT-UHFFFAOYSA-N COc1c(C)c2COC(=O)c2c(O)c1CC(O)C1(C)CCC(=O)O1 Chemical compound COc1c(C)c2COC(=O)c2c(O)c1CC(O)C1(C)CCC(=O)O1 WSNMPAVSZJSIMT-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
- H04N21/44—Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to encoded video stream scene graphs
- H04N21/44008—Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to encoded video stream scene graphs involving operations for analysing video streams, e.g. detecting features or characteristics in the video stream
-
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
- H04N21/435—Processing of additional data, e.g. decrypting of additional data, reconstructing software from modules extracted from the transport stream
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of TV station symbol recognition method and system, methods described includes：After getting frame of video, intercept image block group from described frame of video；Described image block group is input to the input layer of default convolutional neural networks array as |input paramete；Using described convolutional neural networks array, described image block group is identified, to obtain the corresponding television channel of TV station symbol that described frame of video comprises.The present invention is identified the reusability it is achieved that training result by convolutional neural networks array to image block group, it is to avoid training complexity during newly-increased identification species, improves network capacity and recognition accuracy simultaneously.
Description
Technical field
The present invention relates to TV technology, particularly to a kind of TV station symbol recognition method and system.
Background technology
The station symbol of television station is to discriminate between the important mark of different television stations, and TV station symbol recognition technology is to the platform in TV image
Mark a kind of technology being identified, TV station symbol recognition technology is that the application such as video analysis, retrieval and user watched behavioral statisticses carries
Supply important data message.The existing station identification method for distinguishing that can be used for mainly includes two class methods.First kind method is base
The method being identified in the feature of engineer, e.g., the characteristic information such as edge of station symbol color, station symbol shape or station symbol.
Such method is easily affected by video content and station symbol shape, there is accuracy and robustness is not strong.Equations of The Second Kind method is base
In the method for statistical machine learning, the image containing station symbol is sent into neural network module, is automatically learned by neutral net by it
Practise feature representation, it has that network model's autgmentability difference and accuracy rate are low.
Thus prior art could be improved and improves.
Content of the invention
The technical problem to be solved in the present invention is, for the deficiencies in the prior art, provide a kind of TV station symbol recognition method and
System is low to solve the problems, such as network model's autgmentability difference that existing TV station symbol recognition method exists and accuracy rate.
In order to solve above-mentioned technical problem, the technical solution adopted in the present invention is as follows：
A kind of TV station symbol recognition method, it includes：
After getting frame of video, intercept image block group from described frame of video；
Described image block group is input to default convolutional neural networks array input layer, wherein, described convolution as |input paramete
The identification array that neutral net array is made up of with cascade structure some independent convolutional neural networks；
First order convolutional neural networks identification using described convolutional neural networks array is identified to described image block group；
If recognition result is TV station symbol, export the corresponding television channel of described TV station symbol；
If recognition result is other, using next stage convolutional neural networks, described image block group is identified, executes successively
To afterbody convolutional neural networks；
If afterbody convolutional neural networks recognition result is other, judge described convolutional neural networks array None- identified institute
State frame of video, and export unknown result.
Described TV station symbol recognition method, wherein, described after getting frame of video, from described frame of video intercept image block group
Specifically include：
Receive and parse through the video flowing in communications, obtain the video frame images in described video flowing；
Described video frame images are carried out with pretreatment, and some predeterminated positions from pretreatment rear video two field picture intercept figure respectively
As block is to form image block group.
Described TV station symbol recognition method, wherein, described convolutional neural networks array by some independent convolutional neural networks with
The identification array of cascade structure composition is specially：
Described convolutional neural networks array includes some independent convolutional neural networks for identifying inhomogeneity station symbol collection, remembers respectively
For first order convolutional neural networks, second level convolutional neural networks ...., n-th grade of convolutional neural networks；
Connected by cascade system between described some independent convolutional neural networks.
Described TV station symbol recognition method, wherein, if described afterbody convolutional neural networks recognition result is other, judges
Frame of video described in described convolutional neural networks array None- identified, and can also include after exporting unknown result：
It is used for identifying the new convolutional neural networks of the corresponding station symbol of described frame of video according to described frame of video stand-alone training；
After described new convolutional neural networks level is coupled to described convolutional neural networks array, to form new convolutional Neural net
Network array.
Described TV station symbol recognition method, wherein, described convolutional neural networks include amassing god for the master file identifying station symbol feature
Through network and some branch's convolutional neural networks for identifying numeral/character features；Described branch convolutional neural networks bridge
It is connected to the convolutional layer of main convolutional neural networks.
Described TV station symbol recognition method, wherein, by some convolutional layers, space is criticized for the front end of described convolutional neural networks and middle-end
Amount regularization layer, corrects linear unit layer and maximum pond layer composition；Its rear end by some full articulamentums, batch regularization layer and
Correct linear unit layer composition.
Described TV station symbol recognition method, wherein, all includes space batch after each convolutional layer of described convolutional neural networks
Regularization layer and correction linear unit layer.
A kind of TV station symbol recognition system, it includes：
Interception module, for, after getting frame of video, intercepting image block group from described frame of video；
Input module, for described image block group is input to default convolutional neural networks array input layer as |input paramete,
Wherein, the identification array that described convolutional neural networks array is made up of with cascade structure some independent convolutional neural networks；
Identification module, identifies to described image block for the first order convolutional neural networks using described convolutional neural networks array
Group is identified；
First performing module, for being television station's timestamp when recognition result, exports the corresponding television channel of described TV station symbol；
Second performing module, for being other when recognition result, is entered to described image block group using next stage convolutional neural networks
Row identification, repeats identification module, the first performing module and the second performing module up to afterbody convolutional neural networks；
3rd performing module, for when afterbody convolutional neural networks recognition result is other, judging described convolution
Frame of video described in neutral net array None- identified, and export unknown result.
Described TV station symbol recognition system, wherein, described convolutional neural networks include amassing god for the master file identifying station symbol feature
Through network and some branch's convolutional neural networks for identifying numeral/character features；Described branch convolutional neural networks bridge
It is connected to the convolutional layer of main convolutional neural networks.
Described TV station symbol recognition system, wherein, by some convolutional layers, space is criticized for the front end of described convolutional neural networks and middle-end
Amount regularization layer, corrects linear unit layer and maximum pond layer composition；Its rear end by some full articulamentums, batch regularization layer and
Correct linear unit layer composition.
Beneficial effect：Compared with prior art, the invention provides a kind of TV station symbol recognition method and system, methods described bag
Include：After getting frame of video, intercept image block group from described frame of video；Described image block group is inputted as |input paramete
Input layer to default convolutional neural networks array；Using described convolutional neural networks array, described image block group is known
Not, to obtain the corresponding television channel of TV station symbol that described frame of video comprises.The present invention passes through convolutional neural networks array pair
Image block group is identified the reusability it is achieved that training result, it is to avoid training complexity during newly-increased identification species, with
When improve network capacity and recognition accuracy.
Brief description
The flow chart that Fig. 1 is preferably implemented for the TV station symbol recognition method that the present invention provides.
Convolutional neural networks array identification process figure in the TV station symbol recognition method that Fig. 2 provides for the present invention.
The structure principle chart of the TV station symbol recognition method system that Fig. 3 provides for the present invention.
The structure principle chart of another embodiment of TV station symbol recognition method system that Fig. 4 provides for the present invention.
Specific embodiment
The present invention provides a kind of TV station symbol recognition method and system, for making the purpose of the present invention, technical scheme and effect more
Clear, clear and definite, the present invention is described in more detail for the embodiment that develops simultaneously referring to the drawings.It should be appreciated that it is described herein
Specific embodiment only in order to explain the present invention, is not intended to limit the present invention.
In the present invention, using such as " module ", " part " or " unit " for representing element suffix only for favourable
In the explanation of the present invention, itself does not have specific meaning.Therefore, module ", " part " or " unit " can mixedly make
With.
Terminal unit can be implemented in a variety of manners.For example, the terminal described in the present invention can include such as moving
Phone, smart phone, notebook computer, digit broadcasting receiver, PDA (personal digital assistant), PAD (panel computer), PMP
The mobile terminal of (portable media player), guider etc. and such as numeral TV, desk computer etc. consolidate
Determine terminal.However, it will be understood by those skilled in the art that, in addition to being used in particular for the element of mobile purpose, according to this
The construction of bright embodiment can also apply to the terminal of fixed type.
Below in conjunction with the accompanying drawings, by the description to embodiment, content of the invention is described further.
Refer to shown in Fig. 1 and Fig. 2, the flow chart of the preferred embodiment of the TV station symbol recognition method that Fig. 1 provides for the present invention,
Convolutional neural networks array identification process figure in the TV station symbol recognition method that Fig. 2 provides for the present invention.Methods described includes：
S100, after getting frame of video, from described frame of video intercept image block group；
S200, described image block group is input to default convolutional neural networks array input layer as |input paramete, wherein, described
The identification array that convolutional neural networks array is made up of with cascade structure some independent convolutional neural networks；
S300, using described convolutional neural networks array the first order convolutional neural networks identification described image block group is known
Not；
If S400 recognition result is TV station symbol, export the corresponding television channel of described TV station symbol；
If S500 recognition result is other, using next stage convolutional neural networks, described image block group is identified, repeats
Step S300-S500 is up to afterbody convolutional neural networks；
If S600 afterbody convolutional neural networks recognition result is other, judge described convolutional neural networks array
Frame of video described in None- identified, and export unknown result..
Specifically, in described step S100, described frame of video refers to intelligent television and receives in video flowing
Single-frame imagess.Described single-frame imagess are two dimensional image.That is, described frame of video shows for one that intelligent television receives
Image.In actual applications, described get frame of video before can also include monitor intelligent television zapping instruction, receiving
After zapping instruction, first frame of video will be received as described frame of video.Therefore, described after getting frame of video, from institute
State and can include before intercepting image block group in frame of video：
S01, when intelligent television is in open state, the operation of zapping that real-time monitoring users are carried out to intelligent television；
S02, after listening to zapping operation, according to described zapping operational control intelligent television zapping and obtain first receiving
Individual frame of video.
In the present embodiment, obtain image block group after getting frame of video in addition it is also necessary to from described frame of video.Described
Image block group includes multiple equal-sized image blocks, and described image block obtains from the predeterminated position of frame of video.That is, it is pre-
First setting video image obtains the scope of image, and some image blocks corresponding position in described image scope.Described figure
As scope can determine according to the size of smart television display, for example, described image scope and described display screen scope phase
With.
Some positions are pre-set in the range of described image, when intercepting the image block group in image, described by intercepting
The image of some positions is to obtain image block group.The shape size of described some positions is identical, i.e. the shape of the image block of sectional drawing
Identical with size.The interception position of described image block may be located at any position of video frame images, for example, a corner, four
Individual corner and center etc..In actual applications, in order to fast and accurately obtain the image block of protection station symbol, will be described pre-
First some position branches are set to：Four corners and center.
Exemplary, described after getting frame of video, intercept image block group from described frame of video and specifically can include：
S101, the video flowing receiving and parsing through in communications, obtain the video frame images in described video flowing；
S102, described video frame images are carried out with pretreatment, and from some predeterminated positions of pretreatment rear video two field picture respectively
Intercept image block to form image block group.
Specifically, described video frame images are carried out with the described image that pretreatment refers to zoom in and out, at denoising and normalization
Reason.To intercepting multiple images block in the video frame images of pretreatment, the plurality of image block forms image block group.The plurality of figure
As block is the multiple subimages being the same from position intercepting from multiple form and dimensions pre-setting of video frame images.Described
The parts of images content of video frame images described in image block protection.
In described step S200, described convolutional neural networks array can only comprise convolutional neural networks, also may be used
To comprise multiple independent convolutional neural networks.When convolutional neural networks array comprises multiple independent network model, described
Connected by cascade system between multiple independent network modeies.That is, described convolutional neural networks array include some
For identifying the independent convolutional neural networks of inhomogeneity station symbol collection, it is designated as first order convolutional neural networks, second level convolution respectively
Neutral net ...., n-th grade of convolutional neural networks；Connected by cascade system between described some independent convolutional neural networks.
In the present embodiment, the independent network model of described convolutional neural networks array cascade can identify according to it
Content and extend.That is, when described convolutional neural networks array None- identified current image block group, stand-alone training one
New convolutional neural networks model, and by described new convolutional neural networks Cascade in existing convolutional neural networks array
On, as afterbody convolutional neural networks.So can not need to carry out re -training to already present training pattern result
To accurately identify fresh target, solve the problems, such as the neutral net only guarantee accuracy rate in the target of finite number simultaneously.
Exemplarily, described default convolutional neural networks array only comprises convolutional neural networks, is designated as first order volume
Long-pending neutral net, it is used for identifying first station symbol collection (for example comprising multiple TV station symbol species), is designated as A class TV station symbol mould
Type (CNN_A).When assuming to have new station symbol classification collection to need to identify using first order convolutional neural networks, can be to new station symbol
The new convolutional neural networks model of classification collection stand-alone training one, is designated as B class TV station symbol (CNN_B)；Then, by new convolution
After neural network model CNN_B level is associated in original neural network group CNN_A；Finally, new convolutional neural networks array is (for example
Comprise CNN_A and CNN_B) associable for TV station symbol recognition.
In the present embodiment, the front end of described convolutional neural networks and middle-end can be any number of convolutional layers
（convolutional layer）, space batch regularization layer（spatial batch normalization layer）, entangle
Linear positive elementary layer（rectified linear units layer）, and maximum pond layer（max pooling layer）'s
Combination.Rear end is any number of full articulamentum（full connection layer）, batch regularization layer（batch
normalization layer）With correction linear unit layer（rectified linear units layer）Composition.
Described convolutional network carries out the study of weight parameter using stochastic gradient descent method, and constructs nerve with described weight
Network.Wherein, if initial learn speed 0.0001, every 30 wheel iteration are updated to original 1/2, and carry out at random increasing encoding and decoding
Distortion 5%～30%, Random-Rotation -5 spends～5 degree, random 50%～150% size variation disturbance, random left rotation and right rotation, forms number
According to collection, using described data set, convolutional network is trained.Certainly, described convolutional network can also be carried out using other modes
Training, does not just describe in detail here one by one.
Further, described convolutional neural networks comprise main split's convolutional network and Liang Ge sub-branch convolutional network,
Branch is designated as sub-branch convolutional neural networks I and sub-branch convolutional neural networks II.Liang Ge sub-branch convolutional network can be any
Bridge at the convolutional layer of master network.Wherein, described main split convolutional neural networks are responsible for identifying station symbol body feature, described two
Sub-branch's convolutional neural networks are responsible for identifying numerical characteristic and/or character features, and for example, sub-branch convolutional neural networks I is responsible for
Identification numerical characteristic, sub-branch convolutional neural networks II is responsible for recognition character.So pass through to pass through master when carrying out TV station symbol recognition
Branch's convolutional network and Liang Ge sub-branch convolutional network identify respectively, finally three web results are comprehensively chosen, output result.
So master network and sub-network share the parameter in low-level image feature space, can save size and the complexity of network model.
In the present embodiment, described convolutional neural networks can include：
Ground floor, convolutional layer, convolution kernel is 3x3, and convolution step-length is 2；
The second layer, convolutional layer, convolution kernel is 3x3, and convolution step-length is 1；
Third layer, maximum pond layer, pondization is interval to be 2；
4th layer, convolutional layer, convolution kernel is 3x3, and convolution step-length is 1；
Sub-network structure I：
4th layer, convolutional layer, convolution kernel is 3x3, and convolution step-length is 1；
Layer 5, full articulamentum；
Layer 6, full articulamentum（Sub-network I terminates）.
Layer 5, maximum pond layer, pondization is interval to be 2；
Layer 6, convolutional layer, convolution kernel is 3x3, and convolution step-length is 1；
Sub-network structure II：
Layer 6, convolutional layer, convolution kernel is 3x3, and convolution step-length is 1；
Layer 7, full articulamentum（Sub-network II terminates）.
Layer 7, maximum pond layer, pondization is interval to be 2；
8th layer, convolutional layer, convolution kernel is 3x3, and convolution step-length is 1；
9th layer, maximum pond layer, pondization is interval to be 2；
Tenth layer, convolutional layer, convolution kernel is 3x3, and convolution step-length is 1；
Eleventh floor, maximum pond layer, pondization is interval to be 2；
Floor 12, full articulamentum；
13rd layer, full articulamentum（Master network terminates）.
In described convolutional network, there is individual space batch regularization layer after each convolutional layer and correct linear unit
Layer.There is individual spatial regularization layer after each same full articulamentum and correct linear unit layer.The image that convolutional layer will input
Or characteristics of image is by a series of new characteristics of image of linear transformations output.Space batch regularization can be divided with normalization data
Cloth is thus accelerate training process.Correct linear unit the result of input is exported according to the work conversion of approximate people's vision response.Pond
Change layer and multiple input numerical value are mapped as an output numerical value.The upper layer network structure of described convolutional neural networks is according to handled
Objective attribute target attribute have the parameter of specifically different network structures and feature space it can be ensured that the accuracy of network model.Improve
The capacity of network and discrimination.
In described step S300, described using convolutional neural networks array, image block group is identified referring to adopting
Convolutional neural networks array is identified to image block group and detects, its identify station symbol when determine that described station symbol is corresponding
Channel information, described channel information includes the title of described channel, for example, BTV, Shenzhen physical culture etc..
In order to further illustrate convolutional neural networks array to image recognition process, several specific embodiments are given below
It is illustrated.
Embodiment one
Described convolutional neural networks array is the convolutional neural networks group of cascade, and one of neural network group only corresponds to more than one
The convolutional neural networks of the multitask of output.The convolutional neural networks group of the first order identifies one group of initial station symbol.Second level god
Only identify the new station symbol group newly included through network, by that analogy.Its identification process is specifically as follows：Use the convolution of the first order first
Neural network group is identified.If recognition result " other " enters the convolutional neural networks group of next stage, if station symbol is then
Export corresponding television channel.When proceeding to the second level, if recognition result " other " then exports " other ", if station symbol is then
Export corresponding television channel.When carrying out the identification of every one-level, only one of which convolutional neural networks, recognition result is TV frequency
Road title or " other ", this recognition result is directly as the recognition result of this convolutional neural networks group.For example, " Jiangsu physical culture
Channel " is one of television channel that first training data is comprised, and " Shenzhen entertainment channel " belongs to what training data was comprised
One of television channel.For the test image comprising Jiangsu sports channel station symbol, first entered with the convolutional neural networks group of the first order
Row identification.Wherein convolutional neural networks recognition result is " Jiangsu sports channel ", exports " Jiangsu sports channel ".For comprising depth
The test image of ditch between fields entertainment channel station symbol, is first identified with the convolutional neural networks group of the first order.Wherein convolutional neural networks
Recognition result be " other ".Then it is identified with the convolutional neural networks of the second level.The identification of wherein convolutional neural networks
Result is " Shenzhen entertainment channel ".Output " Shenzhen entertainment channel ".
Embodiment two
Convolutional neural networks group using cascade carries out television station channel identification, one-level convolutional neural networks comprise two independent
The convolutional neural networks of the single task of multi output.Wherein, first convolutional neural networks is responsible for identifying main station symbol part, second
Convolutional neural networks identify sub- station symbol part.One group of station symbol of the initial regulation of convolutional neural networks group identification of the first order.Second
Level neural network group only identifies the one group of new station symbol newly included, by that analogy.Concrete identification process can be described as：First with first
The convolutional neural networks group of level is identified.If recognition result " other " enters the convolutional neural networks group of next stage, if
Then export corresponding television channel for station symbol.When carrying out the second level, if recognition result " other " then exports " other ", if
Station symbol then exports corresponding television channel.When being identified with convolutional neural networks group, ground floor convolutional neural networks recognition result
For television station's title or " other ", if ground floor recognition result is " other ", the recognition result of this convolutional neural networks group is " its
He "；If ground floor recognition result is television station, carry out the identification of second layer convolutional neural networks.Second layer convolutional neural networks
Recognition result is channel brief introduction, channel number or " other ".Finally the recognition result of comprehensive two convolutional neural networks obtains
Whole television channel name.For example, " Jiangsu sports channel " is one of television channel that first training data is comprised, Shenzhen
Recreation table belongs to one of channel that second batch training data is comprised.For the test image comprising Jiangsu satellite TV channel station symbol,
First it is identified with the convolutional neural networks group of the first order.Wherein first convolutional neural networks recognition result is " Jiangsu TV
Platform ", second convolutional neural networks recognition result is " physical culture ".Comprehensively first order convolutional neural networks group recognition result can be obtained is
" Jiangsu sports channel ", then exports recognition result.For the test image comprising Shenzhen entertainment channel station symbol, first use the first order
Convolutional neural networks group be identified.The recognition result of wherein first convolutional neural networks is " other ", then with second
The convolutional neural networks of level are identified.In the convolutional neural networks kind of the second level, the knowledge of its first convolutional neural networks
Other result is " Shenzhen TV Station ", and the recognition result of its second convolutional neural networks is " amusement ".Comprehensively the second level can be obtained
The recognition result of convolutional neural networks group is " Shenzhen entertainment channel ".
Embodiment three
Described convolutional neural networks array comprises the convolutional neural networks of the single task of two independent multi output, wherein, first
Individual convolutional neural networks are responsible for identifying main station symbol part, and second convolutional neural networks identifies sub- station symbol part.Newly include when having
Station logo data when coming in, new and old station symbol is put re -training together.For example, " Jiangsu sports channel " is first training data
One of television channel being comprised, Shenzhen recreation table belongs to one of channel that second batch training data is comprised.For " Jiangsu body
Ssd channel ", is identified with convolutional neural networks group.Wherein first convolutional neural networks recognition result is " Jiangsu TV
Platform ", second convolutional neural networks recognition result is " physical culture ".It is " Jiangsu that convolutional neural networks group recognition result comprehensively can be obtained
Sports channel ", then exports recognition result." Shenzhen entertainment channel ", is identified with convolutional neural networks group.Wherein first
The recognition result of convolutional neural networks is " Shenzhen TV Station ", and the recognition result of second convolutional neural networks is " amusement ".Comprehensive
The recognition result that conjunction can obtain convolutional neural networks group is " Shenzhen entertainment channel ".
Present invention also offers a kind of TV station symbol recognition system, as shown in figure 3, it includes：
A kind of TV station symbol recognition system, it includes：
Interception module 100, for, after getting frame of video, intercepting image block group from described frame of video；
Input module 200, for being input to default convolutional neural networks array input using described image block group as |input paramete
Layer, wherein, the identification array that described convolutional neural networks array is made up of with cascade structure some independent convolutional neural networks；
Identification module 300, the first order convolutional neural networks using described convolutional neural networks array identify to described image block
Group is identified；
First performing module 400, for being television station's timestamp when recognition result, output described TV station symbol corresponding TV frequency
Road；
Second performing module 500, for being other when recognition result, using next stage convolutional neural networks to described image block group
It is identified, repeat identification module, the first performing module and the second performing module up to afterbody convolutional Neural net
Network；
3rd performing module 600, for when afterbody convolutional neural networks recognition result is other, judging described volume
Frame of video described in long-pending neutral net array None- identified, and export unknown result..
Described TV station symbol recognition system, wherein, described convolutional neural networks include amassing god for the master file identifying station symbol feature
Through network and some branch's convolutional neural networks for identifying numeral/character features；Described branch convolutional neural networks bridge
It is connected to the convolutional layer of main convolutional neural networks.
Described TV station symbol recognition system, wherein, by some convolutional layers, space is criticized for the front end of described convolutional neural networks and middle-end
Amount regularization layer, corrects linear unit layer and maximum pond layer composition；Its rear end by some full articulamentums, batch regularization layer and
Correct linear unit layer composition.
In another embodiment of the present invention, as shown in figure 4, described system includes：Intelligent television 1000 server
2000；
Described intelligent television 1000 includes interception module 1001 and sending module 1002；
Described interception module 1001, for, after getting frame of video, intercepting image block group from described frame of video；
Described sending module 1002, for sending described image block group to server；
Described server 2000 includes：Receiver module 2001, input module 2002 and identification module 2003；
Described receiver module 2001, for receiving the image block group of intelligent television transmission；
Described input module 2002, for being input to default convolutional neural networks array using described image block group as |input paramete
Input layer；
Described identification module 2003, for being identified to described image block group using described convolutional neural networks array, with
The corresponding television channel of TV station symbol comprising to described frame of video.
Described TV station symbol recognition system, wherein, described interception module specifically includes：
Acquiring unit, for receiving and parsing through the video flowing in communications, obtains the video frame images in described video flowing；
Interception unit for carrying out pretreatment and some default from pretreatment rear video two field picture to described video frame images
Position intercepts image block respectively to form image block group.
Described TV station symbol recognition system, wherein, described default convolutional neural networks array is by some independent convolutional Neural
The identification array that network is formed with cascade structure.
Described TV station symbol recognition system, wherein, described identification module specifically includes：
First recognition unit, for being identified to described image block group using the identification of first order convolutional neural networks；
First output unit, for being television station's timestamp when recognition result, exports the corresponding television channel of described TV station symbol；
Second recognition unit, for when recognition result is for other, being entered to described image block group using next convolutional neural networks
Row identification, executes successively to afterbody convolutional neural networks；
Second output unit, for when the recognition result of afterbody convolutional neural networks is other, exporting unknown result.
Described TV station symbol recognition system, wherein, described convolutional neural networks include rolling up for the main split identifying station symbol feature
Long-pending neutral net and some sub-branch's convolutional neural networks for identifying numeral/character features；Described sub-branch convolution god
Through network bridging in the convolutional layer of main split's convolutional neural networks.
The modules of above-mentioned TV station symbol recognition system are described in detail in the above-mentioned methods, just no longer old one by one here
State.
It should be understood that disclosed system and method in embodiment provided by the present invention, can pass through other
Mode is realized.For example, device embodiment described above is only schematically, for example, the division of described module, it is only
A kind of division of logic function, actual can have other dividing mode when realizing, for example multiple units or assembly can in conjunction with or
Person is desirably integrated into another system, or some features can be ignored, or does not execute.Another, shown or discussed is mutual
Between coupling or direct-coupling or communication connection can be by some interfaces, the INDIRECT COUPLING of device or unit or communication link
Connect, can be electrical, mechanical or other forms.
The described unit illustrating as separating component can be or may not be physically separate, show as unit
The part showing can be or may not be physical location, you can with positioned at a place, or can also be distributed to multiple
On NE.The mesh to realize this embodiment scheme for some or all of unit therein can be selected according to the actual needs
's.
In addition, can be integrated in a processing unit in each functional unit in each embodiment of the present invention it is also possible to
It is that unit is individually physically present it is also possible to two or more units are integrated in a unit.Above-mentioned integrated list
Unit both can be to be realized in the form of hardware, it would however also be possible to employ the form that hardware adds SFU software functional unit is realized.
The above-mentioned integrated unit realized in the form of SFU software functional unit, can be stored in an embodied on computer readable and deposit
In storage media.Above-mentioned SFU software functional unit is stored in a storage medium, including some instructions with so that a computer
Equipment (can be personal computer, server, or network equipment etc.) or processor (processor) execution the present invention each
The part steps of embodiment methods described.And aforesaid storage medium includes：USB flash disk, portable hard drive, read only memory (Read-
Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disc or CD etc. various
Can be with the medium of store program codes.
Finally it should be noted that：Above example only in order to technical scheme to be described, is not intended to limit；Although
With reference to the foregoing embodiments the present invention is described in detail, it will be understood by those within the art that：It still may be used
To modify to the technical scheme described in foregoing embodiments, or equivalent is carried out to wherein some technical characteristics；
And these modification or replace, do not make appropriate technical solution essence depart from various embodiments of the present invention technical scheme spirit and
Scope.
Claims (10)
1. a kind of TV station symbol recognition method is it is characterised in that it includes：
After getting frame of video, intercept image block group from described frame of video；
Described image block group is input to default convolutional neural networks array input layer, wherein, described convolution as |input paramete
The identification array that neutral net array is made up of with cascade structure some independent convolutional neural networks；
First order convolutional neural networks identification using described convolutional neural networks array is identified to described image block group；
If recognition result is TV station symbol, export the corresponding television channel of described TV station symbol；
If recognition result is other, using next stage convolutional neural networks, described image block group is identified, executes successively
To afterbody convolutional neural networks；
If afterbody convolutional neural networks recognition result is other, judge described convolutional neural networks array None- identified institute
State frame of video, and export unknown result.
2. according to claim 1 TV station symbol recognition method it is characterised in that described after getting frame of video, regard from described
Intercept image block group in frequency frame to specifically include：
Receive and parse through the video flowing in communications, obtain the video frame images in described video flowing；
Described video frame images are carried out with pretreatment, and some predeterminated positions from pretreatment rear video two field picture intercept figure respectively
As block is to form image block group.
3. according to claim 1 TV station symbol recognition method it is characterised in that described convolutional neural networks array is by some independences
Convolutional neural networks be specially with the identification array that cascade structure forms：
Described convolutional neural networks array includes some independent convolutional neural networks for identifying inhomogeneity station symbol collection, remembers respectively
For first order convolutional neural networks, second level convolutional neural networks ...., n-th grade of convolutional neural networks；
Connected by cascade system between described some independent convolutional neural networks.
If according to claim 1 TV station symbol recognition method it is characterised in that described afterbody convolutional neural networks identification
Result is other, then judge frame of video described in described convolutional neural networks array None- identified, and exports after unknown result also
Can include：
It is used for identifying the new convolutional neural networks of the corresponding station symbol of described frame of video according to described frame of video stand-alone training；
After described new convolutional neural networks level is coupled to described convolutional neural networks array, to form new convolutional Neural net
Network array.
5. according to the arbitrary described TV station symbol recognition method of claim 1-4 it is characterised in that described convolutional neural networks include for
The main convolutional neural networks of identification station symbol feature and some branch's convolutional neural networks for identifying numeral/character features；
Described branch convolutional neural networks bridge at the convolutional layer of main convolutional neural networks.
6. according to claim 5 TV station symbol recognition method it is characterised in that the front end of described convolutional neural networks and middle-end by
Some convolutional layers, space batch regularization layer, correct linear unit layer and maximum pond layer composition；Its rear end is by some full connections
Layer, batch regularization layer and correction linear unit layer composition.
7. according to claim 6 TV station symbol recognition method it is characterised in that described convolutional neural networks each convolutional layer it
All include space batch regularization layer afterwards and correct linear unit layer.
8. a kind of TV station symbol recognition system is it is characterised in that it includes：
Interception module, for, after getting frame of video, intercepting image block group from described frame of video；
Input module, for described image block group is input to default convolutional neural networks array input layer as |input paramete,
Wherein, the identification array that described convolutional neural networks array is made up of with cascade structure some independent convolutional neural networks；
Identification module, identifies to described image block for the first order convolutional neural networks using described convolutional neural networks array
Group is identified；
First performing module, for being television station's timestamp when recognition result, exports the corresponding television channel of described TV station symbol；
Second performing module, for being other when recognition result, is entered to described image block group using next stage convolutional neural networks
Row identification, repeats identification module, the first performing module and the second performing module up to afterbody convolutional neural networks；
3rd performing module, for when afterbody convolutional neural networks recognition result is other, judging described convolutional Neural
Frame of video described in network array None- identified, and export unknown result.
9. according to claim 8 TV station symbol recognition system it is characterised in that described convolutional neural networks are included for identifying platform
The main convolutional neural networks of mark feature and some branch's convolutional neural networks for identifying numeral/character features；Described point
Prop up the convolutional layer that convolutional neural networks bridge at main convolutional neural networks.
10. according to claim 9 TV station symbol recognition system it is characterised in that the front end of described convolutional neural networks and middle-end
By some convolutional layers, space batch regularization layer, correct linear unit layer and maximum pond layer composition；Its rear end is connected entirely by some
Connect layer, batch regularization layer and correction linear unit layer composition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610931021.4A CN106488313A (en) | 2016-10-31 | 2016-10-31 | A kind of TV station symbol recognition method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610931021.4A CN106488313A (en) | 2016-10-31 | 2016-10-31 | A kind of TV station symbol recognition method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106488313A true CN106488313A (en) | 2017-03-08 |
Family
ID=58272753
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610931021.4A Pending CN106488313A (en) | 2016-10-31 | 2016-10-31 | A kind of TV station symbol recognition method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106488313A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107563434A (en) * | 2017-08-30 | 2018-01-09 | 山东大学 | A kind of brain MRI image sorting technique based on Three dimensional convolution neutral net, device |
| CN108009637A (en) * | 2017-11-20 | 2018-05-08 | 天津大学 | The station symbol dividing method of Pixel-level TV station symbol recognition network based on cross-layer feature extraction |
| CN108509953A (en) * | 2018-04-12 | 2018-09-07 | 南京烽火天地通信科技有限公司 | A kind of TV station symbol detection recognition method |
| CN108734667A (en) * | 2017-04-14 | 2018-11-02 | Tcl集团股份有限公司 | A kind of image processing method and system |
| CN109117768A (en) * | 2018-07-30 | 2019-01-01 | 上海科江电子信息技术有限公司 | A kind of TV station symbol recognition method based on deep learning |
| CN110147462A (en) * | 2019-05-20 | 2019-08-20 | 新联智慧信息技术（深圳）有限公司 | The verification method and Related product of the short-sighted frequency of religion |
| CN110188673A (en) * | 2019-05-29 | 2019-08-30 | 京东方科技集团股份有限公司 | Expression recognition method and device |
| CN110287888A (en) * | 2019-06-26 | 2019-09-27 | 中科软科技股份有限公司 | A kind of TV station symbol recognition method and system |
| CN110309833A (en) * | 2018-03-20 | 2019-10-08 | 国家新闻出版广电总局广播电视规划院 | One kind knowing method for distinguishing, equipment and computer readable storage medium for image identification |
| CN110543884A (en) * | 2018-05-29 | 2019-12-06 | 国际关系学院 | network attack organization tracing method based on image |
| CN110610230A (en) * | 2019-07-31 | 2019-12-24 | 国家计算机网络与信息安全管理中心 | Station caption detection method and device and readable storage medium |
| CN111008738A (en) * | 2019-12-04 | 2020-04-14 | 云南锡业集团（控股）有限责任公司研发中心 | Sn-Bi alloy elongation and tensile strength prediction method based on multi-mode deep learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103607603A (en) * | 2013-11-28 | 2014-02-26 | Tcl集团股份有限公司 | Method and device for identifying station caption |
| CN103679185A (en) * | 2012-08-31 | 2014-03-26 | 富士通株式会社 | Convolutional neural network classifier system as well as training method, classifying method and application thereof |
| CN103902987A (en) * | 2014-04-17 | 2014-07-02 | 福州大学 | Station caption identifying method based on convolutional network |
| CN105956626A (en) * | 2016-05-12 | 2016-09-21 | 成都新舟锐视科技有限公司 | Deep learning based vehicle license plate position insensitive vehicle license plate recognition method |
-
2016
- 2016-10-31 CN CN201610931021.4A patent/CN106488313A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103679185A (en) * | 2012-08-31 | 2014-03-26 | 富士通株式会社 | Convolutional neural network classifier system as well as training method, classifying method and application thereof |
| CN103607603A (en) * | 2013-11-28 | 2014-02-26 | Tcl集团股份有限公司 | Method and device for identifying station caption |
| CN103902987A (en) * | 2014-04-17 | 2014-07-02 | 福州大学 | Station caption identifying method based on convolutional network |
| CN105956626A (en) * | 2016-05-12 | 2016-09-21 | 成都新舟锐视科技有限公司 | Deep learning based vehicle license plate position insensitive vehicle license plate recognition method |
Non-Patent Citations (1)
| Title |
|---|
| 许子立等: "基于递进卷积神经网络的台标识别及其并行化", 《电视技术》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108734667A (en) * | 2017-04-14 | 2018-11-02 | Tcl集团股份有限公司 | A kind of image processing method and system |
| CN107563434A (en) * | 2017-08-30 | 2018-01-09 | 山东大学 | A kind of brain MRI image sorting technique based on Three dimensional convolution neutral net, device |
| CN108009637A (en) * | 2017-11-20 | 2018-05-08 | 天津大学 | The station symbol dividing method of Pixel-level TV station symbol recognition network based on cross-layer feature extraction |
| CN108009637B (en) * | 2017-11-20 | 2021-06-25 | 天津大学 | Station caption segmentation method of pixel-level station caption identification network based on cross-layer feature extraction |
| CN110309833A (en) * | 2018-03-20 | 2019-10-08 | 国家新闻出版广电总局广播电视规划院 | One kind knowing method for distinguishing, equipment and computer readable storage medium for image identification |
| CN110309833B (en) * | 2018-03-20 | 2021-03-05 | 国家新闻出版广电总局广播电视规划院 | Method, equipment and computer-readable storage medium for image identification recognition |
| CN108509953A (en) * | 2018-04-12 | 2018-09-07 | 南京烽火天地通信科技有限公司 | A kind of TV station symbol detection recognition method |
| CN110543884A (en) * | 2018-05-29 | 2019-12-06 | 国际关系学院 | network attack organization tracing method based on image |
| CN109117768A (en) * | 2018-07-30 | 2019-01-01 | 上海科江电子信息技术有限公司 | A kind of TV station symbol recognition method based on deep learning |
| CN110147462A (en) * | 2019-05-20 | 2019-08-20 | 新联智慧信息技术（深圳）有限公司 | The verification method and Related product of the short-sighted frequency of religion |
| CN110188673A (en) * | 2019-05-29 | 2019-08-30 | 京东方科技集团股份有限公司 | Expression recognition method and device |
| CN110188673B (en) * | 2019-05-29 | 2021-07-30 | 京东方科技集团股份有限公司 | Expression recognition method and device |
| CN110287888A (en) * | 2019-06-26 | 2019-09-27 | 中科软科技股份有限公司 | A kind of TV station symbol recognition method and system |
| CN110610230A (en) * | 2019-07-31 | 2019-12-24 | 国家计算机网络与信息安全管理中心 | Station caption detection method and device and readable storage medium |
| CN111008738A (en) * | 2019-12-04 | 2020-04-14 | 云南锡业集团（控股）有限责任公司研发中心 | Sn-Bi alloy elongation and tensile strength prediction method based on multi-mode deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106488313A (en) | A kind of TV station symbol recognition method and system | |
| CN109978893A (en) | Training method, device, equipment and the storage medium of image, semantic segmentation network | |
| CN106156781B (en) | Sort convolutional neural networks construction method and its image processing method and device | |
| CN107657249A (en) | Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again | |
| CN108664999A (en) | A kind of training method and its device, computer server of disaggregated model | |
| CN111368943B (en) | Method and device for identifying object in image, storage medium and electronic device | |
| CN110288082A (en) | Convolutional neural networks model training method, device and computer readable storage medium | |
| CN107644221A (en) | Convolutional neural networks traffic sign recognition method based on compression of parameters | |
| CN112257526B (en) | Action recognition method based on feature interactive learning and terminal equipment | |
| CN107654406A (en) | Fan air-supply control device, fan air-supply control method and device | |
| CN105118027A (en) | Image defogging method | |
| CN108038839A (en) | Twisted-pair feeder lay real-time detection method on a kind of flow production line | |
| CN112950640A (en) | Video portrait segmentation method and device, electronic equipment and storage medium | |
| CN108810648A (en) | A kind of method and terminal carrying out image procossing | |
| CN106446223A (en) | Map data processing method and device | |
| WO2023193521A1 (en) | Video inpainting method, related apparatus, device and storage medium | |
| CN114092920A (en) | Model training method, image classification method, device and storage medium | |
| CN106454195A (en) | Anti-peeping method and system for video chats based on VR | |
| CN108876144B (en) | Transformer substation pre-addressing method based on deep learning algorithm | |
| CN106960188A (en) | Weather image sorting technique and device | |
| CN107729381A (en) | Interactive multimedia resource polymerization method and system based on multidimensional characteristic identification | |
| CN115082800B (en) | Image segmentation method | |
| CN105184809A (en) | Moving object detection method and moving object detection device | |
| CN114882314A (en) | Model training method and related product, image processing method and related product | |
| CN114708575A (en) | Vehicle identification method, device and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | 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 |
Application publication date: 20170308 |
|
| RJ01 | Rejection of invention patent application after publication |