CN107133631A - A kind of method and device for recognizing TV station's icon - Google Patents
A kind of method and device for recognizing TV station's icon Download PDFInfo
- Publication number
- CN107133631A CN107133631A CN201610113656.3A CN201610113656A CN107133631A CN 107133631 A CN107133631 A CN 107133631A CN 201610113656 A CN201610113656 A CN 201610113656A CN 107133631 A CN107133631 A CN 107133631A
- Authority
- CN
- China
- Prior art keywords
- station
- picture
- convolutional neural
- neural networks
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of method and device for recognizing TV station's icon.Identification TV station figure calibration method, including:According to default TV station's icon training sample database, the convolutional neural networks based on deep learning are trained;The convolutional neural networks based on deep learning after test training;Obtain the picture image of TV station to be detected;The picture image of the TV station to be detected is analyzed by the qualified convolutional neural networks based on deep learning;Export TV station's icon corresponding to the picture image of the TV station to be detected.The present invention can pass through training convolutional neural networks, so that convolutional neural networks have the ability of level identification TV station icon, when user submits the picture of TV station to be identified to after convolutional neural networks after the training, TV station's icon in the picture of TV station to be identified can be accurately identified.
Description
Technical field
The present invention relates to image intelligent analysis field, more particularly to a kind of identification TV station's figure calibration method and dress
Put.
Background technology
TV station's icon is one of specific vision content in INVENTIONBroadcast video, contains the platform of the TV station
The important semantic information such as name, type, copyright, is to discriminate between the important mark of broadcast TV channel, Er Qietai
Target identification all has great importance for the program guiding of radio and television, content analysis and retrieval etc..Mesh
Before, in terms of identification TV station icon, it is frequently encountered the background image that TV station's icon appears in complexity
In, and background image can continue change under many circumstances, and background image that is complicated and persistently changing can shadow
Ring the degree of accuracy to identification TV station icon.The quantity of particularly tv stations currently broadcast is very more, according to statistics much
In 300 channels, the similar degree of some of which TV station icon is higher, adds and accurately identifies TV station's figure
Target difficulty.Therefore how to solve the above problems, just become industry problem urgently to be resolved hurrily.
The content of the invention
The present invention provides a kind of method and device for recognizing TV station icon, by training convolutional neural networks come
The thoughtcast that the mankind recognize image is simulated, to reach the purpose in accurately identification TV station icon.
First aspect according to embodiments of the present invention recognizes TV station's figure calibration method there is provided one kind, including:
According to default TV station's icon training sample database, training convolutional neural networks;
The convolutional neural networks after test training, test result are reached the convolution god of default qualified threshold value
It is referred to as qualified convolutional neural networks through network;
Obtain the picture image of TV station to be detected;
The picture image of the TV station to be detected is analyzed by the qualified convolutional neural networks;
Export TV station's icon corresponding to the picture image of the TV station to be detected.
It is described according to default TV station's icon training sample database, training convolutional nerve net in one embodiment
Network, including:
It is each in acquisition two and more than two TV stations in default TV station's icon training sample database
The n width image frames of individual TV station;
According to default interception area, the n width image frames of interception each TV station are corresponding described section
The image frame in region is taken, described image picture is referred to as interception image picture;
It is default standard size by the size conversion of the interception image picture, by the interception after conversion
Image frame is referred to as the first image frame;
Described first image picture is subjected to gray processing processing, the described first image after gray processing is handled is drawn
Face is referred to as standard picture picture;
Using the standard picture picture, the convolutional neural networks are trained.
In one embodiment, the convolutional neural networks after the test training, test result is reached
The convolutional neural networks for presetting qualified threshold value are referred to as qualified convolutional neural networks, including:
The m width image frames of selection each TV station, the m width image frames of each TV station
Do not occur simultaneously with the n width image frame of each TV station;
The m width image frame of each TV station is generated into corresponding standard picture picture;
The m width images of each TV station are analyzed by the convolutional neural networks after the training
Picture;
TV station's icon corresponding to the m width image frames of output each TV station, will be described corresponding
TV station's icon is referred to as test TV station icon;
Calculate the accuracy rate of test TV station icon;
When the accuracy rate of test TV station icon is higher than the default qualified threshold value, the training is determined
The convolutional neural networks afterwards are qualified convolutional neural networks;
When the accuracy rate, which is less than, is equal to the default qualified threshold value, restart to train the convolutional Neural
Network.
In one embodiment, it is described according to default TV station's icon training sample database, training convolutional nerve
Network, in addition to:
The fitting precision during the training convolutional neural networks is detected in real time;
Whether the increasing degree of fitting precision described in real-time judge, which is more than default over-fitting, increases threshold value;
When the increasing degree of the fitting precision increases threshold value higher than default over-fitting, restart training
The convolutional neural networks.
In one embodiment, it is described to restart to train the convolutional neural networks, including:
Reacquire the n width image frames of each TV station;
The n width image frames of each TV station according to described reselect, regenerate each described electricity
The corresponding standard picture picture of n width image frames of television stations;
Using the standard picture picture, convolutional neural networks described in re -training.
Second aspect according to embodiments of the present invention there is provided a kind of device for recognizing TV station icon, including:
Training module, for according to default TV station's icon training sample database, training convolutional neural networks;
Test module, for testing the convolutional neural networks after training, default conjunction is reached by test result
The convolutional neural networks of lattice threshold value are referred to as qualified convolutional neural networks;
Acquisition module, the picture image of TV station to be detected for obtaining;
Analysis module, for analyzing the TV station to be detected by the qualified convolutional neural networks
Picture image;
Output module, for exporting TV station's figure corresponding to the picture image of the TV station to be detected
Mark.
In one embodiment, the training module, including:
Acquisition submodule, in default TV station's icon training sample database, obtain two and two with
On TV station in each TV station n width image frames;
Submodule is intercepted, for according to default interception area, intercepting the n width images of each TV station
Image frame in the corresponding interception area of picture, described image picture is referred to as interception image picture;
Transform subblock, will for being default standard size by the size conversion of the interception image picture
The interception image picture after conversion is referred to as the first image frame;
Gray processing submodule, for described first image picture to be carried out into gray processing processing, gray processing is handled
Described first image picture afterwards is referred to as standard picture picture;
Submodule is trained, for using the standard picture picture, the convolutional neural networks are trained.
In one embodiment, the test module, including:
Submodule is selected, the m width image frames for selecting each TV station, each described TV station
M width image frame and the n width image frame of each TV station do not occur simultaneously;
Submodule is generated, for the m width image frame of each TV station to be generated into corresponding standard picture
Picture;
Submodule is analyzed, for analyzing each described electricity by the convolutional neural networks after the training
The m width image frames of television stations;
Output sub-module, for exporting TV station's figure corresponding to the m width image frames of each TV station
Mark, test TV station icon is referred to as by corresponding TV station's icon;
Calculating sub module, the accuracy rate for calculating test TV station icon;
Determination sub-module, qualified threshold value is preset for the accuracy rate when test TV station icon higher than described
When, it is qualified convolutional neural networks to determine the convolutional neural networks after the training;
First instructs submodule again, for when the accuracy rate be less than be equal to it is described preset qualified threshold value when, again
Start to train the convolutional neural networks.
In one embodiment, the training module, in addition to:
Submodule is monitored, the fitting precision during the training convolutional neural networks is monitored in real time;
Judging submodule, whether the increasing degree for fitting precision described in real-time judge, which is more than default cross, is intended
Close and increase threshold value;
Second instructs submodule again, increases for the increasing degree when the fitting precision higher than default over-fitting
During threshold value, restart to train the convolutional neural networks.
In one embodiment, described first again instruction submodule and second instruct submodule again, be additionally operable to obtain again
Take the n width image frames of each TV station;The n width of each TV station according to described reselect
Image frame, regenerates the corresponding standard picture picture of n width image frames of each TV station;Make
With the standard picture picture, convolutional neural networks described in re -training.
Other features and advantages of the present invention will be illustrated in the following description, also, partly from explanation
Become apparent, or understood by implementing the present invention in book.The purpose of the present invention and other advantages can
Realize and obtain by specifically noted structure in the specification, claims and accompanying drawing write
.
Below by drawings and examples, technical scheme is described in further detail.
Brief description of the drawings
Accompanying drawing is used for providing a further understanding of the present invention, and constitutes a part for specification, with this hair
Bright embodiment is used to explain the present invention together, is not construed as limiting the invention.In the accompanying drawings:
Fig. 1 is a kind of flow of identification TV station figure calibration method shown in an exemplary embodiment of the invention
Figure;
The step of Fig. 2 is a kind of identification TV station's figure calibration method shown in an exemplary embodiment of the invention
S11 flow chart;
The step of Fig. 3 is a kind of identification TV station's figure calibration method shown in an exemplary embodiment of the invention
S12 flow chart;
The step of Fig. 4 is a kind of identification TV station's figure calibration method shown in further example embodiment of the present invention
S11 flow chart;
The step of Fig. 5 is a kind of identification TV station's figure calibration method shown in an exemplary embodiment of the invention
S37 and step S43 flow charts;
Fig. 6 is a kind of block diagram of the device of identification TV station icon shown in an exemplary embodiment of the invention;
Fig. 7 is a kind of training mould of the device of identification TV station icon shown in an exemplary embodiment of the invention
The block diagram of block 61;
Fig. 8 is a kind of test mould of the device of identification TV station icon shown in an exemplary embodiment of the invention
The block diagram of block 62;
Fig. 9 is a kind of training of the device of identification TV station icon shown in further example embodiment of the present invention
The block diagram of module 61.
Embodiment
The preferred embodiments of the present invention are illustrated below in conjunction with accompanying drawing, it will be appreciated that described herein
Preferred embodiment is merely to illustrate and explain the present invention, and is not intended to limit the present invention.
TV station's icon is one of specific vision content in INVENTIONBroadcast video, contains the platform of the TV station
The important semantic information such as name, type, copyright, is to discriminate between the important mark of broadcast TV channel, Er Qietai
Target identification all has great importance for the program guiding of radio and television, content analysis and retrieval etc..Mesh
Before, in terms of identification TV station icon, it is frequently encountered the background image that TV station's icon appears in complexity
In, and background image can continue change under many circumstances, and background image that is complicated and persistently changing can shadow
Ring the precision to identification TV station icon.The quantity of particularly tv stations currently broadcast is very more, is no less than according to statistics
300 channels, the similar degree of some TV station's icons is higher, more reduces the essence of identification TV station icon
Degree.Therefore how to solve the above problems, just become industry problem urgently to be resolved hurrily.
Fig. 1 is a kind of method flow diagram of identification TV station icon according to an exemplary embodiment, such as
Shown in Fig. 1, the identification TV station figure calibration method comprises the following steps S11-S15:
In step s 11, according to default TV station's icon training sample database, training convolutional neural networks;
The concept of deep learning (Deep Learning, abbreviation DL) comes from the research of artificial neural network,
Multilayer perceptron containing many hidden layers is exactly a kind of deep learning structure.Deep learning is by combining low-level feature shape
Into more abstract high-rise expression attribute classification or feature, represented with the distributed nature for finding data.Depth
Study is a field in machine learning research, and its motivation is to set up, simulates human brain progress analytic learning
Neutral net, the mechanism that it imitates human brain explains data, such as image, sound and text.
Convolutional neural networks (Convolutional neural networks, abbreviation CNN) are a kind of depth
Machine learning model under supervised learning, is one kind in deep neural network.Convolutional neural networks, many
On the basis of layer neutral net, the content of feature learning is added, is handled for imitating classification of the human brain to signal
Processing mode.
According to default TV station's icon training sample database, convolutional neural networks can be automatic during training
Learn the recognition capability of simultaneously constantly improve itself.The sample size of training is more, and the convolutional neural networks may
The analysis ability possessed is more powerful.
In step s 12, the convolutional neural networks after test training, default conjunction is reached by test result
The convolutional neural networks of lattice threshold value are referred to as qualified convolutional neural networks;
Convolutional neural networks after test training have had been provided with certain identification to TV station's icon
Ability, at this time, it may be necessary to using the sample in the Sample Storehouse used when being different from training, the training of testing
The recognition capability of convolutional neural networks afterwards.
In step s 13, the picture image of TV station to be detected is obtained;
In step S14, the TV station to be detected is analyzed by the qualified convolutional neural networks
Picture image;
The low layer pictures feature of the picture image of TV station to be detected by obtaining, it is special according to the low layer pictures
Levy to judge the high-rise characteristics of image corresponding to it, and then draw and drawn accordingly by the high-rise characteristics of image
TV station's icon.The number of plies between the low layer pictures feature and the high-rise figure is more than three layers.
In step S15, TV station's icon corresponding to the picture image of the TV station to be detected is exported.
TV station's icon corresponding to the picture image of the TV station to be detected is exported, at the same time it can also carry
For the higher several similar TV station's icons of similarity, with for reference.
Convolutional neural networks used in the present embodiment have the multi-level knot similar to human brain form of thinking
Structure, by recognizing the low layer pictures feature in image, such as pixel.Then gradually it is upgraded to higher level figure
As feature, such as profile, texture.Gradually upgrade again high higher level characteristics of image.It is adjacent it is upper layer by layer
Secondary data the structure that can learn to lower floor in itself.After learning to (n-1)th layer, by n-1 layers of output
As the input of n-th layer, n-th layer is trained, the numerical value of the n is more than 4.
With the increase for the sample data volume that training is completed, the recognition capability of convolutional neural networks also increases therewith
By force.It is without manually participating in regulation during convolutional neural networks are trained.The characteristics of image of low layer
It is also to be manually specified, can be solved by convolutional neural networks itself.
In one embodiment, as shown in Fig. 2 step S11 comprises the following steps S21-S25:
In the step s 21, in default TV station's icon training sample database, two and two or more are obtained
TV station in each TV station n width image frames;
The image frame of TV station is obtained, the quantity of TV station may be greater than 2 any positive integer.Due to
The characteristic of convolutional neural networks, therefore there is provided the image frame of more TV stations for each TV station,
The recognition capability of convolutional neural networks after training can more be improved.
In step S22, according to default interception area, the n width image of interception each TV station is drawn
Image frame in the corresponding interception area in face, described image picture is referred to as interception image picture;
The default interception area is that, according to priori, TV station's icon always occurs from the upper left corner of screen
Or the upper right corner.Therefore, judge that some TV station's icon appears in the upper left corner or the upper right corner, might as well vacation
If TV station's icon appears in the upper left corner, then the image frame of the TV station is only intercepted for the TV station
The upper left corner.And the size of of TV station's icon itself is also conditional, so need
The size that interception area is can also be fixed, can so facilitate and numerous TV stations be made at unified
Reason.The image frame intercepted in interception area out is referred to as interception image picture.
It is default standard size by the size conversion of the interception image picture in step S23, will turns
The interception image picture after changing is referred to as the first image frame;
Size conversion to capturing picture is unified size, and the size is default standard size.Can be square
The processing of continuous link after an action of the bowels.
In step s 24, described first image picture is subjected to gray processing processing, after gray processing is handled
Described first image picture is referred to as standard picture picture;
In step s 25, using the standard picture picture, the convolutional neural networks are trained.
Convolutional neural networks in the present embodiment are on the basis of neutral net, to have imitated human brain to signal
Classification is handled, and adds the function of feature learning, and the convolutional neural networks can be passed through with spontaneous learning characteristic
The training of multilayer, realizes that the identification to the low layer pictures feature of image rises to the high-rise characteristics of image of image
Identification.Between wherein adjacent two layers, the neutral net of lower floor provides input for the neutral net on upper strata, on
The neutral net of layer only receives the characteristics of image of the neutral net of lower floor.
By training substantial amounts of sample, to reach the analysis ability of enhancing convolutional neural networks.In selection sample
Aspect, in order to improve the analysis ability of convolutional neural networks, it is necessary to select in the video pictures with TV station
The image of key frame.The image of so-called key frame refer to for some TV station prepare sample data in, wherein
Diversity factor between each sample data is greater than default diversity factor threshold value.
In one embodiment, as shown in figure 3, step S12 comprises the following steps S31-S37:
In step S31, the m width image frames of selection each TV station, each described TV station
M width image frame and the n width image frame of each TV station do not occur simultaneously;
The test sample storehouse selected by convolutional neural networks after test training, be for selection test sample storehouse
Require.First, sample size can not be very few in test sample storehouse, in general no less than the instruction of half
Practice the sample size of Sample Storehouse;Secondly, the common factor of test sample storehouse and training sample database is empty set.
In step s 32, the m width image frame of each TV station is generated into corresponding standard picture
Picture;
To image frame interception image picture all in test sample storehouse, the image frame is default interception
Image frame in region;The image frame is converted into standard size again;Then gray processing processing is carried out again,
All image frames that can be completed in test sample storehouse are converted to standard picture picture.
In step S33, each described TV is analyzed by the convolutional neural networks after the training
The m width image frames of platform;
In step S34, TV station's figure corresponding to the m width image frames of output each TV station
Mark, test TV station icon is referred to as by corresponding TV station's icon;
In step s 35, the accuracy rate of test TV station icon is calculated;
The pre-recorded image frame of each image frame comes from that TV station in test sample storehouse,
It is known i.e. for the corresponding reality television platform icon of each image frame in test sample database.
By the method for testing in the present embodiment, the test TV station icon of each image frame can obtain.Compare every
Whether the reality television platform icon of individual image frame is consistent with test TV station icon., can by statistics
To draw the accuracy rate of the method prediction TV station icon by the present embodiment.
In step S36, when the accuracy rate of test TV station icon is higher than the default qualified threshold value,
It is qualified convolutional neural networks to determine the convolutional neural networks after the training;
In step S37, when the accuracy rate, which is less than, is equal to the default qualified threshold value, restart instruction
Practice the convolutional neural networks.
Restart to train the convolutional neural networks, it is necessary to choose new TV station's icon training sample database.Can
On the basis of training in last time, continue through new TV station's icon training sample database to train convolution god
Through network;All data produced by last time training can also be removed so that the convolutional neural networks are not appointed
Where last time trains brought influence, then trains the convolution by new TV station's icon training sample database
Neutral net.
In one embodiment, as shown in figure 4, step S11 also comprises the following steps S41-S43:
In step S41, the fitting precision during the training convolutional neural networks is detected in real time;
In step S42, whether the increasing degree of fitting precision described in real-time judge is more than default over-fitting
Increase threshold value;
In step S43, when the increasing degree of the fitting precision increases threshold value higher than default over-fitting,
Restart to train the convolutional neural networks.
In one embodiment, during the method training convolutional neural networks in using the present embodiment,
The situation of change of the fitting precision of the convolutional neural networks is monitored in real time.In the training process to big data,
Fitting precision is typically gradually slowly to be lifted.If there is fitting precision increasing degree whether be more than it is default
Over-fitting increase threshold value situation, then be typically the situation for occurring in that over-fitting.For occurring in that plan
The situation of conjunction, it is however generally that be to need to restart to train the convolutional neural networks.It can continue to training
The convolutional neural networks, by testing the performance of the convolutional neural networks, that is, test the accurate of TV station's icon
Degree, then decide whether to restart to train the convolutional neural networks.
In one embodiment, as shown in figure 5, step S37 and step 43 comprise the following steps S51-S53:
In step s 51, the n width image frames of each TV station are reacquired;
In step S52, the n width image frames of each TV station according to described reselect, again
The corresponding standard picture picture of n width image frames of generation each TV station;
In step S53, using the standard picture picture, convolutional neural networks described in re -training.
In one embodiment, new TV station's icon training sample database is set up again, that is, reacquires each
The n width image frames of TV station.Default section is intercepted to the n width image frames for reacquiring each TV station
The image frame in region is taken, then those image frames are converted into standard size, gray processing is then carried out again
Processing, you can new TV station's icon training sample database is set up in completion again.Wherein, new electricity is set up again
There must be different samples between television stations icon training sample database and TV station's icon training sample database before
Notebook data, and the different sample data is more much better.
In one embodiment, Fig. 6 is a kind of identification TV station icon according to an exemplary embodiment
Device block diagram.As Fig. 6 shows, the device include training module 61, test module 62, acquisition module 63,
Analysis module 64 and output module 65.
The training module 61, for according to default TV station's icon training sample database, training convolutional nerve net
Network;
Test module 62, for testing the convolutional neural networks after training, test result is reached default
The convolutional neural networks of qualified threshold value are referred to as qualified convolutional neural networks;
Acquisition module 63, the picture image of TV station to be detected for obtaining;
Analysis module 64, for analyzing the TV to be detected by the qualified convolutional neural networks
The picture image of platform;
Output module 65, for exporting TV station's figure corresponding to the picture image of the TV station to be detected
Mark.
As shown in fig. 7, the training module 61 includes acquisition submodule 71, interception submodule 72, conversion
Module 73, gray processing submodule 74 and training submodule 75.
Acquisition submodule 71, in default TV station's icon training sample database, obtaining two and two
The n width image frames of each TV station in TV station above;
Submodule 72 is intercepted, for according to default interception area, intercepting the n width figures of each TV station
As the image frame in the corresponding interception area of picture, described image picture is referred to as interception image picture
Face;
Transform subblock 73, for being default standard size by the size conversion of the interception image picture,
The interception image picture after conversion is referred to as the first image frame;
Gray processing submodule 74, for described first image picture to be carried out into gray processing processing, at gray processing
Described first image picture after reason is referred to as standard picture picture;
Submodule 75 is trained, for using the standard picture picture, the convolutional neural networks are trained.
As shown in figure 8, the test module 62 includes selection submodule 81, generation submodule 82, analysis
Module 83, output sub-module 84, calculating sub module 85, determination sub-module 86 and first instruct submodule again
87。
The selection submodule 81, the m width image frames for selecting each TV station, it is described each
The m width image frame of TV station and the n width image frame of each TV station do not occur simultaneously;
Submodule 82 is generated, for the m width image frame of each TV station to be generated into corresponding standard
Image frame;
Analyze submodule 83, for analyzed by the convolutional neural networks after the training it is described each
The m width image frames of TV station;
Output sub-module 84, for exporting the TV corresponding to the m width image frames of each TV station
Platform icon, test TV station icon is referred to as by corresponding TV station's icon;
Calculating sub module 85, the accuracy rate for calculating test TV station icon;
Determination sub-module 86, qualified threshold is preset for the accuracy rate when test TV station icon higher than described
During value, it is qualified convolutional neural networks to determine the convolutional neural networks after the training;
First instructs submodule 87 again, heavy for when the accuracy rate is less than and is equal to the default qualified threshold value
Newly start to train the convolutional neural networks;
First instructs submodule 87 again, is additionally operable to reacquire the n width image frames of each TV station;Root
According to the n width image frames for reselecting each TV station, each TV station is regenerated
The corresponding standard picture picture of n width image frames;Using the standard picture picture, rolled up described in re -training
Product neutral net.
As shown in figure 9, the training module 61 also includes monitoring submodule 91, judging submodule 92 and second
Submodule 93 is instructed again.
Submodule 91 is monitored, the fitting precision during the training convolutional neural networks is monitored in real time;
Whether judging submodule 92, the increasing degree for fitting precision described in real-time judge is more than default mistake
Fitting increases threshold value;
Second instructs submodule 93 again, increases for the increasing degree when the fitting precision higher than default over-fitting
During long threshold value, restart to train the convolutional neural networks.
Second instructs submodule 93 again, is additionally operable to reacquire the n width image frames of each TV station;Root
According to the n width image frames for reselecting each TV station, each TV station is regenerated
The corresponding standard picture picture of n width image frames;Using the standard picture picture, rolled up described in re -training
Product neutral net.
It should be understood by those skilled in the art that, embodiments of the invention can be provided as method, system or meter
Calculation machine program product.Therefore, the present invention can be using complete hardware embodiment, complete software embodiment or knot
The form of embodiment in terms of conjunction software and hardware.Wherein wrapped one or more moreover, the present invention can be used
Containing computer usable program code computer-usable storage medium (include but is not limited to magnetic disk storage and
Optical memory etc.) on the form of computer program product implemented.
The present invention is with reference to the production of method according to embodiments of the present invention, equipment (system) and computer program
The flow chart and/or block diagram of product is described.It should be understood that can by computer program instructions implementation process figure and
/ or each flow and/or square frame in block diagram and the flow in flow chart and/or block diagram and/
Or the combination of square frame.These computer program instructions can be provided to all-purpose computer, special-purpose computer, insertion
Formula processor or the processor of other programmable data processing devices are to produce a machine so that pass through and calculate
The instruction of the computing device of machine or other programmable data processing devices is produced for realizing in flow chart one
The device for the function of being specified in individual flow or multiple flows and/or one square frame of block diagram or multiple square frames.
These computer program instructions, which may be alternatively stored in, can guide computer or the processing of other programmable datas to set
In the standby computer-readable memory worked in a specific way so that be stored in the computer-readable memory
Instruction produce include the manufacture of command device, the command device realization in one flow or multiple of flow chart
The function of being specified in one square frame of flow and/or block diagram or multiple square frames.
These computer program instructions can be also loaded into computer or other programmable data processing devices, made
Obtain and perform series of operation steps on computer or other programmable devices to produce computer implemented place
Reason, so that the instruction performed on computer or other programmable devices is provided for realizing in flow chart one
The step of function of being specified in flow or multiple flows and/or one square frame of block diagram or multiple square frames.
Obviously, those skilled in the art can carry out various changes and modification without departing from this to the present invention
The spirit and scope of invention.So, if these modifications and variations of the present invention belong to the claims in the present invention
And its within the scope of equivalent technologies, then the present invention is also intended to comprising including these changes and modification.
Claims (10)
1. one kind identification TV station figure calibration method, it is characterised in that including:
According to default TV station's icon training sample database, training convolutional neural networks;
The convolutional neural networks after test training, test result are reached the convolution god of default qualified threshold value
It is referred to as qualified convolutional neural networks through network;
Obtain the picture image of TV station to be detected;
The picture image of the TV station to be detected is analyzed by the qualified convolutional neural networks;
Export TV station's icon corresponding to the picture image of the TV station to be detected.
2. the method as described in claim 1, it is characterised in that described according to default TV station's icon
Training sample database, training convolutional neural networks, including:
It is each in acquisition two and more than two TV stations in default TV station's icon training sample database
The n width image frames of individual TV station;
According to default interception area, the n width image frames of interception each TV station are corresponding described section
The image frame in region is taken, described image picture is referred to as interception image picture;
It is default standard size by the size conversion of the interception image picture, by the interception after conversion
Image frame is referred to as the first image frame;
Described first image picture is subjected to gray processing processing, the described first image after gray processing is handled is drawn
Face is referred to as standard picture picture;
Using the standard picture picture, the convolutional neural networks are trained.
3. method as claimed in claim 2, it is characterised in that the convolution after the test training
Neutral net, reaches that the convolutional neural networks of default qualified threshold value are referred to as qualified convolution god by test result
Through network, including:
The m width image frames of selection each TV station, the m width image frames of each TV station
Do not occur simultaneously with the n width image frame of each TV station;
The m width image frame of each TV station is generated into corresponding standard picture picture;
The m width images of each TV station are analyzed by the convolutional neural networks after the training
Picture;
TV station's icon corresponding to the m width image frames of output each TV station, will be described corresponding
TV station's icon is referred to as test TV station icon;
Calculate the accuracy rate of test TV station icon;
When the accuracy rate of test TV station icon is higher than the default qualified threshold value, the training is determined
The convolutional neural networks afterwards are qualified convolutional neural networks;
When the accuracy rate, which is less than, is equal to the default qualified threshold value, restart to train the convolutional Neural
Network.
4. method as claimed in claim 2, it is characterised in that described according to default TV station's icon
Training sample database, training convolutional neural networks, in addition to:
The fitting precision during the training convolutional neural networks is monitored in real time;
Whether the increasing degree of fitting precision described in real-time judge, which is more than default over-fitting, increases threshold value;
When the increasing degree of the fitting precision increases threshold value higher than default over-fitting, restart training
The convolutional neural networks.
5. the method as described in claim 3,4, it is characterised in that described to restart to train the volume
Product neutral net, including:
Reacquire the n width image frames of each TV station;
The n width image frames of each TV station according to described reselect, regenerate each described electricity
The corresponding standard picture picture of n width image frames of television stations;
Using the standard picture picture, convolutional neural networks described in re -training.
6. a kind of device for recognizing TV station's icon, it is characterised in that including:
Training module, for according to default TV station's icon training sample database, training convolutional neural networks;
Test module, for testing the convolutional neural networks after training, default conjunction is reached by test result
The convolutional neural networks of lattice threshold value are referred to as qualified convolutional neural networks;
Acquisition module, the picture image of TV station to be detected for obtaining;
Analysis module, for analyzing the TV station to be detected by the qualified convolutional neural networks
Picture image;
Output module, for exporting TV station's figure corresponding to the picture image of the TV station to be detected
Mark.
7. device according to claim 6, it is characterised in that the training module, including:
Acquisition submodule, in default TV station's icon training sample database, obtain two and two with
On TV station in each TV station n width image frames;
Submodule is intercepted, for according to default interception area, intercepting the n width images of each TV station
Image frame in the corresponding interception area of picture, described image picture is referred to as interception image picture;
Transform subblock, will for being default standard size by the size conversion of the interception image picture
The interception image picture after conversion is referred to as the first image frame;
Gray processing submodule, for described first image picture to be carried out into gray processing processing, gray processing is handled
Described first image picture afterwards is referred to as standard picture picture;
Submodule is trained, for using the standard picture picture, the convolutional neural networks are trained.
8. device according to claim 7, it is characterised in that the test module, including:
Submodule is selected, the m width image frames for selecting each TV station, each described TV station
M width image frame and the n width image frame of each TV station do not occur simultaneously;
Submodule is generated, for the m width image frame of each TV station to be generated into corresponding standard picture
Picture;
Submodule is analyzed, for analyzing each described electricity by the convolutional neural networks after the training
The m width image frames of television stations;
Output sub-module, for exporting TV station's figure corresponding to the m width image frames of each TV station
Mark, test TV station icon is referred to as by corresponding TV station's icon;
Calculating sub module, the accuracy rate for calculating test TV station icon;
Determination sub-module, qualified threshold value is preset for the accuracy rate when test TV station icon higher than described
When, it is qualified convolutional neural networks to determine the convolutional neural networks after the training;
First instructs submodule again, for when the accuracy rate be less than be equal to it is described preset qualified threshold value when, again
Start to train the convolutional neural networks.
9. device according to claim 7, it is characterised in that the training module, in addition to:
Submodule is monitored, the fitting precision during the training convolutional neural networks is monitored in real time;
Judging submodule, whether the increasing degree for fitting precision described in real-time judge, which is more than default cross, is intended
Close and increase threshold value;
Second instructs submodule again, increases for the increasing degree when the fitting precision higher than default over-fitting
During threshold value, restart to train the convolutional neural networks.
10. according to the device of claim 8,9, it is characterised in that
Described first instructs submodule and second again instructs submodule again, is additionally operable to reacquire each described TV station
N width image frames;The n width image frames of each TV station, give birth to again according to described reselect
Into the corresponding standard picture picture of n width image frames of each TV station;Drawn using the standard picture
Face, convolutional neural networks described in re -training.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610113656.3A CN107133631A (en) | 2016-02-29 | 2016-02-29 | A kind of method and device for recognizing TV station's icon |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610113656.3A CN107133631A (en) | 2016-02-29 | 2016-02-29 | A kind of method and device for recognizing TV station's icon |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107133631A true CN107133631A (en) | 2017-09-05 |
Family
ID=59720811
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610113656.3A Pending CN107133631A (en) | 2016-02-29 | 2016-02-29 | A kind of method and device for recognizing TV station's icon |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107133631A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN108304104A (en) * | 2017-10-24 | 2018-07-20 | 腾讯科技(深圳)有限公司 | A kind of data capture method and its equipment, storage medium, terminal |
CN108960412A (en) * | 2018-06-29 | 2018-12-07 | 北京京东尚科信息技术有限公司 | Image-recognizing method, device and computer readable storage medium |
CN109168062A (en) * | 2018-08-28 | 2019-01-08 | 北京达佳互联信息技术有限公司 | Methods of exhibiting, device, terminal device and the storage medium of video playing |
CN109726709A (en) * | 2017-10-31 | 2019-05-07 | 优酷网络技术(北京)有限公司 | Icon-based programming method and apparatus based on convolutional neural networks |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2007102264A (en) * | 2005-09-30 | 2007-04-19 | Toshiba Corp | Character recognition device and method |
CN103607603A (en) * | 2013-11-28 | 2014-02-26 | Tcl集团股份有限公司 | Method and device for identifying station caption |
CN103729678A (en) * | 2013-12-12 | 2014-04-16 | 中国科学院信息工程研究所 | Navy detection method and system based on improved DBN model |
CN103902987A (en) * | 2014-04-17 | 2014-07-02 | 福州大学 | Station caption identifying method based on convolutional network |
CN105095837A (en) * | 2014-05-15 | 2015-11-25 | Tcl集团股份有限公司 | Station logo recognition method and system |
-
2016
- 2016-02-29 CN CN201610113656.3A patent/CN107133631A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2007102264A (en) * | 2005-09-30 | 2007-04-19 | Toshiba Corp | Character recognition device and method |
CN103607603A (en) * | 2013-11-28 | 2014-02-26 | Tcl集团股份有限公司 | Method and device for identifying station caption |
CN103729678A (en) * | 2013-12-12 | 2014-04-16 | 中国科学院信息工程研究所 | Navy detection method and system based on improved DBN model |
CN103902987A (en) * | 2014-04-17 | 2014-07-02 | 福州大学 | Station caption identifying method based on convolutional network |
CN105095837A (en) * | 2014-05-15 | 2015-11-25 | Tcl集团股份有限公司 | Station logo recognition method and system |
Non-Patent Citations (1)
Title |
---|
张博洋等: "基于神经网络的静态台标识别系统设计与实现", 《计算机仿真》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108304104A (en) * | 2017-10-24 | 2018-07-20 | 腾讯科技(深圳)有限公司 | A kind of data capture method and its equipment, storage medium, terminal |
CN108304104B (en) * | 2017-10-24 | 2022-01-28 | 腾讯科技(深圳)有限公司 | Data acquisition method and equipment, storage medium and terminal thereof |
CN109726709A (en) * | 2017-10-31 | 2019-05-07 | 优酷网络技术(北京)有限公司 | Icon-based programming method and apparatus based on convolutional neural networks |
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 |
CN108960412A (en) * | 2018-06-29 | 2018-12-07 | 北京京东尚科信息技术有限公司 | Image-recognizing method, device and computer readable storage medium |
CN109168062A (en) * | 2018-08-28 | 2019-01-08 | 北京达佳互联信息技术有限公司 | Methods of exhibiting, device, terminal device and the storage medium of video playing |
CN109168062B (en) * | 2018-08-28 | 2020-11-24 | 北京达佳互联信息技术有限公司 | Video playing display method and device, terminal equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11605226B2 (en) | Video data processing method and apparatus, and readable storage medium | |
US11967151B2 (en) | Video classification method and apparatus, model training method and apparatus, device, and storage medium | |
CN107133631A (en) | A kind of method and device for recognizing TV station's icon | |
CN106156781A (en) | Sequence convolutional neural networks construction method and image processing method and device | |
CN109784153A (en) | Emotion identification method, apparatus, computer equipment and storage medium | |
CN110148400A (en) | The pronunciation recognition methods of type, the training method of model, device and equipment | |
CN108629772A (en) | Image processing method and device, computer equipment and computer storage media | |
CN109409533A (en) | A kind of generation method of machine learning model, device, equipment and storage medium | |
CN109522963A (en) | A kind of the feature building object detection method and system of single-unit operation | |
CN113435269A (en) | Improved water surface floating object detection and identification method and system based on YOLOv3 | |
CN109753884A (en) | A kind of video behavior recognition methods based on key-frame extraction | |
CN111080746B (en) | Image processing method, device, electronic equipment and storage medium | |
CN115131698B (en) | Video attribute determining method, device, equipment and storage medium | |
CN112231516B (en) | Training method of video abstract generation model, video abstract generation method and device | |
CN109978074A (en) | Image aesthetic feeling and emotion joint classification method and system based on depth multi-task learning | |
CN107135402A (en) | A kind of method and device for recognizing TV station's icon | |
CN110263835A (en) | Rock category automatic identifying method based on deep learning and Bayesian network | |
CN111325212A (en) | Model training method and device, electronic equipment and computer readable storage medium | |
CN113282840B (en) | Comprehensive training acquisition management platform | |
CN105404682B (en) | A kind of book retrieval method based on digital image content | |
CN110427998A (en) | Model training, object detection method and device, electronic equipment, storage medium | |
CN110275820B (en) | Page compatibility testing method, system and equipment | |
CN106960188A (en) | Weather image sorting technique and device | |
CN110555384A (en) | Beef marbling automatic grading system and method based on image data | |
CN105678771A (en) | Determination method and device for determining quality scores of images |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170905 |