CN108985244A - A kind of television program type recognition methods and device - Google Patents
A kind of television program type recognition methods and device Download PDFInfo
- Publication number
- CN108985244A CN108985244A CN201810821306.1A CN201810821306A CN108985244A CN 108985244 A CN108985244 A CN 108985244A CN 201810821306 A CN201810821306 A CN 201810821306A CN 108985244 A CN108985244 A CN 108985244A
- Authority
- CN
- China
- Prior art keywords
- video image
- program
- frame video
- type
- continuous
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/635—Overlay text, e.g. embedded captions in a TV program
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The application provides a kind of television program type recognition methods and device, passes through the continuous N frame video image obtained in current television program;The continuous N frame video image is input to the convolutional neural networks of pre-training, obtains the corresponding program category of every frame video image in the continuous N frame video image of output;Then it is counted the corresponding program category of frame video image every in the continuous N frame video image to obtain the program category of current television program according to preset strategy.The present invention can judge by accident to avoid caused by a small number of video image identification mistakes, to improve the recognition accuracy of TV programme.
Description
Technical field
This application involves TV technology more particularly to a kind of television program type recognition methods and devices.
Background technique
In existing television program type recognizer, existing But most of algorithms is the mark for period and program
Know the identification that information carries out TV programme, however this identification has limitation, if there is no period or mark in one section of picture
Know information, then recognition effect will have a greatly reduced quality.
Another common recognizer is exactly deep learning algorithm, wherein again the most prominent with convolutional neural networks algorithm
Out.Identification for television program type, due to often will appear the discrimination of some pictures not during video broadcasts
Height is also similar to that another program while being similar to certain program.Therefore, the sample of acquirement can not be sufficiently large
In the case of will lead to the not high situation of recognition accuracy.
Summary of the invention
In view of this, accuracy rate is lower when in order to solve the problems, such as the identification of existing program category, the present invention proposes one kind
Television program type recognition methods and device carry out program category knowledge using multi-frame video image of the convolutional neural networks to input
Not, judgement is combined to the corresponding program category of every frame video image identified and finally determines the corresponding section of TV programme
Mesh type, to improve the recognition accuracy of TV programme.
Specifically, the application is achieved by the following technical solution:
According to the embodiment of the present application in a first aspect, providing a kind of television program type recognition methods, which comprises
Obtain the continuous N frame video image in current television program;
The continuous N frame video image is input to the convolutional neural networks of pre-training, obtains the convolutional neural networks
The corresponding program category of every frame video image in the N frame video image of output;
The corresponding program category of frame video image every in the continuous N frame video image is counted according to preset strategy
Obtain the program category of current television program.
As one embodiment, the training method of the convolutional neural networks includes:
Multiple program categories are divided to TV programme;
Obtain the corresponding video sample of each program category;
The image feature data of every frame image in the video sample is extracted as training data;
The training data is input in convolutional neural networks and is trained, convolutional neural networks model is obtained.
As one embodiment, by the corresponding program category of frame video image every in the continuous N frame video image according to
Preset strategy is counted to obtain the program category of current television program, comprising:
From the continuous N frame video image that the convolutional neural networks export, each program category tool is counted respectively
There is the quantity of same program type;It, should if having same program type quantity at most and being more than or equal to the first preset quantity
Program category is exported as type of prediction;
From continuous multiple type of prediction, the quantity that each type of prediction has identical type of prediction is counted respectively;If tool
There is identical type of prediction quantity at most and be more than or equal to the second preset quantity, then using the type of prediction as current television program
Program category.
As one embodiment, the method also includes:
If the N in the continuous N frame video image is more than or equal to first threshold, it is corresponding to be not determined by current television program
Program category, then obtain next frame video image;
If the N in the continuous N frame video image is more than or equal to second threshold, it is corresponding to be not determined by current television program
Program category, then stop obtaining video image, etc. after first time intervals, then start to obtain video image.
As one embodiment, the method also includes:
If the program category of current television program is identical as the program category of the TV programme obtained next time, stopping is obtained
Video image is taken, after waiting the second time interval, then starts to obtain video image.
According to the second aspect of the embodiment of the present application, a kind of television program type identification device is provided, described device includes:
Acquiring unit, for obtaining the continuous N frame video image in current television program;
Input unit, for the continuous N frame video image to be input to the convolutional neural networks of pre-training, described in acquisition
The corresponding program category of every frame video image in the continuous N frame video image of convolutional neural networks output;
Determination unit, for being determined according to the corresponding program category of frame video image every in the continuous N frame video image
The program category of current television program.
As one embodiment, described device further include:
Training unit, for dividing multiple program categories to TV programme;Obtain the corresponding video sample of each program category
This;The image feature data of every frame image in the video sample is extracted as training data;The training data is inputted
It is trained into convolutional neural networks, obtains convolutional neural networks model.
As one embodiment, the determination unit, specifically for the continuous N exported from the convolutional neural networks
In frame video image, the quantity that each program category has same program type is counted respectively;If having same program type number
Amount is then exported the program category as type of prediction at most and more than or equal to the first preset quantity;From continuous multiple prediction classes
In type, the quantity that each type of prediction has identical type of prediction is counted respectively;If have identical type of prediction quantity at most and
More than or equal to the second preset quantity, then using the type of prediction as the program category of current television program.
As one embodiment, described device further include:
First stop unit does not determine if be more than or equal to first threshold for the N in the continuous N frame video image
The corresponding program category of current television program out then obtains next frame video image;If the N in the continuous N frame video image
When more than or equal to second threshold, it is not determined by the corresponding program category of current television program, then stops obtaining video image, waits the
After one time interval, then start to obtain video image.
As one embodiment, described device further include:
Second stop unit, if program category and the program of the TV programme obtained for current television program next time
Type is identical, then stops obtaining video image, after waiting the second time interval, then starts to obtain video image.
As seen from the above-described embodiment, the application can be by obtaining the continuous N frame video image in current television program;It will
The continuous N frame video image is input to the convolutional neural networks of pre-training, obtains in the continuous N frame video image of output
The corresponding program category of every frame video image;Then by the corresponding program of frame video image every in the continuous N frame video image
Type is counted to obtain the program category of current television program according to preset strategy.Compared with the prior art, the present invention can be with
Program category identification is carried out to the multi-frame video image of input using the convolutional neural networks of pre-training, the identification to program category
As a result preliminary prediction is carried out, then the corresponding program category of every frame video image identified is counted according to multiframe strategy
Judgement finally determines the corresponding program category of TV programme, since TV programme have continuity, passes through a TV programme
Multi-frame video image program category carry out Statistic analysis can to avoid caused by a small number of video image identification mistakes judge by accident, from
And improve the recognition accuracy of TV programme.
Detailed description of the invention
Fig. 1 is a kind of the application illustratively embodiment flow chart of television program type recognition methods;
Fig. 2 is the illustrative training program schematic diagram of the application;
Fig. 3-1 is the illustrative first image characteristics extraction schematic diagram of the application;
Fig. 3-2 is the illustrative second image characteristics extraction schematic diagram of the application;
Fig. 3-3 is the illustrative third image characteristics extraction schematic diagram of the application;
Fig. 4 is the illustrative multiframe strategy schematic diagram of the application;
Fig. 5 is one embodiment block diagram of the application television program type identification device;
Fig. 6 is a kind of one embodiment block diagram of computer equipment of the application.
Specific embodiment
Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to
When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment
Described in embodiment do not represent all embodiments consistent with the application.On the contrary, they be only with it is such as appended
The example of the consistent device and method of some aspects be described in detail in claims, the application.
It is only to be not intended to be limiting the application merely for for the purpose of describing particular embodiments in term used in this application.
It is also intended in the application and the "an" of singular used in the attached claims, " described " and "the" including majority
Form, unless the context clearly indicates other meaning.It is also understood that term "and/or" used herein refers to and wraps
It may be combined containing one or more associated any or all of project listed.
It will be appreciated that though various information, but this may be described using term first, second, third, etc. in the application
A little information should not necessarily be limited by these terms.These terms are only used to for same type of information being distinguished from each other out.For example, not departing from
In the case where the application range, the first information can also be referred to as the second information, and similarly, the second information can also be referred to as
One information.Depending on context, word as used in this " if " can be construed to " ... when " or " when ...
When " or " in response to determination ".
Identification for television program type, due to often will appear the identification of some pictures during video broadcasts
Rate is not high, while being similar to certain program, is also similar to that another program.Such as: in sports news, because this is one
A news program, but many sport pictures can be interted in a program, this results in classifier can be the program for originally belonging to news
It is determined as sport and mistake occurs;For example, in the game shows of plot class, in development of action, picture and cartoon section
Mesh is very similar, so that classifier is difficult to judge this picture on earth to be game or cartoon.Identification TV programme are usual at present
It is to be matched according to program sample to judge program category, but since the picture of TV programme is very more, structure is also suitable
Complexity, renewal speed are also very fast.The picture being all likely to occur exhaustive can not all be included for every class TV programme, therefore
Program sample can not accomplish it is exhaustive, so often result in the erroneous judgement of program category, cause the recognition efficiency of program category compared with
It is low, to influence the effect for carrying out picture adjustment according to program category.
In view of the above problems, the application proposes a kind of television program type recognition methods, it can be by obtaining current television
Continuous N frame video image in program;The continuous N frame video image is input to the convolutional neural networks of pre-training, is obtained
The corresponding program category of every frame video image in the N frame video image of output;Then by every frame in the N frame video image
The corresponding program category of video image is counted to obtain the program category of current television program according to preset strategy.Compared to existing
There is technology, the convolutional neural networks that the present invention can use pre-training carry out program category knowledge to the multi-frame video image of input
Not, preliminary prediction is carried out to the recognition result of program category, then to the corresponding program category of every frame video image identified
Statistic analysis is carried out according to multiframe strategy and finally determines the corresponding program category of TV programme, since TV programme have continuously
Property, carrying out Statistic analysis by the program category of the multi-frame video image of a TV programme can know to avoid a small number of video images
Not erroneous judgement caused by mistake, to improve the recognition accuracy of TV programme.
It is as follows, following embodiments are shown, television program type recognition methods provided by the present application is illustrated.
Embodiment one,
It referring to Figure 1, is a kind of the application illustratively embodiment flow chart of television program type recognition methods, the party
Method the following steps are included:
Continuous N frame video image in step 101, acquisition current television program;
In the present embodiment, which can be used for television set or set-top box either calculates
Machine equipment.When playing TV programme, continuous N frame video image can be obtained from current television program, it should be noted that
The N is the positive integer more than or equal to 2.
Step 102, the convolutional neural networks that the continuous N frame video image is input to pre-training, obtain the convolution
The corresponding program category of every frame video image in the continuous N frame video image of neural network output;
In the present embodiment, the continuous N frame video image that can be will acquire is input in the convolutional neural networks of pre-training,
The judgement that can carry out program category to every frame video image according to the training result of convolutional neural networks, then obtains the volume
The corresponding program category of every frame video image in the continuous N frame video image of product neural network output.
The training process of convolutional neural networks is specifically described by embodiment shown below.
Embodiment two,
Referring to FIG. 2, being the illustrative training program schematic diagram of the application, wherein training process specifically includes:
Step 201 divides multiple program categories to TV programme;
In the present embodiment, television program type is mainly divided into following 7 kinds, such as sport, news, trip by the application
Play, animation, variety, film and other program categories, the application can for main program classifications several in TV programme come
Targetedly identified.
Step 202 obtains the corresponding video sample of each program category;
It, can be by targetedly being learned notable feature possessed by the video image in program in the present embodiment
Practise, than more typical video image in available each program category, there are also it is some rule of thumb judge it is more controversial,
But the often more unified video image of type, such as in news image, in fact it could happen that any sport, game, variety, electricity
The picture material of other programs such as shadow, if once it is determined that having apparent news features, it may be considered that the type of the program is
News, it is seen that although the video image controversial of the type is larger, tend to belong to certain a kind of program.By obtaining each section
The corresponding video sample of mesh type can provide more fully training sample during training convolutional neural networks, to be promoted
The recognition accuracy of convolutional neural networks.
Step 203 extracts the image feature data of every frame image in the video sample as training data;
In the present embodiment, the image feature data conduct of every frame image in the video sample can further be extracted
Training data, to obtain training data.Image feature data is usually to represent the more significant picture material of program category.
For example, the feature extraction schematic diagram as shown in Fig. 3-1, Fig. 3-2 and Fig. 3-3, wherein the program class of Fig. 3-1
Type is film, since the characteristics of image of film is black surround upper and lower in film scene, can be extracted black above and below in the figure
Side (region in Fig. 3-1 in black line frame) is used as image feature data;Wherein the program category of Fig. 3-2 is news, due to news
Characteristics of image be the logo and following subtitle in the upper left corner in news scenes, therefore logo and subtitle can be extracted (in Fig. 3-2
Region in black line frame) it is used as image feature data;Wherein the program category of Fig. 3-3 is sport, due to the characteristics of image of sport
It is the grass meadow in scene, therefore meadow (region in Fig. 3-3 in black line frame) can be extracted as image feature data.It is logical
It crosses and the characteristics of image in typical scene is extracted as training data, so as to so that convolutional neural networks are judging program
It can be by the identification to typical scene, to achieve the purpose that type identification when type.
The training data is input in convolutional neural networks and is trained by step 204, obtains convolutional neural networks mould
Type.
The training data is input in convolutional neural networks and carries out targetedly processing and relational learning algorithm
Parameter training obtains trained convolutional neural networks model.Therefore trained model can be used to playing in TV
Whether detected with the particular attribute in each scene in scene, in the guarantee of the higher some prediction results of confidence level
Under, complete the anticipation for the scene program category with preset characteristic feature.In the present embodiment, convolutional neural networks
Structure may include 5 layers of convolutional layer and 3 full articulamentums, or other composite structures, the application is without limitation.
So far, the description to embodiment two is completed.
Step 103, by the corresponding program category of frame video image every in the continuous N frame video image according to preset strategy
It is counted to obtain the program category of current television program.
In the present embodiment, it is corresponding to obtain every frame video image in the continuous N frame video image of convolutional neural networks output
Program category, then by the corresponding program category of frame video image every in the continuous N frame video image according to preset strategy
It is counted to obtain the final program category of current television program.
Specific type judgement method is specifically described by embodiment shown below.
Embodiment three,
In the present embodiment, the predicting strategy used is multiframe strategy, and specifically, which includes two layers, is passed through
First layer multiframe strategy is for the corresponding section of frame video image every in the convolutional neural networks output continuous N frame video image
After mesh type, from the continuous N frame video image that the convolutional neural networks export, each program category tool is counted respectively
There is the quantity of same program type;It, should if having same program type quantity at most and being more than or equal to the first preset quantity
Program category is exported as type of prediction;Second layer multiframe strategy is used for from continuous multiple type of prediction, and statistics is each respectively
Type of prediction has the quantity of identical type of prediction;If having identical type of prediction quantity at most and being more than or equal to the second present count
Amount, then using the type of prediction as the program category of current television program.
For example, Fig. 4 is referred to, is the illustrative multiframe strategy schematic diagram of the application, wherein assuming that according to training
Convolutional neural networks model, the output of the corresponding program category of every frame in continuous N frame video image is obtained, in the present embodiment
In, N is chosen as 10, wherein being divided between every frame 1 second, obtains the corresponding program category of continuous 10 frame video image first, i.e.,
The program category for 10 frame video images that number in Fig. 4 is 1-10 counts in the corresponding program category of 10 frame video image
Each program category have same program type quantity, if one of program category quantity at most and quantity be greater than etc.
In 9, then exporting this program category is first type of prediction, i.e. 1. type of prediction, does not otherwise export.Then next frame is obtained
The corresponding program category of video image, that is, the corresponding program category of video image that number is 11, by the 11st frame and its
Preceding 9 frame forms a new prediction group, in the group or includes the corresponding program category of 10 frame video images (namely number the
The corresponding program category of 2-11 frame video image), program category prediction is carried out to the new prediction group further according to the above method, is obtained
To second type of prediction, i.e., type of prediction 2., and so on, aforesaid operations are repeated since the 11st frame to the 14th frame, are obtained
Continuous 5 type of prediction are counted, and the number that each type of prediction has identical type of prediction is obtained, if one of them
For the number of type of prediction at most and more than or equal to 4, then output is the program category that this type of prediction is the TV programme, no
It does not export then.
Above-mentioned multiframe strategy is not used in the prior art, in order to guarantee stability that image quality show, then picture during 10 frames
The immovable probability of matter parameter is p10, and wherein p is to identify correct probability, is worth less than 1.
If the stable probability of image quality parameter is p10+10*p9* (1-p)=p9* (10-9p) using multiframe strategy.By
In p9* (10-9p) > p10, therefore first layer multiframe strategy can be improved stability.
Similarly, p9* (10-9p) being set as p1, the probability using second layer multiframe strategy is p14* (5-4p1) > p15, because
The stable probability of this image quality parameter can be got higher.
And the promotion for accuracy rate, the recognition correct rate p of single frames result are 93%, and use multiframe strategy accuracy then
It is 99%.
Being 9 frames for can choose the first preset quantity frame number in first layer, can choose the second present count in the second layer
Amount frame number is 4 frames, so that accuracy rate and stability are more preferably.
Since TV programme picture scene is extremely complex, and there is continuous characteristic for a long time, it is quasi- in order to improve identification
True rate, and frequent switching phenomenon will not occur, the application by being combined to recognition result using above-mentioned multiframe strategy, from
And improve the accuracy rate of identification.
So far, the description of embodiment three is completed.
As one embodiment, if the N in the continuous N frame video image is more than or equal to first threshold, it is not determined by
The corresponding program category of current television program, then obtain next frame video image;If the N in the continuous N frame video image is big
When being equal to second threshold, it is not determined by the corresponding program category of current television program, then stops obtaining video image, waits first
After time interval, then start to obtain video image.
Such as first threshold is 15 frames, and when second threshold is 30 frame, then above scheme are as follows:
When identifying 15 frame video images but not obtaining output result, then obtained on the basis of this 15 frame video image
Next frame (the 16th frame) video image is identified, obtains the corresponding program category of the 16th frame video image, (i.e. with its preceding 9 frame
7-15 frame) video image forms a new prediction group, and it still include the corresponding program category of 10 frame video images in the group
(the namely corresponding program category of number 7-16 frame video image) then carries out the new prediction group according to the above method
Program category prediction, and so on, aforesaid operations are repeated since the 16th frame to the 30th frame.If in 30 frame again without knowledge
When other result, then it is assumed that identification is abnormal, illustrates that present television program type is indefinite, because of these first time intervals (such as 30)
Video image is obtained after s again to be identified, until can recognize that result.
As one embodiment, if program category and the last TV programme obtained by obtaining current television program
Program category it is identical, then stop obtain video image, after waiting the second time interval, then start obtain video image;If current
When the program category of acquisition is identical as upper program category twice, then stop obtaining video image, etc. after thirds time interval, then
Start to obtain video image.For example, if can recognize that television program type using multiframe strategy, stop identifying,
It is identified Deng video image is obtained again after the second time interval (such as 30s), if second of recognition result and last
As a result consistent, then interval is extended for 1min, and so on, gradually extend interval time.If occurred twice in identification process
When the result of identification is inconsistent, then interval time gradually extends since 30s again.
Due to when being identified to television program type, since the calculation amount of convolutional neural networks is larger, if always
It is identified, TV EMS memory occupation is larger, therefore the mechanism being spaced by above-mentioned identification can reduce persistently accounting for for TV memory
With.
Since the adjustment of TV programme image quality parameter at this stage is mainly believed according to logo and EPG (electronic program guides)
Breath, and these information cannot guarantee that very high accuracy, and for some programme televised lives be then always using standard image quality or
Person user oneself adjusts image quality parameter, therefore error rate is higher, or needs manual operation.
As one embodiment, the application can carry out picture according to program category after television program type determines automatically
Matter adjustment, therefore the stability that picture is shown can be improved, image quality will not be frequently replaced because of the identification mistake occurred once in a while
Parameter promotes the visual experience of spectators.
It can be seen that the application can be by obtaining the continuous N frame video image in current television program;It will be described continuous
N frame video image is input to the convolutional neural networks of pre-training, obtains every frame video in the continuous N frame video image of output
The corresponding program category of image;Then by the corresponding program category of frame video image every in the continuous N frame video image according to
Preset strategy is counted to obtain the program category of current television program.Compared with the prior art, the present invention can use pre- instruction
Experienced convolutional neural networks carry out program category identification to the multi-frame video image of input, carry out to the recognition result of program category
Preliminary prediction, then it is final according to multiframe strategy progress Statistic analysis to the corresponding program category of every frame video image identified
It determines the corresponding program category of TV programme, since TV programme have continuity, is regarded by the multiframe of a TV programme
The program category of frequency image, which carries out Statistic analysis, to judge by accident to avoid caused by a small number of video image identification mistakes, to improve
The recognition accuracy of TV programme.
Corresponding with the embodiment of aforementioned image processing method, present invention also provides the embodiments of image processing apparatus.
Fig. 5 is referred to, is one embodiment block diagram of the application television program type identification device, which can wrap
It includes:
Acquiring unit 51, for obtaining the continuous N frame video image in current television program;
Input unit 52 obtains institute for the continuous N frame video image to be input to the convolutional neural networks of pre-training
State the corresponding program category of every frame video image in the continuous N frame video image of convolutional neural networks output;
Determination unit 53, for by the corresponding program category of frame video image every in the continuous N frame video image according to
Preset strategy is counted to obtain the program category of current television program.
As one embodiment, described device further include:
Training unit 54, for dividing multiple program categories to TV programme;Obtain the corresponding video of each program category
Sample;The image feature data of every frame image in the video sample is extracted as training data;The training data is defeated
Enter and be trained into convolutional neural networks, obtains convolutional neural networks model.
As one embodiment, the determination unit 53, specifically for the company exported from the convolutional neural networks
In continuous N frame video image, the quantity that each program category has same program type is counted respectively;If having same program type
Quantity is then exported the program category as type of prediction at most and more than or equal to the first preset quantity;From continuous multiple predictions
In type, the quantity that each type of prediction has identical type of prediction is counted respectively;If having identical type of prediction quantity most
And be more than or equal to the second preset quantity, then using the type of prediction as the program category of current television program.
As one embodiment, described device further include:
First stop unit 55, if be more than or equal to first threshold for the N in the continuous N frame video image, not really
The corresponding program category of current television program is made, then obtains next frame video image;If in the continuous N frame video image
When N is more than or equal to second threshold, it is not determined by the corresponding program category of current television program, then stops obtaining video image, etc.
After first time interval, then start to obtain video image.
As one embodiment, described device further include:
Second stop unit 56, if for program category and the last TV obtained by obtaining current television program
The program category of program is identical, then stops obtaining video image, after waiting the second time interval, then starts to obtain video image;If
When the program category currently obtained is identical as upper program category twice, then stop obtain video image, etc. thirds time interval
Afterwards, then start to obtain video image.
In conclusion the application can be by obtaining the continuous N frame video image in current television program;It will be described continuous
N frame video image is input to the convolutional neural networks of pre-training, obtains every frame video in the continuous N frame video image of output
The corresponding program category of image;Then by the corresponding program category of frame video image every in the continuous N frame video image according to
Preset strategy is counted to obtain the program category of current television program.Compared with the prior art, the present invention can use pre- instruction
Experienced convolutional neural networks carry out program category identification to the multi-frame video image of input, carry out to the recognition result of program category
Preliminary prediction, then it is final according to multiframe strategy progress Statistic analysis to the corresponding program category of every frame video image identified
It determines the corresponding program category of TV programme, since TV programme have continuity, is regarded by the multiframe of a TV programme
The program category of frequency image, which carries out Statistic analysis, to judge by accident to avoid caused by a small number of video image identification mistakes, to improve
The recognition accuracy of TV programme.
The function of each unit and the realization process of effect are specifically detailed in the above method and correspond to step in above-mentioned apparatus
Realization process, details are not described herein.
For device embodiment, since it corresponds essentially to embodiment of the method, so related place is referring to method reality
Apply the part explanation of example.The apparatus embodiments described above are merely exemplary, wherein described be used as separation unit
The unit of explanation may or may not be physically separated, and component shown as a unit can be or can also be with
It is not physical unit, it can it is in one place, or may be distributed over multiple network units.It can be according to actual
The purpose for needing to select some or all of the modules therein to realize application scheme.Those of ordinary skill in the art are not paying
Out in the case where creative work, it can understand and implement.
Corresponding with the embodiment of aforementioned image processing method, present invention also provides for executing above-mentioned image processing method
The embodiment of the computer equipment of method.
As one embodiment, referring to FIG. 6, a kind of computer equipment, including processor 61, communication interface 62, storage
Device 63 and communication bus 64;
Wherein, the processor 61, communication interface 62, memory 63 carry out mutual lead to by the communication bus 64
Letter;
The memory 63, for storing computer program;
The processor 61, for executing the computer program stored on the memory 63, the processor 61 is held
The either step of above-mentioned television program type recognition methods is realized when the row computer program.
Each embodiment in this specification is all made of relevant mode and describes, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for computer
For the embodiment of equipment, since it is substantially similar to the method embodiment, so being described relatively simple, related place is referring to side
The part of method embodiment illustrates.
The foregoing is merely the preferred embodiments of the application, not to limit the application, all essences in the application
Within mind and principle, any modification, equivalent substitution, improvement and etc. done be should be included within the scope of the application protection.
Claims (10)
1. a kind of television program type recognition methods, which is characterized in that the described method includes:
Obtain the continuous N frame video image in current television program;
The continuous N frame video image is input to the convolutional neural networks of pre-training, obtains the convolutional neural networks output
The continuous N frame video image in the corresponding program category of every frame video image;
The corresponding program category of frame video image every in the continuous N frame video image is counted to obtain according to preset strategy
The program category of current television program.
2. method described in claim 1, which is characterized in that the training method of the convolutional neural networks includes:
Multiple program categories are divided to TV programme;
Obtain the corresponding video sample of each program category;
The image feature data of every frame image in the video sample is extracted as training data;
The training data is input in convolutional neural networks and is trained, convolutional neural networks model is obtained.
3. method described in claim 1, which is characterized in that by every frame video image pair in the continuous N frame video image
The program category answered is counted to obtain the program category of current television program according to preset strategy, comprising:
From the continuous N frame video image that the convolutional neural networks export, each program category is counted respectively with phase
With the quantity of program category;If having same program type quantity at most and being more than or equal to the first preset quantity, by the program
Type is exported as type of prediction;
From continuous multiple type of prediction, the quantity that each type of prediction has identical type of prediction is counted respectively;If having phase
With type of prediction quantity at most and more than or equal to the second preset quantity, then using the type of prediction as the program of current television program
Type.
4. method described in claim 1, which is characterized in that the method also includes:
If the N in the continuous N frame video image is more than or equal to first threshold, it is not determined by the corresponding section of current television program
Mesh type then obtains next frame video image;
If the N in the continuous N frame video image is more than or equal to second threshold, it is not determined by the corresponding section of current television program
Mesh type then stops obtaining video image, etc. after first time intervals, then start to obtain video image.
5. method described in claim 1, which is characterized in that the method also includes:
If identical as the program category of TV programme that the last time obtains by the program category for obtaining current television program, stop
Video image is only obtained, after waiting the second time interval, then starts to obtain video image;If the program category currently obtained and upper two
When secondary program category is identical, then stop obtaining video image, etc. after thirds time interval, then start to obtain video image.
6. a kind of television program type identification device, which is characterized in that described device includes:
Acquiring unit, for obtaining the continuous N frame video image in current television program;
Input unit obtains the convolution for the continuous N frame video image to be input to the convolutional neural networks of pre-training
The corresponding program category of every frame video image in the continuous N frame video image of neural network output;
Determination unit is used for the corresponding program category of frame video image every in the continuous N frame video image according to default plan
It is slightly counted to obtain the program category of current television program.
7. device as claimed in claim 6, which is characterized in that described device further include:
Training unit, for dividing multiple program categories to TV programme;Obtain the corresponding video sample of each program category;It mentions
Take the image feature data of every frame image in the video sample as training data;The training data is input to convolution
It is trained in neural network, obtains convolutional neural networks model.
8. device as claimed in claim 6, which is characterized in that
The determination unit, specifically for being united from the continuous N frame video image that the convolutional neural networks export respectively
Count the quantity that each program category has same program type;If having same program type quantity at most and being more than or equal to first
Preset quantity is then exported the program category as type of prediction;From continuous multiple type of prediction, each prediction is counted respectively
Type has the quantity of identical type of prediction;If having identical type of prediction quantity at most and being more than or equal to the second preset quantity,
Then using the type of prediction as the program category of current television program.
9. device as claimed in claim 6, which is characterized in that described device further include:
First stop unit is not determined by and works as if be more than or equal to first threshold for the N in the continuous N frame video image
The corresponding program category of preceding TV programme, then obtain next frame video image;If the N in the continuous N frame video image is greater than
When equal to second threshold, it is not determined by the corresponding program category of current television program, then stops obtaining video image, when waiting first
Between be spaced after, then start obtain video image.
10. device as claimed in claim 6, which is characterized in that described device further include:
Second stop unit, if for program category and the TV programme of last acquisition by obtaining current television program
Program category is identical, then stops obtaining video image, after waiting the second time interval, then starts to obtain video image;If currently obtaining
When the program category taken is identical as upper program category twice, then stop obtaining video image, etc. after thirds time interval, then open
Begin to obtain video image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810821306.1A CN108985244B (en) | 2018-07-24 | 2018-07-24 | Television program type identification method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810821306.1A CN108985244B (en) | 2018-07-24 | 2018-07-24 | Television program type identification method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108985244A true CN108985244A (en) | 2018-12-11 |
CN108985244B CN108985244B (en) | 2021-10-15 |
Family
ID=64550039
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810821306.1A Active CN108985244B (en) | 2018-07-24 | 2018-07-24 | Television program type identification method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108985244B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112800919A (en) * | 2021-01-21 | 2021-05-14 | 百度在线网络技术(北京)有限公司 | Method, device and equipment for detecting target type video and storage medium |
CN115119013A (en) * | 2022-03-26 | 2022-09-27 | 泰州可以信息科技有限公司 | Multi-stage data machine control application system |
CN115996300A (en) * | 2021-10-19 | 2023-04-21 | 海信集团控股股份有限公司 | Video playing method and electronic display device |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101807284A (en) * | 2010-03-16 | 2010-08-18 | 许祥鸿 | Service data retrieval method of Internet television |
CN104866843A (en) * | 2015-06-05 | 2015-08-26 | 中国人民解放军国防科学技术大学 | Monitoring-video-oriented masked face detection method |
US20160094812A1 (en) * | 2014-09-30 | 2016-03-31 | Kai Chen | Method And System For Mobile Surveillance And Mobile Infant Surveillance Platform |
US20160292510A1 (en) * | 2015-03-31 | 2016-10-06 | Zepp Labs, Inc. | Detect sports video highlights for mobile computing devices |
CN106228580A (en) * | 2016-07-29 | 2016-12-14 | 李铮 | The detection of a kind of material based on video analysis, power-economizing method and system |
CN106297331A (en) * | 2016-08-29 | 2017-01-04 | 安徽科力信息产业有限责任公司 | Plane cognition technology is utilized to reduce the method and system of junction machine motor-car stop frequency |
CN107194419A (en) * | 2017-05-10 | 2017-09-22 | 百度在线网络技术(北京)有限公司 | Video classification methods and device, computer equipment and computer-readable recording medium |
CN107798313A (en) * | 2017-11-22 | 2018-03-13 | 杨晓艳 | A kind of human posture recognition method, device, terminal and storage medium |
CN108259990A (en) * | 2018-01-26 | 2018-07-06 | 腾讯科技(深圳)有限公司 | A kind of method and device of video clipping |
CN108280406A (en) * | 2017-12-30 | 2018-07-13 | 广州海昇计算机科技有限公司 | A kind of Activity recognition method, system and device based on segmentation double-stream digestion |
-
2018
- 2018-07-24 CN CN201810821306.1A patent/CN108985244B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101807284A (en) * | 2010-03-16 | 2010-08-18 | 许祥鸿 | Service data retrieval method of Internet television |
US20160094812A1 (en) * | 2014-09-30 | 2016-03-31 | Kai Chen | Method And System For Mobile Surveillance And Mobile Infant Surveillance Platform |
US20160292510A1 (en) * | 2015-03-31 | 2016-10-06 | Zepp Labs, Inc. | Detect sports video highlights for mobile computing devices |
CN104866843A (en) * | 2015-06-05 | 2015-08-26 | 中国人民解放军国防科学技术大学 | Monitoring-video-oriented masked face detection method |
CN106228580A (en) * | 2016-07-29 | 2016-12-14 | 李铮 | The detection of a kind of material based on video analysis, power-economizing method and system |
CN106297331A (en) * | 2016-08-29 | 2017-01-04 | 安徽科力信息产业有限责任公司 | Plane cognition technology is utilized to reduce the method and system of junction machine motor-car stop frequency |
CN107194419A (en) * | 2017-05-10 | 2017-09-22 | 百度在线网络技术(北京)有限公司 | Video classification methods and device, computer equipment and computer-readable recording medium |
CN107798313A (en) * | 2017-11-22 | 2018-03-13 | 杨晓艳 | A kind of human posture recognition method, device, terminal and storage medium |
CN108280406A (en) * | 2017-12-30 | 2018-07-13 | 广州海昇计算机科技有限公司 | A kind of Activity recognition method, system and device based on segmentation double-stream digestion |
CN108259990A (en) * | 2018-01-26 | 2018-07-06 | 腾讯科技(深圳)有限公司 | A kind of method and device of video clipping |
Non-Patent Citations (3)
Title |
---|
SUDEEP D. THEPADE: "Video key frame identification using Thepade"s Transform Error Vector Rotation algorithm with Haar transform and assorted similarity measures", 《2015 INTERNATIONAL CONFERENCE ON COMMUNICATIONS AND SIGNAL PROCESSING (ICCSP)》 * |
时勇强: "基于深度神经网络的语音增强算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
王蓉蓉 等: "一种新的利用多帧结合检测视频标题文字的算法", 《计算机研究与发展》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112800919A (en) * | 2021-01-21 | 2021-05-14 | 百度在线网络技术(北京)有限公司 | Method, device and equipment for detecting target type video and storage medium |
CN115996300A (en) * | 2021-10-19 | 2023-04-21 | 海信集团控股股份有限公司 | Video playing method and electronic display device |
CN115119013A (en) * | 2022-03-26 | 2022-09-27 | 泰州可以信息科技有限公司 | Multi-stage data machine control application system |
Also Published As
Publication number | Publication date |
---|---|
CN108985244B (en) | 2021-10-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107862315B (en) | Subtitle extraction method, video searching method, subtitle sharing method and device | |
CN108985244A (en) | A kind of television program type recognition methods and device | |
WO2021129435A1 (en) | Method for training video definition evaluation model, video recommendation method, and related device | |
CN108495120A (en) | A kind of video frame detection, processing method, apparatus and system | |
CN107707931A (en) | Generated according to video data and explain data, data synthesis method and device, electronic equipment | |
CN104837031B (en) | A kind of method of high-speed adaptive extraction key frame of video | |
EP1638321A1 (en) | Method of viewing audiovisual documents on a receiver, and receiver therefore | |
CN103595965B (en) | The mobile video monitoring method controlled based on video flow | |
CN109922334A (en) | A kind of recognition methods and system of video quality | |
CN107948737A (en) | The recommendation method and device of TV programme | |
CN109408672A (en) | A kind of article generation method, device, server and storage medium | |
CN110147469A (en) | A kind of data processing method, equipment and storage medium | |
CN104320670A (en) | Summary information extracting method and system for network video | |
CN107018406A (en) | Video information processing method and device | |
CN108989875A (en) | A kind of method and device generating barrage file | |
CN110830787A (en) | Method and device for detecting screen-patterned image | |
CN107147824A (en) | The output intent and device of multi-angle video | |
CN114026874A (en) | Video processing method and device, mobile device and readable storage medium | |
CN106454426A (en) | Method for identifying analog channel of intelligent television | |
CN101479729A (en) | Method and system of key frame extraction | |
CN111160107A (en) | Dynamic region detection method based on feature matching | |
CN110245696A (en) | Illegal incidents monitoring method, equipment and readable storage medium storing program for executing based on video | |
US20160266726A1 (en) | Display Control Method And Device | |
CN110296448A (en) | A kind of smog recognition effect modification method, system and the kitchen ventilator of kitchen ventilator | |
CN108668129A (en) | A kind of coding method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |