CN106503693A - The offer method and device of video front cover - Google Patents
The offer method and device of video front cover Download PDFInfo
- Publication number
- CN106503693A CN106503693A CN201611059438.2A CN201611059438A CN106503693A CN 106503693 A CN106503693 A CN 106503693A CN 201611059438 A CN201611059438 A CN 201611059438A CN 106503693 A CN106503693 A CN 106503693A
- Authority
- CN
- China
- Prior art keywords
- picture
- data set
- machine learning
- value
- preset
- 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
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/73—Querying
- G06F16/738—Presentation of query results
- G06F16/739—Presentation of query results in form of a video summary, e.g. the video summary being a video sequence, a composite still image or having synthesized frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Probability & Statistics with Applications (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the present application discloses the offer method and device of video front cover, and method includes：The video file that receive user is uploaded, the situation of change according to adjacent content frame in video file determine scene change key frame and the corresponding picture of the scene change key frame are intercepted；It is that the picture being truncated to is given a mark and sorted by the good machine learning model for picture classification of training in advance；Preset width number picture high for score is supplied to as candidate's picture of video front cover by user according to sequence, so that user carries out the selection of video front cover from candidate's picture.With this, can both ensure not omitting all important scenes in video file, the picture multiplicity in the video front cover candidate's picture for providing can be provided again, lift the quality of candidate's picture, the user that is more convenient for therefrom chooses the video front cover being more suitable for.
Description
Technical field
The application is related to field of computer technology, more particularly to the offer method and device of video front cover.
Background technology
When we see video in video website, can see that in related web page each video has a video front cover,
The quality height of the corresponding picture of video front cover is the key factor for attracting user to click on video, fiery short particularly with comparing instantly
For video, the quality of the corresponding picture of video front cover is particularly important.
One video typically (such as, is pressed duration average according to set time point by the Choice of existing video front cover
Be divided into several pieces sub-video, the time point that every one's share of expenses for a joint undertaking video is commenced play out as set time point, etc.), from the video
The middle picture that intercepts therefrom is selected for user as candidate's picture of video front cover, but the picture of the video front cover for so obtaining
The problems such as fuzzy, out of focus often occurs, or picture excessively simple, do not contain significant object or object etc..
With the fast development of depth machine learning techniques, and depth machine learning techniques are in the identification of image and voice
The huge progress obtained in terms of process, in order to solve the problems, such as in above-mentioned selecting video front cover scheme, the base that YouTube is proposed
Scheme is automatically generated in the video thumbnails of depth machine learning techniques, deep neural network (DNN, Deep Neural can be adopted
Network), the picture as video front cover for user being uploaded, will be random from video file used as " high-quality " training set
Then the picture of intercepting is entered using " high-quality " training set and " low quality " training set in advance as " low quality " training set
Training of the row based on the machine learning model of DNN, with the DNN machine learning models for obtaining training.Generate in video thumbnails
During, can be first from video file random intercepting picture (such as, intercepting a frame within one second) then good using above-mentioned training in advance
DNN machine learning models the picture being truncated to is given a mark, then choosing from the picture (possibly some width) of highest scoring
Take a best width picture and be used as video front cover.Through manual evaluation, namely DNN machine learning moulds are contrasted by evaluator
The video front cover that type is produced is thought with the video front cover intercepted according to set time point produced by the scheme of picture, 65% people
The picture of the video front cover that DNN machine learning models are produced is more preferable.
But, this kind of scheme can also have the following disadvantages：
First, the picture for directly uploading user, will be from video according to set time point used as " high-quality " training set
The picture of intercepting can be introduced into substantial amounts of " dirty data " that is, can in the picture of user's upload as " low quality " training set
Can there is the picture of a lot of poor qualities, be likely to exist in the picture intercepted according to set time point from video a lot
The pretty good picture of quality, therefore, this training set comprising " dirty data " can directly result in the machine learning model for training and reach
Less than good classifying quality；
Secondly, when video file duration is longer, this kind of sectional drawing mode can cause the multiplicity of the picture being truncated to compare
Height, the video front cover picture for being finally provided to user are likely to be the higher picture of some multiplicities.
Content of the invention
This application provides the offer method and device of video front cover, can not only ensure not omitting owning in video file
Important scenes, can reduce the picture multiplicity in the video front cover candidate's picture for providing again, lift the quality of candidate's picture, more just
The video front cover that be more suitable for therefrom is chosen in user.
This application provides following scheme：
A kind of offer method of video front cover, including：
The video file that receive user is uploaded, the situation of change according to adjacent content frame in video file determine scene change
Key frame is simultaneously intercepted to the corresponding picture of the scene change key frame；
It is that the picture being truncated to is given a mark side by side by the good machine learning model for picture classification of training in advance
Sequence；
Preset width number picture high for score is supplied to as candidate's picture of video front cover by user according to sequence, to use
Family carries out the selection of video front cover from candidate's picture.
Optionally, also include：
Selection instruction of the receive user to any picture in candidate's picture；
The picture that user selects is defined as video front cover.
Optionally, scene change key frame is determined and to the field according to the situation of change of adjacent content frame in video file
The corresponding picture of scape conversion key frame is intercepted, including：
Judge that whether adjacent two content frames change is beyond preset change threshold in video file；
Frame beyond preset change threshold is defined as scene change key frame；
The corresponding picture of scene change key frame is intercepted, and the picture being truncated to is constituted scene change key frame
Picture set.
Optionally, the training to the machine learning model for picture classification, including：
Determine the image data for machine learning model training；
The image data is done repetitive exercise in the machine learning model of convolutional neural networks CNN, and is instructed in iteration
The weights of convolutional neural networks are adjusted during white silk, to obtain for picture classification on the basis of CNN machine learning models
CNN machine learning models；
The CNN machine learning models for picture classification are estimated；
If assessment passes through, training terminates and using the CNN machine learning models for picture classification as training
The CNN machine learning models for picture classification.
Optionally, also include：
If assessment does not pass through, to being adjusted using the parameter of algorithm in the CNN machine learning models for picture classification
Whole, to continue to do iteration in the CNN machine learning models for picture classification by the image data after parameter adjustment
Training, and the weights of convolutional neural networks are adjusted during repetitive exercise, until the CNN machines for picture classification for obtaining
Learning model assessment passes through.
Optionally, the image data determined for machine learning model training, including：
Obtain basic image data collection；
Obtain the color character parameter value that basic image data concentrates picture；
Basic image data is concentrated the picture for not meeting prerequisite remove according to the color character parameter value, to obtain
The image data of machine learning model training must be used for.
Optionally, the basic image data collection includes：The first data set containing user's uploading pictures and containing by pre-
Put the second data set of the picture that time interval is intercepted at random；
The color character parameter value includes tone value, intensity value and brightness value；
Basic image data is concentrated the picture for not meeting prerequisite remove according to the color character parameter value, to obtain
The image data of machine learning model training must be used for, including：
According to preset color character weight, weighted sum calculating is done to the color character parameter value of every width picture, to obtain
The corresponding color character numerical value of every width picture；
The picture and second data set that color character numerical value in first data set is less than the first preset score value
Middle color character numerical value is removed higher than the picture of the second preset score value, obtains first kind data set and Second Type respectively
Data set, using as the image data for machine learning model training.
Optionally, the basic image data collection includes：The first data set containing user's uploading pictures and containing by pre-
Put the second data set of the picture that time interval is intercepted at random；
The color character parameter value includes tone value, intensity value and rgb value；
Basic image data is concentrated the picture for not meeting prerequisite remove according to the color character parameter value, to obtain
The image data of machine learning model training must be used for, including：
By in picture and second data set of the tone value in first data set less than the first preset hue threshold
Tone value is removed higher than the picture of the second preset hue threshold；
Intensity value in first data set is less than the picture of the first preset saturation threshold value and second data
Intensity value is concentrated to be removed higher than the picture of the second preset saturation threshold value；
The black and white picture in first data set is removed according to the rgb value；
By the picture remained in the first data set and the second data set be identified as first kind data set and
Second Type data set, using as the image data for machine learning model training.
Optionally, by color character numerical value in first data set less than the first preset score value picture and described the
In two data sets color character numerical value higher than the second preset score value picture be removed after, also include：
Respectively the similarity in the first data set and the second data set between remaining picture is judged, and according to judgement
As a result choose a width picture to be retained in the picture for reaching preset similarity threshold from similarity, so as to by the first data set and
The picture remained in second data set is respectively as the first kind data set and Second Type data set.
A kind of offer device of video front cover, including：
Sectional drawing unit, for the video file that receive user is uploaded, and the change according to adjacent content frame in video file
Situation determines scene change key frame and the corresponding picture of the scene change key frame is intercepted；
Marking unit, for being the picture being truncated to by the good machine learning model for picture classification of training in advance
Given a mark and sorted；
Candidate's picture provide unit, for according to sequence using preset width number picture high for score as video front cover candidate
Picture is supplied to user, so that user carries out the selection of video front cover from candidate's picture.
Optionally, also include：
Instruction reception unit, for selection instruction of the receive user to any picture in candidate's picture；
Video front cover determining unit, the picture for selecting user are defined as video front cover.
Optionally, the sectional drawing unit, specifically for：
Judge that whether adjacent two content frames change is beyond preset change threshold in video file；
Frame beyond preset change threshold is defined as scene change key frame；
The corresponding picture of scene change key frame is intercepted, and the picture being truncated to is constituted scene change key frame
Picture set.
According to the specific embodiment that the application is provided, this application discloses following technique effect：
By the embodiment of the present application, after the video file of user's upload is received, can be according to consecutive frame in video file
The situation of change of content determines scene change frame, and the corresponding picture of the scene change frame is intercepted, and then can pass through
The good machine learning model for picture classification of training in advance is that the picture being truncated to is given a mark and sorted, further according to sequence
Preset width number picture high for score is supplied to user as candidate's picture of video front cover, so that user is from candidate's picture
In carry out the selection of video front cover.With this, can both ensure not omitting all important scenes in video file, can reduce again providing
Video front cover candidate's picture in picture multiplicity, lift the quality of candidate's picture, the user that is more convenient for therefrom chooses more
Suitable video front cover.
Certainly, the arbitrary product for implementing the application is it is not absolutely required to while reaching all the above advantage.
Description of the drawings
In order to be illustrated more clearly that the embodiment of the present application or technical scheme of the prior art, below will be to institute in embodiment
The accompanying drawing for using is needed to be briefly described, it should be apparent that, drawings in the following description are only some enforcements of the application
Example, for those of ordinary skill in the art, on the premise of not paying creative work, can be being obtained according to these accompanying drawings
Obtain other accompanying drawings.
Fig. 1 is the method flow diagram that the embodiment of the present application is provided；
Fig. 2 is the training flow process in the method that the embodiment of the present application is provided to the machine learning model for picture classification
Figure；
Fig. 3-1 to Fig. 3-3 is the experimental data schematic diagram in the method that the embodiment of the present application is provided；
Fig. 4 is the schematic device that the embodiment of the present application is provided.
Specific embodiment
Accompanying drawing in below in conjunction with the embodiment of the present application, to the embodiment of the present application in technical scheme carry out clear, complete
Site preparation is described, it is clear that described embodiment is only some embodiments of the present application, rather than whole embodiments.It is based on
Embodiment in the application, the every other embodiment obtained by those of ordinary skill in the art belong to the application protection
Scope.
Referring to Fig. 1, the embodiment of the present application provide firstly a kind of offer method of video front cover, can include following step
Suddenly：
S101, the video file that receive user is uploaded, and determined according to the situation of change of adjacent content frame in video file
Scene change key frame is simultaneously intercepted to the corresponding picture of the scene change key frame.
Under normal circumstances, video website is not only able to for built-in video file in its corresponding server to be supplied to user's sight
See, can be watched so that user is supplied to after the video file for receiving any user upload.In the present embodiment, when receiving
After the video file that user uploads, can first determine the scene change situation in video file (it can be appreciated that camera lens occurs
The situation of switching), the content change situation of consecutive frame in video file in the present embodiment, can be such as obtained, adjacent two are judged
Whether content frame change exceeds preset change threshold, and it is crucial that the frame beyond preset change threshold is defined as scene change
Frame, then can intercept to the corresponding picture of the scene change key frame for determining, and can further by all figures being truncated to
Piece constitutes scene change key frame picture set, used in subsequent step, to ensure not omitting video file midfield with this
Scene (being also believed to important scenes) when scape is converted, can reduce the multiplicity for being truncated to picture again.
In actual applications, also can by video file in code check change judging the scene change feelings in video file
Condition, then carries out the intercepting of picture from video file according to scene change situation, to obtain corresponding sectional drawing during scene change,
By this kind of sectional drawing mode, can ensure as far as possible not omit all important scenes in video file, can reduce again being truncated to figure
The multiplicity of piece.
In addition it is also possible to determine the scene change situation in video file using other modes, for example we can be with
By the grey level histogram feature of picture, Scale invariant features transform (SIFT, Scale-Invariant Feature
Transform) feature etc. is judged to the similarity of picture in video file, such as, first can be intercepted according to preset frequency
Picture (such as, 2 seconds a frame etc.), then according to existing with regard to judging the technology of picture similarity between the picture that is truncated to
Similarity is judged, the picture of similarity height (such as similarity reaches default Similarity value) is only protected according to judged result
One is stayed, the purpose for determining the scene changes situation in video file is so also can reach.
S102, is that the picture being truncated to is given a mark by the good machine learning model for picture classification of training in advance
And sort.
Referring to shown in Fig. 2, in the present embodiment, the training process of the machine learning model to being used for picture classification can be wrapped
Include following steps：
Step 1, determines the image data for machine learning model training.
In the present embodiment, can use depth machine learning model (including supervised learning and unsupervised learning. different
Learning framework under the learning model set up different), can such as adopt deep learning convolutional neural networks (CNN,
Convolutional Neural Network), a kind of machine learning model of the supervised learning of depth, certainly, according to reality
Border needs other suitable depth machine learning models may also be employed.
Under normal circumstances, the data for machine learning model training can be divided into three parts：Training dataset
(training data), test data set (testing data), checking data set (validation data), this three part
The ratio of data may be configured as 80%, 10%, 10%.For the machine learning model of supervised learning, obtain for instructing
Experienced data are one of most important links, and high-quality data are the keys of machine learning model training.
This is based on, when implementing, in order to determine the image data for machine learning model training, base can be first obtained
Plinth image data collection, wherein, the basic image data collection may include：The first data set containing user's uploading pictures and contain
Press the second data set of the picture that preset time interval is intercepted at random.
In existing video website, video front cover has two main generation channels：One be uploaded videos user oneself
Pictures are uploaded as video front cover, another is that aforesaid system intercepts picture at random by preset time interval and therefrom selects
Take some width pictures and be supplied to user's selection, user therefrom chooses a width picture as video front cover.On the one hand, on user oneself
The picture of biography is typically all the reasonable picture of well-chosen matter, but being wherein also not excluded for there are some looks not
It is good picture, we can used as the first data set, (it be the relatively high data of quality also to be understood as using such picture
Collection)；On the other hand, system due to be in a random basis intercepting picture, based on this, there is provided to the picture quality meeting that user does selection
Very different, but wherein also it is not excluded for the presence of the pretty good picture of some quality, and we can be by such picture as second
Data set (it can be appreciated that the relatively low data set of quality), in the present embodiment, we can be by the first data set and
Two data sets are defined as the basic image data for machine learning model training.
After above-mentioned basic image data is got, the color character that basic image data concentrates picture can be further obtained
Parameter value (such as may include tone value, intensity value, brightness value, rgb value etc.), then can be according to the color character parameter value
Basic image data is concentrated the picture for not meeting prerequisite remove, to obtain the picture number for machine learning model training
According to.
The selection of video front cover is the very strong work of a subjectivity, and neither one objectively passes judgment on criterion, pictures
Quality, often related to the subjective factorss of people larger, different people has different viewpoints and a preference, such as enriches
Color, eye-catching people or object, the definition of picture, contrast, saturation etc. are all the factors for affecting width picture quality.
Therefore, in one implementation, we can first obtain the color character parameter that basic image data concentrates picture
Value, the color character parameter value may include HSV (Hue (tone), Saturation (saturation), Luminence (brightness)) value
Deng then the color character numerical value of picture can be calculated by the tone value that gets, intensity value, brightness value etc., such as may be used
The color character numerical value such as color saturation, lightness, contrast including picture.Certainly, according to actual needs, it is also possible to pass through
Obtain HSL (Hue (form and aspect), Saturation (saturation), Luminence (brightness)) value etc. to replace above-mentioned HSV value, to enter
Row subsequent step.
We can carry out color character weight setting previously according to conventional experience to above-mentioned color character parameter value, than
Such as：It is that 1, tone value weight is 0.8 that color saturation weight is 0.7, luminance weights, etc..Then, we just can be according to pre-
The color character weight for first arranging, is weighted to the corresponding color character parameter value of every width picture and calculates, to obtain per width
The corresponding color character numerical value of picture, that is to say, that the corresponding color character numerical value of every width picture.
Next, can be according to the color character numerical value of every width picture, will be low for color character numerical value in first data set
It is removed in the picture (color character numerical value is relatively low, the picture of poor quality) of the first preset score value, to obtain the first kind
Type data set (can such as be quality data), and be higher than second preset point by color character numerical value in second data set
The picture (color character numerical value is higher, the picture that quality is pretty good) of value is removed, to obtain Second Type data set (such as
Can be low quality data), and can be using first kind data set and Second Type data set as machine learning model training
Image data.
In another kind of implementation, we can first obtain the color character parameter value that basic image data concentrates picture,
The color character parameter value may include Hue (tone) value, Saturation (saturation) value, RGB ((Red (red), Green
(green), Blue (indigo plant)) value, then can be respectively by the tone value that gets, intensity value, rgb value to not meeting prerequisite
Picture remove.
When implementing, tone value in first data set can be less than picture and the institute of the first preset hue threshold
The picture that tone value in the second data set is stated higher than the second preset hue threshold is removed, that is, by color in the first data set
The relatively good picture of tone in relatively poor picture and the second data set is adjusted to be removed, to reduce the first data set and the
Picture number in two data sets, and then reduce the operand of machine learning model training, operation time is reduced, computing speed is lifted
Degree, while can also lift the quality of picture in the first data set and the second data set.
Then, can also be by intensity value in first data set less than the picture of the first preset saturation threshold value and described
In second data set, intensity value is removed higher than the picture of the second preset saturation threshold value, that is, by the first data set
In the relatively poor picture of color saturation and the second data set, the relatively good picture of color saturation is removed, to reduce
Picture number in first data set and the second data set, and then reduce the operand of machine learning model training, reduce computing
Time, improving operational speed, while can also lift the quality of picture in the first data set and the second data set.
Additionally, the artwork master in order to further lift the picture quality in the first data set, in the first data set picture
Not we want reservation to piece (being also believed to pure gray scale picture), that is to say, that black and white picture is not that we want to carry
User is used as the picture of video front cover for supply, therefore, carries out the black and white picture in first data set according to the rgb value
Remove, that is, by the first data set, not comprising chrominance information, (such as, in RGB, three component values are three in 0 or RGB
Component value is 255 etc.) black and white picture be removed, with this, the picture number in the first data set can be reduced, and then reduced
The operand of machine learning model training, reduces operation time, improving operational speed, while can also be lifted in the first data set scheming
The quality of piece.
Then, the picture remained in the first data set and the second data set is identified as first kind data
Collection (i.e. quality data collection) and Second Type data set (i.e. low quality data collection), using as machine learning model training
Image data.
With this, " dirty data " fallen in prior art in training set " can be cleaned " and (namely do not meet the figure of prerequisite
Piece), the pretty good picture of quality in the picture that the picture of the poor quality in the picture uploaded including user and system are intercepted at random,
The machine learning model for training is caused not reach ideal sort effect to solve the problems, such as due to there is " dirty data ".
In actual applications, in order to further reduce the operand of machine learning model training, according to preset color
Feature weight, before doing weighted sum calculating to the color character parameter value of every width picture, can also be by the size adjusting of every width picture
For preset size.
As the dimension of picture that system is truncated to is probably than larger, therefore, it can before weighted sum is calculated to figure
Piece carries out resize operations with the Aspect Ratio of unified adjustment picture, to meet the requirement of machine learning model, such as, picture
Original size is 1000*2000, can be 100*200 by resize operations by its size adjusting, and this operation can only change picture
Size, without making picture metamorphopsic distortion, with this, is effectively reduced the operand of machine learning model training, lifts computing speed
Degree.
In actual applications, can there is the very high picture of some similarities due to system in the picture being truncated at random, I
Also picture too high for similarity only can be retained one, to improve the quality of picture in the data set for training, and reduce
Picture number in data set.
In the present embodiment, can be in the picture and institute that color score value in first data set is less than the first preset score value
State color score value in the second data set higher than the second preset score value picture be removed after, to the first data set and the second number
According to concentrating the similarity between remaining picture to be judged, judge than the grey level histogram feature such as by picture picture it
Between similarity, specifically, can first obtain the pixel data of each picture and generate the histogram data of each picture, then to each figure
The histogram data of piece is normalized, and reuses Pasteur's Coefficient Algorithm and histogram data is calculated, finally draw
Each picture Similarity value, its value scope can be between [0,1], wherein, and 0 can represent extremely different, 1 can represent extremely similar (or
Identical), similarity judgement can be carried out according to the Similarity value of each picture for acquiring.
Then, preset similarity threshold (such as Similarity value is not less than 0.8) can be reached from similarity according to judged result
Picture in choose a width picture and retained, that is to say, that only retain a width (i.e. other width figures in the high picture of similarity
Piece is all removed), so as to using the picture remained in the first data set and the second data set as the first kind number
According to collection and Second Type data set, with this, can further in first kind data set and Second Type data set picture quantity,
And it is comprehensive to ensure that picture feature in first kind data set and Second Type data set is covered, and can further improve use
In the data set quality of training, picture number is reduced, and then the operand that machine learning model is trained can be reduced, lift computing speed
Degree.
Step 2, the image data is done repeatedly in the machine learning model of the good convolutional neural networks CNN of training in advance
Generation training, and the weights of convolutional neural networks are adjusted during repetitive exercise, with the good CNN machine learning moulds of training in advance
The CNN machine learning models for picture classification are obtained on the basis of type.
For the training of the machine learning model of large data sets generally requires long time, therefore, we can add and move
Move the thought of study, transfer learning can be carried out using the convolutional neural networks (CNN) of Inception-v3 definition, wherein,
Inception-v3 is the Large Visual Recognition Challenge data for training ImageNet in 2012
Collection, this is a class standard task of computer vision field, and which can be divided into 1000 classifications, Inception- whole image collection
The top5 error rates of v3 are 3.46%.
When implementing, can be in the CNN machine learning models of the Inception-v3 definition for having trained, by not
Disconnected repetitive exercise and the adjustment to neural network weight, to obtain the CNN machine learning for picture classification for suiting the requirements
Model, to increase extensibility and the motility of model.
Step 3, is estimated to the CNN machine learning models.
In the present embodiment, first, can be estimated by above-mentioned 10% checking data set, but, this assessment side
Method possibly cannot learn whether CNN machine learning models have the situation of over-fitting, it is possible to occur in the standard on checking data set
Really rate is very high, but the unsatisfactory problem of effect in actual applications, finally to affect CNN machine learning models to picture classification
Accuracy rate.
Therefore, manual evaluation can be also carried out, can such as randomly selects a video file, and be cut from video file at random
Some width pictures (such as 100 width etc.) are taken, this 100 width picture are given a mark and is ranked up by CNN machine learning models,
Then, high several width (front 8 width in the such as scoring sequence) picture of selection score and the low several width of score are (in such as scoring sequence
Rear 8 width) picture is compared, also will the minimum several pictures of several pictures of model marking highest and marking compared
Compared with being estimated to CNN machine learning models by comparative result.
On the basis of above-mentioned manual evaluation, secondary manual evaluation can also be carried out, such as, can arbitrarily choose a video
File, can intercept a few width picture (such as 8 width pictures) by preset time interval (such as per 2 seconds once), by the random intercepting
8 width pictures and above-mentioned first time manual evaluation during the 8 width pictures of highest scoring chosen be compared, by comparing knot
Fruit is assessed again to machine learning model.
With this, it is estimated with first passing through checking data set, then by way of manual evaluation twice, engineering can be avoided
The situation of model over-fitting is practised, and to realize more efficiently assessment being carried out to CNN machine learning models, obtains preferably assessment effect
Really, so ensure accuracy rate of the CNN machine learning models to picture classification.
Step 4, if assessment passes through, can such as be preset by verifying that the precision that is estimated of data set can reach first
Percentage ratio (such as first preset percentage is 85%), and think to give a mark by CNN machine learning models by manual evaluation
The high score picture for obtaining afterwards is more suitable for can reaching the second preset percentage as the ratio of video front cover, and (such as this is second preset
Percentage ratio is that 90%) as assessment passes through, then training terminates and using the CNN machine learning models as being used for for training
The CNN machine learning models of picture classification.
Step 5, if assessment does not pass through, can such as be preset by verifying the precision not up to first that data set is estimated
Percentage ratio, and think that after CNN machine learning models are given a mark the high score picture that obtains is more suitable for as regarding by manual evaluation
The ratio of frequency front cover is not up to the second preset percentage), as assess and do not pass through.
In such cases, then can to CNN machine learning models be adjusted using the parameter of algorithm, specifically can be according to instruction
Practice the degree of convergence of process, the accuracy situation of training to be adjusted, can such as use the TensorBoard of google intuitively
The situation whether neutral net restrains is obtained, wherein, graphical, visualization tool of the Tensorboard for Tensorflow,
Tensorboard can show the static map being made up of in Tensorflow tensor and flow, and precision in training process, partially
Dynamic Graphs of analysis such as difference etc..
For the adjustment of above-mentioned algorithm parameter, mainly to learning rate (learning rate), batch processing size
(batch size), the isoparametric adjustment of iterationses (step).Such as, in parameter tuning process, if learning rate mistake
Greatly, may be such that convolutional neural networks are not restrained, in concussion state, now need to reduce learning rate；If study speed
Rate is too small, and convergence rate is slower, and more iterationses could cause convolutional neural networks to reach local extremum, can now arrange
Larger iterationses increase learning rate；In addition, batch processing size also influences whether convergence situation, can also pass through to criticizing
The adjustment of reason size is adjusting convergence situation.That is, the details of study can be checked by TensorBoard, analyze
Employed in machine learning model, the parameter setting of algorithm is irrational local and is adjusted correspondingly, by parameter adjustment
Journey, so that machine learning model finally restrains and training for promotion accuracy rate.
After parameter adjustment, will continue to do in CNN machine learning models of the image data after algorithm parameter adjustment
Repetitive exercise, and the weights of convolutional neural networks are adjusted during repetitive exercise, until the CNN for picture classification for obtaining
Machine learning model assessment passes through.
Preset width number picture high for score is supplied to user as candidate's picture of video front cover according to sequence by S103,
So that user carries out the selection of video front cover from candidate's picture.
Wherein, sequence can be ascending order (score is from low to high) or descending (fraction is from high to low), in the present embodiment, can
From being ranked up with descending, high preset width number picture (front 8 in such as sequence of score can be chosen from the forefront of sequence
Width) user is supplied to as candidate's picture of video front cover, so that user chooses a width picture from this 8 width picture as video
Front cover.
When implementing, in user is to above-mentioned candidate's picture, (namely in above-mentioned 8 width picture) any picture is carried out
During clicking operation, selection instruction of the user to the picture is as received, user can be selected according to the selection instruction
Picture is defined as the video front cover of video file.
The present inventor has carried out substantial amounts of experiment in R＆D process, according to the above-mentioned repetitive exercise to machine learning model
Method has obtained 6 editions CNN machine learning models for picture marking, by verifying that the precision that data set is estimated reaches
89.9%, think that the high score picture obtained after CNN machine learning models are given a mark is more suitable for as video by manual evaluation
The ratio of front cover reaches 93.3%, and the picture that is given a mark by CNN machine learning models and provided has high definition, contrast
Good, bright in luster abundant, containing significant object (personage or object etc.) the features such as, than traditional video front cover choosing method
More high-quality and high-efficiency.
Referring to the part comparison diagram (wherein color is not showed that) that Fig. 3-1 to 3-3 is inventor's test, in Fig. 3-1 to 3-3
In, 8 width picture of top is marking 8 width picture of highest, and lower section is 8 minimum width pictures of giving a mark in same video.
By the embodiment of the present application, after the video file of user's upload is received, can be according to consecutive frame in video file
The situation of change of content determines scene change key frame and the corresponding picture of the scene change frame is intercepted, then can lead to
The good machine learning model for picture classification is that the picture being truncated to is given a mark and sorted to cross training in advance, further according to row
Preset width number picture high for score is supplied to user as candidate's picture of video front cover by sequence, so that user is schemed from the candidate
The selection of video front cover is carried out in piece.With this, can both ensure not omitting all important scenes in video file, can reduce again carrying
For video front cover candidate's picture in picture multiplicity, lift the quality of candidate's picture, the user that is more convenient for therefrom chooses more
For the video front cover being suitable for.
Corresponding with the offer method of the video front cover provided in previous embodiment, the embodiment of the present application additionally provides one kind
The offer device of video front cover, referring to Fig. 4, the device can include：
Sectional drawing unit 41, for the video file that receive user is uploaded, and the change according to adjacent content frame in video file
Change situation determines scene change key frame and the corresponding picture of the scene change key frame is intercepted.
When implementing, the sectional drawing unit 41 can be specifically for：
Judge that whether adjacent two content frames change is beyond preset change threshold in video file；
Frame beyond preset change threshold is defined as scene change key frame；
The corresponding picture of scene change key frame is intercepted, and the picture being truncated to is constituted scene change key frame
Picture set.
Marking unit 42, for being the figure being truncated to by the good machine learning model for picture classification of training in advance
Piece is given a mark and is sorted.
Candidate's picture provide unit 43, for according to sequence using preset width number picture high for score as video front cover time
Picture is selected to be supplied to user, so that user carries out the selection of video front cover from candidate's picture.
Additionally, described device, may also include：
Instruction reception unit, for selection instruction of the receive user to any picture in candidate's picture；
Video front cover determining unit, the picture for selecting user are defined as the video front cover of video file.
In the present embodiment, the training to the machine learning model for picture classification used in the marking unit 42
Process, it may include following steps：
Step 1, determines the image data for machine learning model training.
When implementing, basic image data collection can be first obtained, the basic image data collection includes：Containing on user
First data set of blit piece and the second data set containing the picture for pressing the random intercepting of preset time interval.
Then, the color character parameter value that basic image data concentrates picture can be obtained, such as includes the color character
Parameter value includes tone value, intensity value, brightness value, rgb value etc., further according to the color character parameter value by basic picture number
According to concentrating the picture for not meeting prerequisite to remove, to obtain the image data for machine learning model training.
In one implementation, after the color character parameter value of picture being concentrated basic image data is obtained, the color
Color characteristic ginseng value may include HSV, and (Luminence (brightness) value, according to preset for Hue (tone), Saturation (saturation)
Color character weight, the color character parameter value of every width picture is done weighted sum calculating, to obtain the corresponding color of every width picture
Color character numerical value, then color character numerical value in first data set be less than the picture and described second of the first preset score value
In data set, color character numerical value is removed higher than the picture of the second preset score value, obtains first kind data set and the respectively
Two categorical data collection, using as the image data for machine learning model training.
In another kind of implementation, such as, can be in the color character parameter value for obtaining basic image data concentration picture
Afterwards, the color character parameter value may include Hue (tone) value, Saturation (saturation) value, RGB ((Red (red), Green
(green), Blue (indigo plant)) value, the picture and described second that tone value in first data set is less than the first preset hue threshold
In data set, tone value is removed higher than the picture of the second preset hue threshold, is next embezzled first data set again
It is higher than the second preset saturation with intensity value in picture and second data set of the angle value less than the first preset saturation threshold value
The picture of degree threshold value is removed.
Then, also the black and white picture in first data set can be removed according to the rgb value, that is, by
Chrominance information is not included in one data set (such as, in RGB, three component values are three component values in 0 or RGB and are 255 etc.)
Black and white picture be removed, further improve data in picture quality, reduce model training operand, reduce calculate when
Between, improving operational speed.
Finally, the picture remained in the first data set and the second data set is identified as first kind data
Collection and Second Type data set, using as the image data for machine learning model training.
Additionally, for the operand for further reducing model training, improving operational speed can also be according to preset color
Feature weight, before doing weighted sum calculating to the color character parameter value of every width picture, the size adjusting by every width picture is pre-
Size is put, the size that every width picture is adjusted to model needs.
Due to there may be the very high picture of some similarities in the first data set and the second data set, in order to improve number
According to the quality of intensive data, picture number is reduced, reduce the operand of model training, and then improving operational speed, can also incited somebody to action
In first data set, color score value less than color score value in the picture and second data set of the first preset score value is higher than
After the picture of the second preset score value is removed, to the similarity in the first data set and the second data set between remaining picture
Judged, and reached from similarity according to judged result and a width picture is chosen in the picture of preset similarity threshold protected
Stay, so as to using the picture remained in the first data set and the second data set as the first kind data set and the
Two categorical data collection, with this, are obtained that multiplicity is low, better quality data set.
Step 2, the image data is done repeatedly in the machine learning model of the good convolutional neural networks CNN of training in advance
Generation training, and the weights of convolutional neural networks are adjusted during repetitive exercise, with the good CNN machine learning moulds of training in advance
The CNN machine learning models for picture classification are obtained on the basis of type.
Wherein, the convolutional neural networks can be the convolutional neural networks of Inception-v3 definition.
Step 3, is estimated to the CNN machine learning models.
Step 4, if assessment pass through, training terminate and using the CNN machine learning models for picture classification as
The CNN machine learning models for picture classification for training；
Step 5, assessment do not pass through, then to being entered using the parameter of algorithm in the CNN machine learning models for picture classification
Row adjustment, to continue to do in the CNN machine learning models for picture classification by the image data after parameter adjustment
Iteration is instructed and adjusts the weights of convolutional neural networks during repetitive exercise, until the CNN machines for picture classification for obtaining
The assessment of device learning model passes through.
By the embodiment of the present application, after the video file of user's upload is received, can be according to consecutive frame in video file
The situation of change of content determines scene change key frame and the corresponding picture of the scene change frame is intercepted, then can lead to
The good machine learning model for picture classification is that the picture being truncated to is given a mark and sorted to cross training in advance, further according to row
Preset width number picture high for score is supplied to user as candidate's picture of video front cover by sequence, so that user is schemed from the candidate
The selection of video front cover is carried out in piece.With this, can both ensure not omitting all important scenes in video file, can reduce again carrying
For video front cover candidate's picture in picture multiplicity, lift the quality of candidate's picture, the user that is more convenient for therefrom chooses more
For the video front cover being suitable for.
As seen through the above description of the embodiments, those skilled in the art can be understood that the application can
Mode by software plus required general hardware platform is realizing.It is based on such understanding, the technical scheme essence of the application
On part that in other words prior art is contributed can be embodied in the form of software product, the computer software product
Can be stored in storage medium, such as ROM/RAM, magnetic disc, CD etc., use so that a computer equipment including some instructions
(can be personal computer, server, or network equipment etc.) executes some of each embodiment of the application or embodiment
Method described in part.
Each embodiment in this specification is described by the way of going forward one by one, identical similar portion between each embodiment
Divide mutually referring to what each embodiment was stressed is the difference with other embodiment.Especially for system or
For system embodiment, as which is substantially similar to embodiment of the method, so describing fairly simple, related part is referring to method
The part explanation of embodiment.System described above and system embodiment are only schematically wherein described conduct
Separating component explanation unit can be or may not be physically separate, as the part that unit shows can be or
Person may not be physical location, you can be located at a place, or can also be distributed on multiple NEs.Can be with root
Factually border need select some or all of module therein to realize the purpose of this embodiment scheme.Ordinary skill
Personnel are not in the case where creative work is paid, you can to understand and implement.
Offer method and device to video front cover provided herein, is described in detail above, herein should
The principle of the application and embodiment are set forth with specific case, the explanation of above example is only intended to help and manages
Solution the present processes and its core concept；Simultaneously for one of ordinary skill in the art, according to the thought of the application,
Will change in specific embodiment and range of application.In sum, this specification content is should not be construed as to this Shen
Restriction please.
Claims (12)
1. a kind of offer method of video front cover, it is characterised in that include：
The video file that receive user is uploaded, the situation of change according to adjacent content frame in video file determine that scene change is crucial
Frame is simultaneously intercepted to the corresponding picture of the scene change key frame；
It is that the picture being truncated to is given a mark and sorted by the good machine learning model for picture classification of training in advance；
According to sequence preset width number picture high for score is supplied to user as candidate's picture of video front cover, so as to user from
The selection of video front cover is carried out in candidate's picture.
2. method according to claim 1, it is characterised in that also include：
Selection instruction of the receive user to any picture in candidate's picture；
The picture that user selects is defined as video front cover.
3. method according to claim 1, it is characterised in that true according to the situation of change of adjacent content frame in video file
Determine scene change key frame and the corresponding picture of the scene change key frame is intercepted, including：
Judge that whether adjacent two content frames change is beyond preset change threshold in video file；
Frame beyond preset change threshold is defined as scene change key frame；
The corresponding picture of scene change key frame is intercepted, and the picture being truncated to is constituted scene change key frame picture
Set.
4. method according to claim 1, it is characterised in that the training to the machine learning model for picture classification,
Including：
Determine the image data for machine learning model training；
The image data is done repetitive exercise in the machine learning model of convolutional neural networks CNN, and in repetitive exercise mistake
The weights of convolutional neural networks are adjusted in journey, to obtain the CNN machines for picture classification on the basis of CNN machine learning models
Device learning model；
The CNN machine learning models for picture classification are estimated；
If assessment passes through, training terminates and using the CNN machine learning models for picture classification as the use for training
CNN machine learning models in picture classification.
5. method according to claim 4, it is characterised in that also include：
If assessment does not pass through, to being adjusted using the parameter of algorithm in the CNN machine learning models for picture classification, with
Just continue to do repetitive exercise in the CNN machine learning models for picture classification by the image data after parameter adjustment,
And the weights of convolutional neural networks are adjusted during repetitive exercise, until the CNN machine learning for picture classification that obtains
Model evaluation passes through.
6. method according to claim 4, it is characterised in that the determination is used for the picture number of machine learning model training
According to, including：
Obtain basic image data collection；
Obtain the color character parameter value that basic image data concentrates picture；
Basic image data is concentrated the picture for not meeting prerequisite remove according to the color character parameter value, to obtain use
Image data in machine learning model training.
7. method according to claim 6, it is characterised in that the basic image data collection includes：Upload containing user
First data set of picture and the second data set containing the picture for pressing the random intercepting of preset time interval；
The color character parameter value includes tone value, intensity value and brightness value；
Basic image data is concentrated the picture for not meeting prerequisite remove according to the color character parameter value, to obtain use
In machine learning model training image data, including：
According to preset color character weight, weighted sum calculating is done to the color character parameter value of every width picture, to obtain per width
The corresponding color character numerical value of picture；
By color in picture and second data set of the color character numerical value in first data set less than the first preset score value
Color character numerical value is removed higher than the picture of the second preset score value, obtains first kind data set and Second Type data respectively
Collection, using as the image data for machine learning model training.
8. method according to claim 6, it is characterised in that the basic image data collection includes：Upload containing user
First data set of picture and the second data set containing the picture for pressing the random intercepting of preset time interval；
The color character parameter value includes tone value, intensity value and rgb value；
Basic image data is concentrated the picture for not meeting prerequisite remove according to the color character parameter value, to obtain use
In machine learning model training image data, including：
By tone in picture and second data set of the tone value in first data set less than the first preset hue threshold
Value is removed higher than the picture of the second preset hue threshold；
By in picture and second data set of the intensity value in first data set less than the first preset saturation threshold value
Intensity value is removed higher than the picture of the second preset saturation threshold value；
The black and white picture in first data set is removed according to the rgb value；
The picture remained in first data set and the second data set is identified as first kind data set and second
Categorical data collection, using as the image data for machine learning model training.
9. method according to claim 7, it is characterised in that color character numerical value in first data set is being less than
In the picture of the first preset score value and second data set, color character numerical value is gone higher than the picture of the second preset score value
Remove afterwards, also include：
Respectively the similarity in the first data set and the second data set between remaining picture is judged, and according to judged result
Reach from similarity and a width picture is chosen in the picture of preset similarity threshold retained, so as to by the first data set and second
The picture remained in data set is respectively as the first kind data set and Second Type data set.
10. a kind of offer device of video front cover, it is characterised in that include：
Sectional drawing unit, for the video file that receive user is uploaded, and the situation of change according to adjacent content frame in video file
Determine scene change key frame and the corresponding picture of the scene change key frame is intercepted；
Marking unit, for being that the picture being truncated to is carried out by the good machine learning model for picture classification of training in advance
Give a mark and sort；
Candidate's picture provide unit, for according to sequence using preset width number picture high for score as video front cover candidate's picture
User is supplied to, so that user carries out the selection of video front cover from candidate's picture.
11. devices according to claim 10, it is characterised in that also include：
Instruction reception unit, for selection instruction of the receive user to any picture in candidate's picture；
Video front cover determining unit, the picture for selecting user are defined as video front cover.
12. devices according to claim 10, it is characterised in that the sectional drawing unit, specifically for：
Judge that whether adjacent two content frames change is beyond preset change threshold in video file；
Frame beyond preset change threshold is defined as scene change key frame；
The corresponding picture of scene change key frame is intercepted, and the picture being truncated to is constituted scene change key frame picture
Set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611059438.2A CN106503693B (en) | 2016-11-28 | 2016-11-28 | The providing method and device of video cover |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611059438.2A CN106503693B (en) | 2016-11-28 | 2016-11-28 | The providing method and device of video cover |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106503693A true CN106503693A (en) | 2017-03-15 |
| CN106503693B CN106503693B (en) | 2019-03-15 |
Family
ID=58327496
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611059438.2A Active CN106503693B (en) | 2016-11-28 | 2016-11-28 | The providing method and device of video cover |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106503693B (en) |
Cited By (45)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106993215A (en) * | 2017-03-31 | 2017-07-28 | 联想(北京)有限公司 | A kind of information processing method and processing routine device |
| CN107147939A (en) * | 2017-05-05 | 2017-09-08 | 百度在线网络技术（北京）有限公司 | Method and apparatus for adjusting net cast front cover |
| CN107221346A (en) * | 2017-05-25 | 2017-09-29 | 亮风台（上海）信息科技有限公司 | A kind of method and apparatus for the identification picture for being used to determine AR videos |
| CN107392252A (en) * | 2017-07-26 | 2017-11-24 | 上海城诗信息科技有限公司 | Computer deep learning characteristics of image and the method for quantifying perceptibility |
| CN107506736A (en) * | 2017-08-29 | 2017-12-22 | 北京大生在线科技有限公司 | Online education video fineness picture intercept method based on deep learning |
| CN107707967A (en) * | 2017-09-30 | 2018-02-16 | 咪咕视讯科技有限公司 | The determination method, apparatus and computer-readable recording medium of a kind of video file front cover |
| CN107832724A (en) * | 2017-11-17 | 2018-03-23 | 北京奇虎科技有限公司 | The method and device of personage's key frame is extracted from video file |
| CN107832725A (en) * | 2017-11-17 | 2018-03-23 | 北京奇虎科技有限公司 | Video front cover extracting method and device based on evaluation index |
| CN107918656A (en) * | 2017-11-17 | 2018-04-17 | 北京奇虎科技有限公司 | Video front cover extracting method and device based on video title |
| CN107958030A (en) * | 2017-11-17 | 2018-04-24 | 北京奇虎科技有限公司 | Video front cover recommended models optimization method and device |
| CN108377417A (en) * | 2018-01-17 | 2018-08-07 | 百度在线网络技术（北京）有限公司 | Video reviewing method, device, computer equipment and storage medium |
| CN108600781A (en) * | 2018-05-21 | 2018-09-28 | 腾讯科技（深圳）有限公司 | A kind of method and server of the generation of video cover |
| CN108650524A (en) * | 2018-05-23 | 2018-10-12 | 腾讯科技（深圳）有限公司 | Video cover generation method, device, computer equipment and storage medium |
| CN108665769A (en) * | 2018-05-11 | 2018-10-16 | 深圳市鹰硕技术有限公司 | Network teaching method based on convolutional neural networks and device |
| CN108833942A (en) * | 2018-06-28 | 2018-11-16 | 北京达佳互联信息技术有限公司 | Video cover choosing method, device, computer equipment and storage medium |
| CN108965922A (en) * | 2018-08-22 | 2018-12-07 | 广州酷狗计算机科技有限公司 | Video cover generation method, device and storage medium |
| CN108985176A (en) * | 2018-06-20 | 2018-12-11 | 北京优酷科技有限公司 | image generating method and device |
| CN109002812A (en) * | 2018-08-08 | 2018-12-14 | 北京未来媒体科技股份有限公司 | A kind of method and device of intelligent recognition video cover |
| CN109145138A (en) * | 2018-09-10 | 2019-01-04 | 北京点网聚科技有限公司 | A kind of cover choosing method, device, electronic equipment and storage medium |
| CN109257645A (en) * | 2018-09-11 | 2019-01-22 | 传线网络科技（上海）有限公司 | Video cover generation method and device |
| CN109271542A (en) * | 2018-09-28 | 2019-01-25 | 百度在线网络技术（北京）有限公司 | Cover determines method, apparatus, equipment and readable storage medium storing program for executing |
| CN109996091A (en) * | 2019-03-28 | 2019-07-09 | 苏州八叉树智能科技有限公司 | Generate method, apparatus, electronic equipment and the computer readable storage medium of video cover |
| CN110008364A (en) * | 2019-03-25 | 2019-07-12 | 联想(北京)有限公司 | Image processing method, device and system |
| CN110069651A (en) * | 2017-10-23 | 2019-07-30 | 腾讯科技（北京）有限公司 | Picture screening technique and device, storage medium |
| CN110134651A (en) * | 2019-05-09 | 2019-08-16 | 北京金山安全软件有限公司 | Information file processing method and device, electronic equipment and storage medium |
| CN110263213A (en) * | 2019-05-22 | 2019-09-20 | 腾讯科技（深圳）有限公司 | Video pushing method, device, computer equipment and storage medium |
| CN110392306A (en) * | 2019-07-29 | 2019-10-29 | 腾讯科技（深圳）有限公司 | A kind of data processing method and equipment |
| CN110677734A (en) * | 2019-09-30 | 2020-01-10 | 北京达佳互联信息技术有限公司 | Video synthesis method and device, electronic equipment and storage medium |
| CN110881131A (en) * | 2018-09-06 | 2020-03-13 | 武汉斗鱼网络科技有限公司 | Classification method of live review videos and related device thereof |
| WO2020052084A1 (en) * | 2018-09-13 | 2020-03-19 | 北京字节跳动网络技术有限公司 | Video cover selection method, device and computer-readable storage medium |
| CN110909205A (en) * | 2019-11-22 | 2020-03-24 | 北京金山云网络技术有限公司 | Video cover determination method and device, electronic equipment and readable storage medium |
| CN110995999A (en) * | 2019-12-12 | 2020-04-10 | 北京小米智能科技有限公司 | Dynamic photo shooting method and device |
| CN110991373A (en) * | 2019-12-09 | 2020-04-10 | 北京字节跳动网络技术有限公司 | Image processing method, image processing apparatus, electronic device, and medium |
| CN111090778A (en) * | 2019-12-26 | 2020-05-01 | 北京百度网讯科技有限公司 | Picture generation method, device, equipment and storage medium |
| CN111182295A (en) * | 2020-01-06 | 2020-05-19 | 腾讯科技（深圳）有限公司 | Video data processing method, device, equipment and readable storage medium |
| WO2020156171A1 (en) * | 2019-01-29 | 2020-08-06 | 广州市百果园信息技术有限公司 | Video publishing method, apparatus and device, and storage medium |
| CN111523400A (en) * | 2020-03-31 | 2020-08-11 | 易视腾科技股份有限公司 | Video representative frame extraction method and device |
| CN111581435A (en) * | 2020-05-25 | 2020-08-25 | 北京达佳互联信息技术有限公司 | Video cover image generation method and device, electronic equipment and storage medium |
| CN111651633A (en) * | 2020-04-29 | 2020-09-11 | 上海推乐信息技术服务有限公司 | Video cover selection method and system |
| CN111901679A (en) * | 2020-08-10 | 2020-11-06 | 广州繁星互娱信息科技有限公司 | Method and device for determining cover image, computer equipment and readable storage medium |
| CN112004108A (en) * | 2020-08-26 | 2020-11-27 | 深圳创维－Rgb电子有限公司 | Live video recording processing method and device, intelligent terminal and storage medium |
| CN113286171A (en) * | 2020-12-11 | 2021-08-20 | 苏州律点信息科技有限公司 | Video cover determination method and device and cloud server |
| CN113656642A (en) * | 2021-08-20 | 2021-11-16 | 北京百度网讯科技有限公司 | Cover image generation method, device, equipment, storage medium and program product |
| CN114007133A (en) * | 2021-10-25 | 2022-02-01 | 杭州当虹科技股份有限公司 | Video playing start cover automatic generation method and device based on video playing |
| CN115802146A (en) * | 2021-09-07 | 2023-03-14 | 荣耀终端有限公司 | Method for snapping image in video and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104244024A (en) * | 2014-09-26 | 2014-12-24 | 北京金山安全软件有限公司 | Video cover generation method and device and terminal |
| CN105323634A (en) * | 2014-06-27 | 2016-02-10 | Tcl集团股份有限公司 | Method and system for generating thumbnail of video |
| CN106101868A (en) * | 2016-07-18 | 2016-11-09 | 乐视控股（北京）有限公司 | Reduced graph generating method and generating means |
-
2016
- 2016-11-28 CN CN201611059438.2A patent/CN106503693B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105323634A (en) * | 2014-06-27 | 2016-02-10 | Tcl集团股份有限公司 | Method and system for generating thumbnail of video |
| CN104244024A (en) * | 2014-09-26 | 2014-12-24 | 北京金山安全软件有限公司 | Video cover generation method and device and terminal |
| CN106101868A (en) * | 2016-07-18 | 2016-11-09 | 乐视控股（北京）有限公司 | Reduced graph generating method and generating means |
Non-Patent Citations (3)
| Title |
|---|
| WEIGANG ZHANG ET AL: "Web video thumbnail recommendation with content-aware analysis and query-sensitive matching", 《MULTIMEDIA TOOLS AND APPLICATIONS》 * |
| WEILONG YANG ET AL: "Improving YouTube video thumbnails with deep neural nets", 《GOOGLE AI BLOG HTTPS://AI.GOOGLEBLOG.COM/2015/10/IMPROVING-YOUTUBE-VIDEO-THUMBNAILS-WITH.HTML》 * |
| 黄东军 等: "基于HSV空间模型的视频场景切分", 《计算机技术与发展》 * |
Cited By (62)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106993215A (en) * | 2017-03-31 | 2017-07-28 | 联想(北京)有限公司 | A kind of information processing method and processing routine device |
| CN107147939A (en) * | 2017-05-05 | 2017-09-08 | 百度在线网络技术（北京）有限公司 | Method and apparatus for adjusting net cast front cover |
| CN107221346A (en) * | 2017-05-25 | 2017-09-29 | 亮风台（上海）信息科技有限公司 | A kind of method and apparatus for the identification picture for being used to determine AR videos |
| CN107392252A (en) * | 2017-07-26 | 2017-11-24 | 上海城诗信息科技有限公司 | Computer deep learning characteristics of image and the method for quantifying perceptibility |
| CN107506736A (en) * | 2017-08-29 | 2017-12-22 | 北京大生在线科技有限公司 | Online education video fineness picture intercept method based on deep learning |
| CN107707967A (en) * | 2017-09-30 | 2018-02-16 | 咪咕视讯科技有限公司 | The determination method, apparatus and computer-readable recording medium of a kind of video file front cover |
| CN110069651B (en) * | 2017-10-23 | 2023-04-07 | 腾讯科技（北京）有限公司 | Picture screening method and device and storage medium |
| CN110069651A (en) * | 2017-10-23 | 2019-07-30 | 腾讯科技（北京）有限公司 | Picture screening technique and device, storage medium |
| CN107918656A (en) * | 2017-11-17 | 2018-04-17 | 北京奇虎科技有限公司 | Video front cover extracting method and device based on video title |
| CN107958030A (en) * | 2017-11-17 | 2018-04-24 | 北京奇虎科技有限公司 | Video front cover recommended models optimization method and device |
| CN107832725A (en) * | 2017-11-17 | 2018-03-23 | 北京奇虎科技有限公司 | Video front cover extracting method and device based on evaluation index |
| CN107958030B (en) * | 2017-11-17 | 2021-08-24 | 北京奇虎科技有限公司 | Video cover recommendation model optimization method and device |
| CN107832724A (en) * | 2017-11-17 | 2018-03-23 | 北京奇虎科技有限公司 | The method and device of personage's key frame is extracted from video file |
| CN108377417A (en) * | 2018-01-17 | 2018-08-07 | 百度在线网络技术（北京）有限公司 | Video reviewing method, device, computer equipment and storage medium |
| CN108377417B (en) * | 2018-01-17 | 2019-11-26 | 百度在线网络技术（北京）有限公司 | Video reviewing method, device, computer equipment and storage medium |
| CN108665769A (en) * | 2018-05-11 | 2018-10-16 | 深圳市鹰硕技术有限公司 | Network teaching method based on convolutional neural networks and device |
| CN108600781A (en) * | 2018-05-21 | 2018-09-28 | 腾讯科技（深圳）有限公司 | A kind of method and server of the generation of video cover |
| CN108650524A (en) * | 2018-05-23 | 2018-10-12 | 腾讯科技（深圳）有限公司 | Video cover generation method, device, computer equipment and storage medium |
| CN108650524B (en) * | 2018-05-23 | 2022-08-16 | 腾讯科技（深圳）有限公司 | Video cover generation method and device, computer equipment and storage medium |
| CN108985176A (en) * | 2018-06-20 | 2018-12-11 | 北京优酷科技有限公司 | image generating method and device |
| CN108833942A (en) * | 2018-06-28 | 2018-11-16 | 北京达佳互联信息技术有限公司 | Video cover choosing method, device, computer equipment and storage medium |
| CN109002812A (en) * | 2018-08-08 | 2018-12-14 | 北京未来媒体科技股份有限公司 | A kind of method and device of intelligent recognition video cover |
| CN108965922A (en) * | 2018-08-22 | 2018-12-07 | 广州酷狗计算机科技有限公司 | Video cover generation method, device and storage medium |
| CN110881131B (en) * | 2018-09-06 | 2021-07-23 | 武汉斗鱼网络科技有限公司 | Classification method of live review videos and related device thereof |
| CN110881131A (en) * | 2018-09-06 | 2020-03-13 | 武汉斗鱼网络科技有限公司 | Classification method of live review videos and related device thereof |
| CN109145138A (en) * | 2018-09-10 | 2019-01-04 | 北京点网聚科技有限公司 | A kind of cover choosing method, device, electronic equipment and storage medium |
| CN109257645B (en) * | 2018-09-11 | 2021-11-02 | 阿里巴巴（中国）有限公司 | Video cover generation method and device |
| CN109257645A (en) * | 2018-09-11 | 2019-01-22 | 传线网络科技（上海）有限公司 | Video cover generation method and device |
| WO2020052084A1 (en) * | 2018-09-13 | 2020-03-19 | 北京字节跳动网络技术有限公司 | Video cover selection method, device and computer-readable storage medium |
| CN109271542A (en) * | 2018-09-28 | 2019-01-25 | 百度在线网络技术（北京）有限公司 | Cover determines method, apparatus, equipment and readable storage medium storing program for executing |
| WO2020156171A1 (en) * | 2019-01-29 | 2020-08-06 | 广州市百果园信息技术有限公司 | Video publishing method, apparatus and device, and storage medium |
| CN110008364A (en) * | 2019-03-25 | 2019-07-12 | 联想(北京)有限公司 | Image processing method, device and system |
| CN109996091A (en) * | 2019-03-28 | 2019-07-09 | 苏州八叉树智能科技有限公司 | Generate method, apparatus, electronic equipment and the computer readable storage medium of video cover |
| CN110134651A (en) * | 2019-05-09 | 2019-08-16 | 北京金山安全软件有限公司 | Information file processing method and device, electronic equipment and storage medium |
| CN110134651B (en) * | 2019-05-09 | 2021-10-26 | 北京金山安全软件有限公司 | Information file processing method and device, electronic equipment and storage medium |
| CN110263213A (en) * | 2019-05-22 | 2019-09-20 | 腾讯科技（深圳）有限公司 | Video pushing method, device, computer equipment and storage medium |
| CN110263213B (en) * | 2019-05-22 | 2023-07-18 | 腾讯科技（深圳）有限公司 | Video pushing method, device, computer equipment and storage medium |
| CN110392306B (en) * | 2019-07-29 | 2021-11-05 | 腾讯科技（深圳）有限公司 | Data processing method and equipment |
| CN110392306A (en) * | 2019-07-29 | 2019-10-29 | 腾讯科技（深圳）有限公司 | A kind of data processing method and equipment |
| CN110677734A (en) * | 2019-09-30 | 2020-01-10 | 北京达佳互联信息技术有限公司 | Video synthesis method and device, electronic equipment and storage medium |
| CN110909205A (en) * | 2019-11-22 | 2020-03-24 | 北京金山云网络技术有限公司 | Video cover determination method and device, electronic equipment and readable storage medium |
| CN110909205B (en) * | 2019-11-22 | 2023-04-07 | 北京金山云网络技术有限公司 | Video cover determination method and device, electronic equipment and readable storage medium |
| CN110991373A (en) * | 2019-12-09 | 2020-04-10 | 北京字节跳动网络技术有限公司 | Image processing method, image processing apparatus, electronic device, and medium |
| CN110995999A (en) * | 2019-12-12 | 2020-04-10 | 北京小米智能科技有限公司 | Dynamic photo shooting method and device |
| CN111090778B (en) * | 2019-12-26 | 2023-06-27 | 北京百度网讯科技有限公司 | Picture generation method, device, equipment and storage medium |
| CN111090778A (en) * | 2019-12-26 | 2020-05-01 | 北京百度网讯科技有限公司 | Picture generation method, device, equipment and storage medium |
| CN111182295A (en) * | 2020-01-06 | 2020-05-19 | 腾讯科技（深圳）有限公司 | Video data processing method, device, equipment and readable storage medium |
| CN111182295B (en) * | 2020-01-06 | 2023-08-25 | 腾讯科技（深圳）有限公司 | Video data processing method, device, equipment and readable storage medium |
| CN111523400A (en) * | 2020-03-31 | 2020-08-11 | 易视腾科技股份有限公司 | Video representative frame extraction method and device |
| CN111523400B (en) * | 2020-03-31 | 2023-10-13 | 易视腾科技股份有限公司 | Video representative frame extraction method and device |
| CN111651633A (en) * | 2020-04-29 | 2020-09-11 | 上海推乐信息技术服务有限公司 | Video cover selection method and system |
| CN111581435B (en) * | 2020-05-25 | 2023-12-01 | 北京达佳互联信息技术有限公司 | Video cover image generation method and device, electronic equipment and storage medium |
| CN111581435A (en) * | 2020-05-25 | 2020-08-25 | 北京达佳互联信息技术有限公司 | Video cover image generation method and device, electronic equipment and storage medium |
| CN111901679A (en) * | 2020-08-10 | 2020-11-06 | 广州繁星互娱信息科技有限公司 | Method and device for determining cover image, computer equipment and readable storage medium |
| CN112004108A (en) * | 2020-08-26 | 2020-11-27 | 深圳创维－Rgb电子有限公司 | Live video recording processing method and device, intelligent terminal and storage medium |
| CN113286171A (en) * | 2020-12-11 | 2021-08-20 | 苏州律点信息科技有限公司 | Video cover determination method and device and cloud server |
| CN113656642A (en) * | 2021-08-20 | 2021-11-16 | 北京百度网讯科技有限公司 | Cover image generation method, device, equipment, storage medium and program product |
| CN113656642B (en) * | 2021-08-20 | 2024-05-28 | 北京百度网讯科技有限公司 | Cover image generation method, device, apparatus, storage medium and program product |
| CN115802146A (en) * | 2021-09-07 | 2023-03-14 | 荣耀终端有限公司 | Method for snapping image in video and electronic equipment |
| CN115802146B (en) * | 2021-09-07 | 2024-04-02 | 荣耀终端有限公司 | Method for capturing images in video and electronic equipment |
| CN114007133A (en) * | 2021-10-25 | 2022-02-01 | 杭州当虹科技股份有限公司 | Video playing start cover automatic generation method and device based on video playing |
| CN114007133B (en) * | 2021-10-25 | 2024-02-23 | 杭州当虹科技股份有限公司 | Video playing cover automatic generation method and device based on video playing |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106503693B (en) | 2019-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106503693A (en) | The offer method and device of video front cover | |
| US8897604B2 (en) | Image quality analysis for searches | |
| Redi et al. | The beauty of capturing faces: Rating the quality of digital portraits | |
| CN107808132A (en) | A kind of scene image classification method for merging topic model | |
| CN106709453A (en) | Sports video key posture extraction method based on deep learning | |
| CN109948566A (en) | A kind of anti-fraud detection method of double-current face based on weight fusion and feature selecting | |
| CN107330360A (en) | A kind of pedestrian's clothing colour recognition, pedestrian retrieval method and device | |
| El Khoury et al. | Color and sharpness assessment of single image dehazing | |
| CN109472193A (en) | Method for detecting human face and device | |
| CN103164687A (en) | Method and system for detecting pornographic image | |
| Mancas et al. | A rarity-based visual attention map-application to texture description | |
| CN108647696B (en) | Picture color value determining method and device, electronic equipment and storage medium | |
| CN106910195A (en) | A kind of web page layout monitoring method and device | |
| CN110889718B (en) | Scheme screening method, scheme screening device, medium and electronic equipment | |
| Wu et al. | Automatic colorization of Chinese ink painting combining multi-level features and generative adversarial networks | |
| CN109741315B (en) | Non-reference image quality objective evaluation method based on deep reinforcement learning | |
| JP7336033B2 (en) | Data augmentation-based matter analysis model learning device and method | |
| CN114845158A (en) | Video cover generation method, video publishing method and related equipment | |
| CN111651633A (en) | Video cover selection method and system | |
| JP7362924B2 (en) | Data augmentation-based spatial analysis model learning device and method | |
| Yuan et al. | Color image quality assessment with multi deep convolutional networks | |
| CN112115824B (en) | Fruit and vegetable detection method, fruit and vegetable detection device, electronic equipment and computer readable medium | |
| US20210374916A1 (en) | Storage medium storing program, image processing apparatus, and training method of machine learning model | |
| Hu et al. | Pedestrian traffic light detection in complex scene using adaboost with multi-layer features | |
| CN111950565B (en) | Abstract picture image direction identification method based on feature fusion and naive Bayes |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP03 | Change of name, title or address | ||
| CP03 | Change of name, title or address |
Address after: Room 222, Floor 2, Building 1, Yard A23, North Third Ring West Road, Haidian District, Beijing 100098 Patentee after: Beijing Douyin Information Service Co.,Ltd. Address before: 100098 Building 1, AVIC Plaza, No. 43, North Third Ring West Road, Haidian District, Beijing Patentee before: BEIJING BYTEDANCE TECHNOLOGY Co.,Ltd. |