CN107194419A - Video classification methods and device, computer equipment and computer-readable recording medium - Google Patents

Video classification methods and device, computer equipment and computer-readable recording medium Download PDF

Info

Publication number
CN107194419A
CN107194419A CN201710325322.7A CN201710325322A CN107194419A CN 107194419 A CN107194419 A CN 107194419A CN 201710325322 A CN201710325322 A CN 201710325322A CN 107194419 A CN107194419 A CN 107194419A
Authority
CN
China
Prior art keywords
picture
classification
training
pre
probability
Prior art date
Application number
CN201710325322.7A
Other languages
Chinese (zh)
Inventor
常鑫
Original Assignee
百度在线网络技术(北京)有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 百度在线网络技术(北京)有限公司 filed Critical 百度在线网络技术(北京)有限公司
Priority to CN201710325322.7A priority Critical patent/CN107194419A/en
Publication of CN107194419A publication Critical patent/CN107194419A/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6217Design or setup of recognition systems and techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
    • G06K9/6256Obtaining sets of training patterns; Bootstrap methods, e.g. bagging, boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6267Classification techniques

Abstract

The present invention provides a kind of video classification methods and device, computer equipment and computer-readable recording medium.Its methods described includes:Multiframe picture is obtained from target video, the picture of each frame is identified according to the picture classification model of training in advance, the probability of the corresponding each pre-set categories of picture of each frame is predicted;According to the probability of the corresponding each pre-set categories of the picture of each frame, the target classification of target video is obtained.Technical scheme, the poor technical problem of accuracy that the reply for being directly based upon video title and video in the prior art can be overcome to classify video content, pass through the probability of the corresponding each pre-set categories of multiframe picture in the picture classification model prediction target video based on training in advance, so as to obtain the target classification of target video, the accuracy classified to target video can be effectively improved.

Description

Video classification methods and device, computer equipment and computer-readable recording medium

【Technical field】

The present invention relates to field of computer technology, more particularly to a kind of video classification methods and device, computer equipment with Computer-readable recording medium.

【Background technology】

With the rapid growth of internet, it has turned into the important channel that user obtains information.But at the same time, information Blast also generates many users and not wanted to see that or uninterested content.Especially in recent years, shown according to the interest of user Commending contents have been increasingly becoming the basic function of internet product.

For example, in existing video field in the suggested design of video, can according to video title and user to regarding Reply of frequency etc., classifies to video.Then the classification based on video recommends video to user again.But prior art In, the topmost source of editor of video title and video replies is all domestic consumer.The video replies that domestic consumer edits With very strong randomness, and video title is more the mood of prominent publisher, such as too severe, makes laughs very much it Class, so that cause the reply based on video title and video effectively can not classify to video content, or even meeting Utilized by some vulgar video content producers.

Therefore, in the prior art, the accuracy that the reply based on video title and video is classified to video content It is poor.

【The content of the invention】

The invention provides a kind of video classification methods and device, computer equipment and computer-readable recording medium, for improving video The accuracy of classification.

The present invention provides a kind of video classification methods, and methods described includes:

Multiframe picture is obtained from target video;

The picture of each frame is identified according to the picture classification model of training in advance, the described of each frame is predicted The probability of the corresponding each pre-set categories of picture;

According to the probability of the corresponding each pre-set categories of the picture of each frame, the target classification of the target video is obtained.

Still optionally further, in method as described above, according to the general of the corresponding each pre-set categories of the picture of each frame Rate, obtains the target classification of the target video, specifically includes:

According to the probability of the corresponding each pre-set categories of the picture of each frame, from the corresponding default class of the picture of each frame Not the maximum pre-set categories of middle acquisition probability as the target video target classification;

Or the probability of the corresponding each pre-set categories of the picture according to each frame, to the default class of the picture of each frame Other probability is weighted processing, obtains the probability of the corresponding each pre-set categories of the target video;From the target video pair In each pre-set categories answered the maximum pre-set categories of acquisition probability as the target video target classification.

Still optionally further, in method as described above, according to the picture classification model of training in advance to each frame The picture is identified, before the probability of the corresponding each pre-set categories of the picture for predicting each frame, in addition to:

Gather the training picture of several known class, generation picture training storehouse;

Using several Zhang Xunlian pictures in the picture training storehouse, the picture classification model is trained.

Still optionally further, in method as described above, using several Zhang Xunlian pictures in the picture training storehouse, The picture classification model is trained, is specifically included:

Each Zhang Suoshu training pictures are inputted into the picture classification model successively so that the picture classification model is defeated Go out probability of the corresponding training picture in each pre-set categories;

According to probability of the corresponding training picture in each pre-set categories, the prediction class of the training picture is determined Not;

Detect whether the known class of the corresponding training picture is consistent with the prediction classification of the training picture;

When the known class of the correspondence training picture and the inconsistent prediction classification of the training picture, adjustment is described The parameter of picture classification model so that the known class of the training picture is consistent with the prediction classification of the training picture;

Above-mentioned steps are repeated, until several Zhang Xunlian pictures training are finished, and the known class of the training picture It is consistent not with the prediction classification of the training picture, the parameter of the picture classification model is determined, so that it is determined that the picture point Class model.

Still optionally further, in method as described above, according to the general of the corresponding each pre-set categories of the picture of each frame After rate, the target classification for obtaining the target video, in addition to:

User classification interested is detected according to the historical behavior of user;

Whether detect in user classification interested includes the target classification;

If including recommending the target video to the user.

The present invention also provides a kind of visual classification device, and described device includes:

Picture acquisition module, for obtaining multiframe picture from target video;

Prediction module, the picture of each frame is identified for the picture classification model according to training in advance, Predict the probability of the corresponding each pre-set categories of the picture of each frame;

Classification acquisition module, for the probability of the corresponding each pre-set categories of the picture according to each frame, obtains the mesh Mark the target classification of video.

Still optionally further, in device as described above, the classification acquisition module, specifically for:

According to the probability of the corresponding each pre-set categories of the picture of each frame, from the corresponding default class of the picture of each frame Not the maximum pre-set categories of middle acquisition probability as the target video target classification;

Or the probability of the corresponding each pre-set categories of the picture according to each frame, to the default class of the picture of each frame Other probability is weighted processing, obtains the probability of the corresponding each pre-set categories of the target video;From the target video pair In each pre-set categories answered the maximum pre-set categories of acquisition probability as the target video target classification.

Still optionally further, in device as described above, in addition to:

Acquisition module, the training picture for gathering several known class, generation picture training storehouse;

Training module, for using several Zhang Xunlian pictures in the picture training storehouse, training the picture classification Model.

Still optionally further, in device as described above, the training module, specifically for:

Each Zhang Suoshu training pictures are inputted into the picture classification model successively so that the picture classification model is defeated Go out probability of the corresponding training picture in each pre-set categories;

According to probability of the corresponding training picture in each pre-set categories, the prediction class of the training picture is determined Not;

Detect whether the known class of the corresponding training picture is consistent with the prediction classification of the training picture;

When the known class of the correspondence training picture and the inconsistent prediction classification of the training picture, adjustment is described The parameter of picture classification model so that the known class of the training picture is consistent with the prediction classification of the training picture;

Above-mentioned steps are repeated, until several Zhang Xunlian pictures training are finished, and the known class of the training picture It is consistent not with the prediction classification of the training picture, the parameter of the picture classification model is determined, so that it is determined that the picture point Class model.

Still optionally further, in device as described above, in addition to:

Detection module, for detecting user classification interested according to the historical behavior of user;

Whether the detection module, being additionally operable to detect in the classification that the user is interested includes the target classification;

Recommending module, if detecting that user classification interested includes the target class for the detection module Not, the target video is recommended to the user.

The present invention also provides a kind of computer equipment, and the equipment includes:

One or more processors;

Memory, for storing one or more programs,

When one or more of programs are by one or more of computing devices so that one or more of processing Device realizes video classification methods as described above.

The present invention also provides a kind of computer-readable medium, is stored thereon with computer program, the program is held by processor Video classification methods as described above are realized during row.

The video classification methods and device, computer equipment and computer-readable recording medium of the present invention, by being obtained from target video Multiframe picture, the picture of each frame is identified according to the picture classification model of training in advance, predicts that the picture of each frame is corresponding The probability of each pre-set categories;According to the probability of the corresponding each pre-set categories of the picture of each frame, the target classification of target video is obtained. Technical scheme, can overcome the reply for being directly based upon video title and video in the prior art to enter video content The poor technical problem of the capable accuracy classified, by many in the picture classification model prediction target video based on training in advance The probability of the corresponding each pre-set categories of frame picture, so as to obtain the target classification of target video, can be effectively improved to target The accuracy of visual classification.

【Brief description of the drawings】

Fig. 1 is the flow chart of the video classification methods embodiment one of the present invention.

Fig. 2 is the flow chart of the video classification methods embodiment two of the present invention.

Fig. 3 is the structure chart of the visual classification device embodiment one of the present invention.

Fig. 4 is the structure chart of the visual classification device embodiment two of the present invention.

Fig. 5 is the structure chart of the computer equipment embodiment of the present invention.

A kind of exemplary plot for computer equipment that Fig. 6 provides for the present invention.

【Embodiment】

In order that the object, technical solutions and advantages of the present invention are clearer, below in conjunction with the accompanying drawings with specific embodiment pair The present invention is described in detail.

Fig. 1 is the flow chart of the video classification methods embodiment one of the present invention.As shown in figure 1, the video of the present embodiment point Class method, specifically may include steps of:

100th, multiframe picture is obtained from target video;

The executive agent of the video classification methods of the present embodiment is visual classification device, the visual classification device of the present embodiment It can be an electronic entity device, it would however also be possible to employ the device of Integrated Simulation.

In the video classification methods of the present embodiment, multiframe picture can be obtained from target video first.Such as multiframe Picture is not preferably continuous multiframe, such as the time difference of the corresponding time frame of any two frames picture needs to be more than in multiframe picture Regular hour length threshold, to ensure the picture of the discontinuous frame of two frames picture so that the selection of each frame picture is all tried one's best Do not influenceed by former frame picture, be all independent, so as to ensure the accuracy of visual classification.Multiframe picture in the present embodiment Quantity, can set according to demand within 10 or more than 10, or even tens.

101st, the picture of each frame is identified according to the picture classification model of training in advance, predicts the picture correspondence of each frame Each pre-set categories probability;

Which the picture classification model of the training in advance of the present embodiment can belong to according to each frame picture to each frame picture The probability of individual pre-set categories is predicted.Picture classification model in the present embodiment needs to pre-set what picture can be classified Pre-set categories, pre-set categories here can only include the classification of one layer of classification;Category can also not only be included, also simultaneously Including two grades of classifications.For the ease of representing, the picture classification model of the present embodiment is according to each frame picture, the frame picture institute of prediction The probability for belonging to classification can be using vectorial form.The vectorial dimension is equal to the quantity of all pre-set categories pre-set, The numerical value of each element is equal to the probable value of the corresponding pre-set categories in the position in the vector.For example pre-set the picture energy The quantity for the pre-set categories being enough classified has 30, then in prediction, picture classification model can predict the probability of certain frame picture For one 30 dimension vector, vector in each element numerical value such as p18For 0.3, if it is corresponding at the 18th position in vector Pre-set categories are amusement, p18For the classification of the 0.3 expression frame picture, to belong to the probability of amusement be 0.3.

102nd, according to the probability of the corresponding each pre-set categories of picture of each frame, the target classification of target video is obtained.

According to above-mentioned steps, the probability of the affiliated pre-set categories of each frame obtained in video can be obtained, then can be integrated Consider that the picture of each frame obtained in video belongs to the probability of each pre-set categories, obtain the target classification of target video.

For example the step 102 " according to the probability of the corresponding each pre-set categories of the picture of each frame, obtains the target of target video Classification ", can specifically include the following two kinds mode:

First way:It is corresponding pre- from the picture of each frame according to the probability of the corresponding each pre-set categories of the picture of each frame If in classification the maximum pre-set categories of acquisition probability as target video target classification;

In the embodiment, the probability of each pre-set categories belonging to each frame picture is set out to come, probability is therefrom taken most Big pre-set categories as target video target classification.For example, pre-set categories have 20, the first frame picture is corresponding each default The probability of classification is respectively P1,1=0.1, P1,2=0.3 ..., P1,10=0.3 ... P1,19=0.2, P1,20=0.1.P1,1= The probability that 0.1 the first frame picture of expression belongs to the 1st pre-set categories is 0.1, P1,2=0.3 the first frame picture of expression, which belongs to the 2nd, to be preset The probability of classification is 0.3, P1,10The probability that=0.3 the first frame picture of expression belongs to the 10th pre-set categories is 0.3, P1,19=0.2 table Show that the probability that the first frame picture belongs to the 19th pre-set categories is 0.2, P1,20=0.1 the first frame picture of expression belongs to the 20th default class Other probability is 0.1;And the probability that the 1st frame picture belongs to other pre-set categories is 0.Similarly, the second frame picture of correspondence correspondence The probability of each pre-set categories be respectively P2,1=0.2, P2,2=0.6 ..., P2,11=0.1 ... P2,19=0.1.P2,1= The probability that 0.2 the second frame picture of expression belongs to the 1st pre-set categories is 0.2, P2,2=0.6 the 2nd frame picture of expression, which belongs to the 2nd, to be preset The probability of classification is 0.6, P2,11The probability that=0.1 the 2nd frame picture of expression belongs to the 11st pre-set categories is 0.1, P2,19=0.1 table Show that the probability that the 2nd frame picture belongs to the 19th pre-set categories is 0.1.If also obtaining other frame pictures from the target video, Step 101 is all used for each frame picture, the probability of the corresponding each pre-set categories of picture of the frame can be got.Specifically The maximum of the probability of the corresponding pre-set categories of each frame picture can be obtained;Then by the maximum probability value of all frame pictures Corresponding pre-set categories as the target video target classification.For example, in the present embodiment, if other frame pictures are corresponding general Rate maximum is both less than 0.6, then it is considered that the corresponding classification of probability 0.6 is the corresponding target classification of the target video, i.e., It is considered that the target classification of the target video is the 2nd pre-set categories.

The second way is:According to the probability of the corresponding each pre-set categories of the picture of each frame, to the default of the picture of each frame The probability of classification is weighted processing, obtains the probability of the corresponding each pre-set categories of target video;It is corresponding each from target video In pre-set categories the maximum pre-set categories of acquisition probability as target video target classification.

The technical scheme of the present embodiment is right after the probability for the corresponding each pre-set categories of picture for getting each frame The probability of the pre-set categories of the picture of each frame is weighted processing.If, target video gets 3 frame pictures.Similarly, the first frame The probability of the corresponding each pre-set categories of picture is respectively P1,1=0.1, P1,2=0.3 ..., P1,10=0.3 ... P1,19= 0.2、P1,20=0.1;Remaining pre-set categories probability is 0.The probability of the corresponding each pre-set categories of second frame picture is respectively P2,1= 0.2、P2,2=0.6 ..., P2,11=0.1 ... P2,19=0.1;Remaining pre-set categories probability is 0.3rd frame picture is corresponding The probability of each pre-set categories is respectively P3,1=0.7, P3,10=0.2, P3,20=0.1;Remaining pre-set categories probability is 0.By each frame The probability of pre-set categories of picture be weighted after processing, the probability that target video belongs to the 1st pre-set categories is P1=P1,1 +P2,1+P3,1=0.1+0.2+0.7=1.0, the probability that target video belongs to the 2nd pre-set categories is P2=P1,2+P2,2+P3,2= 0.3+0.6+0=0.9, the probability that target video belongs to the 10th pre-set categories is P10=P1,10+P2,10+P3,10=0.3+0+0.2= 0.5, the probability that target video belongs to the 11st pre-set categories is P11=P1,11+P2,11+P3,11=0+0.1+0=0.1, target video The probability for belonging to the 19th pre-set categories is P19=P1,19+P2,19+P3,19=0.2+0.1+0=0.3, it is pre- that target video belongs to the 20th If the probability of classification is P20=P1,20+P2,20+P3,20=0.1+0+0.1=0.2.Then it is corresponding each default from target video again The maximum pre-set categories of acquisition probability are as the target classification of target video in classification, and wherein target video belongs to the 1st default class Other probability is P1=1.0 be maximum, now it is considered that the target classification of target video is the 1st pre-set categories.

Wherein mesh of the target classification for the target video that the second way is obtained than target video that first way is obtained Mark classification more accurate.

The video classification methods of the present embodiment, by obtaining multiframe picture from target video, according to the figure of training in advance The picture of each frame is identified piece disaggregated model, predicts the probability of the corresponding each pre-set categories of picture of each frame;According to each frame The corresponding each pre-set categories of picture probability, obtain target video target classification.The technical scheme of the present embodiment, can with gram The poor skill of accuracy that the reply that clothes are directly based upon video title and video in the prior art is classified to video content Art problem, passes through the corresponding each pre-set categories of multiframe picture in the picture classification model prediction target video based on training in advance Probability, so as to obtain the target classification of target video, the accuracy classified to target video can be effectively improved.

Fig. 2 is the flow chart of the video classification methods embodiment two of the present invention.The video classification methods of the present embodiment, upper On the basis of the technical scheme for stating embodiment illustrated in fig. 1, technical scheme is further introduced in further detail.Such as Fig. 2 Shown, the video classification methods of the present embodiment specifically may include steps of:

200th, the training picture of several known class, generation picture training storehouse are gathered;

201st, using the number Zhang Xunlian pictures in picture training storehouse, picture classification model is trained;

For example the step 201 " utilizing the number Zhang Xunlian pictures in picture training storehouse, training picture classification model ", specifically may be used To comprise the following steps:

(a1) each Zhang Xunlian pictures are inputted into picture classification model successively so that the output of picture classification model is corresponding Train probability of the picture in each pre-set categories;

(a2) probability according to corresponding training picture in each pre-set categories, it is determined that the prediction classification of training picture;

(a3) detect whether the known class of corresponding training picture is consistent with the prediction classification of training picture;When correspondence instruction When the known class for practicing picture and the inconsistent prediction classification for training picture, step (a4) is performed;When correspondence has trained picture When knowing that classification is consistent with the prediction classification of training picture, return to step (a1) is inputted to picture classification using next Zhang Xunlian pictures It is trained in model.

(a4) parameter of picture classification model is adjusted so that the known class of training picture and the prediction classification of training picture Unanimously;

Above-mentioned steps (a1)-(a4) is repeated, until number Zhang Xunlian picture training is finished, and the known class of training picture It is consistent not with the prediction classification of training picture, the parameter of picture classification model is determined, so that it is determined that picture classification model.

Before training, the parameter of the picture classification model is initial value, when to the picture classification mode input first When training picture, picture classification model is based on initial parameter, can predict the training picture in each pre-set categories Probability;Then according to the corresponding probability for training picture in each pre-set categories, it is determined that the prediction classification of training picture, for example, The pre-set categories of the training picture can be used as using the maximum pre-set categories of select probability.Then corresponding training picture is detected Know whether classification is consistent with the prediction classification of training picture.If for example, certain training picture is in prediction, in the 1st pre-set categories Probability is that the probability of the 0.6, the 2nd pre-set categories is 0.2, is 0.2 also in the probability of the 9th pre-set categories.And the training picture is It is the 2nd pre-set categories to know classification, i.e., be 1.0 in the probability of the 2nd pre-set categories, should now adjust the ginseng of picture classification model Number so that the known class of training picture is consistent with the prediction classification of training picture, that is, causes the picture classification model prediction The training picture increases in the probability of the 2nd pre-set categories towards 1.0 direction, and causes the training of the picture classification model prediction Picture successively decreases in the probability of the 1st pre-set categories and the probability of the 9th pre-set categories towards probability for 0 direction, and now the Zhang Xunlian schemes Piece training is finished, next using the next Zhang Xunlian figures of parameter training of the picture classification model after upper one training picture adjustment Piece, the like, the parameter of picture classification model is adjusted again;Continue to train next, reusing down next Zhang Xunlian pictures Picture classification model, adjusts the parameter of picture classification model, until number Zhang Xunlian picture training is finished, and trains the known of picture Classification and the prediction classification of training picture reach unanimity, and the parameter of picture classification model are determined, so that it is determined that picture classification model.

In practical application, after several Zhang Xunlian pictures are taken turns to picture classification model training one, if training picture is known Classification and with picture classification model to train picture prediction classification it is still inconsistent, can now use several Zhang Xunlian pictures to figure Piece disaggregated model retraining one is taken turns or taken turns more, until training the known class of picture and the prediction classification of training picture to tend to one Cause, the parameter of picture classification model is determined, so that it is determined that picture classification model.

The quantity of the training picture gathered in picture training storehouse in the present embodiment can be with thousands of, or even hundreds of thousands , the quantity for the training picture that picture training storehouse includes is more, and the parameter of the picture classification model of training is more accurate, the training Each picture of picture classification model prediction afterwards is more accurate in the probability of each pre-set categories.

202nd, multiframe picture is obtained from target video;

203rd, the picture of each frame is identified according to the picture classification model of training in advance, predicts the picture correspondence of each frame Each pre-set categories probability;

204th, according to the probability of the corresponding each pre-set categories of picture of each frame, to the probability of the pre-set categories of the picture of each frame Processing is weighted, the probability of the corresponding each pre-set categories of target video is obtained;

205th, the maximum pre-set categories of acquisition probability are used as target video from target video corresponding each pre-set categories Target classification;

Above-mentioned steps 202-205 implementation may be referred to the related record of above-mentioned embodiment illustrated in fig. 1 in detail, herein no longer Repeat.

206th, user's classification interested is detected according to the historical behavior of user;

207th, whether target classification is included in detection user classification interested;If including performing step 208;If otherwise Do not include, continue whether include target classification in the next user of detection classification interested;

208th, target video is recommended to user.

Still optionally further, also include recommending target video to user according to the target classification of target video in the present embodiment Application scenarios.For example, in the video library of some video playback platform, a video has been newly increased, can be according to above-mentioned step Rapid 202-204 mode, gets the target classification of target video.Then each user using the Video Applications can be detected Historical behavior, judge each user using interested in the target video.For example, first can be examined according to the historical behavior of user Survey which user's classification interested has, then judge whether include the target of the target video in user's classification interested Classification, if including the target video can be recommended to the user, so as to which once new addition user is interested in video library New video, its video interested can be recommended to user in time, the using experience degree of user is improved, so as to keep this here User in video playback platform.If detecting does not include target classification in certain user classification interested, necessarily avoid to this User recommends the target video, to prevent the dislike of user.Detection can now be continued and use the next of the video playback platform Individual user, until all users to the video playback platform detect, and to all to target class of the video playback platform User not interested recommends the target video.

The video classification methods of the present embodiment, can be overcome by the above method and be directly based upon video title in the prior art And the poor technical problem of the accuracy classified to video content of the reply of video, pass through the picture based on training in advance The probability of the corresponding each pre-set categories of multiframe picture in disaggregated model prediction target video, so as to obtain the target of target video Classification, can effectively improve the accuracy classified to target video.But also can further promptly and accurately to interested User recommend the target video, improve the Experience Degree of user.

Fig. 3 is the structure chart of the visual classification device embodiment one of the present invention.As shown in figure 3, the video of the present embodiment point Class device, can specifically include:Picture acquisition module 10, prediction module 11 and classification acquisition module 12.

Wherein picture acquisition module 10 is used to from target video obtain multiframe picture;

Prediction module 11 is used for each frame obtained according to the picture classification model of training in advance to picture acquisition module 10 Picture is identified, and predicts the probability of the corresponding each pre-set categories of picture of each frame;

Classification acquisition module 12 is used for the general of the corresponding each pre-set categories of picture for each frame predicted according to prediction module 11 Rate, obtains the target classification of target video.

The visual classification device of the present embodiment, the realization principle and technology of visual classification are realized by using above-mentioned module Effect is identical with realizing for above-mentioned related method embodiment, the record of above-mentioned related method embodiment is may be referred in detail, herein Repeat no more.

Fig. 4 is the structure chart of the visual classification device embodiment two of the present invention.As shown in figure 4, the video of the present embodiment point Class device, on the basis of the technical scheme of above-mentioned embodiment illustrated in fig. 3, can also further include following technical scheme.

In the visual classification device of the present embodiment, classification acquisition module 12 specifically for:

According to the probability of the corresponding each pre-set categories of the picture of each frame, obtained from the corresponding pre-set categories of the picture of each frame The pre-set categories of maximum probability as target video target classification;

Or the probability of the corresponding each pre-set categories of picture according to each frame, to the probability of the pre-set categories of the picture of each frame Processing is weighted, the probability of the corresponding each pre-set categories of target video is obtained;From the corresponding each pre-set categories of target video The maximum pre-set categories of acquisition probability as target video target classification.

Still optionally further, as shown in figure 4, in the visual classification device of the present embodiment, in addition to:

Acquisition module 13 is used for the training picture for gathering several known class, generation picture training storehouse;

Training module 14 is used to, using the number Zhang Xunlian pictures in picture training storehouse, train picture classification model.

Accordingly, prediction module 11 is used to obtain mould to picture according to the picture classification model of the training in advance of training module 14 The picture for each frame that block 10 is obtained is identified, and predicts the probability of the corresponding each pre-set categories of picture of each frame.

Still optionally further, in the visual classification device of the present embodiment, training mould 14 specifically for:

Each Zhang Xunlian pictures are inputted into picture classification model successively so that picture classification model exports corresponding training Probability of the picture in each pre-set categories;

According to probability of the corresponding training picture in each pre-set categories, it is determined that the prediction classification of training picture;

Whether the known class of the corresponding training picture of detection is consistent with the prediction classification of training picture;

When the known class of correspondence training picture and the inconsistent prediction classification of training picture, picture classification model is adjusted Parameter so that training picture known class and train picture prediction classification it is consistent;

Above-mentioned steps are repeated, until number Zhang Xunlian picture training is finished, and known class and the training of training picture The prediction classification of picture is consistent, the parameter of picture classification model is determined, so that it is determined that picture classification model.

Still optionally further, as shown in figure 4, in the visual classification device of the present embodiment, in addition to:

Detection module 15 is used to detect user's classification interested according to the historical behavior of user;

Detection module 15 is additionally operable in detection user classification interested whether include the mesh that classification acquisition module 12 is obtained Mark the target classification of video;

If recommending module 16, which is used for detection module 15, detects that user's classification interested includes target classification, pushed away to user Recommend target video.

The visual classification device of the present embodiment, the realization principle and technology of visual classification are realized by using above-mentioned module Effect is identical with realizing for above-mentioned related method embodiment, the record of above-mentioned related method embodiment is may be referred in detail, herein Repeat no more.

Fig. 5 is the structure chart of the computer equipment embodiment of the present invention.As shown in figure 5, the computer equipment of the present embodiment, Including:One or more processors 30, and memory 40, memory 40 are used to store one or more programs, work as memory The one or more programs stored in 40 are performed by one or more processors 30 so that one or more processors 30 are realized such as The video classification methods of upper any embodiment.In embodiment illustrated in fig. 5 exemplified by including multiple processors 30.

For example, a kind of exemplary plot for computer equipment that Fig. 6 provides for the present invention.Fig. 6 is shown suitable for being used for realizing this The exemplary computer device 12a of invention embodiment block diagram.The computer equipment 12a that Fig. 6 is shown is only an example, Any limitation should not be carried out to the function of the embodiment of the present invention and using range band.

As shown in fig. 6, computer equipment 12a is showed in the form of universal computing device.Computer equipment 12a component can To include but is not limited to:One or more processor 16a, system storage 28a, connection different system component (including system Memory 28a and processor 16a) bus 18a.

Bus 18a represents the one or more in a few class bus structures, including memory bus or Memory Controller, Peripheral bus, graphics acceleration port, processor or the local bus using any bus structures in a variety of bus structures.Lift For example, these architectures include but is not limited to industry standard architecture (ISA) bus, MCA (MAC) Bus, enhanced isa bus, VESA's (VESA) local bus and periphery component interconnection (PCI) bus.

Computer equipment 12a typically comprises various computing systems computer-readable recording medium.These media can be it is any can The usable medium accessed by computer equipment 12a, including volatibility and non-volatile media, moveable and immovable Jie Matter.

System storage 28a can include the computer system readable media of form of volatile memory, for example, deposit at random Access to memory (RAM) 30a and/or cache memory 32a.Computer equipment 12a may further include it is other it is removable/ Immovable, volatile/non-volatile computer system storage medium.Only as an example, storage system 34a can be used for reading Write immovable, non-volatile magnetic media (Fig. 6 is not shown, is commonly referred to as " hard disk drive ").Although not shown in Fig. 6, It can provide for the disc driver to may move non-volatile magnetic disk (such as " floppy disk ") read-write, and to removable non-easy The CD drive of the property lost CD (such as CD-ROM, DVD-ROM or other optical mediums) read-write.In these cases, each Driver can be connected by one or more data media interfaces with bus 18a.System storage 28a can be included at least One program product, the program product has one group of (for example, at least one) program module, and these program modules are configured to hold The function of row above-mentioned each embodiments of Fig. 1-Fig. 4 of the invention.

Program with one group of (at least one) program module 42a/utility 40a, can be stored in such as system and deposit In reservoir 28a, such program module 42a include --- but being not limited to --- operating system, one or more application program, The reality of network environment is potentially included in each or certain combination in other program modules and routine data, these examples It is existing.Program module 42a generally performs the function and/or method in above-mentioned each embodiments of Fig. 1-Fig. 4 described in the invention.

Computer equipment 12a can also be with one or more external equipment 14a (such as keyboard, sensing equipment, display 24a etc.) communication, the equipment communication interacted with computer equipment 12a can be also enabled a user to one or more, and/or With any equipment (such as network interface card, tune for enabling computer equipment 12a to be communicated with one or more of the other computing device Modulator-demodulator etc.) communication.This communication can be carried out by input/output (I/O) interface 22a.Also, computer equipment 12a can also by network adapter 20a and one or more network (such as LAN (LAN), wide area network (WAN) and/or Public network, such as internet) communication.As illustrated, network adapter 20a by bus 18a and computer equipment 12a its Its module communicates.It should be understood that although not shown in the drawings, can combine computer equipment 12a uses other hardware and/or software Module, includes but is not limited to:Microcode, device driver, redundant processor, external disk drive array, RAID system, tape Driver and data backup storage system etc..

Processor 16a is stored in program in system storage 28a by operation, thus perform various function application and Data processing, for example, realize the video classification methods shown in above-described embodiment.

The present invention also provides a kind of computer-readable medium, is stored thereon with computer program, the program is held by processor The video classification methods as shown in above-mentioned embodiment are realized during row.

The computer-readable medium of the present embodiment can be included in the system storage 28a in above-mentioned embodiment illustrated in fig. 6 RAM30a, and/or cache memory 32a, and/or storage system 34a.

With the development of science and technology, the route of transmission of computer program is no longer limited by tangible medium, can also be directly from net Network is downloaded, or is obtained using other modes.Therefore, the computer-readable medium in the present embodiment can not only include tangible Medium, can also include invisible medium.

The computer-readable medium of the present embodiment can use any combination of one or more computer-readable media. Computer-readable medium can be computer-readable signal media or computer-readable recording medium.Computer-readable storage medium Matter for example may be-but not limited to-system, device or the device of electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor, or Combination more than person is any.The more specifically example (non exhaustive list) of computer-readable recording medium includes:With one Or the electrical connections of multiple wires, portable computer diskette, hard disk, random access memory (RAM), read-only storage (ROM), Erasable programmable read only memory (EPROM or flash memory), optical fiber, portable compact disc read-only storage (CD-ROM), light Memory device, magnetic memory device or above-mentioned any appropriate combination.In this document, computer-readable recording medium can Be it is any include or storage program tangible medium, the program can be commanded execution system, device or device use or Person is in connection.

Computer-readable signal media can be included in a base band or as the data-signal of carrier wave part propagation, Wherein carry computer-readable program code.The data-signal of this propagation can take various forms, including --- but It is not limited to --- electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also be Any computer-readable medium beyond computer-readable recording medium, the computer-readable medium can send, propagate or Transmit for being used or program in connection by instruction execution system, device or device.

The program code included on computer-readable medium can be transmitted with any appropriate medium, including --- but do not limit In --- wireless, electric wire, optical cable, RF etc., or above-mentioned any appropriate combination.

It can be write with one or more programming languages or its combination for performing the computer that the present invention is operated Program code, described program design language includes object oriented program language-such as Java, Smalltalk, C++, Also include conventional procedural programming language-such as " C " language or similar programming language.Program code can be with Fully perform, partly perform on the user computer on the user computer, as independent software kit execution, a portion Divide part execution or the execution completely on remote computer or server on the remote computer on the user computer. Be related in the situation of remote computer, remote computer can be by the network of any kind --- including LAN (LAN) or Wide area network (WAN)-be connected to subscriber computer, or, it may be connected to outer computer (is for example carried using Internet service Come for business by Internet connection).

In several embodiments provided by the present invention, it should be understood that disclosed system, apparatus and method can be with Realize by another way.For example, device embodiment described above is only schematical, for example, the unit Divide, only a kind of division of logic function there can be other dividing mode when actually realizing.

The unit illustrated as separating component can be or may not be it is physically separate, it is aobvious as unit The part shown can be or may not be physical location, you can with positioned at a place, or can also be distributed to multiple On NE.Some or all of unit therein can be selected to realize the mesh of this embodiment scheme according to the actual needs 's.

In addition, each functional unit in each embodiment of the invention can be integrated in a processing unit, can also That unit is individually physically present, can also two or more units it is integrated in a unit.Above-mentioned integrated list Member can both be realized in the form of hardware, it would however also be possible to employ hardware adds the form of SFU software functional unit to realize.

The above-mentioned integrated unit realized in the form of SFU software functional unit, can be stored in an embodied on computer readable and deposit In storage media.Above-mentioned SFU software functional unit is stored in a storage medium, including some instructions are to cause a computer Equipment (can be personal computer, server, or network equipment etc.) or processor (processor) perform the present invention each The part steps of embodiment methods described.And foregoing storage medium includes:USB flash disk, mobile hard disk, read-only storage (Read- Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disc or CD etc. it is various Can be with the medium of store program codes.

The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention God is with principle, and any modification, equivalent substitution and improvements done etc. should be included within the scope of protection of the invention.

Claims (12)

1. a kind of video classification methods, it is characterised in that methods described includes:
Multiframe picture is obtained from target video;
The picture of each frame is identified according to the picture classification model of training in advance, the picture of each frame is predicted The probability of corresponding each pre-set categories;
According to the probability of the corresponding each pre-set categories of the picture of each frame, the target classification of the target video is obtained.
2. according to the method described in claim 1, it is characterised in that according to the corresponding each pre-set categories of the picture of each frame Probability, obtains the target classification of the target video, specifically includes:
According to the probability of the corresponding each pre-set categories of the picture of each frame, from the corresponding pre-set categories of the picture of each frame The maximum pre-set categories of acquisition probability as the target video target classification;
Or the probability of the corresponding each pre-set categories of the picture according to each frame, to the pre-set categories of the picture of each frame Probability is weighted processing, obtains the probability of the corresponding each pre-set categories of the target video;It is corresponding from the target video In each pre-set categories the maximum pre-set categories of acquisition probability as the target video target classification.
3. according to the method described in claim 1, it is characterised in that according to the picture classification model of training in advance to each frame The picture be identified, before the probability of the corresponding each pre-set categories of the picture for predicting each frame, methods described is also wrapped Include:
Gather the training picture of several known class, generation picture training storehouse;
Using several Zhang Xunlian pictures in the picture training storehouse, the picture classification model is trained.
4. method according to claim 3, it is characterised in that utilize several Zhang Xunlian figures in the picture training storehouse Piece, trains the picture classification model, specifically includes:
Each Zhang Suoshu training pictures are inputted into the picture classification model successively so that the picture classification model output pair Probability of the training picture answered in each pre-set categories;
According to probability of the corresponding training picture in each pre-set categories, the prediction classification of the training picture is determined;
Detect whether the known class of the corresponding training picture is consistent with the prediction classification of the training picture;
When the known class of the correspondence training picture and the inconsistent prediction classification of the training picture, the picture is adjusted The parameter of disaggregated model so that the known class of the training picture is consistent with the prediction classification of the training picture;
Repeat above-mentioned steps, until several Zhang Xunlian pictures training finish, and the known class of the training picture with The prediction classification of the training picture is consistent, the parameter of the picture classification model is determined, so that it is determined that the picture classification mould Type.
5. according to any described methods of claim 1-4, it is characterised in that corresponding each default according to the picture of each frame After the probability of classification, the target classification for obtaining the target video, methods described also includes:
User classification interested is detected according to the historical behavior of user;
Whether detect in user classification interested includes the target classification;
If including recommending the target video to the user.
6. a kind of visual classification device, it is characterised in that described device includes:
Picture acquisition module, for obtaining multiframe picture from target video;
Prediction module, the picture of each frame is identified for the picture classification model according to training in advance, prediction The probability of the corresponding each pre-set categories of the picture of each frame;
Classification acquisition module, for the probability of the corresponding each pre-set categories of the picture according to each frame, obtains the target and regards The target classification of frequency.
7. device according to claim 6, it is characterised in that the classification acquisition module, specifically for:
According to the probability of the corresponding each pre-set categories of the picture of each frame, from the corresponding pre-set categories of the picture of each frame The maximum pre-set categories of acquisition probability as the target video target classification;
Or the probability of the corresponding each pre-set categories of the picture according to each frame, to the pre-set categories of the picture of each frame Probability is weighted processing, obtains the probability of the corresponding each pre-set categories of the target video;It is corresponding from the target video In each pre-set categories the maximum pre-set categories of acquisition probability as the target video target classification.
8. device according to claim 6, it is characterised in that described device also includes:
Acquisition module, the training picture for gathering several known class, generation picture training storehouse;
Training module, for using several Zhang Xunlian pictures in the picture training storehouse, training the picture classification model.
9. device according to claim 8, it is characterised in that the training module, specifically for:
Each Zhang Suoshu training pictures are inputted into the picture classification model successively so that the picture classification model output pair Probability of the training picture answered in each pre-set categories;
According to probability of the corresponding training picture in each pre-set categories, the prediction classification of the training picture is determined;
Detect whether the known class of the corresponding training picture is consistent with the prediction classification of the training picture;
When the known class of the correspondence training picture and the inconsistent prediction classification of the training picture, the picture is adjusted The parameter of disaggregated model so that the known class of the training picture is consistent with the prediction classification of the training picture;
Repeat above-mentioned steps, until several Zhang Xunlian pictures training finish, and the known class of the training picture with The prediction classification of the training picture is consistent, the parameter of the picture classification model is determined, so that it is determined that the picture classification mould Type.
10. according to any described devices of claim 6-9, it is characterised in that described device also includes:
Detection module, for detecting user classification interested according to the historical behavior of user;
Whether the detection module, being additionally operable to detect in the classification that the user is interested includes the target classification;
Recommending module, if detecting that user classification interested includes the target classification for the detection module, to The user recommends the target video.
11. a kind of computer equipment, including memory, processor and storage are on a memory and the meter that can run on a processor Calculation machine program, it is characterised in that the side as described in any in claim 1-5 is realized during the computing device described program Method.
12. a kind of computer-readable medium, is stored thereon with computer program, it is characterised in that the program is executed by processor Methods of the Shi Shixian as described in any in claim 1-5.
CN201710325322.7A 2017-05-10 2017-05-10 Video classification methods and device, computer equipment and computer-readable recording medium CN107194419A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710325322.7A CN107194419A (en) 2017-05-10 2017-05-10 Video classification methods and device, computer equipment and computer-readable recording medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710325322.7A CN107194419A (en) 2017-05-10 2017-05-10 Video classification methods and device, computer equipment and computer-readable recording medium

Publications (1)

Publication Number Publication Date
CN107194419A true CN107194419A (en) 2017-09-22

Family

ID=59872910

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710325322.7A CN107194419A (en) 2017-05-10 2017-05-10 Video classification methods and device, computer equipment and computer-readable recording medium

Country Status (1)

Country Link
CN (1) CN107194419A (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108090497A (en) * 2017-12-28 2018-05-29 广东欧珀移动通信有限公司 Video classification methods, device, storage medium and electronic equipment
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
CN108965920A (en) * 2018-08-08 2018-12-07 北京未来媒体科技股份有限公司 A kind of video content demolition method and device
CN109308490A (en) * 2018-09-07 2019-02-05 北京字节跳动网络技术有限公司 Method and apparatus for generating information
CN109902547A (en) * 2018-05-29 2019-06-18 华为技术有限公司 Action identification method and device
WO2020087974A1 (en) * 2018-10-30 2020-05-07 北京字节跳动网络技术有限公司 Model generation method and device
WO2020107625A1 (en) * 2018-11-27 2020-06-04 北京微播视界科技有限公司 Video classification method and apparatus, electronic device, and computer readable storage medium

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101359372A (en) * 2008-09-26 2009-02-04 腾讯科技(深圳)有限公司 Training method and device of classifier, and method apparatus for recognising sensitization picture
CN101853377A (en) * 2010-05-13 2010-10-06 复旦大学 Method for identifying content of digital video
CN102073864A (en) * 2010-12-01 2011-05-25 北京邮电大学 Football item detecting system with four-layer structure in sports video and realization method thereof
CN103761295A (en) * 2014-01-16 2014-04-30 北京雅昌文化发展有限公司 Automatic picture classification based customized feature extraction algorithm for art pictures
CN105184313A (en) * 2015-08-24 2015-12-23 小米科技有限责任公司 Classification model construction method and device
CN105550699A (en) * 2015-12-08 2016-05-04 北京工业大学 CNN-based video identification and classification method through time-space significant information fusion
CN105893930A (en) * 2015-12-29 2016-08-24 乐视云计算有限公司 Video feature identification method and device
CN105913072A (en) * 2016-03-31 2016-08-31 乐视控股(北京)有限公司 Training method of video classification model and video classification method
CN106294783A (en) * 2016-08-12 2017-01-04 乐视控股(北京)有限公司 A kind of video recommendation method and device
CN106612457A (en) * 2016-11-09 2017-05-03 广州视源电子科技股份有限公司 Method and system for video sequence alignment

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101359372A (en) * 2008-09-26 2009-02-04 腾讯科技(深圳)有限公司 Training method and device of classifier, and method apparatus for recognising sensitization picture
CN101853377A (en) * 2010-05-13 2010-10-06 复旦大学 Method for identifying content of digital video
CN102073864A (en) * 2010-12-01 2011-05-25 北京邮电大学 Football item detecting system with four-layer structure in sports video and realization method thereof
CN103761295A (en) * 2014-01-16 2014-04-30 北京雅昌文化发展有限公司 Automatic picture classification based customized feature extraction algorithm for art pictures
CN105184313A (en) * 2015-08-24 2015-12-23 小米科技有限责任公司 Classification model construction method and device
CN105550699A (en) * 2015-12-08 2016-05-04 北京工业大学 CNN-based video identification and classification method through time-space significant information fusion
CN105893930A (en) * 2015-12-29 2016-08-24 乐视云计算有限公司 Video feature identification method and device
CN105913072A (en) * 2016-03-31 2016-08-31 乐视控股(北京)有限公司 Training method of video classification model and video classification method
CN106294783A (en) * 2016-08-12 2017-01-04 乐视控股(北京)有限公司 A kind of video recommendation method and device
CN106612457A (en) * 2016-11-09 2017-05-03 广州视源电子科技股份有限公司 Method and system for video sequence alignment

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108090497A (en) * 2017-12-28 2018-05-29 广东欧珀移动通信有限公司 Video classification methods, device, storage medium and electronic equipment
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
CN109902547A (en) * 2018-05-29 2019-06-18 华为技术有限公司 Action identification method and device
CN108965920A (en) * 2018-08-08 2018-12-07 北京未来媒体科技股份有限公司 A kind of video content demolition method and device
CN109308490A (en) * 2018-09-07 2019-02-05 北京字节跳动网络技术有限公司 Method and apparatus for generating information
WO2020087974A1 (en) * 2018-10-30 2020-05-07 北京字节跳动网络技术有限公司 Model generation method and device
WO2020107625A1 (en) * 2018-11-27 2020-06-04 北京微播视界科技有限公司 Video classification method and apparatus, electronic device, and computer readable storage medium

Similar Documents

Publication Publication Date Title
US9336483B1 (en) Dynamically updated neural network structures for content distribution networks
US9514389B1 (en) Training a neural network to detect objects in images
CN105210064B (en) Classifying resources using deep networks
CN105283848B (en) Application tracking is carried out with distributed object
CN104246656B (en) It is recommended that video editing automatic detection
US20150294216A1 (en) Cognitive distributed network
US10671679B2 (en) Method and system for enhanced content recommendation
US10528973B2 (en) Network-aware product rollout in online social networks
US20170357720A1 (en) Joint heterogeneous language-vision embeddings for video tagging and search
US9942358B2 (en) Recommending applications
CN103064826B (en) A kind of method, equipment and system for input of expressing one's feelings
AU2014251221A1 (en) Identifying content in electronic images
CN103718206B (en) Recommend the mthods, systems and devices of other users or object
CN107544670A (en) The computing device of non-vision response with power triggering
CN105046515B (en) Method and device for sorting advertisements
US20110106799A1 (en) Measuring web site satisfaction of information needs
CN105869633A (en) Cross-lingual initialization of language models
CN104040584A (en) Selecting content formats based on predicted user interest
CN102799890A (en) Image clustering method
CN110430455A (en) It distinguishes local media and overflows the method, apparatus and storage medium of media
CN109155006A (en) The audio analysis based on frequency is carried out using neural network
JP6267344B2 (en) Content selection using quality control
US20090187520A1 (en) Demographics from behavior
AU2012316659A1 (en) Referral program for businesses
US8756178B1 (en) Automatic event categorization for event ticket network systems

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