CN101650740B - Method and device for detecting television advertisements - Google Patents

Method and device for detecting television advertisements Download PDF

Info

Publication number
CN101650740B
CN101650740B CN2009101672490A CN200910167249A CN101650740B CN 101650740 B CN101650740 B CN 101650740B CN 2009101672490 A CN2009101672490 A CN 2009101672490A CN 200910167249 A CN200910167249 A CN 200910167249A CN 101650740 B CN101650740 B CN 101650740B
Authority
CN
China
Prior art keywords
camera lens
video
detected
database
advertisement
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.)
Expired - Fee Related
Application number
CN2009101672490A
Other languages
Chinese (zh)
Other versions
CN101650740A (en
Inventor
陈向文
朱明�
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Science and Technology of China USTC
Original Assignee
University of Science and Technology of China USTC
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 University of Science and Technology of China USTC filed Critical University of Science and Technology of China USTC
Priority to CN2009101672490A priority Critical patent/CN101650740B/en
Publication of CN101650740A publication Critical patent/CN101650740A/en
Application granted granted Critical
Publication of CN101650740B publication Critical patent/CN101650740B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Studio Devices (AREA)

Abstract

The invention discloses a method and a device for detecting television advertisements, wherein the method adopts the technical schemes comprising the following steps: generating at least one shot according to a video to be detected; extracting image features of the shot and establishing a database by using the image features, wherein the database stores advertisement shots and corresponding related information; and searching the related information stored in the database so as to detect the television advertisements that appear repeatedly. The device for detecting the television advertisements comprises a shot generating module, a data control module and an advertisement searching module. By adopting the technical schemes, the invention can detect repeated advertisements in the video and improves the experiences of users.

Description

A kind of television advertising detection method and device
Technical field
The present invention relates to the video identification technology field, specifically, relate to a kind of television advertising detection method and device.
Background technology
At present, along with rapid development of multimedia, the status that television advertising is occupied in daily life becomes more and more important.Because what television advertising repeated appears in the TV programme, has not only influenced the user experience of spectators' TV receptions, has also taken a large amount of sdi videos.
At the characteristics of present television advertising, it is no longer suitable that retromercial detects the tactful detection mode based on station symbol and black/silent frame (black/silent frame) that proposes.On the one hand be because TV station can not conceal station symbol usually in playing advertisements now, and the form of station symbol also is tending towards complicated gradually, so brought a difficult problem based on the detection mode of station symbol.On the other hand, always do not have the black/silent frame when general programs and advertising segment conversion, the black/silent frame can be according to certain montage needs insertion at random, and this just directly causes the failure based on the black/silent frame detection method.
Commercial detection method based on camera lens is can represent the feature of advertising segment by extract some from camera lens, and utilizes these features that television cameras are divided into general programs camera lens and advertisement camera lens.Because this method is only carried out simple classification to camera lens usually, not only there is not to consider how to eliminate the wrong influence that divides the advertisement camera lens, there is not to consider how to merge the problem that the advertisement camera lens obtains advertising segment yet, and now a lot of technology are not all considered the consistance of ad content, thereby are difficult to improve the effect that detects.So,, but when detecting blanking or being fade-in fade-out camera lens, will encounter problems even this method is obtaining good effect aspect the detection shearing lens.
Based on the commercial detection method of database is by the feature of the advertisement video section of area definition in advance in database, utilizes this database identification to be embedded in the advertisement section of TV programme the inside.This method must dispose an enough big database and store known characteristic of advertisement, and maintenance workload is huge, just detects not come out for the content that is not stored in the advertisement video section feature in the database in advance.
In realizing process of the present invention, the inventor finds that how can detect television advertising fast in multitude of video is the technical barrier that will solve at present.
Summary of the invention
The technical problem to be solved in the present invention is, a kind of television advertising detection method and device are provided, and can the advertisement that repeat in the video flowing be detected, and improves user's experience.
Technical scheme of the present invention is as described below:
The invention provides a kind of television advertising detection method, technical scheme comprises:
Generate at least one camera lens according to video to be detected;
Characteristics of image to described camera lens extracts, and makes up database with described characteristics of image, and described database storing has advertisement camera lens and corresponding relevant information; Described structure database specifically comprises:
Extract the eigenmatrix of all pictures at least one camera lens, with the key frame eigenmatrix of each camera lens characteristics of image as this camera lens;
According to described characteristics of image, calculate in the video to be detected distance between the storage camera lens in the camera lens and database respectively;
When distance is greater than predetermined threshold between the storage camera lens in camera lens and the database in the video to be detected, there is camera lens in the specified data storehouse with video lens coupling to be detected, and upgrade the relevant information corresponding of preserving in the described database with this camera lens, otherwise, there is not camera lens in the specified data storehouse with video lens coupling to be detected, add the relevant information of this camera lens correspondence in described database, described relevant information comprises the filename of video to be detected, the eigenmatrix of each camera lens, position and in multiplicity one or more of this camera lens in video to be detected;
The described relevant information of storing in the described database is retrieved, so that detect the television advertising that repeats; The described relevant information of storing in the described database is retrieved, is specifically comprised:
Multiplicity in the described database equated and extract greater than one one group of cinestrip and corresponding relevant information;
By predetermined condition described relevant information is judged, when the described cinestrip that extracts all satisfies predetermined condition, determined that described cinestrip is same section advertisement, and determine the position of this advertisement according to the positional information of described cinestrip; When comprising the camera lens that does not satisfy predetermined condition in the described cinestrip, then determine from next initial camera lens of this camera lens be the initial camera lens of another section advertisement, distinguish different advertisements according to the positional information of this camera lens, and definite every section advertisement the position; Described predetermined condition comprises and counts up to identically entirely on the relevant position of filename array of each camera lens in the container that and the corresponding start frame numerical value in the pos array of each end frame in the pos array of a back camera lens and previous camera lens differs and is no more than 500.
Further, generate at least one camera lens, specifically comprise according to video to be detected:
Video slicing to be detected is become at least one picture;
According to the camera lens partitioning algorithm described at least one picture is generated at least one camera lens.
Further, described method also comprises:
Preestablish the camera lens density threshold, after generating at least one camera lens, when described camera lens density during greater than described camera lens density threshold according to video to be detected, determine that this camera lens is a TV programme, and TV programme deleted from video to be detected, otherwise, determine that this camera lens is that advertising programme keeps.
The invention provides a kind of television advertising pick-up unit, technical scheme comprises:
The camera lens generation module is used for generating at least one camera lens according to video to be detected;
The Data Control module, the characteristics of image that is used for described camera lens that described camera lens generation module is generated extracts, and makes up database with described characteristics of image, and described database storing has the relevant information of advertisement camera lens and correspondence; Described Data Control module specifically comprises:
Feature extraction unit is used for extracting the eigenmatrix of at least one all picture of camera lens, with the key frame eigenmatrix of each camera lens characteristics of image as this camera lens;
Similarity calculated is used for according to described characteristics of image, calculates in the video to be detected distance between the storage camera lens in the camera lens and database respectively;
The judgment processing unit, be used for when distance is greater than predetermined threshold between the storage camera lens in video camera lens to be detected and the database, there is camera lens in the specified data storehouse with video lens coupling to be detected, upgrade the relevant information corresponding of preserving in the described database with this camera lens, otherwise, there is not camera lens in the specified data storehouse with video lens coupling to be detected, add the relevant information of this camera lens correspondence in described database, described relevant information comprises the filename of video to be detected, the eigenmatrix of each camera lens, position and in multiplicity one or more of this camera lens in video to be detected;
The advertisement retrieval module is used for the described relevant information that described database is stored is retrieved, so that detect the television advertising that repeatability occurs; Described advertisement retrieval module specifically comprises:
The data pick-up unit is used for described database multiplicity equated and extracts greater than one one group of cinestrip and corresponding relevant information;
The Data Detection unit, be used for described relevant information being judged by predetermined condition, when the described cinestrip that extracts all satisfies predetermined condition, determine that described cinestrip is same section advertisement, and determine the position of this advertisement according to the positional information of described cinestrip; When comprising the camera lens that does not satisfy predetermined condition in the described cinestrip, then determine to be the initial camera lens of another section advertisement from next initial camera lens of this camera lens, positional information according to this camera lens is distinguished different advertisements, and determine every section advertisement the position, described predetermined condition comprises and counts up to identically entirely on the relevant position of filename array of each camera lens in the container that and the corresponding start frame numerical value in the pos array of each end frame in the pos array of a back camera lens and previous camera lens differs and is no more than 500.
Preferably, described camera lens generation module specifically comprises:
Video processing unit is used for video slicing to be detected is become at least one picture;
The camera lens cutter unit is used for according to the camera lens partitioning algorithm at least one picture of described video processing unit cutting being generated at least one camera lens.
Preferably, described device also comprises:
The camera lens filtering module, be used to preestablish the camera lens density threshold, after generating at least one camera lens according to video to be detected, when described camera lens density during greater than described camera lens density threshold, determine that this camera lens is a TV programme, and TV programme deleted from video to be detected, otherwise, determine that this camera lens is that advertising programme keeps.
Beneficial effect:
The present invention adopts the detection method based on advertisement repeatability, relevant information by each advertisement camera lens of when making up database, preserving, so long as the advertisement camera lens that repeats can detect, solved and used the detection of existing database detection method less than the shortcoming that is not stored in the advertisement camera lens sample in the database in advance.
Description of drawings
Fig. 1 is a kind of television advertising detection method of embodiment of the invention process flow diagram;
Fig. 2 is a kind of television advertising pick-up unit of embodiment of the invention structural representation;
Fig. 3 is a Video Detection process flow diagram in the embodiment of the invention one;
Fig. 4 is image region segmentation synoptic diagram during image characteristics extraction in the embodiment of the invention one.
Embodiment
Below in conjunction with specific embodiment technical scheme of the present invention is described further.
A kind of television advertising detection method of the embodiment of the invention and device are according to video image characteristic, detect at the advertising programme that repeats in the video flowing, can reduce the number of times that the identical content advertisement repeats in same video, improve user's experience.
Because the technical scheme of the embodiment of the invention is based on the basis of extracting characteristics of image, does not need as the commercial detection method based on camera lens, extracts a large amount of characteristic of advertisement with the shot classification in the video.Because existing commercial detection method, need a large amount of advertisement camera lens of storage in database in advance, the problem that can't detect for the advertisement camera lens that does not store in advance in the database based on database.The foundation of embodiment of the invention database and purposes of commercial detection process are almost carried out simultaneously, so after having upgraded database, content in the database is retrieved, and the advertisement camera lens and the corresponding position information that can obtain to repeat be convenient to this camera lens is deleted or shielded.
As shown in Figure 1, a kind of television advertising detection method of the embodiment of the invention can be achieved through the following technical solutions:
Step 100: generate at least one camera lens according to video to be detected;
Step 200: the characteristics of image to described camera lens extracts, and makes up database with described characteristics of image, and described database storing has advertisement camera lens and corresponding relevant information;
Step 300: the described relevant information of storing in the described database is retrieved, so that detect the television advertising that repeats.
In a specific embodiment, step 100 specifically can be achieved through the following technical solutions:
(1) video slicing to be detected is become at least one picture;
(2) according to the camera lens partitioning algorithm described at least one picture is generated at least one camera lens.
Concrete, in the time of need detecting video flowing, video flowing is intercepted into some pictures, obtain the video data of discretize, can be convenient to like this content in this video is detected.
To intercept the some pictures that obtain by the camera lens split plot design in the embodiment of the invention and cut into camera lens, comprise several pictures in each camera lens, can obtain the point of contact file of each camera lens last frame by the camera lens split plot design, so that obtain the positional information of each camera lens in this video.Wherein, the camera lens split plot design specifically can realize by technical scheme as described below:
The camera lens that the camera lens partitioning algorithm that is adopted in the embodiment of the invention is based on color histogram is cut apart.The R of difference statistical picture, G, the histogram of three components of B, the interval with 256 is divided into 64 sections and adds up, and 64bins is promptly arranged, so three components one total 192bins, availability vector is expressed as: B=(b 1, b 2..., b 192).The difference degree of the color histogram of two width of cloth images can be calculated by following formula and obtain:
d A , B = Σ i = 1 192 ( a i - b i ) 2
In embodiments of the present invention the difference threshold value is got 5 * 10 8, as d during, illustrate that this two width of cloth image is the point of contact of camera lens greater than this threshold value, can obtain the position of this camera lens in video to be detected by this point of contact.
In a specific embodiment, a kind of television advertising detection method also comprises:
Preestablish the camera lens density threshold, after generating at least one camera lens, when described camera lens density during greater than described camera lens density threshold according to video to be detected, determine that this camera lens is a TV programme, and TV programme deleted from video to be detected, otherwise, determine that this camera lens is that advertising programme keeps.
Concrete, the camera lens greater than the camera lens density threshold must be a TV programme usually, but the camera lens density of the TV programme that has also can be less than this threshold value; And the camera lens density of advertisement is less than this threshold value certainly, so the camera lens that generates is carried out primary filter by the camera lens density threshold, the workload in the testing process be can significantly reduce, but probably a part of TV programme and whole advertisements in the video that is staying, also comprised.
In a specific embodiment, step 200 specifically can be achieved through the following technical solutions:
(1) extracts the eigenmatrix of all pictures at least one camera lens, with the key frame eigenmatrix of each camera lens characteristics of image as this camera lens;
(2), calculate in the video to be detected distance between the storage camera lens in the camera lens and database respectively according to described characteristics of image;
(3) when distance is greater than predetermined threshold between the storage camera lens in camera lens and the database in the video to be detected, there is camera lens in the specified data storehouse with video lens coupling to be detected, and upgrade the relevant information corresponding of preserving in the described database with this camera lens, otherwise, there is not camera lens in the specified data storehouse with video lens coupling to be detected, add the relevant information of this camera lens correspondence in described database, described relevant information comprises the filename of video to be detected, the eigenmatrix of each camera lens, position and in multiplicity one or more of this camera lens in video to be detected.
Concrete, at above-mentioned steps (1) and (2), selection is based on the characteristics of image of the RGB color space characteristics of image as each camera lens in the embodiment of the invention.The value space (0,255) of R, G, three color components of B is divided into 16 intervals respectively.Form between one 16 * 16 * 16 chromatic zones, add up the ratio that pixel between each chromatic zones accounts for entire image respectively and form one 16 * 16 * 16 eigenmatrix (double type).In order to improve the sensitivity of this feature, the error that lower quantization is brought, the embodiment of the invention do not adopt and quantize or method for normalizing, but directly with the eigenmatrix of this double type as characteristics of image.In order to improve the sensitivity of feature, the embodiment of the invention becomes 5 zones with each width of cloth image segmentation, each zone is calculated 16 * 16 * 16 eigenmatrix respectively.
Because the embodiment of the invention is based on camera lens to the processing of video flowing, so advertisement phase knowledge and magnanimity calculate and return that to make a thorough investigation of the end be the calculating of shot similarity, i.e. the distance of camera lens.After having determined characteristics of image, be converted into the distance of image for distance with camera lens, according to the little reason of various attribute change of each picture frame in a camera lens, the embodiment of the invention is defined as key frame corresponding attribute average of all frames in the camera lens.Therefore in the embodiment of the invention with the mean value of the eigenmatrix of all images frame in each camera lens key frame eigenmatrix as this camera lens.
When extracting the key frame eigenmatrix of a camera lens, extract the eigenmatrix of all images frame in the camera lens earlier, average then, suppose that the eigenmatrix of key frame is A, the n frame is arranged in the camera lens, the eigenmatrix of i frame is A i
Then
A = A 1 + A 2 + . . . + A n n
Calculate between two camera lenses apart from the time, introduce the notion of vector distance, vectorial L 1And L 2Distance be:
d = L 1 · L 2 | L 1 | | L 2 |
With the eigenmatrix of key frame as the vector that is one 16 * 16 * 16 dimension.Then the distance of camera lens A and B is:
d AB = A · B | A | | B |
Wherein, A · B = Σ i = 0 15 Σ j = 0 15 Σ k = 0 15 a [ i ] [ j ] [ k ] * b [ i ] [ j ] [ k ] , | A | = Σ i = 0 15 Σ j = 0 15 Σ k = 0 15 ( a [ i ] [ j ] [ k ] ) 2
| B | = Σ i = 0 15 Σ j = 0 15 Σ k = 0 15 ( b [ i ] [ j ] [ k ] ) 2
Because the embodiment of the invention becomes 5 parts with image segmentation, thus calculate two camera lenses apart from the time, the distance of 5 parts calculates 5 respectively apart from d 1, d 2, d 3, d 4, d 5, judge that to this two similar conditions of camera lens have two kinds:
First kind: get d = d 1 + d 2 + d 3 + d 4 + d 5 5 , Judge that then whether d is greater than a certain threshold value L;
Second kind: judge d 1, d 2, d 3, d 4, d 5Whether all greater than some threshold value L;
The embodiment of the invention is in order to improve accuracy and sensitivity, and preferred second method is got threshold value L=0.95.
Concrete, the embodiment of the invention is by above-mentioned similarity calculating method, calculate the distance between the camera lens of storing in each camera lens and database in the video to be detected respectively, come whether to include in the specified data storehouse with video to be detected in the camera lens that mates, database upgraded original lens data or to add the processing of new lens data.
Concrete, at step 200, the process of setting up of database can be set up algorithm by database as described below and realizes in the embodiment of the invention:
Step 101: the filename of importing current video to be detected;
Step 102: each camera lens to this video extracts the key frame eigenmatrix;
Step 103: by calculating the distance between the camera lens of storing in each camera lens and database in the video to be detected, determine in database, whether to comprise the camera lens that is complementary with video lens to be detected, if exist, with camera lens that is complementary in the video to be detected and relevant information, the relevant information of corresponding camera lens in the new database more, be specially: data cell count (multiplicity) value that will mate the camera lens correspondence adds 1, and in filename (filename) array, add the filename of this video, wherein, filename is defined as the int type, adds the position of this camera lens in video to be detected simultaneously in pos (position) array;
If do not exist, then in database, add the data cell corresponding with this coupling camera lens, wherein, the count value is set to 1, adds the filename of current video in filename, adds the position of this camera lens in video in the pos array.
Before video flowing finishes, all need to search for by step 103 pair video lens to be detected, finish up to video flowing.
About the structure of data cell, can define in the following way:
typedef?struct{
Keyframe shot; // key frame eigenmatrix
Int count; The number of times statistics that // camera lens occurs
Int filename[50]; // with storing the filename that this camera lens occurs
Matchpair pos[50]; // be used for storing the position of camera lens at video
Int shotbegin; // when the start frame of pre-treatment camera lens in current video
Int shotend; // when the end frame of pre-treatment camera lens in current video
}shotcharacter；
In a specific embodiment, step 300 specifically can realize by technical scheme as described below:
(1) multiplicity in the described database is equated and extract greater than one one group of cinestrip and corresponding relevant information;
(2) by predetermined condition described relevant information is judged, when the described cinestrip that extracts all satisfies predetermined condition, determined that described cinestrip is same section advertisement, and determine the position of this advertisement according to the positional information of described cinestrip; When comprising the camera lens that does not satisfy predetermined condition in the described cinestrip, then determine from next initial camera lens of this camera lens be the initial camera lens of another section advertisement, distinguish different advertisements according to the positional information of this camera lens, and definite every section advertisement the position.
Specifically, step 300 is processes that retrieval repeats advertisement in the database after renewal.Usually because the minimum data unit of storing in the database is a camera lens, generate advertising segment so these camera lenses will be merged.After finishing these, can utilize the relevant information (the video file name of the continuity of the count attribute in the data cell, similar camera lens and filename) of the camera lens of preserving in the database to determine detect whether comprise the advertisement that repeats in the video.In database, find out the continuous camera lens of the identical a slice of count value,, determine the position of the advertisement that repeats in each video file then according to filename among the filename and the lens location among the pos.The detailed process of step 300 can realize by purposes of commercial detection algorithm as described below:
Step 201: search database, find out that the count value equates and a set of shots of (be interrupted in being meant continuously in the embodiment of the invention camera lens that falls can not above 2) continuously of count＞1, be pressed in the interim container; When count＞1, just can determine those camera lenses and be the camera lens of the advertisement that repeats in the embodiment of the invention;
Step 202: the data in the temporary container are judged that Rule of judgment is as follows:
A. on the relevant position of the filename array of each camera lens in the container count up to identical entirely;
Corresponding start frame numerical value in each end frame in the pos array of a b. back camera lens and the pos array of previous camera lens differs and is no more than 500;
If camera lens all satisfies above-mentioned condition a and b, illustrate that then the continuous camera lens of this group in the container all is same section advertisement, because the camera lens of storing in the database all is independently, so independently camera lens this group is continuous according to predetermined condition couples together, be combined into one section advertisement, and determine the position of this advertisement section, and export the position of this advertisement according to the positional information of pos array in the relevant information of each camera lens;
If there is camera lens not satisfy wherein any one condition, illustrate that then this next camera lens that begins with camera lens is the another one advertisement, and the boundary between the advertisement of two ends is exactly that camera lens that can not satisfy these two conditions simultaneously, two different advertisements can be split thus, and according to the positional information of the pos array in the relevant information of the camera lens that does not satisfy predetermined condition, determine the position of every section advertisement, and output;
Step 203: empty temporary container, get back to the place continuation execution that step 201 is stopped, finish up to Data Detection.
By a kind of commercial detection method of the invention described above, can reach 94.572% by the recall rate that realizes the proof advertising image, rate of accuracy reached 93.472% can realize the advertising segment that occurs more than twice or twice in the video flowing is detected.
As shown in Figure 2, based on the method embodiment of above-mentioned Fig. 1, the embodiment of the invention proposes a kind of television advertising pick-up unit, can be achieved through the following technical solutions:
Camera lens generation module 11 is used for generating at least one camera lens according to video to be detected;
Data Control module 22, the characteristics of image that is used for described camera lens that described camera lens generation module 11 is generated extracts, and makes up database with described characteristics of image, and described database storing has the relevant information of advertisement camera lens and correspondence;
Advertisement retrieval module 33 is used for the described relevant information that described database is stored is retrieved, so that detect the television advertising that repeatability occurs.
In a specific embodiment, described camera lens generation module 11 specifically comprises:
Video processing unit 111 is used for video slicing to be detected is become at least one picture;
Camera lens cutter unit 112 is used for according to the camera lens partitioning algorithm at least one picture of described video processing unit cutting being generated at least one camera lens.
In a specific embodiment,, described Data Control module 22 specifically comprises:
Feature extraction unit 221 is used for extracting the eigenmatrix of at least one all picture of camera lens, with the key frame eigenmatrix of each camera lens characteristics of image as this camera lens;
Similarity calculated 222 is used for according to described characteristics of image, calculates in the video to be detected distance between the storage camera lens in the camera lens and database respectively;
Judgment processing unit 223, be used for when distance is greater than predetermined threshold between the storage camera lens in video camera lens to be detected and the database, there is camera lens in the specified data storehouse with video lens coupling to be detected, upgrade the relevant information corresponding of preserving in the described database with this camera lens, otherwise, there is not camera lens in the specified data storehouse with video lens coupling to be detected, add the relevant information of this camera lens correspondence in described database, described relevant information comprises the filename of video to be detected, the eigenmatrix of each camera lens, position and in multiplicity one or more of this camera lens in video to be detected.
In a specific embodiment,, described advertisement retrieval module 33 specifically comprises:
Data pick-up unit 331 is used for described database multiplicity equated and extracts greater than one one group of cinestrip and corresponding relevant information;
Data Detection unit 332, be used for described relevant information being judged by predetermined condition, when the described cinestrip that extracts all satisfies predetermined condition, determine that described cinestrip is same section advertisement, and determine the position of this advertisement according to the positional information of described cinestrip; When comprising the camera lens that does not satisfy predetermined condition in the described cinestrip, then determine from next initial camera lens of this camera lens be the initial camera lens of another section advertisement, distinguish different advertisements according to the positional information of this camera lens, and definite every section advertisement the position.
In a specific embodiment,, described device also comprises:
Camera lens filtering module 44, be used to preestablish the camera lens density threshold, after generating at least one camera lens according to video to be detected, when described camera lens density during greater than described camera lens density threshold, determine that this camera lens is a TV programme, and TV programme deleted from video to be detected, otherwise, determine that this camera lens is that advertising programme keeps.
In view of the embodiment of the invention obtains in view of the described method embodiment of above-mentioned Fig. 1, the technical scheme that is specifically related to can be given unnecessary details no longer one by one referring to above-mentioned Fig. 1 embodiment.
Embodiment one:
Describe below in conjunction with a pair of technical scheme of the present invention of embodiment.
In conjunction with method and apparatus illustrated in figures 1 and 2, for example to detect 3 segment length and be 30 minutes to 40 minutes video, the filename of these 3 sections videos is filel, file2 and file3, the form of the location advertising of output is filename[a, b], wherein filename is a filename, a is the start frame of advertising segment, and b is the end frame of advertising segment.
Be illustrated in figure 3 as the Video Detection process flow diagram in the embodiment of the invention one, specific as follows:
Step 501: first section video to be detected by ffmpeg software, according to the speed intercepting picture of 25 frames/s, obtained the video of discretize, and the size of each image is 448 * 336, and form is jpg;
Step 502: utilize the camera lens partitioning algorithm, these pictures are carried out the camera lens cutting, obtain the point of contact file of the last frame of each camera lens, wherein, the point of contact file is the .txt form, has wherein write down the last frame of each camera lens; The detailed process of camera lens split plot design sees also the step 100 among the said method embodiment:
Step 503: utilize the characteristic of the camera lens switch speed of advertisement much larger than the existing-quality television program, by preestablishing the threshold value of camera lens density, the point of contact file that scanning obtains, when camera lens density (in the unit interval camera lens switch number) greater than threshold value, determine that then these camera lenses are TV programme, can reject; So scanning obtains a start frame and a new camera lens file of end frame that has write down all camera lenses except disallowable camera lens until complete point of contact file of scanning;
Step 504: the camera lens in the new camera lens file is extracted characteristics of image (key frame eigenmatrix), and it is 5 parts that Fig. 4 is illustrated in when carrying out image characteristics extraction image region segmentation.The feature sizes of camera lens is 5 * 16 * 16 * 16 * 32=640kb, according to the distance between the storage camera lens in each camera lens and the database in the calculating video to be detected, whether exist in the specified data storehouse with video to be detected in the camera lens that mates, specifically can set up algorithm and handle (detailed process sees also said method embodiment step 200) according to database, when the characteristics of image of new camera lens exists in database, upgrade the relevant information of this camera lens correspondence in the described database with described characteristics of image, when not existing, determine that this video lens is new camera lens, set up algorithm by database the relevant information of new camera lens is kept in the database; In this step, will carry out the processing that database is set up algorithm to 3 sections videos to be detected respectively;
Step 505: by above-mentioned advertisement searching algorithm (detailed process sees also step 300 among the said method embodiment), the relevant information corresponding with camera lens to be detected of preserving in the database handled, finally the location advertising that in video to be detected, repeats of the advertisement video that will obtain to repeat.
Based on the step described in the foregoing description one, when detecting the advertisement video that repeats, can guarantee user's experience at these repeated content shielding or deletion, accuracy rate is higher.
The above; only for the preferable embodiment of the present invention, but protection scope of the present invention is not limited thereto, and anyly is familiar with those skilled in the art in the technical scope that the present invention discloses; the variation that can expect easily or replacement all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.

Claims (6)

1. a television advertising detection method is characterized in that, comprising:
Generate at least one camera lens according to video to be detected;
Characteristics of image to described camera lens extracts, and makes up database with described characteristics of image, and described database storing has advertisement camera lens and corresponding relevant information; Described structure database specifically comprises:
Extract the eigenmatrix of all pictures at least one camera lens, with the key frame eigenmatrix of each camera lens characteristics of image as this camera lens;
According to described characteristics of image, calculate in the video to be detected distance between the storage camera lens in the camera lens and database respectively;
When distance is greater than predetermined threshold between the storage camera lens in camera lens and the database in the video to be detected, there is camera lens in the specified data storehouse with video lens coupling to be detected, and upgrade the relevant information corresponding of preserving in the described database with this camera lens, otherwise, there is not camera lens in the specified data storehouse with video lens coupling to be detected, add the relevant information of this camera lens correspondence in described database, described relevant information comprises the filename of video to be detected, the eigenmatrix of each camera lens, position and in multiplicity one or more of this camera lens in video to be detected;
The described relevant information of storing in the described database is retrieved, so that detect the television advertising that repeats; The described relevant information of storing in the described database is retrieved, is specifically comprised:
Multiplicity in the described database equated and extract greater than one one group of cinestrip and corresponding relevant information;
By predetermined condition described relevant information is judged, when the described cinestrip that extracts all satisfies predetermined condition, determined that described cinestrip is same section advertisement, and determine the position of this advertisement according to the positional information of described cinestrip; When comprising the camera lens that does not satisfy predetermined condition in the described cinestrip, then determine from next initial camera lens of this camera lens be the initial camera lens of another section advertisement, distinguish different advertisements according to the positional information of this camera lens, and definite every section advertisement the position; Described predetermined condition comprises and counts up to identically entirely on the relevant position of filename array of each camera lens in the container that and the corresponding start frame numerical value in the pos array of each end frame in the pos array of a back camera lens and previous camera lens differs and is no more than 500; Described filename array is used to store the video file name that described camera lens occurs; Described pos array is used for storing the position of described camera lens at described video.
2. according to the described detection method of claim 1, it is characterized in that, generate at least one camera lens, specifically comprise according to video to be detected:
Video slicing to be detected is become at least one picture;
According to the camera lens partitioning algorithm described at least one picture is generated at least one camera lens.
3. detection method according to claim 1 is characterized in that, described method also comprises:
Preestablish the camera lens density threshold, after generating at least one camera lens, when camera lens density during greater than described camera lens density threshold according to video to be detected, determine that this camera lens is a TV programme, and TV programme deleted from video to be detected, otherwise, determine that this camera lens is that advertising programme keeps.
4. a television advertising pick-up unit is characterized in that, comprising:
The camera lens generation module is used for generating at least one camera lens according to video to be detected;
The Data Control module, the characteristics of image that is used for described camera lens that described camera lens generation module is generated extracts, and makes up database with described characteristics of image, and described database storing has the relevant information of advertisement camera lens and correspondence; Described Data Control module specifically comprises:
Feature extraction unit is used for extracting the eigenmatrix of at least one all picture of camera lens, with the key frame eigenmatrix of each camera lens characteristics of image as this camera lens;
Similarity calculated is used for according to described characteristics of image, calculates in the video to be detected distance between the storage camera lens in the camera lens and database respectively;
The judgment processing unit, be used for when distance is greater than predetermined threshold between the storage camera lens in video camera lens to be detected and the database, there is camera lens in the specified data storehouse with video lens coupling to be detected, upgrade the relevant information corresponding of preserving in the described database with this camera lens, otherwise, there is not camera lens in the specified data storehouse with video lens coupling to be detected, add the relevant information of this camera lens correspondence in described database, described relevant information comprises the filename of video to be detected, the eigenmatrix of each camera lens, position and in multiplicity one or more of this camera lens in video to be detected;
The advertisement retrieval module is used for the described relevant information that described database is stored is retrieved, so that detect the television advertising that repeatability occurs; Described advertisement retrieval module specifically comprises:
The data pick-up unit is used for described database multiplicity equated and extracts greater than one one group of cinestrip and corresponding relevant information;
The Data Detection unit, be used for described relevant information being judged by predetermined condition, when the described cinestrip that extracts all satisfies predetermined condition, determine that described cinestrip is same section advertisement, and determine the position of this advertisement according to the positional information of described cinestrip; When comprising the camera lens that does not satisfy predetermined condition in the described cinestrip, then determine to be the initial camera lens of another section advertisement from next initial camera lens of this camera lens, positional information according to this camera lens is distinguished different advertisements, and determine every section advertisement the position, described predetermined condition comprises and counts up to identically entirely on the relevant position of filename array of each camera lens in the container that and the corresponding start frame numerical value in the pos array of each end frame in the pos array of a back camera lens and previous camera lens differs and is no more than 500; Described filename array is used to store the video file name that described camera lens occurs; Described pos array is used for storing the position of described camera lens at described video.
5. according to the described pick-up unit of claim 4, it is characterized in that described camera lens generation module specifically comprises:
Video processing unit is used for video slicing to be detected is become at least one picture;
The camera lens cutter unit is used for according to the camera lens partitioning algorithm at least one picture of described video processing unit cutting being generated at least one camera lens.
6. according to the described pick-up unit of claim 4, it is characterized in that described device also comprises:
The camera lens filtering module, be used to preestablish the camera lens density threshold, after generating at least one camera lens according to video to be detected, when camera lens density during greater than described camera lens density threshold, determine that this camera lens is a TV programme, and TV programme deleted from video to be detected, otherwise, determine that this camera lens is that advertising programme keeps.
CN2009101672490A 2009-08-27 2009-08-27 Method and device for detecting television advertisements Expired - Fee Related CN101650740B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2009101672490A CN101650740B (en) 2009-08-27 2009-08-27 Method and device for detecting television advertisements

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2009101672490A CN101650740B (en) 2009-08-27 2009-08-27 Method and device for detecting television advertisements

Publications (2)

Publication Number Publication Date
CN101650740A CN101650740A (en) 2010-02-17
CN101650740B true CN101650740B (en) 2011-09-21

Family

ID=41672978

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2009101672490A Expired - Fee Related CN101650740B (en) 2009-08-27 2009-08-27 Method and device for detecting television advertisements

Country Status (1)

Country Link
CN (1) CN101650740B (en)

Families Citing this family (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2579258A1 (en) * 2011-10-04 2013-04-10 Thomson Licensing Method of automatic management of a collection of images and corresponding device
CN102404064B (en) * 2011-11-30 2018-03-23 上海博泰悦臻电子设备制造有限公司 Shielding radio advertisement method and system
CN103425474B (en) 2012-05-24 2016-08-17 国际商业机器公司 For the method and apparatus obtaining the content in Snipping Tool
CN103237233B (en) * 2013-03-28 2017-01-25 深圳Tcl新技术有限公司 Rapid detection method and system for television commercials
CN103248953B (en) * 2013-04-24 2017-04-19 Tcl集团股份有限公司 Method and system for processing TV advertisement time and TV programme playing terminal
CN104185017B (en) * 2013-05-23 2017-02-08 中国科学院深圳先进技术研究院 Video matching method and system
CN104424436B (en) * 2013-08-28 2019-02-15 腾讯科技（深圳）有限公司 A kind of malice advertisement recognition method and device
CN104636940A (en) * 2013-11-07 2015-05-20 深圳市腾讯计算机系统有限公司 Advertisement delivery method, advertisement delivery system, advertisement screening method, advertisement screening device, and server
CN103596016B (en) * 2013-11-20 2018-04-13 韩巍 A kind of multimedia video data treating method and apparatus
CN106375847A (en) * 2015-07-23 2017-02-01 无锡天脉聚源传媒科技有限公司 Template generation method, template generation device, video updating method and video updating device
CN106372092A (en) * 2015-07-23 2017-02-01 无锡天脉聚源传媒科技有限公司 Method and device for generating template, and video update method and device
JP6524899B2 (en) * 2015-12-02 2019-06-05 富士通株式会社 Secret data matching device, secret data matching program, and secret data matching method
CN105760472A (en) * 2016-02-06 2016-07-13 中国农业大学 Video retrieval method and system
CN107133266B (en) * 2017-03-31 2020-02-18 北京奇艺世纪科技有限公司 Method and device for detecting video shot type and method and device for updating database
CN107135401B (en) * 2017-03-31 2020-03-27 北京奇艺世纪科技有限公司 Key frame selection method and system
CN107133264A (en) * 2017-03-31 2017-09-05 北京奇艺世纪科技有限公司 The detection method and device of camera lens and the update method of advertising database and device
CN107169004B (en) * 2017-03-31 2021-07-30 北京奇艺世纪科技有限公司 Feature database updating method and device
CN107729489A (en) * 2017-10-17 2018-02-23 北京京东尚科信息技术有限公司 Advertisement text recognition methods and device
CN108153882A (en) * 2017-12-26 2018-06-12 中兴通讯股份有限公司 A kind of data processing method and device
CN108470290A (en) * 2018-03-28 2018-08-31 百度在线网络技术（北京）有限公司 Commercial detection method, device and server
CN108540833A (en) * 2018-04-16 2018-09-14 北京交通大学 A kind of television advertising recognition methods based on camera lens
CN109189991B (en) * 2018-08-17 2021-06-08 百度在线网络技术（北京）有限公司 Duplicate video identification method, device, terminal and computer readable storage medium
CN111008868A (en) * 2019-12-05 2020-04-14 秒针信息技术有限公司 Advertisement monitoring method, system and computer readable storage medium

Also Published As

Publication number Publication date
CN101650740A (en) 2010-02-17

Similar Documents

Publication Publication Date Title
CN101650740B (en) Method and device for detecting television advertisements
US8295611B2 (en) Robust video retrieval utilizing audio and video data
US6366699B1 (en) Scheme for extractions and recognitions of telop characters from video data
CN100499751C (en) Moving image division apparatus, caption extraction apparatus and method
US8316301B2 (en) Apparatus, medium, and method segmenting video sequences based on topic
US9087125B2 (en) Robust video retrieval utilizing video data
US7983340B2 (en) Extracting key frames from video using a triangle model of motion based on perceived motion energy
CN101427250B (en) Data summarization system and method for summarizing a data stream
US20130163864A1 (en) Video detection system and methods
JP2004357302A (en) Method and system for identifying position in video by using content-based video timeline
WO2003055202A1 (en) Family histogram based techniques for detection of commercials and other video content
WO2009069840A1 (en) Method and system for providing information based on logo included in digital contents
WO2008074877A1 (en) Method for creating a new summary of an audiovisual document that already includes a summary and reports and a receiver that can implement said method
KR100612842B1 (en) An apparatus and method for deciding anchor shot
WO2010078629A1 (en) A system for real time near-duplicate video detection
Yu et al. A multi-resolution video segmentation scheme for wipe transition identification
KR20050033075A (en) Unit for and method of detection a content property in a sequence of video images
¯ Cerneková et al. Entropy metrics used for video summarization
Quénot TREC-10 Shot Boundary Detection Task: CLIPS System Description and Evaluation.
KR20100109345A (en) Apparatus and method for extracting and decision-making of spatio-temporal feature in broadcasting and communication systems
Kim et al. An efficient graphical shot verifier incorporating visual rhythm
Detyniecki et al. Video rushes summarization by adaptive acceleration and stacking of shots
Heng et al. The implementation of object-based shot boundary detection using edge tracing and tracking
KR100262344B1 (en) Detectiong method of dissolving gradual scene changes in video data analysis
Kawai et al. A Localization Method of Video Text in TV Program Video

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20110921

Termination date: 20150827

EXPY Termination of patent right or utility model