JP2007267211A - Commercial detection method, commercial detection device and medium for recording commercial detection program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an inexpensive commercial detection method which can accurately detect a commercial, even if a short commercial such as a program propaganda is inserted between commercials. <P>SOLUTION: In this solution, commercial switching timing is detected. When a time interval from one commercial to the next commercial is 15 seconds or 30 seconds equivalent to a time interval for single commercial broadcasting, the interval is selected as a commercial candidate interval. When the time interval is 5 seconds equivalent to the other kind of a commercial broadcasting time such as a program propaganda, the interval is selected as a quasi-commercial candidate interval. For example, when the length of a block composed of a commercial candidate interval and a quasi-commercial candidate interval exceeds 60 seconds, the interval is determined as a commercial interval to absolutely detect the commercial interval even if the other kind of a short commercial like a program propaganda is inserted between two commercials. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a commercial detection method and apparatus for automatically detecting a commercial included in a broadcast signal.

  For example, it automatically detects commercials (hereinafter referred to as CMs) included in broadcast signals typified by television broadcasts, monitors whether CMs are being broadcast as contracted, or registers new CMs in the database. Conventionally, a broadcast signal is recorded by automatically skipping CMs.

  FIG. 11 is a first conventional example relating to a method for detecting a CM inserted in a main program of a television broadcast (Patent Document 1). This is because there is a short period of silence when switching from the main program to CM or from one CM to another CM, and one CM has a certain length of time such as 15 seconds or 30 seconds. The CM is detected by paying attention to the fact that there is a silent section detecting unit for detecting a silent section from the audio signal of the TV signal, and when the time interval of the adjacent silent section corresponds to the time of the CM, the section is CM. Among the blocks in which a plurality of CM candidate sections are continuously arranged, and a block having the largest number of CM candidate sections included in the block, a block having the largest number of CM candidate sections is determined as a CM block. It is composed of a CM broadcast section determination unit and a continuity evaluation unit. In addition, a cut detection unit that detects a cut based on the detection result of the feature amount of the received image using the fact that the screen is cut when the CM is switched is provided, and the reliability of the extracted CM candidate section is improved. It also describes the technology to be used.

  FIG. 12 shows a second conventional example for detecting a CM included in a television broadcast signal (Patent Document 2). In this example, one CM has a length of 15 seconds or 30 seconds, cuts often occur in the CM, and the CM section often appears continuously for 1 minute or more. The CM is detected by utilizing the property peculiar to the CM that there is a silent section at the time of switching. Specifically, the cut sequence Ct that detects the time when the screen cut occurs and the silent segment sequence S that detects the silent interval are obtained, and the time when both occur simultaneously, that is, the silent interval and the screen is cut. When the time interval between the occurrences is 15 seconds or 30 seconds, the section is taken out as a CM candidate section. When the length of consecutive CM candidate sections is 60 seconds or more, the section is determined as a CM section.

  Patent Document 3 describes a third conventional example relating to a CM detection method. This is because CM audio data is registered in the database in advance, and from the time when both the silent section and screen cut included in the received broadcast signal occur simultaneously to the next time when the silent section and screen cut occur simultaneously. , That is, when the time interval in which the screen cut and the silent section occur at the same time coincides with a predetermined value as the number of CM broadcast seconds, that period is set as a CM candidate section, and the CM data in which the sound data of the CM candidate section is stored in advance By collating with audio data, erroneous detection of CM is eliminated and detection reliability is improved.

Furthermore, as the simplest and classic CM detection method, there is a method using an audio multiplexing method (Patent Document 4). This is a method of judging a stereo sound part as a CM by paying attention to the fact that CM uses stereo sound and the main part of the program uses monaural sound.
JP 2003-047031 A Japanese Unexamined Patent Publication No. 09-219835 Japanese Patent Laid-Open No. 03-289829 Japanese Patent Laid-Open No. 02-081344

  In any of the above first to third conventional examples, a silent section occurs when switching CMs, and screen cuts occur frequently. The broadcast time of one CM is 15 seconds or 30 seconds. Since the CM block composed of a plurality of continuous CMs is a method using the characteristic of CM that it continues for one minute or longer, if short information such as program advertisement is inserted between CMs, the above-mentioned There is a drawback that the CM detection fails because the condition is not satisfied. Further, since it is necessary to detect not only a silent section but also a screen cut, the processing cost is high, and there is a disadvantage that it cannot be easily used for a hard disk or the like. In addition, since the number of stereo broadcasts has increased recently in the main program, the fourth conventional example can hardly be used.

  The CM detection method of the present invention detects a time point when a CM is switched, and when a time interval between adjacent CM switching time points corresponds to a broadcasting time of one CM such as 15 seconds or 30 seconds, the CM candidate section (first As a commercial candidate section), and when the time interval corresponds to another type of CM broadcasting time such as a program advertisement of 5 seconds, for example, detects as a semi-CM candidate section (second commercial candidate section), If the length of a block formed by both the CM candidate section and the quasi-CM candidate section exceeds 60 seconds, for example, the block is determined as a CM section. As described above, when another type of CM such as a program advertisement is inserted between the CMs by detecting not only the conventional CM candidate section but also the quasi-CM candidate section. However, CM can be detected accurately.

  The second CM detection method of the present invention detects a CM switching time point, and when a time interval between adjacent CM switching time points corresponds to a broadcasting time of one CM such as 15 seconds or 30 seconds, Detecting, extracting feature amounts of CM contents such as CM images and CM sounds for each CM candidate section, classifying each CM candidate section into a CM cluster according to the similarity of the feature quantities, A CM candidate section in a CM cluster that includes only one is considered not to be a CM and is deleted. If the length of a block formed by the remaining CM candidate sections exceeds, for example, 60 seconds, the block is defined as a CM section. It is determined. In this way, since erroneous detection of CM can be prevented by using clustering based on the feature amount of content (sound or image) included in the CM candidate section, a large processing amount such as scene cut detection is not required. CM detection accuracy can be improved. Further, by registering the CM classified into the CM cluster in the database, the CM database can be created without manpower, so the database can be created economically.

  As described above, according to the CM detection method of the present invention, a CM block can be accurately detected even when another type of CM such as a program advertisement having a shorter time than the CM is inserted between CMs. Can do. Further, without requiring a large amount of processing such as scene cut detection, it is possible to reduce CM false detection and improve CM detection accuracy. Further, by storing the CM classification result in the database, the CM database can be constructed at low cost without manpower.

First Embodiment FIG. 1 is a CM detection procedure that is the gist of the present invention, and FIG. 2 is an example of an input signal that is an operation target of the present invention. In FIG. 2, 50 is a program main part, 51 is a normal CM corresponding to the first commercial (CM1), 52 is another type of CM such as a program advertisement broadcasted between normal CMs 51. 2 commercials (CM2). In most cases, a plurality of CMs and program advertisements are collectively inserted into a program. 53 is a CM 51 (first commercial) and a program advertisement 52 (second commercial) inserted between the main program 50. This is a CM block to be formed. Usually, a plurality of CMs 51 and a program advertisement 52 are broadcast continuously. The number of broadcasts and the broadcast order of the CMs 51 and 52 in the CM block 53 vary, but the present invention can be applied to either case. It is an object of the present invention to detect the CM block 53 from the received broadcast signal. CM 51 is broadcast in a predetermined time (first predetermined length) such as 15 seconds or 30 seconds, and program advertisement 52 is shorter than CM 51, for example, a predetermined time (second predetermined time) such as 5 seconds. Broadcast). The CM block 53 is detected using the fact that the broadcast times of the CM 51 and the program advertisement 52 are determined in advance. Here, the description will be given for television broadcasting and the like, but of course it is not limited to this, and it is of course applicable to radio broadcasting and moving image distribution using the Internet. In short, any transmission medium may be used as long as the signal format shown in FIG.

  In FIG. 1, S1 is a CM boundary detecting step for detecting a CM boundary that is the time when the CM 51 or the program advertisement 52 is switched, S2 is a step for detecting a CM candidate section that is a section of the CM 51 based on the CM boundary, and S3 is a CM boundary. A step of detecting a quasi-CM candidate section which is a section of the program advertisement 52 inserted between the CM 51 and the CM 51 based on the above, S4 is a CM block 53 inserted in the main program based on the CM candidate section and the quasi-CM candidate section Is a step of detecting. The order of the CM candidate section detection step S2 and the quasi-CM candidate section detection step S3 does not matter. As shown in FIG. 1, the CM candidate section detection step S2 may be performed first, and conversely, the quasi-CM candidate section detection step S3 may be performed first. However, as described later, the CM candidate section detection step S2 and the quasi-CM candidate detection step are performed. S3 is preferably executed as a series of operations.

  In the CM boundary detection step S1, the CM boundary can be detected by detecting silence or detecting a scene change (screen cut). Specifically, various methods can be used. For example, in the silent section detection method, the received voice signal is digitized, the sum of squares of voice samples is calculated for each section of about 500 milliseconds, which is a typical CM switching time, and the value is smaller than a threshold value. The section can be determined as a silent section. As a method for detecting screen cuts, if the difference between two frame images acquired at intervals of several tens to several hundreds of milliseconds is calculated and the absolute value of the difference is greater than a threshold value, the scene has changed. Can be determined. Further, when the CM screen fades in and out, the variance of the pixel values of the entire screen can be calculated for each frame, and a place where the variance is small can be detected as the CM boundary. As described in the conventional example, the time when the silent section and the screen cut appear at the same time can be determined as the CM boundary.

  In CM candidate section detection step S2, a CM candidate section for detecting CM 51, which is the first commercial, is detected using the information on the CM boundary detected in step S1. Since CM51 is generally manufactured at time intervals of 15 seconds, 30 seconds, 45 seconds, etc., if the interval between two CM boundaries coincides with 15 seconds, 30 seconds, 45 seconds, etc., the two CM boundaries The section divided by is determined as a CM candidate section. Of course, when calculating the interval between CM candidate sections, it is desirable to make a determination with a width in consideration of an error of about several hundred milliseconds.

  In the quasi-CM candidate section detection step S3, a quasi-CM candidate section for detecting the program advertisement 52, which is the second commercial, is detected using the information on the CM boundary detected in step S1. Since the program advertisement 52 is produced at an interval shorter than the CM 51, for example, at a time interval of 5 seconds, for example, when the interval between two CM boundaries is 5 seconds, the interval is determined as a quasi-CM candidate interval.

  In the CM block detection step S4, the CM block 53 is detected based on the information of the CM candidate section detected in Step S2 and the quasi-CM candidate section detected in Step S3. Although the details will be described later, the outline of this step is as follows. That is, since the CM block 53 has a property that it often lasts about 60 seconds even if it is short, when the CM candidate section or the quasi-CM candidate section continues for 60 seconds or more, the CM candidate section and the quasi-CM candidate section A column can be detected as a CM block. This is the invention described in claim 2. In addition, since this detection operation is performed for each CM boundary, a plurality of CM blocks are detected in most cases. However, if there are CM blocks that overlap each other, only the CM block having the maximum section length is included. Is adopted, and other CM blocks are deleted. This is based on an empirical rule that when there are overlapping CM blocks, the CM block with the longest length is correct, and this is the invention described in claim 3. Note that, instead of the length of the CM block, a CM block having a large number of CM candidate sections included in the CM block may be employed, and this is the invention described in claim 4.

  Since the program advertisement 52, that is, the quasi-CM candidate section is often inserted between a plurality of CMs 51, that is, between the CM candidate sections, the quasi-CM candidate section does not continue for a certain number of times, or the head of the CM block. Alternatively, it is possible to reduce the processing load by suppressing the excessive detection of CM blocks using the property that there is no quasi-CM candidate section at the end.

  FIG. 3 is a diagram for explaining the principle of the CM detection method of the present invention. Here, it is assumed that, for example, t1, t2 to t11 are detected as CM boundaries detected by the CM boundary detection step S1 using detection of a silent section or a screen cut. CM boundary = [t1, t2,... T11]. In order to make the explanation easy to understand, an example of a specific time is used. For example, CM boundaries = [00:00, 00:07, 00:15, 00:30, 00:35, 00:37, 00: 45, 01:00, 01:05, 01:07, 01:15]. (00:07 means 0 minutes and 7 seconds.) Also, two types of lengths Δ1 and Δ2 are assumed as the time length of the normal commercial 51 that is the first commercial. Of course, there may be three or more types. As a specific example, it is assumed that Δ1 = 15 seconds and Δ2 = 30 seconds. As a time length of the program advertisement 52 which is the second commercial, δ shorter than Δ1 and Δ2 is assumed. Of course, there may be two or more types. As a specific example of δ, 5 seconds is assumed.

  First, at the first CM boundary t1, it is searched whether there is a nearest CM boundary whose time length to the next CM boundary corresponds to Δ1, Δ2, and δ. Here, since the interval up to t3 (00: 15-00: 00) corresponds to Δ1 (15 seconds), the CM candidate section 61 having the length Δ1 is extracted. Note that the interval up to t4 (00: 30-00: 00) corresponds to Δ2 (30 seconds), but since the section up to t3 has already been extracted as the CM candidate section, the section up to t4 is not extracted. Further, since the quasi-CM candidate section is detected independently from the CM candidate section, if there is a CM boundary after δ from t1, it is extracted as a quasi-CM candidate section, but here there is no CM boundary corresponding to it. Therefore, the quasi-CM candidate section is not extracted at t1. At the next CM boundary t2, a CM candidate section 66 having a length Δ2 (30 seconds) up to t6 is extracted in the same manner as described above. However, since this CM candidate section 66 is not continuous with the CM candidate section 61 extracted at t1, it becomes a CM block of another series. This will be described later. At t3, the CM candidate section 62 having the length Δ1 up to t4 is extracted in the same manner as t1. Since this is continuous with the CM candidate section 61 extracted at t1, it becomes a CM candidate section of the same series. At t4, a CM candidate section 63 having a length Δ1 up to t7 is extracted and a quasi-CM candidate section 67 up to t6 is extracted. The above operation is repeated until t10. As a result, several CM candidate sections having two types of lengths, Δ1 and Δ2, and a quasi-CM candidate section having a length δ are extracted. The above is the operation in the CM candidate section detection step S2 and the quasi-CM candidate section detection step S3. Thus, these two steps are preferably performed as a series of flows.

  Next, a CM block is detected based on the extracted CM candidate section and quasi-CM candidate section. First, considering the CM boundary t1 as a starting point, CM candidate sections 61 of length Δ1 extracted at t1, t3, t4, t7, and t8, respectively, are formed continuously in time. For easy understanding, reference numerals 61 to 65 are assigned to these CM candidate sections in FIG. The time length of this block is t1 to t11, which exceeds a predetermined time length, for example, 60 seconds, and is extracted as one CM block 57-1. Next, CM candidate sections 61 and 62 having a length Δ1 extracted at t1 and t3, a quasi-CM candidate section 67 extracted at t4, and a CM candidate section 69 having a length Δ2 extracted at t5 are temporally continuous. Since the time length t1 to t9 during this period also exceeds 60 seconds, this is extracted as another CM block 57-2. Since there are only two CM blocks starting from t1, the next CM block starting from t2 is detected. Here, CM candidate sections 66 and 68 of length Δ2 extracted at t2 and t6 are continuous in time, and this section exceeds 60 seconds, so this is extracted as another CM block 57-3. . Since this is the only CM block starting from t2, a CM block starting from t3 is next detected. Here, CM candidate sections 62, 63, 64, and 65 of length Δ1 extracted at t3, t4, t6, and t8 are continuous in time, and since this is 60 seconds, it can be extracted as a CM block. Since it is included in the CM block 57-1 already detected at the time of starting t1, illustration is omitted.

  In the case of t4 starting point, there is a connection between a CM candidate section of length Δ1 extracted at t4, t7 and t8, and a connection between a quasi-CM candidate section extracted at t4 and a CM candidate section of length Δ2 extracted at t6. None of these are extracted as CM blocks because the time length does not reach 60 seconds. Even at t5 starting point, there is only one CM candidate section of length Δ2, and since it does not reach 60 seconds, it is not extracted as a CM block. If the above process is repeated until t10, three effective CM blocks can be finally extracted. Since these blocks overlap in time, the CM block 57-1 having the longest time is determined as the correct CM section, and the CM block 57-2 and the CM block 57-3 are rejected. This is the operation of the CM block detection step S4, whereby the CM block can be detected.

  FIG. 4 and FIG. 5 show the detection method shown in FIG. 3 in detail as an operation algorithm. FIG. 4 shows the entire operation procedure, which is an algorithm for detecting a CM block by inputting the CM boundary detected in the CM boundary detection step S1. In S10, the CM boundary ti (i = 1 to N) extracted in the CM boundary detection step S1 is input, and the output variable is reset. In S11, by setting i = 1, CM block extraction processing is performed for the first CM boundary t1. S12 is a subroutine part for searching for a CM block, and the detailed algorithm is shown in FIG. Since this subroutine searches for the CM block for the i-th CM boundary and stores the search result in the variable Block, if i = 1, the CM block is searched for t1. CM blocks can be expressed as an array of CM boundary times. For example, the CM block 57-1 shown in FIG. 3 can be expressed as [00:00, 00:15, 00:30, 00:45, 01:00, 01:15]. In S12, it is checked whether the length of the CM block taken out in S12, that is, the length of the variable Block exceeds 60 seconds. If it does not exceed, the block is judged not to be a CM block. A process of incrementing i by 1, that is, a process of advancing the CM boundary by one is performed. The length of the variable Block can be calculated by taking the time difference between the end and the beginning of the stored time sequence.

  If the length of the block searched in S12 is 60 seconds or more, since the block may be a CM, it is checked in S14 whether there is an overlap with the already extracted CM block. If there is no overlap, it has become clear that an already extracted block (a block that has already been stored in the variable prev_block) does not overlap with the CM block for the subsequent CM boundary. The block is taken out as a CM block, and the CM block extracted in the current process in S17 is made an already extracted block, and then the process moves to the next CM boundary process in S18. If the block searched this time (stored in variable Block) in S14 overlaps the already extracted block (stored in variable prev_block), the lengths of both blocks are compared in S16. If it is long, it is moved to the already extracted block, and then the process proceeds to the next CM boundary processing in S18. If the already extracted block is longer, it is rejected as the CM block searched this time is not a CM block. To do. In S19, it is determined whether or not the above processing has been executed for all CM boundaries. If not completed, the process returns to S12. If completed, the block already extracted in S20 (the block stored in the variable prev_block) is added to the block list. to add.

  Through the above steps, a variable blocklist that is a CM block can be output. For example, in the case of the example shown in FIG. 3, the CM block 57-1 having the longest length is extracted from the three CM blocks 57-1, 57-2, 57-3 output in S12. Therefore, the blocklist = [00:00, 00:15, 00:30, 00:45, 01:00, 01:15] corresponding to this is obtained. In S13, whether or not the block is a CM block is determined based on the length (time interval) of the searched block. However, the determination is based on the number of CM candidate sections included in the block instead of the time interval. You can also. In this case, in S13, it is a condition for determining whether or not the number of CM candidate sections included in the block is, for example, three or more. The determination condition in S16 can also be based on the number of CM candidate sections included in the block instead of the length of the block as in S13.

  FIG. 5 shows the detailed algorithm of the block search subroutine S12. The first half of this flow is a CM block search part starting from a CM candidate section, and the second half is a CM block search part starting from a quasi-CM candidate section. Here, a procedure for searching for a CM block starting from the CM boundary ti is shown. Therefore, since the time interval between the CM boundary after ti + 1 and ti is checked, the time interval between ti between S1201 and S1203 corresponds to any of 15 seconds, 30 seconds, and 45 seconds. Check whether there is a CM boundary to be checked. If there is no corresponding CM boundary, it is determined that no CM candidate section has been found, and the process moves to a determination point as to whether or not to search for a quasi-CM candidate section. If there is a corresponding CM boundary, that section is extracted as a CM candidate section, and the CM candidate section (ti, tj) is added to the block of the CM block in S1205. In addition, in order to repeat the same process for the subsequent CM boundaries starting from the CM boundary tj, this function is recursively executed in S1204. At this time, not only the CM candidate section but also the CM block starting from the quasi-CM candidate section is searched. In this way, until a CM candidate section cannot be found, a process for searching for a CM candidate section is performed while increasing the CM boundary tj in S1206, and the result is added to the variable block of the CM block in S1205.

  For example, in the example shown in FIG. 3, when this function is executed with the CM boundary t1 as the starting point, first, j = 2 is set in S1201, that is, whether the time interval between t2 and t1 is 15 seconds, 30 seconds, 45 seconds or not. Check in. Since this time is 7 seconds, the above condition is not satisfied, so the process proceeds to S1203. Since the condition of S1203 is not satisfied, the process returns to S1201 and the CM boundary t3 in which j is increased by 1 is checked in the same manner. Since the time interval between t3 and t1 is 15 seconds, the condition of S1202 is satisfied and the routine proceeds to S1204, and t1 is substituted into the variable block. In S1204, the same processing is repeated starting from t3, and CM candidate sections up to t4 are found. Therefore, t3 is added to the variable block, and block = [t1, t3]. Further, the same processing is repeated after t4 to extract one CM block 57-1. The variable block at this time is [t1, t3, t4, t7, t8, t11].

Returning to FIG. 5, S1207 is a determination as to whether or not to search for a CM block starting from the quasi-CM candidate section, and if so, the process proceeds to S1208. The basic algorithm is the same as in the case of CM candidate section extraction. The difference is that the determination condition of the time interval in S1209 is 5 seconds, and the variable representing the discovered quasi-CM candidate section is block2. It has only changed. In S1213, the length of the CM block starting from the CM candidate section is compared with the length of the CM block starting from the quasi-CM candidate section. If the latter is longer, it is determined that the latter is the correct CM block, and the latter is determined in S1214. This is a CM block. If the above processing is executed for all the CM boundaries, in this subroutine, for example, as already described in the examples of FIGS. 3 and 4, three CM blocks 57-1, 57-2, 57-3 are extracted. be able to.
Second Embodiment The first embodiment is a CM detection method that uses only time information such as a silent section of an audio signal and a screen switching time. However, a classification result of CM contents such as a CM image and audio is used. If used together, the CM detection accuracy can be further improved. First, the CM content classification method on which this method is based will be described. FIG. 6 shows a content classification procedure. In S30, the feature amount of the content is calculated from the audio and image data included in the CM candidate section CMi extracted in the CM candidate section detection step S2 shown in FIG. There are various methods for calculating the feature amount. For example, in the case of speech, the number of zero crossings, short-time power spectrum, LPC cepstrum, MFCC, etc. can be used. Since it is not a feature of the present invention, description thereof is omitted. In S31, each CM candidate section CMi is classified (clustered) in the feature space based on the feature amount calculated in S30. As a classification method, an existing method such as a k-average method can be used. As a simple method, the maximum radius α of each cluster is set, and if the distance between the representative vector that is the average of the feature vectors of each cluster and the feature vector of each CM candidate section CMi is less than or equal to α, There is a method for determining that CMi belongs to the cluster.

  FIG. 7 schematically shows the classification process. Since CM1, CM2, and CM3 are included in a hypersphere having a radius α or less in the feature space, it is determined that they belong to one cluster CMC1. On the other hand, since CM4 is not included in the hypersphere of CMC1, another cluster CMC2 is newly generated. Returning to FIG. 6, in S32, all the generated clusters CMCj (j = 1 ·· N) are output. CM candidate sections belonging to the same cluster can be determined to be the same CM broadcast at different times. In addition, although the example which extracts the feature-value from the whole CM candidate area was described in S31, only a part of CM candidate area can also be used. For example, since the advertiser's company name, logo, etc. are often displayed at the end of the CM, by extracting the feature amount from the last 0.5 second of the CM candidate section, They can also be categorized into company units rather than product units. Further, if a company logo is registered in advance, classification can be performed based on the degree of similarity with the logo.

  FIG. 8 shows a procedure for improving the CM detection accuracy using the above content classification method. This is a process executed after the classification shown in FIG. 6, and is executed after the CM candidate section detected in the CM candidate detection step S2 shown in FIG. 1 is classified into each cluster by the process shown in FIG. The S40 is a step of setting the first CM candidate section CM1. In S41, the cluster CMCj to which the CM candidate section CMi belongs is searched. At first, the cluster to which CM1 belongs is searched. In S42, it is checked whether or not there are a plurality of CM candidate sections belonging to the cluster CMCj. If there are no CM candidate sections, the CM candidate section CMi is deleted from the CM candidate sections in S43. This is because in the television program, it is rare that the exact same video is broadcast twice, whereas the commercial has the property that it is repeatedly broadcast a plurality of times. In S44 and S45, the same processing is repeated until the last CM candidate section. As described above, it is desirable to perform re-sorting using the content classification method for all detected CM candidate sections, but it is also possible to select only the first or last CM candidate section of the CM block as the target of the sorting process.

  FIG. 9 is a configuration example of a CM detection apparatus that implements the CM detection method of the present invention described above. This is a configuration corresponding to both the first embodiment and the second embodiment. 1 is a tuner unit that receives, for example, a television broadcast signal and outputs a video signal and an audio signal, 2 is a CM boundary detection unit that detects a CM boundary signal from the signal output by the tuner unit, and 3 is based on the CM boundary signal CM candidate section detecting unit for detecting a CM candidate section, 4 is a quasi-CM candidate section detecting unit for detecting a quasi-CM candidate section based on a CM boundary signal, and 5 is a CM based on the detected CM candidate section and quasi-CM candidate section. These are CM block detection units for detecting blocks, and these are components necessary for carrying out the CM detection method of the first embodiment. Reference numeral 6 denotes a CM information accumulating unit that accumulates detected CM information, for example, information on CM candidate sections and quasi-CM candidate sections, information on CM blocks, and the like. A flag for identifying a candidate section or a quasi-CM candidate section, a CM appearance frequency, and the like are stored.

  7 is a CM classifying unit that classifies CM candidate sections into clusters based on feature quantities of contents such as images, and 8 is a CM cluster storing unit that stores the classified CM candidate sections. These are the CM detection units according to the second embodiment. Additional components necessary to carry out the method. 9 is a control unit that controls the operation of the apparatus, 10 is an encoding unit that encodes video and audio signals output from the tuner unit 1, 11 is a video storage unit that stores the encoded video, and 12 is each component. It is a signal bus that connects the parts. The encoding unit 10 compresses the received video signal and audio signal and converts them into a signal format suitable for storage, but is not related to the operation of the present invention. Since the tuner 1 has an appropriate function according to the type of the received wave, it can cope with various broadcasting forms such as video distribution using CATV and moving image distribution using the Internet as well as television broadcasting. Hereinafter, the operation of the present apparatus will be described taking the case of television broadcasting as an example.

  The received broadcast wave is input to the tuner 1, and a video signal and an audio signal of a specific channel are output from the tuner 1. By including an AD conversion function in the tuner 1, a digital signal can be output. The audio signal and the video signal are compressed and multiplexed by the encoding unit 10 and stored in the video storage unit 11. The encoding unit 10 can use, for example, an MPEG encoder. The video signal and audio signal output from the tuner 1 are input to the CM boundary detection unit 2.

  The CM boundary detection unit 2 includes, for example, a silence detection unit 2-1 and a cut detection unit 2-2, and an AND unit 2-3 that takes a logical product of the outputs. The silence detector 2-1 averages the power of the voice data input in time series over a period of several hundred milliseconds, and detects the silence interval when the value is smaller than a predetermined threshold. The cut detection unit 2-2 takes a difference for each pixel of the frame image input as a time series, and outputs a signal indicating the cut point when the absolute value or the sum of the squares of the difference is larger than a predetermined threshold value. To do. The AND section 2-3 takes the logical product of the outputs of the silence detection section 2-1 and the cut detection section 2-2 and outputs a CM boundary detection signal as a CM boundary when both cut and silence are detected simultaneously. Output.

  The CM candidate section detection unit 3 processes the CM boundary detection signal input from the CM boundary detection unit 2 to detect a CM candidate section and records it in the CM information storage unit 6. Here, the time at which the CM boundary is detected is accumulated for a certain time (for example, 60 seconds), and the section delimited by the CM boundary when the CM boundary appears at any interval of 15 seconds, 30 seconds, or 45 seconds Is output as a CM candidate section.

  The quasi-CM candidate section detector 4 detects the quasi-CM candidate section on the same principle as the CM candidate section detector 3. That is, a CM boundary that appears at a time interval shorter than the CM section such as 5 seconds or 10 seconds is detected.

  The CM information storage unit 6 can be configured as a relational database table, for example. The CM classification unit 7 is activated when the detected CM candidate section is written in the CM information storage unit 6, and is nearest to the newly written CM candidate section among the clusters stored in the CM cluster storage unit 8. If the distance to the cluster is smaller than a predetermined threshold, the cluster is classified into that cluster. Otherwise, it is determined that a new CM cluster has been found, and the information is stored in the CM cluster storage unit 8. to add.

  The CM block detection unit 5 scans the CM information storage unit 6 and, when a CM candidate section or a quasi-CM candidate section appears continuously for a predetermined time or more, outputs the section as a CM block. This detailed operation has already been described.

  The CM detection method of the present invention can be applied to a DVD recorder or a hard disk recorder. For example, it is possible to record a television program or a moving image distributed over the Internet, and when viewing the recorded program, cut the CM and reproduce only the main program. Also, by removing the CM when dubbing video onto a storage medium such as a DVD, it is possible to efficiently record a large number of videos on a limited capacity storage medium. Also, in CM monitoring work, information regarding CMs can be efficiently obtained by displaying the detected CM candidate sections in a timeline or icon.

  FIG. 10 shows an example of a screen for CM monitoring. 71 is a video display area for displaying recorded video, 72 is a CM information list for displaying CM information extracted according to the present invention, 73 is a channel selection button for selecting a broadcast channel to be received, and 74 is for displaying CMs by time. , 75 is a timeline display area. The CM information list 72 displays a start time, an end time, a time length, a frequency, and the like as an example of the extracted CM section information. This frequency is the number of times the same CM is broadcast within a certain time, and can be acquired as a result of the CM classification process. It is also possible to register related information in the remarks column. In the timeline display area 74, the time axis is displayed on the vertical axis, the time in minutes in each time zone is displayed on the horizontal axis, and the appearance of CMs can be visualized by spatially displaying CM sections. Can be displayed as a list. The broadcast time of the CM selected in the CM information list can be displayed using the marker 75.

It is CM detection procedure of this invention. It is an example of a structure of the broadcast signal which is a process target of this invention. This is the principle of the CM detection method of the present invention. It is an operation algorithm of the CM detection method of the present invention. It is a detailed algorithm of a CM block search subroutine included in the operation algorithm of the CM detection method of the present invention. This is a CM content classification procedure. It is a conceptual diagram of clustering of CM candidate sections. It is the re-selection procedure of CM candidate area of this invention using clustering. It is a structure of CM detection apparatus of this invention. It is an example of the screen of CM monitoring apparatus which is an application example of this invention. It is the 1st prior art example of CM detection method. It is the 2nd prior art example of CM detection method.

Explanation of symbols

2 CM boundary detection unit 3 CM candidate detection unit 4 Quasi-CM candidate detection unit 5 CM block detection unit 6 CM information storage unit 7 CM classification unit 8 CM cluster storage unit 9 Control unit (CPU)
51 Regular CM (1st commercial)
52 Another type of commercials (second commercial)
S1 CM boundary detection step S2 CM candidate section detection step S3 Quasi-CM candidate section detection step S4 CM block detection step

Claims (12)

A commercial block composed of a first commercial having a first predetermined length and a second commercial having a second predetermined length different from the first length is inserted into the broadcast program information A method for receiving a broadcast signal and detecting the commercial block from the broadcast signal,
Detect the commercial switching time, which is the time when the commercial switches,
A first commercial candidate section in which a time interval between adjacent commercial switching time points corresponds to the first length is detected for each commercial switching time point;
A second commercial candidate section in which a time interval between adjacent commercial switching points corresponds to the second length is detected for each commercial switching point;
A commercial detection method, wherein a block formed by the first commercial candidate section and the second commercial candidate section satisfies a predetermined condition, and determines that the block is a commercial block.   The commercial detection according to claim 1, wherein the predetermined condition is that a length of the first commercial candidate section and the second commercial candidate section included in the block is equal to or greater than a predetermined threshold value. Method.   3. The commercial detection method according to claim 1, wherein when a plurality of blocks satisfying the predetermined condition are detected, the longest block is determined to be a commercial block.   When a plurality of blocks satisfying the predetermined condition are detected, it is determined that a block having the largest total number of the first commercial candidate sections and the second commercial candidate sections included in the block is a commercial block. The commercial detection method according to claim 1, wherein: A method of receiving a broadcast signal in which a commercial block formed by a plurality of commercials having a predetermined length is inserted into broadcast program information, and detecting the commercial block from the broadcast signal,
Detect the commercial switching time, which is the time when the commercial switches,
A commercial candidate section in which a time interval between adjacent commercial switching points corresponds to the predetermined length is detected for each commercial switching point,
For each commercial candidate section, the feature amount of the commercial content is extracted, and each commercial candidate section is classified into a commercial cluster according to the similarity of the feature amount,
A commercial candidate section included in a commercial cluster including a predetermined number of commercial candidate sections is regarded as a regular commercial candidate section,
A commercial detection method characterized in that, when a block formed by a commercial candidate section regarded as a regular commercial candidate section satisfies a predetermined condition, the block is determined to be a commercial block.   The commercial detection method according to claim 5, wherein the predetermined condition is that a length of a block formed by a commercial candidate section is equal to or greater than a predetermined threshold value. The commercial block includes a second commercial having a length different from the commercial,
A second commercial candidate section corresponding to the second commercial is detected for each commercial switching point;
The commercial detection method according to claim 5, wherein when a block including the second commercial candidate section satisfies a predetermined condition, the block is determined to be a commercial block. A commercial block composed of a first commercial having a first predetermined length and a second commercial having a second predetermined length different from the first length is inserted into the broadcast program information A commercial detection device that receives a broadcast signal and detects the commercial block from the broadcast signal,
Commercial boundary detection means for detecting a commercial switching time point when the commercial is switched;
First commercial candidate detection means for detecting, for each commercial switching time, a first commercial candidate section in which a time interval between adjacent commercial switching time points corresponds to the first length;
Second commercial candidate detecting means for detecting a second commercial candidate section in which a time interval between adjacent commercial switching time points corresponds to the second length for each commercial switching time point;
When the block formed by the first commercial candidate section and the second commercial candidate section satisfies a predetermined condition, the block is composed of commercial block detecting means for determining that the block is a commercial block. Commercial detection device. A commercial detection device that receives a broadcast signal in which a commercial block formed by a plurality of commercials having a predetermined length is inserted into broadcast program information, and detects the commercial block from the broadcast signal,
Commercial boundary detection means for detecting a commercial switching time point when the commercial is switched;
Commercial candidate detection means for detecting a commercial candidate section whose time interval between adjacent commercial switching points corresponds to the predetermined length for each commercial switching point;
Commercial classification means for extracting feature quantities of commercial content for each commercial candidate section, and classifying each commercial candidate section into a commercial cluster according to the similarity of the feature quantities;
Commercial cluster storage means for storing commercial candidates classified into each commercial cluster;
A commercial candidate section included in a commercial cluster including a predetermined number of commercial candidate sections is regarded as a regular commercial candidate section, and a block formed by the commercial candidate sections regarded as the regular commercial candidate section satisfies a predetermined condition A commercial detection apparatus comprising: commercial block detection means for determining that the block is a commercial block. The commercial block includes a second commercial having a predetermined length different from the commercial,
A second commercial candidate detecting means for detecting a second commercial candidate section corresponding to the second commercial for each commercial switching time point;
10. The commercial detection apparatus according to claim 9, wherein the commercial block detection unit determines that the block including the second commercial candidate section satisfies the predetermined condition as a commercial block. A commercial block composed of a first commercial having a first predetermined length and a second commercial having a second predetermined length different from the first length is inserted into the broadcast program information A recording medium on which a program for receiving a broadcast signal and detecting the commercial block from the broadcast signal is recorded,
Detecting a commercial switching time, which is a time when the commercial is switched,
Detecting for each commercial switching time point a first commercial candidate section in which a time interval between adjacent commercial switching time points corresponds to the first length;
Detecting for each commercial switching time point a second commercial candidate section in which a time interval between adjacent commercial switching time points corresponds to the second length;
And a step of determining that the block formed by the first commercial candidate section and the second commercial candidate section satisfy a predetermined condition as a commercial block. Recording medium. A recording medium on which is recorded a program for receiving a broadcast signal in which a commercial block formed by a plurality of commercials having a predetermined length is inserted into broadcast program information and detecting the commercial block from the broadcast signal,
Detecting a commercial switching time, which is a time when the commercial is switched,
Detecting a commercial candidate section in which a time interval between adjacent commercial switching points corresponds to the predetermined length for each commercial switching point;
Extracting a feature amount of commercial content for each commercial candidate section, and classifying each commercial candidate section into a commercial cluster according to the similarity of the feature amount; and
A commercial candidate section included in a commercial cluster including a predetermined number of commercial candidate sections is regarded as a regular commercial candidate section, and a block formed by the commercial candidate sections regarded as the regular commercial candidate section satisfies a predetermined condition And a step of determining that the block is a commercial block.

Images