JP2011039776A - Moving image content detection device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a technique which can accurately and quickly detect a content even when the content is an illegally distributed content in which edition in a time-base direction such as segmentation of a part of a content of which the free distribution is not granted by a copyright holder or the like or an illegally distributed content of which the whole is deteriorated due to compression noise or the like. <P>SOLUTION: A moving image content detection device 1 includes: a shot boundary detection part 10 for detecting a shot boundary of a moving image content; a shot boundary feature value extraction part 20 for extracting feature values from front and rear frames of the shot boundary detected by the shot boundary detection part 10; and a feature value collation part 40 for collating the shot boundary feature value which is a feature value concerned with one moving image content and extracted by the shot boundary feature value extraction part 20 with shot boundary feature values concerned with a plurality of moving image contents stored in a database 50. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a moving image content detection apparatus.

  With the spread of broadband in recent years and the increase in storage capacity of HDDs (Hard Disk Drives), DVDs (Digital Versatile Disks), Blu-ray discs, etc., digital content has become copyright holders and content providers (hereinafter referred to as "Copyrights"). It has become easy to share and publish via a network without obtaining permission from a right holder or the like), and such illegal sharing and publishing has become a problem. To deal with such problems, the digital content fingerprint (feature value) is used to automatically detect a specific content that the copyright holder or the like does not permit free distribution from among a plurality of digital content. Techniques have been proposed (see Patent Documents 1 and 2 and Non-Patent Document 1).

  In Patent Document 1, a feature amount of a content is described using a three-dimensional frequency analysis and a principal component analysis, and a specific content is detected. This method performs 3D frequency analysis by adding frequency analysis (FFT) in the time axis direction to the coefficient obtained by spatial frequency analysis (DCT), and then the coefficient obtained by 3D frequency analysis by principal component analysis. Feature values are extracted from. In Patent Document 2, the specific content similar to the distributed content is narrowed down using the feature amount used in Patent Document 1, and when it cannot be narrowed down, the most similar to the distributed content using the phase-only correlation method Specific content is determined, and it is determined whether or not the same content is based on a threshold value.

  In Non-Patent Document 1, first, an average absolute error (Motion intensity) of luminance values of adjacent frames is obtained from a video, and a frame in which the average absolute error takes an extreme value is set as a key frame. Next, a feature point called a corner is detected from each key frame by a Harris detector, and a feature amount is extracted from the periphery using a Gaussian derivative. Thereafter, matching and voting are performed between each feature quantity and the database, and content with a large number of votes is detected as illegally distributed content. In this method, unauthorized distribution content can be detected even when video is temporally edited.

JP 2005-18675 A JP 2006-285907 A

J. Law-To et al., “Video Copy Detection: A Comparative Study,” in Proc. ACM CIVR’07, pp.371-378, 2007.

  However, in the methods disclosed in Patent Documents 1 and 2, since one feature amount is extracted from one moving image content, for example, it can be detected when editing in the time axis direction such as dividing the moving image content is performed. There is a problem of disappearing. On the other hand, the method disclosed in Non-Patent Document 1 has the following problems. First, a key frame is selected based on the motion intensity. However, there is a problem that the extreme value of the motion intensity is unstable with respect to noise, and the accuracy is lowered due to the shift of the key frame. In addition, the number of key frames extracted by the motion intensity varies depending on the scene, and there is a problem that processing time increases due to the extraction of redundant key frames. Furthermore, since the extracted Gaussian derivative feature amount is relatively sensitive to compression noise or the like, there is a problem that accuracy is lowered when such noise is added.

  The present invention has been made in view of such circumstances, and illegally distributed content that has been edited in the time axis direction, such as cutting out a part of content that the copyright holder or the like does not permit free distribution, It is an object of the present invention to provide a technology capable of accurately and rapidly detecting even illegally distributed content deteriorated due to compression noise or the like.

  In order to solve the above problem, a video content detection apparatus according to an aspect of the present invention includes a shot boundary detection unit that detects a shot boundary of video content, and frames before and after the shot boundary detected by the shot boundary detection unit. A shot boundary feature amount extraction unit that extracts feature amounts, and a plurality of moving images in which a shot boundary feature amount that is a feature amount related to one moving image content and is extracted by the shot boundary feature amount extraction unit is stored in the storage unit And a feature amount matching unit for matching with a shot boundary feature amount related to the content.

  In the moving image content detection apparatus, the shot boundary detection unit determines whether a shot boundary exists within a certain interval using information of a predetermined frame that is included in the moving image content and is present at certain intervals. Then, the shot boundary of the moving image content may be detected by determining whether or not each frame within a certain interval determined to have a shot boundary is a shot boundary.

  In the moving image content detection apparatus, the predetermined frame may be a frame that can be decoded without referring to another frame among frames constituting the compressed moving image content.

  In the moving image content detection apparatus, the shot boundary feature amount extraction unit may extract the shot boundary feature amount based on a correlation between frames before and after the shot boundary. Further, the shot boundary feature amount extraction unit divides the frames before and after the shot boundary into a plurality of blocks, creates a set of a certain number of blocks from the plurality of blocks, and based on the correlation between the sets of the blocks, A shot boundary feature amount may be extracted.

  In the moving image content detection apparatus, the shot boundary feature amount extraction unit divides each frame before and after the shot boundary into a plurality of blocks, creates a set of a predetermined number of blocks from the plurality of blocks, and averages the set of blocks You may make it extract a shot boundary feature-value based on at least one of the magnitude relationship of a brightness | luminance, motion intensity | strength, and edge amount.

  In the moving image content detection apparatus, the feature amount matching unit calculates a distance between the shot boundary feature amount related to one moving image content and the shot boundary feature amount related to the moving image content stored in the storage unit, and based on the distance Thus, the moving image content stored in the storage unit and the moving image content stored in the storage unit may be collated, and in the distance calculation, the magnitude relationship between the sets of blocks having similar average luminance, motion intensity, and edge amount may not be used.

  According to the present invention, illegal distribution content that has been edited in the time axis direction, such as cutting out a part of content that the copyright holder or the like has not permitted free distribution, or illegal distribution content that has deteriorated due to compression noise or the like as a whole. Even if it exists, it becomes possible to detect with high accuracy and high speed.

It is a functional block diagram of the moving image content detection apparatus 1 by one Embodiment of this invention. 4 is a flowchart illustrating an example of the operation of the shot boundary detection unit 10. FIG. 6 is an explanatory diagram for explaining the operation of the shot boundary detection unit 10. FIG. 6 is an explanatory diagram for explaining an operation of a shot boundary feature quantity extraction unit 20. FIG. 6 is an explanatory diagram for explaining an operation of a feature amount matching unit 40.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings. The moving image content detection apparatus 1 according to the embodiment of the present invention includes a feature amount of a moving image content (query content) to be inspected and specific content (hereinafter referred to as “reference content”) that the copyright holder or the like does not permit free distribution. The query content that is presumed to be illegally distributed content is detected using the feature amount of As shown in FIG. 1A, the moving image content detection apparatus 1 includes a shot boundary detection unit 10, a shot boundary feature amount extraction unit 20, a feature amount registration unit 30, a feature amount collation unit 40, and a database (storage unit) 50. Prepare.

  The shot boundary detection unit 10 detects shot boundaries of moving image content (reference content and query content). Specifically, first, the shot boundary detection unit 10 determines whether or not a shot boundary exists within a certain interval by using information of predetermined frames that are frames constituting the moving image content and exist at every certain interval. judge. The predetermined frame is, for example, a frame that can be decoded without referring to another frame among frames constituting the compressed moving image content, that is, a frame that can be decoded alone. An I frame (Intra Picture) in a GOP (Group of Picture) corresponds to the predetermined frame.

  Subsequently, when the shot boundary detection unit 10 determines that a shot boundary exists within a certain interval, the shot boundary detection unit 10 further determines whether each frame corresponds to a shot boundary for each frame within the certain interval. . On the other hand, the shot boundary detection unit 10 does not further determine whether or not each frame corresponds to a shot boundary for each frame within a certain interval determined that there is no shot boundary. In other words, the shot boundary detection unit 10 determines whether or not there is a shot boundary at each fixed interval using information of a predetermined frame, and detects a shot boundary only from the fixed interval at which it is determined that a shot boundary exists.

  The shot boundary detection unit 10 that has detected the shot boundary extracts two frames before and after the shot boundary as key frames. Hereinafter, the two front and rear frames extracted as key frames are also referred to as key frame pairs. The shot boundary detection unit 10 that has extracted the key frame pair supplies the extracted key frame pair to the shot boundary feature quantity extraction unit 20.

Hereinafter, the operation of the shot boundary detection unit 10 will be described in detail with reference to FIGS. Note that the feature amount used in the shot boundary detection unit 10 may be, for example, the feature amount described in Reference Document 1 below, but the chi-square between color histograms that are considered to have the highest accuracy alone for speeding up. Value.
(Reference 1) K. Matsumoto, M. Naito, K. Hoashi, and F. Sugaya, “SVM-Based Shot Boundary Detection With a Novel Feature,” in Proc. Of ICME'06, pp.1837-1840, 2006 .

  As shown in the flowchart of FIG. 2, the shot boundary detection unit 10 first extracts an I frame related to a certain GOP in the moving image content (reference content and query content) (step S10). For example, each time this process is executed, the shot boundary detection unit 10 extracts I frames in order from the top of the moving image content.

  The shot boundary detection unit 10 determines whether or not a shot boundary exists in the GOP including the I frame (step S20). Hereinafter, processing for determining whether or not a shot boundary exists in a certain GOP is referred to as GOP level shot boundary detection processing.

  FIG. 3A is a conceptual diagram showing a feature amount extraction method used for the GOP level shot boundary determination process. For example, the shot boundary detection unit 10 extracts a feature amount from each of N I frames before and after the GOP to be determined in the GOP level shot boundary determination process, that is, the GOP including the I frame.

Specifically, first, the shot boundary detection unit 10 equally divides each I frame into X × Y regions, and extracts a color histogram from each region after the equal division. In the above-mentioned reference 1, a histogram in the Ohta color space is used, but a histogram in the YCbCr color space is used in order to save the amount of calculation necessary for the color space conversion. Then, the shot boundary detector 10, as a histogram distance between the same region of the adjacent I frames, calculates the chi-square value d _X using the following equation (1).

Next, the shot boundary detection unit 10 calculates V _inter that is a chi-square value d _X for all regions between all adjacent I frames, using the following equation (2) as a feature amount used for SVM discrimination. To do.

  The shot boundary detection unit 10 uses the above-described feature amount, uses the content that has been previously labeled with the shot boundary, uses a feature in which a shot boundary exists in the GOP including the I frame as a positive example, and other features. The GOP level shot boundary determination process is realized by learning as a negative example.

  As a result of the GOP level shot boundary detection process, the shot boundary detection unit 10 determines that a shot boundary exists in the GOP including the I frame (step S20: Yes), and a certain frame in the GOP becomes a shot boundary. It is determined whether it is applicable (step S30). For example, each time this process is executed, the shot boundary detection unit 10 determines whether or not the shot boundary corresponds to the first frame of the GOP in order. Hereinafter, the process of determining whether a certain frame corresponds to a shot boundary is referred to as a frame level shot boundary detection process. In other words, the shot boundary detection unit 10 executes the frame level shot boundary detection process for the GOP that is determined to have a shot boundary in the GOP in the GOP level shot boundary detection process.

FIG. 3B is a conceptual diagram illustrating a feature amount extraction method used for frame level shot boundary determination processing. The feature quantity used in the frame level shot boundary determination process is substantially the same as the feature quantity used in the GOP level boundary detection process, as shown in FIG. However, if the frame from which the feature amount is extracted exists outside the determination target GOP, the chi-square value d _X between the frame existing outside the determination target GOP is set to prevent generation of unnecessary decoding processing. An average value of chi-square values between frames that are not actually obtained and that are not shot boundaries is used.

  The shot boundary detection unit 10 extracts key frame pairs before and after the frames when it is determined that the frames in the GOP correspond to the shot boundaries as a result of the frame level shot boundary detection processing (step S30: Yes). (Step S40). The shot boundary detection unit 10 that has extracted the key frame pair is associated with the time between the frames determined to correspond to the shot boundary (for example, the time from the beginning of the content; hereinafter referred to as “shot boundary time”). The key frame pair is temporarily stored.

  When the shot boundary detection unit 10 determines that the frames in the GOP do not correspond to the shot boundaries (step S30: No) or extracts a key frame pair (step S40), the shot boundary detection unit 10 extracts all the frames in the GOP. It is determined whether or not frame level shot boundary determination processing has been performed between frames (step S50). When the shot boundary detection unit 10 determines that the frame level shot boundary determination process is not performed between all frames in the GOP (step S50: No), the process returns to step S30 to return to the next in the GOP. It is determined whether or not a frame boundary corresponds to a shot boundary (step S30).

  When it is determined that there is no shot boundary in the GOP including the I frame (step S20: No), the shot boundary detection unit 10 performs a frame level shot boundary determination process for all frames in the GOP. If it is determined that the GOP level shot boundary determination process has been performed on all GOPs in the content (step S50: Yes), it is determined (step S60). When the shot boundary detection unit 10 determines that the GOP level shot boundary determination process has not been performed for all GOPs in the content (step S60: No), the process returns to step S10 to return to the next in the content An I frame related to the GOP is extracted (step S10).

  When the shot boundary detection unit 10 determines that the GOP level shot boundary determination process has been performed on all GOPs in the content (step S60: Yes), the shot temporarily stored together with the content ID for identifying the content The boundary time and the key frame pair are supplied to the shot boundary feature quantity extraction unit 20, and this flowchart ends.

  As described above, in shot boundary detection, attention is paid to GOP which is a basic structure of compressed content, and GOP level shot boundary detection processing is executed prior to frame level shot boundary determination processing. Accordingly, the processing time for decoding and the like is reduced, and the shot boundary detection process is accelerated. Note that there is a method for detecting shot boundaries at high speed without decoding all frames by using encoded information of video content, but this method uses encoded information that depends on a specific codec. Therefore, it can be applied only to moving image content compressed with a specific codec, and is not general purpose.

  The shot boundary feature quantity extraction unit 20 acquires a content ID, a shot boundary time, and a key frame pair from the shot boundary detection unit 10. The shot boundary feature amount extraction unit 20 extracts a feature amount (hereinafter referred to as “shot boundary feature amount”) from the key frame pair acquired from the shot boundary detection unit 10.

  Specifically, the shot boundary feature amount extraction unit 20 extracts a shot boundary feature amount based on the correlation of each key frame. For example, the shot boundary feature amount extraction unit 20 divides each key frame into a plurality of blocks, creates a set of a certain number of blocks from the plurality of blocks, and creates shot boundary features based on the correlation between the sets of blocks. Extract the amount. For example, the shot boundary feature amount extraction unit 20 extracts a shot boundary feature amount based on at least one of the magnitude relationship between the average luminance, the motion intensity, and the edge amount of a block set.

  The shot boundary feature quantity extraction unit 20 that has extracted the shot boundary feature quantity of the reference content supplies shot information (content ID, shot boundary time, shot boundary feature quantity) related to the reference content to the feature quantity registration unit 30. The shot boundary feature amount extraction unit 20 that has extracted the shot boundary feature amount of the query content supplies the feature amount matching unit 40 with the shot information (content ID, shot boundary time, shot boundary feature amount) related to the query content. Note that the shot boundary feature amount extraction unit 20 may determine whether the content is reference content or query content based on an input from the user, for example. For example, the moving image content detection apparatus includes a mode selection receiving unit (not shown) that receives a mode selection from the user, and receives a registration mode for registering the shot boundary feature quantity of the reference content in the database 50 via the mode selection receiving unit. In this case, when the shot boundary feature quantity extraction unit 20 determines that the content is reference content and receives a matching mode for matching the query content and the reference content, the shot boundary feature quantity extraction unit 20 Determine that the content is query content.

  Hereinafter, an example in which the shot boundary feature amount extraction unit 20 extracts a shot boundary feature amount based on the magnitude relationship between the edge amounts of a set of blocks will be described. First, the shot boundary feature quantity extraction unit 20 divides each key frame into N × M areas. Next, the shot boundary feature quantity extraction unit 20 calculates an edge quantity E (i, j) of each divided area using the following formula (3) or the following formula (4).

Next, as shown in FIG. 4, the shot boundary feature quantity extraction unit 20 calculates an N × M bit shot boundary feature quantity B (i, j) represented by the following equations (5) and (6). . However, the edge amount E ⁻ (i, j) is a key frame before the shot boundary, and the edge amount E ⁺ (i, j) is a key frame after the shot boundary.

  As described above, in the extraction of the shot boundary feature value, since the shot boundary feature value expressed in bits is generated, the distance between the feature values can be calculated at high speed by XOR. Further, the registration (accumulation) cost of the database 50 can be reduced. Note that there are a plurality of methods for representing a bit string of an image, but the main problem is robustness. Robustness is a characteristic in which a feature amount does not change as much as possible when some modification is made to an image. The conventional method has a problem that the feature amount largely changes when a pattern such as a logo or a caption is inserted. In the method of the present embodiment, feature amounts are not extracted from one frame, but feature amounts are extracted by using correlation information between key frame pairs, that is, two key frames. It is possible to extract feature amounts that are robust to various modifications.

  As a specific example, the example in which the shot boundary feature amount extraction unit 20 extracts the shot boundary feature amount based on the magnitude relationship between the edge amounts of the block sets has been described, but based on the magnitude relationship between the average luminance and the motion intensity. The same applies to the case where the boundary feature amount is extracted.

The feature amount registration unit 30 acquires shot information (content ID, shot boundary time, shot boundary feature amount) related to the reference content from the shot boundary feature amount extraction unit 20. The feature amount registration unit 30 that has acquired the shot information related to the reference content registers (stores) the shot information related to the reference content in the database 50. The feature amount registration unit 30 calculates a hash value of the shot boundary feature amount, and stores each shot information in a plurality of hash tables using the hash value as a hash key. For hashing, for example, Local Sensitive Hashing described in Reference Document 2 below may be used.
(Reference 2) Datar, M., N. Immorlica, P. Indyk and V. Mirrokni, “Locality-Sensitive Hashing Scheme Based on p-Stable Distributions,” Proceedings of the 20th Symposium on Computational Geometry, pp.253-262 , 2004.

  As described above, in registering feature amounts, each piece of shot information is registered in a plurality of hash tables based on shot boundary feature amounts, for example, by Local Sensitive Hashing.

  The database 50 registers (stores) shot information (content ID, shot boundary time, shot boundary feature amount) related to the reference content. As described above, each piece of shot information is registered in a plurality of hash tables using the shot boundary feature quantity as a hash key.

  The feature amount matching unit 40 acquires shot information (content ID, shot boundary time, shot boundary feature amount) related to the query content from the shot boundary feature amount extraction unit 20. The feature amount collation unit 40 that has acquired the shot information related to the query content collates the shot boundary feature amount related to the query content with the shot boundary feature amount related to a plurality of reference contents stored in the database 50. That is, when acquiring the shot boundary feature amount of the query content, the feature amount matching unit 40 refers to the database 50 in which the shot boundary feature amount of the reference content is registered in advance, and the query content includes at least a part of the reference content. Check whether it corresponds to the copied one.

  Specifically, the feature amount matching unit 40 first estimates the reference content section most similar to the query content based on the shot information (shot boundary time, shot boundary feature amount) of each shot boundary of the query content. Then, it is determined whether the query content is a copy of the reference content using the similarity and threshold value of the section. The estimation of the similar reference section is realized by collecting matching between query content and reference content shot boundaries by voting. Further, the similarity of the section is calculated based on the number of votes in the section.

  Hereinafter, the operation of the feature amount matching unit 40 will be specifically described with reference to FIG. It is assumed that shot information (content ID, shot boundary time, shot boundary feature value) related to a plurality of reference contents is registered in the database 50.

The feature amount matching unit 40 acquires shot information (content ID, shot boundary time, shot boundary feature amount) related to the query content Q shown in FIG. 5A from the shot boundary feature amount extraction unit 20. The shot boundary time t ₁ is the shot boundary time of the shot boundary Q ₁ , the shot boundary time t ₂ is the shot boundary time of the shot boundary Q ₂ , and the shot boundary time t ₃ is the shot boundary time of the shot boundary Q _3. To do.

The feature amount matching unit 40 that has acquired the shot information related to the query content Q, the shot related to the reference content having a shot feature amount similar to each shot feature amount of each shot boundary Q ₁ , Q ₂ , Q ₃ of the query content Q. The boundary is searched from the database 50 by matching. Specifically, the feature amount matching unit 40 is predetermined from a shot boundary in which the distance (similarity) between the shot feature amounts of the query content and the reference content is equal to or less than a certain value, or the distance between the shot feature amounts is shorter. Are searched as similar shot boundaries.

  Specifically, the feature amount matching unit 40 determines that the distance between the above-described shot feature amounts is, for example, N × M bits that are shot boundary feature amounts related to query content, and N is a shot boundary feature amount related to reference content. The Hamming distance between × M bits can be calculated at high speed by a simple bit operation. For example, when the shot boundary feature amount B related to the N × M bit query content is expressed by the following equation (7), and the shot boundary feature amount B ′ related to the N × M bit reference content is expressed by the following equation (8): The feature amount matching unit 40 generates an N × M-bit bit string (B XOR B ′) and calculates the number of 1 included in the bit string (B XOR B ′) as the Hamming distance between B and B ′.

However, the feature amount matching unit 40 may exclude the size relationship between sets of blocks having similar average luminance, motion intensity, and edge amount from the distance calculation target. For example, when excluding the size relationship of a set of blocks having close edge amounts from the calculation target of the distance, the edge amount of each key frame block (i, j) before and after the shot boundary of the query content is set to E ⁻ (i, j) and ^E + (i, when a ^{j), | E + (i} , j) -E - (i, j) | is not used in the distance calculation with respect to bits corresponding to a predetermined number of blocks from the smallest You may do it. This is because a block having a small chi-square value d _X (i, j) has a low bit value reliability because the magnitude relationship between E ⁻ (i, j) and E ⁺ (i, j) is easily changed.

The distance for the calculation of ^{| E + (i, j)} -E - (i, j) | when not using the bits corresponding to a predetermined number of blocks from the smaller, first, the feature checker 40, A mask feature amount H (i, j) represented by the following equation (9) is created. Then, the feature amount matching unit 40 uses a bit string ((B XOR B ′) by using an N × M bit bit string H represented by the following equation (10) instead of the above bit string (B XOR B ′). ) AND H) is generated, and the number of 1 included in the generated bit string ((B XOR B ′) AND H) is calculated as the Hamming distance between B and B ′.

As described above, the feature amount matching unit 40, for example, as shown in FIG. 5B, each shot boundary A ₁ , A ₂ , A ₃ , A ₄ , A ₅ , and reference content B of the reference content A Each shot boundary B ₁ , B ₂ , B ₃ is searched.

  The feature amount matching unit 40 that has searched for the shot boundary related to the reference content votes for (shot time related to the reference content−time of shot boundary related to the query content) for all the matched shot boundary pairs. Do. The vote is an estimate of the beginning of the copy section. That is, as shown in FIG. 5C, the voting is concentrated at the time estimated to be the head of the actual copy section if the matching is correct, and dispersed if the matching is not correct. Therefore, the feature amount matching unit 40 determines whether or not the number of votes at the time when the most votes are concentrated is greater than or equal to a threshold, and if it is greater than or equal to the threshold, the time is estimated to be the head of the copy section. To do. Note that the feature amount matching unit 40, which has estimated that the time when the most votes are concentrated, is the head of the copy section, outputs information indicating that the query content is unauthorized distribution content to the outside. Further, the feature amount matching unit 40, for example, instead of or in addition to the information that the query content is estimated to be unauthorized distribution content, for example, shot information related to the query content and the reference content, and the start position of the copy section Or the like may be output to the outside.

  As described above, the moving image content detection apparatus 1 detects query content that is presumed to be illegally distributed content using the feature amount of the query content and the feature amount of the reference content. The moving image content detection apparatus 1 determines whether the query content is unauthorized distribution content including at least a part of the reference content.

  In the moving image content detection apparatus 1, since key frames are extracted from shot boundaries, key frames that are robust against noise and not redundant are extracted. In addition, since the feature amount calculated based on the correlation between the two key frames before and after the shot boundary is used, robust matching can be performed with respect to the content change (for example, editing, modification, noise). In addition, since it is not necessary to decode all frames in the content, it can be detected at high speed. That is, according to the moving image content detection apparatus 1, illegally distributed content that has been edited in the time axis direction, such as cutting out a part of the content that the copyright holder or the like has not permitted to distribute freely, or the entire content is deteriorated due to compression noise or the like Even illegally distributed content can be detected accurately and at high speed.

  As shown in FIG. 1A, the moving image content detection apparatus 1 includes a feature amount registration unit 30 and a database 50 in addition to the shot boundary detection unit 10, the shot boundary feature amount extraction unit 20, and the feature amount comparison unit 40. Although the example provided is demonstrated, it is not limited to the said structure. For example, the moving image content detection apparatus 1 may include only the shot boundary detection unit 10, the shot boundary feature amount extraction unit 20, and the feature amount comparison unit 40 as illustrated in FIG. In the case of the configuration shown in FIG. 1B, the moving image content detection device 1 is a shot related to reference content by an external device including the above-described shot boundary detection unit 10, shot boundary feature amount extraction unit 20, and feature amount registration unit 30. Referring to the external database 50 in which information is registered, query content that is presumed to be illegally distributed content is detected.

  Note that a program for executing each process of the moving image content detection apparatus 1 according to an embodiment of the present invention is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read by a computer system. By executing, the above-described various processes related to each process of the moving image content detection apparatus 1 according to the embodiment of the present invention may be performed. Here, the “computer system” may include an OS and hardware such as peripheral devices. Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used. The “computer-readable recording medium” means a flexible disk, a magneto-optical disk, a ROM, a writable nonvolatile memory such as a flash memory, a portable medium such as a CD-ROM, a hard disk built in a computer system, etc. This is a storage device.

  Further, the “computer-readable recording medium” means a volatile memory (for example, DRAM (Dynamic DRAM) in a computer system that becomes a server or a client when a program is transmitted through a network such as the Internet or a communication line such as a telephone line. Random Access Memory)), etc., which hold programs for a certain period of time. The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

  The embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to this embodiment, and includes designs and the like that do not depart from the gist of the present invention.

DESCRIPTION OF SYMBOLS 1 Movie content detection apparatus 10 Shot boundary detection part 20 Shot boundary feature-value extraction part 30 Feature-value registration part 40 Feature-value collation part 50 Database (memory | storage part)

Claims (7)

A shot boundary detector for detecting the shot boundary of the video content;
A shot boundary feature amount extraction unit that extracts feature amounts from frames before and after the shot boundary detected by the shot boundary detection unit;
A feature that is a feature amount related to one moving image content, and that matches a shot boundary feature amount extracted by the shot boundary feature amount extraction unit with the shot boundary feature amount related to a plurality of moving image contents stored in a storage unit A video content detection apparatus comprising: a quantity verification unit. The shot boundary detection unit
It is determined whether or not the shot boundary exists within a fixed interval using information of a predetermined frame that is a frame constituting video content and exists at fixed intervals, and the fixed interval at which the shot boundary is determined to exist The designated moving image content detection apparatus according to claim 1, wherein a shot boundary of moving image content is detected by determining whether or not each of the frames is a shot boundary. The predetermined frame is:
3. The designated moving image content detection apparatus according to claim 2, wherein the specified moving image content detection apparatus is a frame that can be decoded without referring to other frames among the frames constituting the compressed moving image content. The shot boundary feature amount extraction unit
4. The designated moving image content detection apparatus according to claim 2, wherein the shot boundary feature amount is extracted based on a correlation between frames before and after the shot boundary. The shot boundary feature amount extraction unit
The frame before and after the shot boundary is divided into a plurality of blocks, a set of a predetermined number of blocks is created from the plurality of blocks, and the shot boundary feature amount is extracted based on the correlation of the block sets. The designated moving image content detection apparatus according to claim 2, wherein the moving image content detection apparatus is a specified moving image content detection apparatus. The shot boundary feature amount extraction unit
Each of the frames before and after the shot boundary is divided into a plurality of blocks, a set of a certain number of blocks is created from the plurality of blocks, and at least one of the magnitude relationships of the average luminance, motion intensity, and edge amount of the block set The specified moving image content detection apparatus according to claim 2, wherein the shot boundary feature amount is extracted based on. The feature amount matching unit
A distance between the shot boundary feature amount related to the one moving image content and the shot boundary feature amount related to the moving image content stored in the storage unit is calculated, and the one moving image content and the storage are calculated based on the distance Check the video content stored in the
4. The designated moving image content detection apparatus according to claim 2, wherein the distance calculation does not use a magnitude relationship between a set of blocks having close average brightness, motion intensity, and edge amount.

Images