Moving image identification apparatus and moving image identification method and moving image identification program

Abstract

To detect a moving image that relies on a target moving image appropriately from among multiple public moving images.SOLUTION: A moving image identification apparatus 10 detects a moving image that relies on a target moving image from among multiple public moving images RMs. The moving image identification apparatus 10 has a target moving image recording part 11, a public moving image selection part 12, and a determination part 13. The target moving image recording part 11 records, from the target moving images, for each composition unit in a time axis direction of the target moving image, an image of a frame included in the composition unit as a representative image TMx. The public moving image selection part 12 selects, for each of the multiple public moving images RM, an image of a frame from the public moving image RM at each predetermined time interval as a cut-out image RMx. The determination part 13 determines whether or not the public moving image RM is a moving image that may rely on the target moving image based on the representative image TMx of the target moving image and the cut-out image RMx of the public moving image RM.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a technique for detecting a video that depends on a target video from a plurality of publicly available videos.

Background Art

[0002] On the Internet, a lot of video content is publicly available. There is an illegal act of stealing someone else's video and publishing it on a website to obtain advertising revenue. Legitimate owners of videos need to take measures such as discovering such illegal websites and requesting the deletion of the discovered illegal videos.

[0003] Patent Document 1 discloses a technique for calculating the mutual similarity of content submitted by individual users and determining whether the submitted content is copyright-appropriate based on the calculated mutual similarity.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Content providers who provide their own video content such as dramas and movies through TV, movies, or the Internet will suffer damage to their interests if their video content is stolen by others and illegally published. Therefore, when their video content is illegally published on a video sharing site or the like, content providers want to take measures such as finding it as quickly as possible and stopping the publication.

[0006] However, the inappropriate content detection method disclosed in Patent Document 1 calculates the similarity between posted content and detects groups of content that are similar to each other as copyright-inappropriate content. Therefore, the method in Patent Document 1 is not suitable for the purpose of content providers detecting plagiarism by others targeting their specific content.

[0007] One of the purposes of this disclosure is to provide a technology that appropriately detects videos that rely on a target video from among multiple publicly available videos. [Means for solving the problem]

[0008] A video identification device according to one embodiment of the present invention is a video identification device that detects a video that depends on a target video from among a plurality of publicly available videos, and comprises: a target video image recording unit that records images of frames included in each time axis unit of the target video as representative images; a publicly available video image selection unit that, for each of the plurality of publicly available videos, selects images of frames from the publicly available video at predetermined time intervals as cropped images; and a determination unit that determines whether or not the publicly available video is a video that may depend on the target video based on the representative images of the target video and the cropped images of the publicly available videos. [Effects of the Invention]

[0009] According to the present invention, it becomes possible to appropriately detect videos that rely on a target video from among multiple publicly available videos. [Brief explanation of the drawing]

[0010] [Figure 1] A block diagram showing the functional configuration of a video recognition device. [Figure 2] A block diagram showing the hardware configuration of a video recognition device. [Figure 3] A block diagram showing the functional configuration of the determination unit. [Figure 4] A flowchart of the overall process. [Figure 5]An illustrative diagram to explain the overall process. [Figure 6] A flowchart of the decision-making process. [Figure 7] Flowchart for the secondary decision process [Modes for carrying out the invention]

[0011] Embodiments of the present invention will be described below with reference to the drawings.

[0012] Figure 1 is a block diagram showing the functional configuration of the video identification device. Figure 2 is a block diagram showing the hardware configuration of the video identification device.

[0013] The video identification device 10 includes a target video image recording unit 11, a public video image selection unit 12, a determination unit 13, and a display unit 14.

[0014] The video identification device 10 in this embodiment is a device that detects videos that rely on a target video TM (see Figure 4) from among multiple publicly available videos RM. Publicly available videos RM are videos that companies, individuals, etc., publish on a communication network 90 such as the Internet through web pages such as video sharing sites provided by the web server 91. The target video TM is a copyrighted video that is searched for on the Internet to see if it has been plagiarized. It is assumed that some publicly available videos RM contain videos that rely on the target video TM. Videos created based on the target video TM may be considered to rely on the target video TM. Creating a video that relies on the target video TM without permission from the copyright holder of the target video TM may constitute plagiarism.

[0015] The video identification device 10 searches for publicly available video RM that are presumed to be identical to the target video TM or that are presumed to rely on the target video TM, in order to find publicly available video RM that have plagiarized the target video TM.

[0016] The target video image recording unit 11 records, from the target video TM, the images of the frames included in each scene, which is a constituent unit in the time axis direction of the target video TM, as representative images TMx representing the scene. That is, the target video TM includes a plurality of representative images TMx. The constituent unit may be, in addition to the scenes constituting the target video TM, a cut or a shot. The frame is, for example, the entire area of a still image constituting the target video TM.

[0017] The target video TM may be multiple types of video content. The video content may be, for example, videos such as movies, dramas, literature, animations, sports, etc. As an example, the representative image TMx may be the image at the beginning of each scene of the target video TM, or the representative image TMx may be a characteristic image selected from the target video TM by the user. The representative image TMx may include, for example, a thumbnail image.

[0018] The public video image selection unit 12 selects, for each of the plurality of public videos RM, an image of a frame cut out every predetermined time from the public video RM as an extracted image RMx. That is, the public video RM includes a plurality of extracted images RMx. The predetermined time may be, for example, a fixed time such as 10 seconds. Here, while the representative image TMx is selected from the scenes of the target video TM, the extracted image RMx is acquired from the public video RM every predetermined time. This is because when the public video RM is illegally created, it may be difficult to distinguish the scene boundaries due to processing or deterioration.

[0019] The public video image selection unit 12 may select, as the extracted image RMx, an image that is not an inserted image inserted between the frames of the public video RM from the public video RM. The inserted image is an image unrelated to the content inserted between the frames of the public video RM to avoid detection by someone other than the copyright owner of the target video TM.

[0020] The determination unit 13 determines whether the publicly released video RM is likely to depend on the target video TM by comparing the representative image TMx of the target video TM with the clipped image RMx of the publicly released video RM. At this time, the determination unit 13 first excludes the publicly released video RM with extremely low likelihood of depending on the target video TM by means of a simple primary determination, and then performs a detailed secondary determination on the remaining publicly released video RM to determine whether it is likely to depend on the target video TM. This enables both reduction of the processing time required for the determination and improvement of the accuracy of the determination result.

[0021] FIG. 3 is a block diagram showing the functional configuration of the determination unit.

[0022] More specifically, the determination unit 13 includes a feature point extraction unit 131, a first similar video determination unit 132, and a second similar video determination unit 133.

[0023] The feature point extraction unit 131 extracts the feature points of the representative image TMx of the target video TM and the clipped image RMx of the publicly released video RM, respectively. Also, the feature point extraction unit 131 specifies the feature amounts of the extracted feature points. At this time, the feature point extraction unit 131 may specify the feature amounts from the luminance gradients around the respective feature points of the representative image TMx of the target video TM and the clipped image RMx of the publicly released video RM.

[0024] The first similar video determination unit 132 creates a plurality of groups by clustering all the feature points of all the representative images TMx of the target video TM, classifies the feature points of the representative image TMx of the target video TM and the clipped image RMx of the publicly released video RM extracted by the feature point extraction unit 131 into the groups, and determines the publicly released video RM similar to the target video TM based on the classification result. Incidentally, the first similar video determination unit 132 may use the bag-of-features method as an example of category recognition. The first similar video determination unit 132 may classify the feature points of the representative image TMx of the target video TM and the clipped image RMx of the publicly released video RM extracted by the feature point extraction unit 131 according to the feature amounts based on the luminance gradients around the feature points.

[0025] The first similar video determination unit 132 classifies the feature points of the representative image TMx of the target video TM and the cropped image RMx of the publicly available video RM, extracted by the feature point extraction unit 131, to create histograms. Based on the Bhattacharya distance between the histograms of the representative image TMx and the cropped image RMx, it determines whether the representative image TMx and the cropped image RMx are similar. This allows for a simple comparison of publicly available videos RM that are similar to the target video TM.

[0026] The second similar video determination unit 133 excludes combinations of target video TM and public video RM that were determined to be dissimilar in the first similarity determination unit 132. For the remaining combinations of target video TM and public video RM, it calculates the similarity between images by comparing the feature points of the representative image TMx of the target video TM and the cropped image RMx of the public video RM. Based on the similarity between images, it calculates the similarity between the target video TM and the public video RM, and determines whether the public video RM is based on the target video TM based on the similarity between the videos.

[0027] In this process, the second similar video determination unit 133 first calculates the similarity between representative image TMx and cropped image RMx (feature point similarity) for each combination of target video TM and public video RM that were not excluded by the first similar video determination unit 132, by performing a full-cycle test of all possible combinations of representative image TMx and cropped image RMx. Furthermore, for each full-cycle test of all possible combinations of representative image TMx and cropped image RMx, the second similar video determination unit 133 identifies pairs of feature points (similar feature point pairs) whose feature point similarity exceeds a predetermined threshold. In addition, the second similar video determination unit 133 identifies combinations of representative image TMx and cropped image RMx (feature point pairing image pairs) where the number of similar feature point pairs satisfies a predetermined pairing condition. The pairing condition is a condition for associating representative image TMx and cropped image RMx that may be from the same scene. The second similar video determination unit 133 then calculates a similarity score between videos based on the number of similar feature point pairs in the feature point pairing image pairs, and performs a secondary determination based on the score to determine whether the publicly available video RM may be based on the target video TM.

[0028] The display unit 14 displays the determination result from the determination unit 13 on the screen of the display device 26, which will be described later.

[0029] According to this embodiment, the determination is made using the images TMx and RMx that constitute the target video TM and the publicly available video RM, respectively. This makes it possible to appropriately and quickly detect the publicly available video RM that is based on the target video TM from among multiple publicly available video RMs.

[0030] In particular, the determination is made using a representative image TMx included in the content unit of the target video TM, and cropped images RMx selected at predetermined time intervals from each of the multiple publicly available videos RM. Therefore, the determination, which takes into account the degree of similarity as video content, can be achieved by comparing the images TMx and RMx.

[0031] Furthermore, since images other than the inserted images are selected from the publicly available video RM and used for the determination, it becomes possible to appropriately evaluate the degree of similarity between the target video TM and the publicly available video RM even if images unrelated to the publicly available video RM are inserted.

[0032] The video identification device 10 according to this embodiment can also be realized by having a computer execute a software program that defines the processing procedures for each part shown in Figure 1. Figure 2 shows an example of the hardware configuration of a computer that realizes the video identification device.

[0033] Referring to Figure 2, the video identification device 10 can connect to a web server 91 via a communication network 90 such as the Internet. In terms of hardware configuration, the video identification device 10 has a processing unit 21, a main memory 22, a storage device 23, a communication device 24, an input device 25, and a display device 26, which are connected to a bus 27.

[0034] The recording device 23 is capable of recording data in a writable and readable format. This storage device 23 stores data of publicly available videos RM (extracted images RMx) and target videos TM (representative images TMx) used by the video identification device 10 for processing. For example, data of publicly available videos RM (extracted images RMx) collected from multiple web servers 91 is stored in the recording device 23. In addition, data of target videos TM (representative images TMx) that are the target of the search to determine whether or not the video has been plagiarized is also recorded in the recording device 23.

[0035] The processing unit 21 is a processor that reads data recorded in the storage device 23 into the main memory 22 and uses the main memory 22 to execute the processing of the software program. The processing unit 21 realizes the target video image recording unit 11, the public video image selection unit 12, the determination unit 13, and the display unit 14 shown in Figure 1.

[0036] The communication device 24 transmits information processed by the processing unit 21 via the communication network 0, which includes priority, wireless, or both, and also transmits information received via the communication network 90 to the processing unit 21. The received information is used by the processing unit 21 for software processing.

[0037] The input device 25 is a device that receives information from operator input such as a keyboard or mouse. The information input to the input device 25 is used for software processing by the processing unit 21.

[0038] The display device 26 is a device that displays video TMx,RM, image TMx,RMx, and text information on a display screen in accordance with software processing by the processing device 21.

[0039] Figure 4 is a flowchart of the overall process. Figure 5 is an illustrative diagram to explain the overall process.

[0040] Referring to Figure 4, the video identification device 10 first performs target video image recording processing using the target video image recording unit 11 (step S401). In the target video image recording processing, the target video image recording unit 11 first calculates the difference between images between adjacent frames of the target video TM to be searched for plagiarism, and identifies the location where the calculated difference exceeds a predetermined threshold as the scene boundary. Furthermore, the target video image recording unit 11 selects an image representing the scene from the images of the frames included in each scene as the representative image TMx, and stores the data of the selected representative image TMx (TM1, TM2, TM3… in Figure 5) in the storage device 23. For example, the image of the frame at a predetermined time position in the scene may be used as the representative image TMx for that scene. For example, the image of the first frame of the scene may be used as the representative image TMx, or the image of a frame that has elapsed a predetermined time from the beginning may be used as the representative image TMx.

[0041] Next, the video identification device 10 performs public video image selection processing using the public video image selection unit 12 (step S402). In the public video image selection processing, the public video image selection unit 12 first crawls each website of the web server 91 via the communication network 90, collects (crawls) data of public video RM published on the web pages included in the websites, and stores the collected public video RM data in the recording device 23. Next, the public video image selection unit 12 selects cropped images RMx (RM1, RM2, RM3… in Figure 5) from the public video RM at regular time intervals, and stores the data of the selected cropped images RMx in the recording device 23.

[0042] In this embodiment, an example is shown in which the public video image selection process is performed after the target video image recording process, but the processing order is not limited to this. The public video image selection process may be performed before the target video image recording process.

[0043] Next, the video identification device 10 performs a determination process using the determination unit 13 (step S403). The determination process determines whether or not the publicly available video RM is a video that is likely to be based on the target video TM, based on the representative image TMx of the target video TM and the cropped image RMx of the publicly available video RM. Details of the determination process will be described later.

[0044] Next, the video identification device 10 displays the determination result from the determination unit 13 on the screen using the display unit 14 (step S404). For publicly available videos RM that are presumed to have been created by plagiarizing the target video TM, the similarity score used for the determination may be displayed. Alternatively, for example, the target video TM and the publicly available video RM that are presumed to have been created by plagiarizing the target video TM may be displayed side by side.

[0045] Next, the video identification device 10 notifies the publisher of the publicly available video RM, which is presumed to have been created by plagiarizing the target video TM, of a warning (step S405). Alternatively, the user may determine the result of the determination unit 13 and, based on that determination, notify the publisher of the publicly available video RM, which is presumed to have been created by plagiarizing the target video TM, of the warning. In this case, the display of the determination result by the determination unit 13 in step S304 may be omitted.

[0046] Figure 6 is a flowchart of the decision-making process.

[0047] Referring to Figure 6, in the determination process, the determination unit 13 first acquires data of all representative images TMx from the storage device 23, extracts feature points from each representative image TMx, and identifies the feature quantities of the feature points (step S601). The method for extracting feature points from the representative images TMx is not particularly limited, but for example, the corner points of the representative image TMx may be extracted as feature points. Similarly, the method for identifying the feature quantities of the feature points of each representative image TMx is not particularly limited, but the feature quantities may be identified from the brightness gradient of each feature point.

[0048] Next, the determination unit 13 acquires data of all cropped images RMx from the storage device 23, extracts feature points from each cropped image RMx, and identifies the feature quantities of the feature points (step S602). The feature points of the cropped image RMx can be extracted using the same method as the feature points of the representative image TMx. Also, the feature quantities of the feature points of the cropped image RMx can be identified using the same method as the feature quantities of the feature points of the representative image TMx.

[0049] Next, the determination unit 13 creates multiple groups by clustering all feature points of all representative images TMx according to their features (step S603).

[0050] Next, the determination unit 13 classifies the feature points of the representative image TMx into groups by nearest neighbor search with respect to the cluster centroid in the feature space, and creates a histogram of the feature points for each representative image TMx (step S604).

[0051] Next, the determination unit 13 classifies the feature points of the cropped image RMx into groups by nearest neighbor search with the cluster centroid in the feature space, and creates a histogram of the feature points for each cropped image RMx (step S605).

[0052] Next, the determination unit 13 performs a preliminary determination of the similarity between the target video TM and the publicly available video RM by comparing the histogram of feature points of the representative image TMx with the histogram of feature points of the cropped image RMx (step S606).

[0053] In this process, the determination unit 13 first calculates the similarity of the feature point histograms (hereinafter also referred to as "inter-image feature point classification similarity") by exhaustively examining all representative images TMx of the target video TM and all cropped images RMx of the publicly available video RM. The histogram similarity may be determined based on the Bhattacharya distance. In this case, the shorter the Bhattacharya distance, the higher the similarity. Next, the determination unit 13 identifies combinations of representative images TMx and cropped images RMx (hereinafter also referred to as "feature point classification similar image pairs") in which the inter-image feature point classification similarity exceeds a predetermined threshold, or where the Bhattacharya distance is shorter than a predetermined threshold. The determination unit 13 may also identify the top N feature point classification similar image pairs in terms of inter-image feature point classification similarity. Furthermore, the determination unit 13 performs a preliminary determination of the similarity between the target video TM and the publicly available video RM based on the number of feature point classification similar image pairs. This initial assessment is intended to quickly exclude unrelated publicly available videos (RM) that are highly unlikely to rely on the target video (TM) by using a simple histogram comparison. For example, if there are zero similar image pairs in the feature point classification, the publicly available video RM may be judged as highly unlikely to rely on the target video (TM), i.e., dissimilar.

[0054] Next, the determination unit 13 performs a secondary determination of the similarity between the target video TM and the publicly available video RM by comparing their respective feature points for the combinations of the target video TM and the publicly available video RM that were determined not to be dissimilar in the primary determination (step S607). The process in step S607 will hereafter be referred to as the secondary determination process.

[0055] Figure 7 is a flowchart of the secondary decision process.

[0056] Referring to Figure 7, the determination unit 13 first calculates the similarity between feature points (hereinafter also called "feature point similarity") for each combination of target video TM and public video RM that were determined not to be dissimilar in the initial determination, by exhaustively examining the feature points in each combination of representative image TMx and cropped image RMx (step S701). The feature point similarity may be determined, for example, based on the Euclidean distance between features. In that case, the shorter the Euclidean distance, the higher the feature point similarity.

[0057] Next, the determination unit 13 identifies the combination of feature points that maximizes the similarity between feature points of the representative image TMx and the cropped image RMx, i.e., the combination of feature points that minimizes the Euclidean distance (hereinafter referred to as "similar feature point pair") (step S702). This process searches for corresponding feature points between the representative image TMx and the cropped image RMx. Note that, for any feature point p, if the distance of the combination with the closest similarity between feature points of the representative image TMx and the cropped image RMx is smaller than the distance of the second closest combination multiplied by n (n<1), the determination unit 13 may identify the pair with the closest feature point as a similar feature point pair and delete similar feature point pairs that do not satisfy this condition (ratio test).

[0058] Next, the determination unit 13 identifies a pair of representative image TMx and cropped image RMx that satisfy a predetermined pairing condition in terms of the number of similar feature point pairs (hereinafter also referred to as a "feature point pairing image pair") (step S703). Here, the pairing condition is, for example, that the number of similar feature point pairs exceeds a predetermined threshold and that this number is the maximum for the cropped image RMx.

[0059] Next, the determination unit 13 calculates the similarity between the target video TM and the publicly available video RM in the combination (step S704). In this process, the determination unit 13 first calculates the similarity between the videos by combining the similarity between the images of the target video TM and the publicly available video RM that make up each feature point pairing image pair, and the average value of that combined value. Hereafter, this combined value representing the similarity between the target video TM and the publicly available video RM may be referred to as the total score. Also, the average value representing the similarity between the target video TM and the publicly available video RM may be referred to as the average score.

[0060] The similarity between the target video TM and the publicly available video RM, which form the feature point pairing image pairs described above, is determined based on the number of similar feature point pairs. For example, the ratio of the number of similar feature points to the total number of feature points in the representative image TMx may be used as the similarity between the representative image TMx and the cropped image RMx, which form the feature point pairing image pairs. Here, the ratio of the number of similar feature points to the total number of feature points in the representative image TMx is used as the similarity between the representative image TMx and the cropped image RMx, which form the feature point pairing image pairs. Other configurations are also possible. For example, the ratio of the number of similar feature points to the total number of feature points in the publicly available image RMx may be used as the similarity between the representative image TMx and the cropped image RMx, which form the feature point pairing image pairs.

[0061] Next, the determination unit 13 performs a secondary determination of whether the publicly available video RM may depend on the target video TM, based on the total score and average score between the video of the representative image TMx and the cropped image RMx that form the feature point pairing image pair (step S705).

[0062] In this case, if both the total score and the average score exceed their respective thresholds, the determination unit 13 may determine that the publicly available video RM may be based on the target video TM. If the publicly available video RM is based on the target video TM, similar scenes exist, and detection is possible by using the cumulative value for determination. Also, if the target video TM and the publicly available video RM are long videos, the number of scenes will be large, so even if the publicly available video RM is not based on the target video TM, the similarity between images of coincidentally similar scenes may be accumulated, and the cumulative value (total score) may exceed the threshold. In such cases as well, by using the average score in combination, it is possible to suppress the erroneous detection of publicly available video RM that is not based on the target video TM by making a determination based on the average value. Furthermore, if there are scenes shot in the same or similar locations or scenes that shot the same or similar objects, there may be cases where only certain scenes are similar even if the publicly available video RM is not based on the target video TM. However, even in such cases, determination based on the average score can be used to exclude them.

[0063] The thresholds mentioned above may be predetermined fixed values, or they may be values ​​that can be set or changed by the user.

[0064] The embodiments of the present invention described above are illustrative for the purpose of explaining the invention and are not intended to limit the scope of the invention to those embodiments only. Those skilled in the art can implement the present invention in various other forms without departing from the spirit of the invention.

[0065] For example, in the embodiment described above, the frame images of the target video TM were used directly as the representative image TMx, but other configurations are also possible. In order to improve the detectability of publicly available videos RM that have been plagiarized by modifying the target video TM, it is also possible to create a modified image by modifying the frame images of the target video TM in advance, and then use the modified image in the same way as the representative image TMx to perform the above-described process.

[0066] The processed images include images in which the representative image TMx has been cropped to remove parts that have little impact on the content, images that have been flipped horizontally, images that have been divided, and images that have been corrected (pixels reduced). It is preferable to prepare processed images of the representative image TMx using processing methods that are expected when the representative image TMx is processed and plagiarized.

[0067] According to this configuration, those who commit fraud may modify the stolen video in some way before publishing it to make it harder to detect. A modified video will have a reduced similarity to the original target video TM, making identification difficult. With this configuration, it is possible to easily determine whether a published video RM is a modified version of the target video TM.

[0068] Furthermore, there are video theft techniques that attempt to evade detection by embedding the target video TM into a portion of the screen of the publicly available video RM. The above-described embodiment may be modified to enhance the detectability of such embedding.

[0069] In that case, the determination unit 13 identifies feature point pairing image pairs in step S703, then processes to identify the region in the cropped image RMx where the image of the plagiarized video is embedded, and in step 704 calculates the similarity between the videos based on the number of similar feature point pairs present in the identified region in the cropped image RMx.

[0070] Specifically, a projection transformation matrix is ​​created that associates the region in the representative image TMx with the region in the cropped image RMx where the plagiarized video is embedded, based on the coordinates of similar feature point pairs in the first feature point pairing image pair. For the second and subsequent feature point pairing image pairs, the similarity between the videos can be calculated based on the number of similar feature point pairs between the representative image TMx and the region in the cropped image RMx where the plagiarized video is embedded.

[0071] Furthermore, in the embodiment described above, for example, the public video image selection unit 12 selected a cropped image RMx from public video RMs that companies, individuals, etc., have published on a communication network 90 such as the Internet via a web page provided by the web server 91. However, the public video image selection unit 12 may also select a public video RM from a specific web page that publishes multiple public video RMs. The specific web page may be, for example, a web page of a video posting site where individuals can post videos. This makes it possible to quickly detect a public video RM that depends on the target video TM from among multiple public video RMs.

[0072] The public video image selection unit 12 may select public video RMs from other web pages that link to a specific web page that publishes multiple public video RMs. This makes it possible to improve the accuracy of detecting public video RMs that depend on the target video TM from among multiple public video RMs. [Explanation of symbols]

[0073] 10...Video identification device, 11...Target video image recording unit, 12...Public video image selection unit, 13...Determination unit, 131...Feature point extraction unit, 132...First similar video unit, 133...Second similar video unit, RM...Public video, RMx...Cropped image, TM...Target video, TMx...Representative image

Claims

1. A video identification device that detects a video that is based on a target video from among a plurality of public videos, a target moving image recording unit that records, from the target moving image, an image of a frame included in each structural unit in the time direction of the target moving image as a representative image; a public video image selection unit that selects, for each of the plurality of public videos, a frame image from the public video at predetermined time intervals as a cut-out image; A determination unit that determines whether the public video is a video that may be dependent on the target video based on the representative image of the target video and the cropped image of the public video; A video identification device having the above.

2. The determination unit Extracting feature points from the representative image of the target video and the cut-out image of the public video; The feature points of the representative images are clustered by feature amounts to create multiple groups, and for each representative image, the feature points of the representative image are classified into the groups, and for each cropped image, the feature points of the cropped image are classified into the groups, and for each combination of the target video and the public video, pairs of representative images and cropped images whose inter-image feature point classification similarity based on the classification of the feature points into each group exceeds a predetermined threshold or whose inter-image feature point classification similarity is in the top N rankings are identified as feature point classification similar image pairs, and a primary judgment of the similarity of the target video and the public video is made based on the feature point classification similar image pairs, For each combination of a target video and a public video that is determined to be not dissimilar in the primary determination, a pair of feature points for which the feature point similarity between the representative image and the cutout image in each combination of the representative image and the cutout image is greatest is identified as a similar feature point pair, and a pair of the representative image and the cutout image for which the number of similar feature point pairs satisfies a predetermined pairing condition is identified as a feature point paired image pair, and a secondary determination is made that the public video may be dependent on the target video based on the number of similar feature point pairs in the feature point paired image pair. The moving image identification device according to claim 1 .

3. The determination unit Identifying the feature amount of the feature point based on the brightness gradient around the feature point of each of the representative image of the target video and the cut-out image of the public video; The moving image identification device according to claim 2 .

4. the determination unit represents a distribution of frequencies of feature points classified into each group for each of the target video and the public video in a histogram, and identifies the pair of images with similar feature point classification using the Bhattacharyya distance of the histogram as the inter-image feature point classification similarity; The moving image identification device according to claim 2 .

5. The determination unit calculates an inter-image feature point similarity, which is a similarity between the feature point pairing image pairs, based on the number of similar feature point pairs in the feature point pairing image pairs, and performs a secondary determination that the public video may be dependent on the target video based on an integrated value and an average value of the inter-image feature point similarities. The moving image identification device according to claim 2 .

6. the inter-image feature point similarity is a ratio of the number of the similar feature point pairs to the total number of the feature points of the representative image of the target video or the total number of the feature points of the cropped image of the public image; The moving image identification device according to claim 5 .

7. The determination unit performs a secondary determination that the public video may be dependent on the target video if the cumulative value is equal to or greater than a first threshold and the average value is equal to or greater than a second threshold. The moving image identification device according to claim 5 .

8. the target video image recording unit records a representative image constituting the target video and a processed image obtained by performing a predetermined processing on the representative image; The determination unit determines whether the public video is a video that may be dependent on the target video based on the cut-out image selected from the public video and the representative image and the processed image of the target video. The moving image identification device according to claim 1 .

9. The processed image is a degraded image obtained by degrading the representative image. The moving image identification device according to claim 9.

10. The public video image selection unit selects an image from the public video that is not an inserted image inserted between frames of the public video. The moving image identification device according to claim 1 .

11. The constituent units are scenes that constitute the target video. The moving image identification device according to claim 1 .

12. The determination unit identifies an area in the cut-out image in which an image of the public video that may be based on the target video is embedded, and calculates the inter-image feature point similarity based on the number of similar feature point pairs present in the identified area in the cut-out image. The moving image identification device according to claim 5 .

13. The determination unit creates a projective transformation matrix that associates the representative image with an area in the cropped image in which an image of the public video that may be based on the target video is embedded, based on the coordinates of the similar feature point pair in the feature point pairing image pair. The moving image identification device according to claim 12.

14. a target moving image recording step of recording, as a representative image, an image of a frame included in each constituent unit in a time axis direction of the target moving image from the target moving image; a public video image selection step of selecting, for each of a plurality of public videos, frame images at predetermined time intervals from the public videos as cut-out images; A determination step of determining whether the public video is a video that may be dependent on the target video based on the representative image of the target video and the cropped image of the public video; A video identification method having the following.

15. a target video image recording unit provided in the computer is caused to record, from the target video, an image of a frame included in each constituent unit in a time axis direction of the target video as a representative image; a public video image selection unit included in the computer is caused to select, for each of a plurality of public videos, frame images at predetermined time intervals from the public videos as cut-out images; A video identification program that causes a judgment unit provided in the computer to determine whether the public video is a video that may be dependent on the target video based on the representative image of the target video and the cropped image of the public video.

Images