JP2010026981A - Particular scene learning system and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a particular scene learning system and its program which convert metadata edited as a particular scene into a predicate format for machine learning and extract the definition of the particular scene. <P>SOLUTION: The particular scene learning system includes: a digest editing device 11 which accumulates meta information used in editing in a first knowledge base 12 while adding to the meta information additional information making it possible to determine whether or not the meta information has been edited as a particular scene; a definition knowledge extraction device 14 for converting the meta information with the additional information accumulated in the first knowledge base 12 into a predicate format, allowing machine learning of the information converted in the predicate format, extracting knowledge definition data to be edited as the particular scene, and accumulating the data in a second knowledge base 15; and a digest creation device 16 which uses the knowledge definition data accumulated in the second knowledge base 15 as a search key and extracts the meta information corresponding to the search key from within the meta information accumulated in the first knowledge base 12. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a system for automatically learning and storing a definition of a specific scene (for example, a highlight scene) using information on a scene such as a sports video created by a user. The present invention relates to an apparatus that supports automatic production of a specific scene.

  In recent years, with the spread of digital broadcasting and CS (Communications Satellite) broadcasting, the multi-channel era has come, and content providers such as broadcasting stations are required to produce many high-quality programs, and efficient programs Realization of production methods is indispensable. While there are programs that are produced by making full use of the advanced know-how of the creators, there are many standardized programs. Therefore, a technology is desired that realizes a support system that can automate the production of such a program to some extent, reduces the burden on the creator, and allows the creator to focus on more advanced work.

  For example, digest video used for sports news is also a type of content that producers make use of their know-how, but the “scenes” during the game are considered to be shared among producers to some extent. If the creator's concept about the “rising scene” can be structured and stored in the database, it is possible to easily search for highlight scene candidates and automatically produce the digest video itself. It becomes like this.

  Therefore, the concept related to sports games is structured in advance as an ontology, and the user who produces the digest video handles the concept using GUI (Graphical User Interface) etc., thereby defining and extracting important scenes during the game. There has been proposed a technique that makes it possible to perform the above processing (see, for example, Patent Document 1).

  On the other hand, a technique for automatically detecting a highlight scene of a sports broadcast using audio information in content has been proposed for automatic extraction of highlights (see, for example, Patent Document 2).

  In these techniques, the production know-how is not in a form of knowledge expression that can be understood by a third party, so digest or highlight editing is performed based on the creator's unique knowledge.

JP 2008-131170 A JP 2004-23354 A

  However, in the technique of Patent Document 1, it is necessary for the user who produces or edits the digest video to combine the concepts while considering what kind of scene will be proactively raised. Although it has become easier to search for scenes with intent, it does not significantly reduce the burden on the creator, and there is a need for a system that can automatically extract highlight scenes.

  In the technique of Patent Document 2, only speech recognition of input speech is used, and language data such as an announcer's comment is required, and it is expected that the result depends on the recognition rate of speech recognition. The For this reason, it is considered that using the game metadata that is currently generally distributed is better for generalizing information used for highlight extraction.

  Therefore, an object of the present invention is to establish a definition of a specific scene (for example, a highlight scene) using metadata to which additional information that can identify a specific scene (highlight information) selected by a content creator is added. It is another object of the present invention to provide a specific scene learning system and program for supporting digest generation.

  That is, the specific scene learning system according to the present invention receives meta information associated with predetermined image information from the outside, performs digest editing of the specific scene, and meta information used for editing the specific scene, A digest editing device having means for accumulating in the first knowledge base with additional information for identifying whether or not the scene has been edited, and a meta with additional information stored in the first knowledge base. Definition having means for converting information into a predicate format, and means for causing machine learning of information converted into the predicate format, extracting knowledge definition data to be edited as a specific scene, and storing the data in a second knowledge base For the digest extraction of the knowledge extraction device and arbitrary image information, the knowledge definition data stored in the second knowledge base is used as a search key. Characterized in that it comprises a digest generation unit having a means for extracting the meta information corresponding to the search key from the meta information stored in the first knowledge base.

  In the specific scene learning system according to the present invention, the meta information compiled as the specific scene is referred to the meta information stored in the first knowledge base with reference to an ontology defined in advance to be edited as the specific scene. It further comprises a knowledge metadata generation device having means for accumulating advanced meta information by adding a description based on the ontology to the data, and storing the meta information in place of the meta information edited as the specific scene. .

  In the specific scene learning system according to the present invention, the definition knowledge extraction device converts metadata into a predicate format that can be used in an inductive logic programming process, and the digest generation device converts the knowledge definition data into a predicate. It is characterized in that it is generated by data converted into a query language format based on the format itself or a SPARQL (sparqL protocol and RDF Query Language) knowledge base.

  Also, in the specific scene learning system according to the present invention, the knowledge metadata generation device has means for sending a scene group corresponding to the corresponding meta information extracted by the search key to the digest editing device. To do.

  Also, in the specific scene learning system according to the present invention, the knowledge metadata generation device includes means for generating a digest video by combining scene groups corresponding to corresponding meta information extracted by the search key. To do.

  In the specific scene learning system according to the present invention, the knowledge metadata generation device further includes means for recording the generated digest video or a list of corresponding metadata on a recording device or displaying the list on a display device. It is characterized by.

  Furthermore, the present invention provides a computer configured as a digest editing apparatus according to the present invention by receiving meta information associated with predetermined image information from the outside and performing digest editing of a specific scene, It is also characterized as a program for causing the meta information used for editing to add additional information that can identify whether or not the specific scene has been edited to be accumulated in the first knowledge base.

  Furthermore, the present invention provides a computer configured as the definition knowledge extracting apparatus of the present invention, means for converting meta information with additional information accumulated in the first knowledge base into a predicate format, and information converted into a predicate format. Is also characterized as a program for executing the steps of extracting knowledge definition data to be edited as a specific scene and storing it in the second knowledge base.

  Furthermore, the present invention provides a computer configured as the digest generation apparatus of the present invention, using the knowledge definition data stored in the second knowledge base as a search key for digest editing of arbitrary image information. It is also characterized as a program for executing a step of extracting meta information corresponding to the search key from meta information stored in one knowledge base.

  According to the present invention, the digest scene creation know-how possessed by the content creator can be accumulated as knowledge. As a result, it is easy for producers to create digest videos for sports news.

  In addition, by incorporating this function in a content receiver such as a TV tuner, a digest video according to the content viewer's preference can be generated in the receiver, and a new content viewing style can be provided. .

  Hereinafter, a specific scene learning system using a digest generation apparatus according to an embodiment of the present invention will be described.

[System configuration using digest generator]
FIG. 1 is a partial functional block diagram according to the present invention of a digest editing apparatus in a specific scene learning system according to an embodiment of the present invention. FIG. 2 is a functional block diagram of the knowledge metadata generation device 13 in the specific scene learning system according to the embodiment of the present invention. FIG. 3 is a functional block diagram of the definition knowledge extracting apparatus in the specific scene learning system according to the embodiment of the present invention. FIG. 4 is a functional block diagram of the digest generation device in the specific scene learning system according to the embodiment of the present invention.

  The specific scene learning system of the present embodiment shown in FIGS. 1 to 4 includes a digest editing device 11, a definition knowledge extracting device 14, and a digest generating device 16. Preferably, the knowledge metadata generating device 13 is further provided. Prepare.

  The digest editing apparatus 11 receives meta information associated with predetermined image information from the outside, performs digest editing of the specific scene, and whether or not the meta information used for editing the specific scene has been edited as the specific scene. It has a function of accumulating in the knowledge base (KB1) 12 with additional information that can be identified.

  The knowledge metadata generation device (KMG) 13 refers to the meta information stored in the knowledge base (KB1) 12 with reference to an ontology defined in advance to be edited as a specific scene, and the metadata edited as the specific scene. Is added with a description based on the ontology, and the advanced meta information is stored in the knowledge base (KB1) 12 instead of the meta information edited as a specific scene.

  The definition knowledge extracting device 14 converts the meta information with additional information stored in the knowledge base (KB1) 12 into a predicate format, machine-learns the information converted into the predicate format, and knowledge to be edited as a specific scene. It has a function of extracting definition data and storing it in the definition knowledge base (KB2) 15.

  The digest generation device (DG) 16 uses the knowledge definition data stored in the definition knowledge base (KB2) as a search key for digest editing of arbitrary image information, and stores the meta data stored in the knowledge base (KB1) 12 as a search key. It has a function of extracting meta information corresponding to the search key from the information.

  Hereinafter, each device will be described in more detail.

  The specific scene learning system according to the present embodiment is a system that enables the knowledge metadata generation device 13 to structure the know-how of the creator who uses the digest editing device 11, and the existing metadata processing for processing the metadata. It can be installed in the system. In such a metadata processing system for sports video, the specific scene learning system of the present embodiment uses a recursive logic programming processing system (hereinafter also referred to as an ILP processing system) or the like for the game metadata to which highlight information is added. Convert to available predicate format, establish and store definition of specific scene (eg highlight scene) in predicate format output from ILP processing system, and highlight sports using stored highlight scene definition Search scenes or automatically generate digest video.

  In the conventional highlight generation technique using video data and audio data, the reason why the scene is selected by the creator cannot be obtained in an information format that can be understood by humans. In other words, in the specific scene learning system of this embodiment, in the case of actual digest production, the creator can answer the reason clearly to the question “Why did you choose the scene as a highlight?” However, by acquiring the reason by machine learning of the digest generation device 13 of the present embodiment, it is possible to construct a program production knowledge base in which knowledge to be called production know-how is accumulated.

  That is, the digest generation device 13 in the specific scene learning system of the present embodiment acquires the definition of the highlight scene using the metadata to which the content creator has added highlight information of a certain game, and the highlight based on the definition. The digest of the light scene can be provided to the digest editing apparatus 11 used by the content creator. The definition of the highlight scene can be described using a predicate format as to what kind of sports situation (highlight scene) the sports scene is exciting. The digest generation apparatus 13 of the present embodiment can also automatically extract a highlight scene from, for example, metadata of another game using this definition.

  Referring to FIG. 1, a broadcast station program producer (content producer) operates a digest editing apparatus 11 to view a scene that swells while watching a recorded game video (generally also referred to as image information). Extract and edit to produce a digest video for sports news. In recent years, sports game metadata has been distributed to broadcasting stations and Web portal sites, and program producers can arbitrarily obtain image information and its metadata.

  For example, a program producer such as a broadcast station can usually receive game situation metadata from a data distribution company or the like in order to superimpose information such as score and SBO on a game video. . Metadata (m) provided by a sports data distribution company or a known video / audio analysis system is stored in the knowledge base (KB1) 12 in a table format or an XML document format. FIG. 5 shows an example of game metadata as an example of metadata (m) described in the XML format. This metadata describes a part of a certain game, and shows a situation in which a pitcher throws a straight and a batter hits it at one bat and results in 1 point. In this example, a series of flows such as “throwing, hitting, scoring” is a minimum scene unit of “1 game”. All highlight scenes are extracted in units of this “one game”. It is also possible to add advanced information (high-level information such as “a goodbye home run” or “reversal hit” for a certain scene) by metadata using a knowledge metadata generation device (KMG) (for example, JP 2007-249907).

  That is, the digest editing apparatus 11 inputs image information and metadata thereof, and information indicating which scene the program producer has selected is selected as a highlight scene (for example, flag value 1). ) / Not selected (for example, flag value 0) "(hereinafter also referred to as metadata with additional information) is stored in the knowledge base (KB1) 12 functioning as a database for storing as knowledge.

  This content creator can edit a recorded video group (scene group) and create a digest video by connecting the exciting scenes. Generally, metadata stored in the knowledge base (KB1) 12 describes a link to a video. When a content creator edits a specific scene, the digest editing apparatus 11 Using the link information, highlight information identification information (additional information on whether or not the scene is selected as a highlight scene) can be added to the metadata in the knowledge base (KB1) 12.

  Referring to FIG. 2, the knowledge metadata generation device (KMG) 13 includes ontology information reference means 131 and high level information addition means 132, and operates metadata with additional information accumulated by the digest editing apparatus 11. The content of the metadata with additional information is explained in more detail by the high level information adding means 132 with reference to the ontology previously defined by the ontology information referring means 131. Is added to the knowledge base (KB1) 12 to generate or update advanced metadata (hereinafter referred to as advanced metadata).

  Note that it is possible to arbitrarily select whether or not to upgrade the metadata in the knowledge base (KB1) 12 by the knowledge metadata generation device (KMG) 13; Learning performance can be improved. It should be noted that the metadata stored in the knowledge base (KB1) 12 is still described in the XML format in which image information and the metadata can be obtained arbitrarily.

  Referring to FIG. 3, the definition knowledge extraction device 14 functions as, for example, a highlight knowledge extraction device (HKE) that extracts knowledge of a highlight scene, and includes a predicate format conversion unit 141 and a machine learning unit 142. The definition knowledge extracting device 14 reads the advanced metadata (or metadata used for the digest editing device 11) stored in the knowledge base (KB1) 12 by an operation of the operator or automatically, The predicate format conversion unit 141 converts the XML format metadata into a predetermined inductive logic programming processing format, and the machine learning unit 142 performs machine learning using the converted metadata for editing. Definition knowledge data (for example, concept definition data of a highlight scene) is acquired as knowledge data for defining the concept of this, and this definition knowledge data is stored in a dedicated knowledge base (definition knowledge base (KB2) 15). In this way, the definition knowledge extraction device 14 acquires the definition of the highlight scene using, for example, metadata to which the content creator adds highlight information of a certain game, and the definition knowledge base (KB2) 15 ).

  FIG. 6 shows an example of a predicate generated by converting the metadata accumulated in the knowledge base (KB1) 12 into a predicate format by the definition knowledge extracting device 14 functioning as a highlight knowledge extracting device (HKE). As described above, since the metadata (for example, game metadata) stored in the knowledge base (KB1) 12 is described in a table format or an XML format, all the contents are “subject-predicate-object”. Can be defined to be described in a collection of the form “(this form is hereinafter referred to as“ triple ”). The triple format can be uniquely converted into the predicate format. For example, the game metadata in the XML format shown in FIG. 5 can be converted into the predicate format shown in FIG. This predicate format is a Prolog program itself, and by converting metadata into this format, machine learning by the ILP processing system of the machine learning means 142 at the subsequent stage is enabled.

  Here, an example of learning and accumulation of highlight scene definitions by inductive logic programming will be described in more detail.

[Learn Highlight Scene Definition by Inductive Logic Programming]
PROGOL and the like are well known as ILP processing of inference processing by inductive logic programming (for example, Koichi Furukawa et al., “Inductive Logic Programming”, Kyoritsu Publishing, published in May 2001). Therefore, the definition knowledge extraction device 14 functioning as a highlight knowledge extraction device (HKE) can use the knowledge related to the learning target concept as background knowledge of inductive logic programming, and the predicate format conversion means 141 can It is possible to generate a predicate format that can be used as an input to an ILP processing system.

  In other words, the ILP processing system used for the machine learning unit 142 can learn a concept common to positive examples mainly by using positive examples, negative examples, and background knowledge as inputs. An example of a positive example, a negative example, and background knowledge regarding a baseball game is shown in FIG. The positive example input to the ILP processing is “scene selected as highlight scene” (FIG. 7A), and the negative example is “scene not selected as highlight scene” (FIG. 7B). The background knowledge describes the game situation related to each scene described above (FIG. 7C). FIG. 11 shows an example in which game metadata in XML format is converted into a predicate format on a line-by-line basis. As described above, classification into “positive example”, “negative example”, and “background knowledge” is suitable for enhancing the identification performance of highlight information for machine learning of the ILP processing system 142. As background knowledge, not only a simple game situation but also an ontology describing various concepts related to baseball can be given.

  The ILP processing system performs inductive inference using these pieces of information, and outputs “highlight scene definition” in predicate form as output. An output example of the ILP processing system is shown in FIG.

[Accumulation of highlight scene definitions]
The output of the ILP processing system is accumulated in the definition knowledge base (KB2) 15 as “highlight definition knowledge”. The storage format may be the predicate format itself, or may be stored after being converted into a knowledge-based query language format such as SPARQL (sparq Protocol and RDF Query Language). For example, FIG. 9 shows a correspondence example between the predicate format and the sparql query.

  Referring to FIG. 4, the digest generation device (DG) 16 includes a definition scene extraction unit 161, a definition scene combination unit 162, and a definition scene recording / playback unit 163. The definition scene extraction unit 161 extracts a query language such as SPARQL from the highlight knowledge stored in the definition knowledge base (KB2) 15, reads metadata in the knowledge base (KB1) 12, and applies the query language. Then, a scene group corresponding to definition knowledge (for example, highlight definition) is extracted, and the extracted scene group is provided to the digest editing apparatus 17 used by the content creator as a “highlight scene candidate”. The definition scene combining unit 162 combines scene groups extracted by the definition scene extraction unit 161 by an operation of the content creator or automatically, and automatically generates a digest video. The definition scene recording / reproducing unit 163 causes the recording device 18 to record the digest video generated by the definition scene combining unit 162 or a list of corresponding metadata by an operation of the content creator or a display device. 19 is controlled to be reproduced and displayed.

  For example, a usage example of the highlight knowledge stored in the definition knowledge base (KB2) 15 will be described in more detail.

[Use of highlight definition knowledge]
A user (for example, a program producer of a broadcasting station) uses a digest generation device (DG) 16 to highlight highlights from any sport video from the knowledge base (KB1) 12 and the definition knowledge base (KB2) 15. A scene can be extracted. For example, the digest generation device (DG) 16 extracts a query language such as SPARQL from the highlight definition stored in the definition knowledge base (KB2) 15, and uses the query language as a search key in the knowledge base (KB1) 12. The meta information can be searched to extract a scene group corresponding to the highlight definition. The obtained scene group can be automatically supported by adding identification information as a “highlight scene candidate” and providing it to the content creator, or by combining the obtained scene groups. A digest video is generated.

  As described above, the specific scene learning system of the present embodiment automatically acquires the concept definition of the highlight scene using the metadata to which the highlight information is added and the inductive logic programming processing system, and the definition knowledge base ( KB2) 15.

  Taking a baseball game as an example, the game metadata describes the status of the game (score, SBO count, progress, etc.) and the information about the event that occurred, in a perceptible table format or XML document format. Stored in the knowledge base (KB1) 12. The specific scene learning system of this embodiment converts this information into a predicate format such as a Prolog program, and performs inductive inference processing by an inductive logic programming processing system. In other words, the specific scene learning system of the present embodiment uses various features of the highlight scene (game status, events that occurred, information on players related to the event) and various features of the scene that was not selected as the highlight scene. Inductive inference is performed, and the definition of what features the highlight scene has is automatically acquired by machine learning. The definition obtained is stored in the definition knowledge base (KB2) 15 as “highlight knowledge”, and the creator searches for a scene that easily rises using that knowledge, or the system automatically produces a digest video. Such a function can be realized.

  Next, an implementation example 1 in which the specific scene learning system using the digest generation apparatus according to the embodiment of the present invention is configured as a digest video production support system will be described.

[Example 1]
FIG. 10 shows a digest video production support system using a digest generation apparatus according to an embodiment of the present invention. This system is assumed to be used in content providers such as broadcasting stations.

  A broadcast station or the like usually has a game situation meta-data from a data distribution company (for example, the metadata automatic generation system 20 of a sports data distribution company) to superimpose information such as a score and SBO on a game video. Data (m) is provided. This metadata (m) is stored in a game knowledge base 12a (corresponding to the above-described knowledge base 12) in the broadcasting station by a metadata storage device (MDSS) 21 that sequentially stores received metadata, for example. The game knowledge base 12a can add more advanced metadata by a knowledge metadata generation device (KMG) 13a having a sports ontology 131a (corresponding to the ontology information calculation means 131 described above).

  The program producer operates the digest editing system (DPS) 22 (corresponding to the digest editing device 11 described above), while watching the recorded sports game video stored together with the meta information in the game knowledge base 12a. Search for highlight scenes manually, stitch them together and edit them to produce a digest video for specific sports news. Since the game metadata includes scene time information, the digest editing system (DPS) 22 selects the scene and game selected by the creator when the game metadata is stored in the game knowledge base 12a. Using the link information with the metadata, highlight information h (whether or not the scene is selected as a highlight scene) is added to the game metadata. The digest editing system (DPS) 22 may control the knowledge metadata generation device (KMG) 13a to add more advanced metadata.

  The game metadata (m) to which the highlight information is added corresponds to the highlight knowledge extraction device (HKE) 14a (the definition knowledge extraction device (HKE) 14 described above) in units of specific image information such as every game and every season. Are converted into “positive examples”, “negative examples”, and “background knowledge” for an ILP processing system (ILPS) 142a (corresponding to the machine learning means 142 described above). The ILP processing system (ILPS) 142a performs inductive inference processing using the information, and “highlight definition knowledge” (corresponding to the definition knowledge data described above) as to what features the highlight scene has. Send to highlight extractor (HKE) 14a. The highlight extraction device 14a stores the obtained highlight definition knowledge in the highlight knowledge base (KB2) 15a, and allows the creator to operate the digest editing system (DPS) 22 so that it can be used freely. deep.

  Even in the digest video production support system constructed in this way, when creating a highlight video, the creator selects the scene by himself while viewing the game video and connects them by editing. Can be used, but the digest generation device 16a can automatically present “a video group that is a candidate for a highlight scene” on the digest editing system (DPS) 22 screen. The program producer can create a digest video by selecting and connecting the displayed highlight candidate videos without viewing all the videos.

  According to this implementation example, the work of producing a digest video can be significantly improved. Further, as described above, it is also possible to automatically connect the highlight scenes extracted by this system and generate a digest video.

  Next, an implementation example 2 in which the specific scene learning system using the digest generation apparatus according to the embodiment of the present invention is configured as a digest video providing system will be described.

[Example 2]
FIG. 12 shows a digest video providing system using a digest generating apparatus according to an embodiment of the present invention. This system is assumed to be built in a receiver or a PC in the user's home. Components similar to those in FIG. 10 are described with the same reference numerals.

  The sports data distribution company's automatic metadata generation system 20 distributes game metadata (m) to receivers and PCs in each home via broadcast waves and the Internet. The metadata storage device stores the received metadata (m) in the game knowledge base 12a in real time. The game knowledge base 12a can be given more advanced metadata by the knowledge metadata generation device 13a.

  A user of this system for home environment operates a known video viewing system (VPS) 23 (corresponding to the digest editing device 11 described above) built in a receiver, a PC, or the like to display a normal TV program or the like. Suppose you are watching. A known video viewing system has a built-in video storage system such as an HDD so that a recorded sports program can be viewed later. When a user views a recorded program while fast-forwarding / rewinding, the viewing information by the user that the user has watched the scene in real time or skipped fast-forwarding is the game knowledge base as highlight information h. (KB1) When sent to 12a and accumulates game metadata in the game knowledge base 12a, highlight information (whether or not the scene is selected as a highlight scene) is added to the game metadata. .

  The game metadata (m) to which the highlight information h is added is sent to the highlight knowledge extraction device (HKE) 14a in units of specific image information such as every game and every season, where the ILP processing system (ILPS) 142a. Are converted into “positive examples”, “negative examples” and “background knowledge”. The ILP processing system (ILPS) 142a performs inductive inference processing using these pieces of information, and “highlight definition knowledge” (corresponding to the definition knowledge data described above) indicating what characteristics the highlight scene has. To the highlight knowledge extractor (HKE) 14a. The highlight knowledge extraction device (HKE) 14a stores the obtained highlight definition knowledge in the highlight knowledge base (KB2) 15a, and the user operates the video viewing system (VPS) 23 (other devices). (Also) Make the situation freely accessible.

  The digest generation device 16a extracts a highlight scene from the metadata using information in the highlight knowledge base 15a, and automatically generates a digest video. The produced digest video can be displayed on the video viewing scene system (VPS) 23 such as “Today's Team A vs. Team B digest” in a manner allowing a plurality of items to be selected. It is possible to select and view the corresponding digest video arbitrarily.

  As one aspect of the present invention, each of the digest editing device 11, the knowledge metadata generating device 13, the definition knowledge extracting device 14, and the digest generating device 16 can be configured as a computer, and the above-described means and functions are realized. Is stored in a storage unit (not shown) provided inside or outside each computer. Such a storage unit can be realized by an external storage device such as an external hard disk or an internal storage device such as ROM or RAM. The control unit that executes the program can be realized by a central processing unit (CPU) or the like. That is, the CPU can appropriately read from the storage unit a program in which the processing content for realizing the function of each component is described, and implement each device on the computer. Here, the function of any means may be realized by all or part of the hardware.

  In the embodiment described above, the program describing the processing contents for realizing the function of each component can be recorded on a computer-readable recording medium. As the computer-readable recording medium, any recording medium such as a magnetic recording device, an optical disk, a magneto-optical recording device, and a semiconductor memory may be used.

  In addition, the program describing the processing contents can be distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM, and such a program can be distributed on a network such as an IP. The program can be distributed by storing the program in a storage area of the server and transferring the program from the server to another computer via the network.

  In addition, a computer that executes such a program can temporarily store, for example, a program recorded on a portable recording medium or a program transferred from a server in a storage unit of each computer. As another embodiment of the program, the computer may directly read the program from a portable recording medium and execute processing according to the program, and each time the program is transferred from the server to the computer. In addition, the processing according to the received program may be executed sequentially. Note that the program in this aspect includes information provided for processing of an electronic computer and equivalent to the program (data that is not a direct command to the computer but has a property that defines the processing of the computer). .

  Although the highlight scene has been described as a representative example of the above-described embodiments, it will be apparent to those skilled in the art that many changes and substitutions can be made within the spirit and scope of the present invention. For example, the creator may desire to generate a digest of a specific player based on a clear intention, and the present invention can be applied by replacing a highlight scene with a scene of a specific player. Accordingly, the invention should not be construed as limited by the embodiments described above, but only by the claims.

  According to the specific scene learning system using the digest generation device of the present invention, information necessary for digest generation can be automatically extracted from the provided meta information, which is useful for applications where digest generation is desired. .

It is a partial functional block diagram concerning the present invention of a digest editing device in a specific scene learning system of one example by the present invention. It is a functional block diagram of the knowledge metadata production | generation apparatus 13 in the specific scene learning system of one Example by this invention. It is a functional block diagram of the definition knowledge extraction apparatus in the specific scene learning system of one Example by this invention. It is a functional block diagram of the digest production | generation apparatus in the specific scene learning system of one Example by this invention. It is a figure which shows an example of the game metadata of an XML format. It is a figure which shows an example which converted the game metadata of the XML format shown in FIG. 5 into the predicate format. It is a figure which shows the example of the positive example, negative example, and background knowledge as input information of the ILP processing system in the specific scene learning system of one Example by this invention. It is a figure which shows the example of an output of the ILP process type | system | group in the specific scene learning system of one Example by this invention. It is a figure which shows the example of a correspondence of the predicate form of the definition knowledge in a specific scene learning system of one Example by this invention, and a sparql query. It is a figure which shows the digest video production support system using the digest production | generation apparatus of the one implementation example by this invention. It is a figure which shows an example which converted the game metadata of the XML format into the predicate format. It is a figure which shows the digest video provision system using the digest production | generation apparatus of the one implementation example by this invention.

Explanation of symbols

11 Digest Editing Device 12 Knowledge Base (KB1)
13 Knowledge metadata generator (KMG)
14 Definition Knowledge Extraction Device 15 Definition Knowledge Base (KB2)
16 Digest generator (DG)
20 Metadata automatic generation system 21 Metadata storage device (MDSS)
22 Digest Editing System (DPS)
23 Video viewing system (VPS)
131 ontology information reference means 132 high-level information addition means 141 predicate format conversion means 142 machine learning means 161 definition scene extraction means 162 definition scene combination means 163 definition scene recording / playback means 12a game knowledge base 13a knowledge metadata generation device (KMG)
14a Highlight Knowledge Extraction Device (HKE)
15a Highlight Knowledge Base (KB2)
16a Digest generator 131a Sports ontology 142a ILP processing system (ILPS)

Claims (9)

The meta information associated with the predetermined image information can be received from the outside, and the digest editing of the specific scene can be identified and whether the meta information used for editing the specific scene has been edited as a specific scene can be identified A digest editing device having means for storing additional information in the first knowledge base;
Means for converting meta-information with additional information stored in the first knowledge base into a predicate format, and machine learning of the information converted into the predicate format to extract knowledge definition data to be edited as a specific scene And a definition knowledge extraction device having means for accumulating in the second knowledge base,
Corresponds to the search key from the meta information stored in the first knowledge base using the data of the knowledge definition stored in the second knowledge base as a search key for digest editing of arbitrary image information A digest generating device having means for extracting meta information to be
A specific scene learning system comprising:   The meta information stored in the first knowledge base is referred to an ontology defined in advance to be edited as a specific scene, and a description based on the ontology is attached to the metadata edited as the specific scene. The specific scene learning system according to claim 1, further comprising a knowledge metadata generation device having means for storing sophisticated meta information in place of the meta information edited as the specific scene. . The definition knowledge extraction device converts the metadata into a predicate format that can be used in an inductive logic programming process,
The digest generating apparatus generates the data of the knowledge definition from data converted into a predicate format itself or a knowledge base query language format of SPARQL (sparql Protocol and RDF Query Language). The specific scene learning system according to 1 or 2.   4. The knowledge metadata generating apparatus according to claim 1, further comprising means for sending a scene group corresponding to corresponding metadata extracted by the search key to the digest editing apparatus. The specific scene learning system according to the item.   The knowledge metadata generation apparatus includes means for generating a digest video by combining scene groups corresponding to corresponding metadata extracted by the search key. The specific scene learning system according to the item.   6. The knowledge metadata generation device according to claim 5, further comprising means for recording the generated digest video or a list of corresponding metadata on a recording device or displaying the list on a display device. Specific scene learning system. A computer configured as the digest editing device according to claim 1,
Receiving meta information associated with predetermined image information from the outside, and performing digest editing of a specific scene;
A program for executing the step of accumulating in the first knowledge base with additional information identifiable as to whether or not the meta information has been edited as a specific scene is added to the meta information used for editing the specific scene in the step . A computer configured as the definition knowledge extraction device according to claim 1,
Means for converting meta-information with additional information stored in the first knowledge base into a predicate format, and machine learning of the information converted into the predicate format to extract knowledge definition data to be edited as a specific scene And a program for executing the step of accumulating in the second knowledge base. A computer configured as the digest generation device according to claim 1,
Corresponds to the search key from the meta information stored in the first knowledge base using the data of the knowledge definition stored in the second knowledge base as a search key for digest editing of arbitrary image information A program for executing a step of extracting meta information to be performed.

Images