JP3023461B2 - Database device for non-coded information - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to non-coded information such as drawings, pictures, moving images, etc., which is not coded.
The present invention relates to a technique for creating a database of the information and enabling a highly accurate search. The field of application of the present invention is a database system for drawings, pictures, moving images, etc., which can be searched with arbitrary words, a multimedia system,
There is an electronic file system that can be searched by words, an automatic selection recording and retrieval system of television images, a classification and retrieval system of video recorded images, an interactive movie, and the like.

[0002]

2. Description of the Related Art Conventionally, when non-coded information such as drawings and images is stored in a database, each is simply assigned a different number, a separate comparison table is prepared, and the database is stored. Or searching the database. Alternatively, humans look at non-coded information such as drawings and images, add the most appropriate keywords and search information to the meaning (contents) represented by the non-coded information, classify them, and accumulate a database. Or searching its database. For example, in an image file system or the like, a human inputs a title and search information later on an input image image. In any case, artificial judgment was essential.

[0003]

For non-coded information such as drawings, pictures and moving images, it is extremely difficult to simply extract features of the information. Therefore, when traditionally classifying or organizing such uncoded information,
Either they were numbered mechanically and a separate look-up table was prepared, or humans looked at the uncoded information and added appropriate titles and search information. Most of these operations were artificial. In the case of mechanical numbering, a comparison table had to be prepared and searched, and the search operation was extremely difficult. In addition, when adding a title or search information, the work itself of adding them is also difficult. As described above, it takes a lot of time and effort to create a database, and furthermore, since the search information is attached to humans, the subjectivity enters, and there is a problem that it is difficult to create an objective and accurate database.

The present invention has been made in view of such circumstances, and since it is not a character code such as a drawing, a picture, or a moving image, it is originally very difficult to extract features from the character. in among difficult uncoded information, or in the vicinity of such a non-coded information, focused on the language information of the added by which characters for explaining the non-coded information, such language information It is an object of the present invention to automatically create a database of non-coded information such as drawings, pictures, and moving images, and to enable high-precision search. In other words, an object of the present invention is to automate the creation of a database of non-coded information, which conventionally required enormous manual work to create the database, and to save labor. In addition, conventionally, it has been difficult to search a database of non-coded information that has not been classified or assigned a keyword, but it is another object of the present invention to facilitate the search.

[0005]

According to the present invention, there is provided an apparatus for creating a database of non-coded information , the most characteristic of which is that a keyword extracted from character information included in the non-coded information is weighted to a certain level or more. The point is that a weighted keyword is added to the non-coded information portion to create a database. In other words, a sentence containing uncoded information is read and
Reading means to make child, character data part and non-coding
Image processing means for separating an information part, and an uncoded information part
Character recognition processing to encode characters in and around minutes
Means and a plurality of keywords from the encoded character information.
Language processing means for extracting keyword candidates, and keyword candidates
Character information corresponding to the original non-coded information
The closer to the location and the higher the frequency of appearance, the greater
Scoring processing means for assigning points
Score judgment processing means for judging whether or not
It is judged that the score has reached the reference point by the mark number judgment processing means.
The non-code
And additional processing means for adding to the conversion information portion .

[0006]

According to the present invention, a score is assigned to a keyword extracted from coded character information, and a keyword above a reference point is added to a separated non-coded information part to form a database. The association between the coded information part and the keyword for classifying and searching the coded information part becomes close.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments according to the present invention will be described below in detail with reference to the drawings.

First Embodiment This first embodiment relates to creation of a database for figures in a document. FIG. 1 is a block diagram showing a configuration of a database system for non-coded information according to the first embodiment.

A document 1 containing a figure which is non-coded information is read as image data by a scanner 2 which is a reading means, and the text information and the figure information are digitized. Next, the image data is subjected to required image processing 3 by the image processing means, thereby separating the text data portion 4 and the drawing data portion 5. The figure data portion 5 corresponds to the non-coded information.
Furthermore, statements for text data portion 4 and Figure portion 5
The character recognition processing unit 6 performs the character recognition processing 6 to encode the characters in the text data part 4 and the characters in the figure data part 5, respectively. Although was created by this encoding is character information 7, language processing hand against the character information 7
By performing the language processing 8 in the stage, a plurality of keywords 9
Is extracted. These candidates for the keyword 9 correspond to the original document 1.

[0010] Then, the extracted climate of more keywords 9
The scoring processing means 10 performs scoring processing 10 on the complement . The scoring process 10 is performed as follows.
In other words, the score is given to a keyword candidate extracted from a sentence that is separated from other sentences near the side of the figure and that is described in parallel with the side of the figure. The next largest score is given to the keyword candidates extracted from the characters in the figure. The next largest score is given to a keyword candidate extracted from a sentence existing around the figure. In this case, the largest score is given in the order of position closer to the figure. Furthermore, a keyword with a higher appearance frequency is given a higher score. As described above, a unique score is given to each of various keyword 9 candidates .

Next, a point discriminating process 1 is performed by the point discriminating processing means.
It is determined whether or not the score assigned to each candidate for the keyword 9 has reached the reference point according to 1, and the candidate for the keyword 9 having the score equal to or more than the reference point is determined as the keyword 9 for the corresponding figure data portion 5. Figure database by performing additional processing 12 for adding an additional processor Te 1
Create 3.

When classifying the database 13 of the diagram created as described above, the classification is performed according to the keyword 9 described above. When searching the database 13, the searcher inputs words related to the figure to be searched, and reads out the figure data portion 5 to which a keyword 9 matching the input word or a keyword 9 close thereto is added. . In the case of this search, the search is performed in order from the keyword 9 having the highest assigned score.

Second Embodiment This second embodiment relates to the creation of a database for moving images. FIG. 2 is a block diagram showing a configuration of a non-coded information search method according to the second embodiment.

By performing a division process 22 on a moving image 21 with sound, which is a prominent example of non-coded information,
The moving image 21 is divided for an arbitrary fixed time or for each scene, and a divided moving image 23 is obtained. Next, voice recognition processing 24 is performed on the voice accompanied by the divided moving image 23, and the conversation and the contents of the voice spoken in the divided moving image 23 are recognized. Voice information 25 is obtained by the voice recognition processing 24, and a plurality of keywords 27 are extracted by performing language processing 26 on the voice information 25. These keywords 27 correspond to the divided moving images 23.

Next, a scoring process 28 is performed on the plurality of extracted keywords 27. This scoring process 28
, A keyword with a higher appearance frequency is given a higher score. As a result, a unique score is assigned to each keyword 27.

Next, it is determined by a score determination process 29 whether or not the score assigned to each keyword 27 has reached a reference point, and the keyword 27 having a score equal to or greater than the reference point is assigned to the corresponding divided moving image 23. Or its divided video 2
A moving image database 31 is created by performing an additional process 30 for adding to a portion in which sound is recorded in addition to the portion 3.

In order to classify the moving picture database 31 created as described above, the above-described keyword 27 is used.
Classify according to. When searching the database 31, the searcher inputs words related to the moving image to be searched, and the divided moving image 23 to which the keyword 27 matching the input word or the keyword 27 close thereto is added. read out. In the case of this search, the search is performed sequentially from the keyword 27 having the highest assigned score.

According to the first and second embodiments described above, there are the following advantages.

By completely automating the creation of the database, it is possible to significantly reduce labor.

Since the keyword is automatically added, there is no room for human subjectivity to enter, and a uniform and highly accurate database without variation in keywords can be created.

Since the keyword is automatically added, the speed can be greatly increased as compared with the case of manual operation.

An objective keyword can be added to information that is difficult for a human to understand or judge.

By using a lot of linguistic information existing before and after, a wide range of keywords can be added, and classification and search without omission can be performed.

Third Embodiment This third embodiment relates to a search of a multimedia database. More specifically, this is an effective search method shown in a database that has already been created like a CD-ROM and can be read but cannot be written. C
This is a system for searching a multimedia database recorded in a D-ROM or the like for a part of a specific drawing, image, or moving image. FIG. 3 is a block diagram showing a configuration of a non-coded information search method according to the third embodiment.

The searcher 41 inputs words and sentences related to the information he or she wants as a natural language 42 into the system.
The system automatically performs a plurality of appropriate keywords 44 from the natural language 42 by performing the language processing 43 on the input natural language 42. And
Using the keyword 44, the coded character information 45a is first searched in a multimedia database 45 such as a CD-ROM, and character data having the same, similar, or similar meaning as the keyword 44 is extracted.

Next, the system searches for the image information 45b located in a position close to the position of the extracted character data in the order of the closest. Alternatively, a moving image at a portion temporally closer to the extracted character data is cut out and extracted in a fixed scene. The length of a scene to be cut out can be arbitrarily set according to an instruction of the searcher 41, and can be arbitrarily changed after extraction.
In the extraction of drawings, images, and moving images, a matching process 46 between the extracted one and the keyword 44 is performed, and if the matching is performed to a certain degree or more, an image search process 47 is performed to extract an object to be extracted. Are finally narrowed down, and the extraction target is finally provided to the searcher 41 by the image extraction processing 48.

If the information obtained by the above processing is not satisfactory or if more detailed information is desired, the searcher 41 instructs the system to perform the next step. The system analyzes the image information 45b in the database 45 by the image recognition processing 49, and converts image-like characters into character codes by character recognition. Further, the database 45 is executed by the voice recognition process 50.
The voice information 45c is analyzed and the voice is converted into a character code by voice recognition. Then, based on at least one of the newly generated character codes, the database 45 is searched in the same manner as described above, and the image information 4 located at a position that is close in position or time to related character data is read.
5b and a part of the scene of the moving image are extracted and provided to the searcher 41 through the same procedure as described above.

According to the search method of the third embodiment, the CD
Fast and accurate retrieval of any non-coded information required by the searcher from any database not classified or keyworded, not just ROM.

Fourth Embodiment This fourth embodiment relates to creation and retrieval of an interactive movie. FIG. 4 is a conceptual diagram of an interactive movie database, and FIG. 5 is a diagram showing a keyword image comparison table. The database 51 has an image data recording unit 52 and a keyword image comparison table 53. The image data recording unit 52 has a plurality of image data 52a, 52b,.
… Is recorded. In the keyword image comparison table 53, not only the correspondence between the keyword and the image data number, but also a keyword link that associates with the keyword to be extracted and to be extracted next.

This method is very similar to the second embodiment. First, the voice information recorded simultaneously with the moving image information is analyzed to extract voice parts such as narrations and conversations, and are converted into character codes by voice recognition processing. Next, language processing is performed on the converted character code string to extract keywords. Next, image processing is performed on the moving image information, and a moving image in which character information is recorded on the screen is cut out. Then, for one piece of character information, an image in which the character information is recorded most clearly for each character information is automatically extracted one by one. Next, character recognition processing is performed on the image to recognize characters in the screen and convert them to character codes. Next, language processing is performed on the character code string converted in the same manner as described above to extract a keyword. As described above, a keyword based on audio information and a keyword based on image information are extracted.

An identification signal for searching is recorded for a portion where a keyword is extracted from voices and characters in the moving image information, and the keyword and the identification signal (before or after the moving image information in the database 51). Image data number)
And a keyword image comparison table 53 composed of a combination of

When an interactive movie is created, one or a plurality of pieces of moving image information created in this way are combined and created. The user inputs the story or scene that he / she wants to see at that time in a natural language. Then, the system performs linguistic processing to extract keywords,
A keyword having the same, similar, or similar meaning as the keyword is selected from the keyword image comparison table 53. As a result, the moving image information closest to the content desired by the user is extracted, and the reproduction of the moving image information from the head or the necessary part is started.

As described above, the user can interactively extract the moving image information of the content he wants to see. Further, by inputting a plurality of keywords sequentially by the user, a movie of an arbitrary story can be created by connecting a plurality of moving images.

Further, by making various associations called keyword links among a plurality of keywords extracted from the moving image information, it is possible to create a complicated interactive movie.

[0035]

According to the present invention , since the non-coded information is automatically converted into a database, it is possible to largely reduce labor as compared with the conventional method which generally relies on artificial work. Yes, and high-precision search became possible.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a database system for non-coded information according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a database system for non-coded information according to a second embodiment of the present invention.

FIG. 3 is a block diagram showing a configuration of an uncoded information search method according to a third embodiment of the present invention.

FIG. 4 is a conceptual diagram of an interactive movie database according to a fourth embodiment of the present invention.

FIG. 5 is a diagram showing a keyword image comparison table in a fourth embodiment.

[Explanation of symbols]

1 ... Documents containing figures, 2 ... Scanners, 3 ... Image processing, 4 ...
Sentence data part, 5 ... Figure data part, 6 ... Character recognition processing, 7 ... Character information, 8 ... Language processing, 9 ... Keyword, 1
0: scoring processing, 11: score discrimination processing, 12: addition processing, 13: database of figures, 21: moving image, 22: division processing, 23: divided moving image, 24: voice recognition processing, 25
... voice information, 26 ... language processing, 27 ... keywords, 28
... Scoring processing, 29 ... Scoring determination processing, 30 ... Addition processing, 31 ... Video database, 41 ... Searcher, 42
... natural language, 43 ... language processing, 44 ... keywords, 45
... Multimedia database, 45a ... Character information, 4
5b image information, 45c audio information, 46 matching processing, 47 image search processing, 48 image extraction processing, 4
9: Image recognition processing, 50: Voice recognition processing, 51: Interactive movie database, 52: Image data recording unit, 53: Keyword image comparison table

Claims (1)

(57) [Claims] 1. A method for reading a sentence including non-coded information,
Reading means for converting the image into a child data ,
Encoding and management means, the characters in the periphery and in the non-coded information part
Character recognition processing means, and a plurality of keyword candidates from the coded character information.
Language processing means for extracting keyword information and character information corresponding to each keyword candidate
Nearer to the non-coded information part
Scoring processing means that gives a higher score as the score is higher
And a score determination process for determining whether the score has reached a reference point.
And management means, when the number in the point number judgment processing means has reached the reference point determine
Using the separated keyword candidate as a keyword,
Database apparatus of a non-coded information, characterized by comprising the additional processing means for adding the de-information portion.