JP2000076287A - Image search method, image search device, and recording medium storing image search program - Google Patents

Abstract

(57) [Summary] [Problem] Define a semantic hierarchical relationship between keywords,
By centrally managing the feature vector types and weights corresponding to keywords, it is easy to add and delete keywords, and to add databases.Furthermore, it eliminates differences in search conditions in each database management system, and uses common keywords Search a plurality of image databases with different search conditions. A query sentence is analyzed to extract a keyword (S4), a search condition including a feature vector type of an image for the extracted keyword and its weight is determined (S5), and the search condition is set as an access target image. Convert to search conditions that can access the database (S6, S
7) A search command corresponding to each image database is generated based on the converted search conditions (S8), and each image database is accessed to search for image data (S9). By making the search condition of the upper keyword inherited by the lower keyword, multiple management of the feature vector type and its weight is not required.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image search method, an image search apparatus, and a recording medium storing an image search program for searching a plurality of different image databases based on a keyword. Extract feature vectors and weights of images to be searched based on keywords specified by keyword search by input or category, and convert the extracted feature vectors and weights to feature vectors and weights corresponding to each image database management system By acquiring key image names and feature vector values from keywords, and dynamically generating and searching for a search command corresponding to each image database from the obtained information, a desired image can be obtained from a plurality of image databases. Image search method and images that enable integrated search of images It relates a recording medium recording a search device and an image search program.

[0002]

2. Description of the Related Art In recent years, due to the sophistication and price reduction of personal computers and the spread of open networks such as the Internet, intranets and extranets, the environment in which image data can be easily browsed and processed by end users has rapidly increased. It is getting ready. In parallel with this,
In order to efficiently manage the stored image data, research has been actively conducted to centrally manage the image data and to search for similar image data using image data and feature vector values as search keys. ing.

In fact, the top vendors in relational (relational) database management systems have shifted their targets to object relational databases and have begun providing cartridges that can easily handle image data as cartridges.

[0004] On the other hand, most user interfaces found in Internet search engines are mainly searched by keywords. The conventional image data search method using a keyword is a method in which character or numerical information serving as a keyword is added to image data in advance, and image data is searched based on this information.

[0005] The present applicant discloses in Japanese Patent Application Laid-Open No. Hei 10-91503 an information resource for collecting meta information held by a database management system in each database system and meta information from a unique information resource management tool and managing the meta information. An information resource management control method and system that can be automatically collected and managed in a management dictionary is proposed. According to the information resource management control method and system, basic management information for collecting information of a database system connected to a network is registered in advance in an information resource management dictionary for managing meta information of the entire database system. In addition, meta information of the individual database system or meta information is collected from the individual information resource tool at a timing designated by the basic information for information collection, and the collected meta information is reflected in the information resource management dictionary.

The applicant of the present invention has proposed, in Japanese Patent Application Laid-Open No. Hei 10-143439, an information search method and an information search system in which information can be easily searched for a plurality of different database systems connected to a network. . This information search method and information search system store in advance reference information for accessing a plurality of database systems, and when an information search request is obtained, analyze the information search request and refer to the reference information. In addition to specifying in which database system the information corresponding to the search request is stored, information on how the information can be obtained is obtained, and the storage position of the information corresponding to the information search request and An information search command is generated based on the information indicating the acquisition method, a search request is made to a database system corresponding to the information search request, and a search result is provided to a user.

The present applicant discloses in Japanese Patent Application Laid-Open No. Hei 10-187742 that each database can be freely searched in its own expression format even if the data item expressions in each database are different despite the same meaning. A database heterogeneity resolving and retrieving apparatus is proposed. This database heterogeneity resolving search device searches a local domain management table for a database storing search data of an inquiry sentence from a user and a local domain corresponding to the search data, and searches for a domain group including the local domain. A global domain is searched from the domain management table, and if the searched local domain is different from the global domain, a conversion function for converting the local domain into the global domain is searched from the conversion function management table, and each conversion function is searched using this conversion function. The query is translated into a query sentence suitable for the database, and each database is searched using the translated query.

[0008]

However, in a conventional database management system, a search interface for similar image data using image data and feature vector values as keys is used to specify image data, feature vector types, weighting of feature vector types, and the like. At present, the method is not standardized and each database management system provides its own interface. In this case, in order to search for integrated similar image data in an environment where a plurality of image databases exist, it is necessary to describe a search program unique to each system.

Further, in the conventional image data search method using a keyword, it is necessary to assign character and numerical information as a keyword to image data in advance and manage the image data. For this reason, there is a problem that it becomes difficult to add or delete a keyword or to add a database.

Further, in the conventional image data search method using keywords, there is a problem that search cannot be performed if a keyword that matches the keyword input by the searcher is not registered. For this reason, when a keyword equivalent to the keyword input by the searcher is registered, a function of replacing the keyword input by the searcher with the registered keyword (automatic unification of search terms) is desired. Was. Also, for the entered keyword,
There has been a demand for a function capable of selecting and specifying a more appropriate keyword by presenting a keyword of a higher concept or a keyword of a lower concept.

SUMMARY OF THE INVENTION The present invention has been made to solve such a problem, and aims at image data retrieval based on keywords, defines a meaningful hierarchical relationship between keywords, and unites the feature vector type and weighting of keywords. Management eliminates the problem of adding and deleting keywords and adding databases, and the differences in the management methods of image data, feature vector types, and feature vector type weights in each database management system. Image search method and image search that can solve the problem of having to write a search program unique to each system by providing a mechanism to solve the problem and dynamically generating a search statement for each image database To provide a recording medium in which an apparatus and an image retrieval program are recorded To.

[0012]

In order to solve the above-mentioned problems, an image search technique according to claim 1, 11 or 21, wherein a feature vector type and a feature vector type of an image to be searched based on a keyword for image search. The search conditions are determined by weighting the search conditions, the search conditions are converted into search conditions that can access the image database to be accessed, and the image database is accessed using the converted search conditions. As a result, an image search can be performed for a plurality of image databases having different search conditions using a common keyword.

In the image retrieval technique according to the second, twelfth, and twenty-second aspects, a vertical relationship is set in advance between keywords for image retrieval, and a feature vector type and a weight of the feature vector type are registered for the keyword. When there is no keyword, the search condition is set using the feature vector type registered for the higher keyword of the keyword and the weighting of the feature vector type. By providing a function for inheriting the feature vector type and the weight registered for the high-order keyword to the low-order keyword, it is not necessary to manage the feature vector type and the weight in multiples.

In the image search technique according to the third, thirteenth, and twenty-third aspects, a vertical relationship is set in advance between keywords for image search, and a feature vector type and a weight of the feature vector type are registered for the keyword. If not, the search condition is set using the feature vector type registered for the higher keyword of the keyword and the weight of the feature vector type, and the search condition can be accessed to the image database to be accessed. It is characterized in that it is converted into a search condition and an image database is accessed with the converted search condition. As a result, an image search can be performed for a plurality of image databases having different search conditions using a common keyword. Further, by providing a function of allowing the lower-level keywords to inherit the feature vector type and weight registered for the higher-level keyword, it is not necessary to manage the feature vector type and weight more than once.

According to a fourth aspect of the present invention, there is provided an image search technique in which words that are the same as or similar to a search keyword are associated with the search keyword, and a search is performed based on the associated keyword. It is characterized in that image search conditions are determined. Even when the searcher does not know the search keyword, the image data can be searched by inputting a term having the same meaning or a similar term as the search keyword. Therefore, the degree of freedom of keyword specification is expanded.

In the image search technique according to the fifth, fifteenth, and twenty-fifth aspects, a hierarchical relationship is set in advance between keywords for image search, and a higher rank of a keyword input by a searcher and / or And a lower keyword is presented to a searcher so that a keyword for image search can be selected. Since the upper and lower keywords are presented for the input keyword, a more appropriate keyword can be designated by referring to the upper and lower keywords. Therefore, it is possible to efficiently and effectively perform an image search using a keyword.

An image search technique according to claim 6, 16 or 26, comprising: a keyword feature management table for managing a feature vector type and a weight for each keyword in a conceptual keyword tree in which a conceptual hierarchical relationship is applied to the keyword; A feature conversion function table that manages a conversion function that converts a feature vector type and a weight of the feature vector type associated with the feature vector type into a feature vector type and a weight of the feature vector type for each database management system; and an image database for each keyword. A key conversion function table that manages a conversion function that returns a key image name and a feature vector value, and a keyword feature management table that applies an inheritance relationship to the definition of the feature vector type and the weight of the feature vector type for each keyword in the concept keyword tree. Using the keyword A keyword feature extraction unit that extracts weights of vectors and feature vectors, and a feature that converts the feature vector types and weights of keywords into feature vector types and weights for each database management system using a conversion function that manages the feature conversion table. It is characterized by comprising conversion means and key conversion means for converting a keyword into a key image name or a feature vector value of each image database using a key conversion function managed by a key conversion function table. Thereby, the feature vector types and weights of the keywords and the feature vector types and the weights of the feature vector types for each database management system are integrally managed, and the feature vector types and the weights of the keywords are compared with the feature vector types for each database management system. It can be converted into a feature vector type weight.

The keyword feature extracting means applies the inheritance relationship to the definition of the feature vector type and the weight for each keyword in the concept keyword tree, and extracts the keyword feature vector and the weight using the keyword feature management table. This eliminates the need to manage multiple feature vectors and weights for keywords.

The feature conversion means uses a feature conversion function managed by a feature conversion table, and uses the feature vector types and weights of the keywords extracted by the keyword feature extraction means to search each database management system for image data. It is converted into a feature vector type and weighting, respectively. This makes it possible to map the feature vector types and weights of the keywords managed in the keyword feature management table to the feature vector types and weights used for searching image data in each database management system. .

The key conversion means converts a keyword into a key image name and a characteristic vector value for each image database using a key conversion function managed by a key function table managed by a key function table. As a result, a key image name and a feature vector value serving as a search key can be obtained.

The image search technique according to claims 7, 17 and 27 is the image search technique according to claims 6, 16 and 26, wherein the terms managed in the keyword feature management table are synonymous with the terms of synonyms. A keyword alias management table for managing as a keyword, a searcher searches the keyword alias management table for a keyword on the concept keyword tree for a keyword specified as a search request, and searches the keyword on the concept keyword tree obtained by the search. It is characterized by being used. As a result, it is possible to set a search condition for image data even if a keyword is used as a synonymous term.

The image retrieval technique according to claim 8, 18, 28 is a local database management table for managing a data management system type and a data structure of each image database, a feature vector type and weighting for each database management system, and a key image. A search statement generation function table that manages a function for dynamically generating a search statement for each image database from a database management system type and a data structure managed by the name and feature vector value and the local database management table; A local database information acquisition unit for acquiring a database management system type and an image database name from a database management table; and a feature for converting a keyword feature vector type and weight into a feature vector type and weight for each database management system. Conversion means, key conversion means for converting a keyword into a key image name or a feature vector value of each image database, and a search statement generation function managed by a search statement generation function table, obtained by the feature conversion means. Search for each image database based on the keyword feature vector type and weight, the key image name or feature vector value obtained by the key conversion means, and the database management system type and data structure obtained by the local database information obtaining means A search statement generation means for dynamically generating a statement; and image data access means for searching image data from each image database based on the search statement generated by the search statement generation means. And Thus, image data can be integratedly searched from a plurality of image databases having different search conditions by specifying a keyword.

The local database information obtaining means obtains a database management system type for managing each image database and each image database name from the local database management table. Thus, mapping of each function for converting the feature vector type and the weight into the feature vector type and the weight used for managing the image in each database management system by the feature conversion unit, and the key image for each image database from the keyword by the key conversion unit Mapping of each function to be converted into a name or a feature vector value becomes possible.

The search statement generation means uses the search statement generation function managed in the search statement generation function table and obtains the feature vector type and weight of the keyword obtained by the characteristic conversion means and the weight obtained by the key conversion means. Using the key image name and feature vector value, and the database management system type and data structure obtained by the local database information obtaining means, a search command sentence for each image database is generated.
This makes it possible to generate a search command sentence for each database management system from the keyword feature vector type and weight, the key image name and the feature vector value.

The image database access means retrieves the target image data from each image database according to the search command sentence generated by the search command sentence generation means. This makes it possible to provide an integrated image data search by specifying a keyword.

The image retrieval technique according to the ninth, nineteenth, and twenty-ninth aspects provides a feature conversion function table for managing a conversion function for converting a feature vector type and a weight of a keyword into a feature vector type and a weight for each database management system.
Database management using a key conversion function table that manages a conversion function that returns a key image name and a feature vector value of each image database for each keyword, and a conversion function that manages the feature vector type and weight of the keyword in the feature conversion function table Feature conversion means for converting the feature vector type and weight for each system, and key conversion means for converting a keyword into a key image name and a feature vector value of each image database using a key conversion function managed in a key conversion function table. Inherited in the keyword feature management table that manages the feature vector value type and weight for each keyword for a conceptual keyword tree that applies a conceptual hierarchical relationship to keywords, and the definition of feature vector type and weight for each keyword in the concept keyword tree Apply relationship and kiwa Keyword feature extraction means for extracting a feature vector type and weight of a keyword using a feature management table; a local database management table for managing a database management system type and a data structure of each image database; and features for each database management system Manages a function that dynamically generates a search statement for each image database based on the vector type and weight, the key image name and feature vector value, and the database management system type and data structure managed in the local database management table. Search statement generation function table, local database information acquisition means for acquiring the database management system type and image database name from the local database management table, and search statement generation function managed by the search statement generation function table Using the feature vector type and weighting of the keyword obtained by the feature converting means, the key image name and feature vector value obtained by the key converting means, and the database management system type and data structure obtained by the local database information obtaining means. Search statement generation means for dynamically generating a search statement for each image database based on the image database, and image database access means for searching image data from each image database based on the search statement generated by the search statement generation means And characterized in that: Thereby, the feature vector type and the weight of the keyword are converted into the feature vector type and the weight for each database management system, the search command sentence for each image database is dynamically generated, and the image data is searched from each image database. Can be. Further, since a function is provided that allows the lower-level keywords to inherit the feature vector types and weights registered for the higher-order keywords, it is not necessary to manage the feature vector types and the weights more than once.

The image retrieval technique according to the tenth, twentieth, and thirty aspects of the present invention is the image retrieval technique according to the ninth, nineteenth, and twenty-ninth aspects, wherein the keyword managed in the keyword feature management table is synonymous with the term of a synonym. A keyword alias management table for managing as a keyword is provided, and a keyword on the concept keyword tree for the keyword specified by the searcher as a search request is searched from the keyword alias management table, and the keyword on the concept keyword tree obtained by the search is used. It is characterized by doing. This makes it possible to set a search condition for image data even by using a keyword and a synonymous term, and to search for image data in an integrated manner from a plurality of image databases having different search conditions.

[0028]

Embodiments of the present invention will be described below with reference to the accompanying drawings. FIG. 1 is a block configuration diagram of an image search device according to the present invention.

In FIG. 1, 1 is an image retrieval device, 2 is an application program, and 3A to 3N are image database devices. Each image database device 3A to 3N
Consists of database management units 4A to 4N and image databases 5A to 5N. Each image database device 3A ~
3N may be configured to be directly connected to the image search device 1 or may be configured to be indirectly connected to the image search device 1 via a network.

In the present embodiment, each of the image database devices 3A to 3N has a different data structure and a different search method for accessing image data. Note that not all of the search methods for the image database devices 3A to 3N are different, and some image database devices may be accessible by a common search method.

Upon receiving a search request input from a user (searcher) using the application program 2, the image search device 1 accesses each of the image database devices 3A to 3N based on the search request, and performs a search. The result is supplied to the application program 2.

The image search device 1 comprises a syntax analysis unit 10, a keyword conversion unit 20, a keyword search unit 30, and a database access unit 40. The keyword conversion unit 20 includes a local database management unit 21, a keyword feature management unit 22, a keyword feature extraction unit 23, a feature conversion function unit 24, a feature conversion unit 25, a key conversion function unit 26, and a key conversion unit. 27. The keyword search unit 30 includes a search command control unit 31, a search command sentence generation function unit 32, a search command sentence generation unit 33, and a search result processing unit 3.
4 is provided.

FIG. 2 is a flowchart showing the overall operation of the image search apparatus according to the present invention, and FIG. 3 is an explanatory diagram showing a specific example of a conceptual keyword tree. 4 is an explanatory diagram showing a specific example of the contents of the keyword feature management table, FIG. 5 is an explanatory diagram showing a specific example of the contents of the keyword alias management table, and FIG. 6 is a specific example of the contents of the feature conversion function table. FIG. 7 is an explanatory diagram showing a specific example of the contents of a key conversion function table. FIG. 8 is an explanatory diagram showing a specific example of the contents of a local database management table. It is an explanatory view showing a specific example of contents.

As shown in FIG. 2, the image search device 1 has two stages, a preparation phase (steps S1 to S3) for defining a concept keyword tree and the like before performing a search, and a search phase for performing an actual search. Perform processing.

First, the preparation phase will be described. In the preparation phase, the concept keyword tree shown in FIG. 3 is defined in step S1, the conversion and the creation of a search command sentence generation function are performed in step S2, and the data is added to the various tables shown in FIGS. 4 to 8 in step S3. Perform

Here, it is assumed that the system administrator performs the conversion, the search command generation function, and the addition of data to various data tables.

In the concept keyword tree (step S1), a meaningful hierarchical relationship is defined for keywords, a keyword tree is defined as tree structure data, and an inheritance function is applied to the definition of the feature vector type and weight for each keyword. are doing. Therefore, the feature vector type and weight of the upper keyword are carried over to the feature vector type and weight of the lower keyword. Also, when the feature vector type and weight to be inherited are defined in the feature vector type and weight of the lower keyword, the feature is not inherited.

For example, as shown in FIG. 4, the keyword “soccer ball” has upper keywords “/ sports goods” and “/ sports goods / ball” as upper keywords, and the feature vector type and weighting of “soccer ball” Is the top keyword "/ sports goods / ball: feature vector type = layout, weight = 20
% "And determined only by the keyword" soccer ball "" feature vector type = pattern, weight =
According to 60%, feature vector type = color, weight = 30%, the feature vector type and weight of the keyword “soccer ball” are “feature vector type = layout, weight = 20%, feature vector type = pattern, weight =
60%, feature vector type = color, weight = 30% ”.

For the keyword “tennis ball”, “feature vector type = layout, weight = 10
%, Feature vector type = pattern, weight = 10%, feature vector type = color, weight = 70% ", are defined for the top keywords“ / sports equipment ”and“ / sports equipment / ball ”. Since the same feature vector type and the same weight as the feature vector type and the weight are defined, the feature vector type and the weight are not inherited.

In step S2, a feature conversion function is created,
Creates a key conversion function and a search statement generation function. In the creation of the feature conversion function, as shown in FIG. 6, a function for converting the feature vector type and the weight of the keyword into the feature vector type and the weight for each database management system to be searched is created. For example, as a feature conversion function, a function “vector_color” that converts a feature vector type “color” into a feature vector type of a database management system with a DBMS name “DBMS1”
1 () ”and its weight conversion“ weight_co ”
lor1 () "is created.

In the creation of the key conversion function, for example, a function for returning a key image name or a feature vector value symbolized by the keyword "soccer ball" is created for each image database. As shown in FIG. 7, for example, as a key conversion function, a keyword “soccer ball” is assigned to a key image “/ hostname /
A-DB / image / soccer_ball. jp
Create a function called "A-key ()" that returns "g".

In the creation of the search statement generating function, a function for generating a search statement for each image database is created from the information for each database management system, the feature vector type and weight of the keyword, the key image name and the feature vector value. As shown in FIG. 8, for example, as the database management system information, the DBMS name is “DBMS1” and the image D
B name is “A-DB” and image table name is “A1-TAB”
LE ”, and further, by inputting a feature vector type, a weight, and a key image name, as shown in FIG.
"SQL_DBM" to generate a search statement for "S1"
S1 () "is created.

When each function is created in step S2, data is added to each management table in step S3. The keyword feature management unit 22 shown in FIG.
5 shows the keyword feature management table 51 shown in FIG.
And the keyword alias management table 52 shown in FIG.

As shown in FIG. 4, in the keyword feature management table 51, data of the concept keyword tree is managed as a keyword identifier (keyword ID), a keyword (keyword name), an upper keyword, a layout, a pattern, a color, and the like. For example, the keyword feature management table 5
1, the keyword ID as "soccer ball" is 3,
The keyword (keyword name) is “soccer ball”, the top keyword is “/ sports equipment / ball”, and the feature vector type and weight are “feature vector type = layout, weight = Null, feature vector type = pattern, weight = 60% , And feature vector type = color, weight = 30% ”.

As shown in FIG. 5, in the keyword alias management table 52, a keyword identifier (keyword I
D) Manage alias keywords. For example, in the keyword alias management table 52, “soccer” is set as an alias keyword of “soccer ball” having a keyword ID of 3.
"ball", "SOCCER BALL", and "sackball" are added.

The feature conversion function unit 24 shown in FIG.
The feature conversion function table 53 shown in FIG. In the feature conversion function table 53, the feature vector type, DBMS
Manage names, feature conversion functions, and weighting conversion functions. For example, in the feature conversion function table 53, the DBMS name is "DB
“MS1” is a function for converting to the feature vector type and a function for converting the weight to the feature vector type is: feature vector type = color feature vector type conversion function vector_color
1 () Weight conversion function weight_color1 () Feature vector type = pattern Feature vector type conversion function vector_textu
re1 () weighting conversion function weight_texture
1 () Feature vector type = layout Feature vector type conversion function vector_struct
cure1 () weighting conversion function weight_structure
Add as 1 ().

The key conversion function unit 26 shown in FIG.
The key conversion function table 54 shown in FIG. The key conversion function table 54 includes an image DB name, a key image conversion function,
Manages feature vector value conversion function names. For example, in the key conversion function table 54, the image database is "AD
The key image conversion function name for “B” is “A-key ()”.
Add as

The local database management section 21 shown in FIG. 1 manages the local database management table 55 shown in FIG. Local database management table 5
Reference numeral 5 manages a DBMS name, an image DB name, and an image table name. For example, the local database management table 55
The image DB name managed by the database management system “DBMS1” is “A-DB”, and the table name managed by “A-DB” is “A1-TABLE, A2-TAB”.
LE, A3-TABLE. "

The search statement generating function 32 shown in FIG.
Manages the search statement generation function table 56 shown in FIG. The search statement generation function table 56 is a DBMS
Manages name and search statement generation functions. For example, "DBM
The search statement generation function managed in “S1” is set to “SQL_DB”.
MS1 () ".

Next, the search phase will be described. In the search phase, the query sentence is analyzed in step S4 shown in FIG. 2 to extract a keyword, and in step S5, the feature vector type and weight for the extracted keyword are extracted. Then, the feature vector types and weights extracted in step S6 are converted into feature vector types and weights for each database management system, and in step S7, they are converted into key images and feature vector values for each image database. Next, in step S8, a search statement for each database management system is generated, and in step S9, an image database is searched using the search statement.

The search method is to use the keyword alias management table to unify keywords and to search for image data based on unified keywords, and to determine keywords by tracing keywords in the concept keyword tree up and down. There is a method of searching for image data based on the keyword. First, the former method will be described.

The extraction of the keyword is performed by the syntax analyzer 10 shown in FIG. When receiving the user's search request from the application program 2, the syntax analysis unit 10 analyzes the search request, extracts a keyword, and supplies the extracted keyword to the keyword feature extraction unit 23. When receiving the search request “keyword = soccer ball” from the user, the syntax analysis unit 10 extracts the keyword “soccer ball” and passes it to the keyword feature extraction unit 23.

The keyword feature extraction unit 23 stores the keywords extracted by the syntax analysis unit 10 in the keyword feature management unit 22.
The keyword ID is extracted by searching from the keyword alias table 52 managed by. Then, the keyword ID
From the keyword feature management table 51 to extract the feature vector type and weighting of the designated keyword. Then, the keyword feature extraction unit 23 supplies the extracted feature vector type and weight to the feature conversion unit 25 and supplies the keyword to the key conversion unit 27.

For example, the keyword “soccer ball” extracted by the syntax analysis unit 10 is stored in the keyword alias table 5.
2 and extract the keyword ID “3”. Further, the keyword ID “3” is searched from the keyword feature management table 51, and “keyword = soccer ball, upper keyword = / sporting goods / ball, layout =
Null, pattern = 60%, color = 30%. " “Soccer ball” inherits the feature vector types and weights of the top keywords “/ sports goods” and “/ sports goods / ball”, and finally assigns the keyword “soccer ball” and its feature vector types and weights. "Feature vector type = layout, weight = 20"
%, Feature vector type = pattern, weight = 60%, feature vector type = color, weight = 30% ”. These feature vector types and weights are passed to the feature converter 25, and the keywords are passed to the key converter 27.

The feature conversion unit 25 manages the feature vector type and the weight received from the keyword feature extraction unit 23 in the feature conversion table 5 managed by the feature conversion unit 25, respectively.
Using the feature conversion function and the weighting conversion function of No. 3, the data is converted into the feature vector type and the weight for each database management system, and the search result generation unit 33 is provided with the feature vector type and the weight of the conversion result.

For example, for the keyword “soccer ball” received from the keyword feature extraction unit 23, “feature vector type = layout, weight = 20%, feature vector type = pattern, weight = 60%, feature vector type =
“Color, weight = 30%” indicates that the conversion of the DBMS name to DBMS1 is performed for the database management system. Type = pattern Function name: vector_texture1 () Feature vector type = texture Function name: weight_texture1 () Weight = 0.6 Feature vector type = color Function name: vector_color1 () Feature vector type = globalcolor Function name: weight1_weight_weight (weight1_weight)
Perform as in 0.3.

The key conversion unit 27 extracts the image database name for each database management system from the local database management table 55, and extracts the keyword feature extraction unit 23
Is converted into a key image and a feature vector value for each image database using a key image conversion function and a feature vector value conversion function of a key conversion function table 54 managed by the key conversion unit 27, and a search command The key image and the characteristic vector value resulting from the conversion are passed to the sentence generation unit 33.

For example, the image database name of each database management system is extracted from the local database management table 55, and the keyword “soccer ball” received from the keyword feature extraction unit 23 is changed to the DBMS name “D”.
"A-DB" of the database management system of "BMS1"
Key image conversion function using “A-key ()” for the key word feature extraction unit 2
3 with the key image “/ h”
ostname / A-DB / image / soccer
_Ball. jpg ”, and the search statement generation unit 33
Pass the result to.

The search command section generation section 33 includes information on the database management system in the local database management table 55, the feature vector type and weight for each database management system converted by the feature conversion section 5, and the key conversion section 27.
Receiving the key image and the characteristic vector value of each image database converted by the above, and using the search statement generation function of the search statement generation function table 56 managed by the search statement generation function unit 32, for each image database. A search statement is generated, and the search statement is passed to the database access unit 40.

For example, the search statement generation unit 33 manages the DBM managed by the local database management table 55.
"A1-TAB" of "A-DB" whose S name is "DBMS1"
For the table “LE”, the following retrieval statement is generated from the feature vector and the weight of the keyword “soccer ball” received from the feature conversion unit 25 and the key image received from the key conversion unit 27. Weight =
0.2 ・ Feature vector type = texture, weight = 0.
6 • Feature vector type = globalcolor, weight = 0.3 • Search statement selectMatch (a.asset, b.as
set, 0.3, 0, 0.6, 0.2), a. path
namefrom Al-TABLE a, Al-T
Able bwhere a. pathname = '/
hostname / A-DB / image / source
r ball. jpg 'order by 1;
Pass to

The database access unit 40 receives each image database search command sentence from the search command sentence generation unit 33 and executes a search for each image database. Then, the database access unit 40 supplies the search result to the application program 2 via the search result processing unit 34. The delivery of the search command sentence and the delivery of the search result are controlled by the search command control unit 31.

For example, a search command sentence is received from the search command sentence generation unit 33, and each image database is searched, so that the DBMS1.
set, 0.3, 0, 0.6, 0.2), a. path
namefrom Al-TABLE a, Al-T
Able bwhere a. pathname = '/
hostname / A-DB / image / source
r ball. jpg 'order by 1; ・ Search result / hostname / A-DB / image / socc
er1_ball. jpg / hostname / A-DB / image / soc
er2_ball. jpg DBMS2 ・ Search statement select T0. parentfilename
from p14a. object T0where
T0 in similar (RESULT.soc
er / soccer. C00003. L255. A23
73. Xs5. Ys4. Xe57. Y356, p14a
C, 0.3, L1, p14ac, 0.6, L1, p14
aA, 0.2, L1) ・ Search result / hostname / C-DB / image / socc
er3_ball. jpg / hostname / C-DB / image / socc
er4_ball. jpg A search result as exemplified above is obtained from each image database.

Next, a method of determining a keyword by tracing the keyword up and down using the concept keyword tree and performing a search will be described. Note that the processing after the keyword determination is the same as the processing in the search phase described above. As an inquiry from the user, the keyword search provides two types of operations of acquiring a keyword immediately above a designated keyword and acquiring all keywords immediately below a designated keyword.

In obtaining the keyword immediately above the designated keyword, this operation is received from the keyword feature management table 51 of the keyword feature management unit 22, and a keyword higher than the keyword is searched in the concept keyword tree of the keyword feature management table 51. Obtain and return the top keyword to the user as a result of this operation.

In acquiring the keyword immediately below the designated keyword, this operation is received from the keyword feature management table 51 of the keyword feature management unit 22, and all lower keywords of the keyword are searched in the concept keyword tree of the keyword feature management table 51. The lower keyword is returned to the user as a result of this operation.

The user obtains a target keyword by interactively following the concept keyword tree up and down by combining the two types of operations.

The feature management table has an abstract weight value. This is a numerical value that does not depend on the DBMS, and indicates the weight of the word (keyword) itself. For example, for the keyword "apple"
As imagined as “red and round fruit”, the search is performed at 70% using the degree of the color (red) as the feature vector type as a weight. In addition, the inheritance inherits the feature vector type and the weight every time a keyword is narrowed down. For example, a ball can only be determined to have a round shape, but a tennis ball has a color feature vector type added thereto. By setting this to inheritance, the setting operation of the feature type and the weight can be reduced.

[0068]

According to the present invention, as described above,
The image search technology according to the first aspect determines a search condition of an image to be searched based on a keyword, converts the search condition into a search condition that can access an image database to be accessed, and accesses the image database with the converted search condition. Therefore, an image search can be performed for a plurality of image databases having different search conditions using a common keyword.

In the image search technique according to the second, twelfth, and twenty-second aspects, a vertical relationship is set in advance between keywords for image search, and the feature vector types and the weights of the feature vector types are not registered for the keywords. In this case, the feature vector types registered for the higher-order keywords of the keywords and the weights of the feature vector types are inherited, so that it is not necessary to manage the feature vector types and the weights in a multiplexed manner. Therefore, construction and maintenance of the image search system are facilitated.

The image search technique according to the third, thirteenth, and twenty-third sets a search condition including a feature vector type and a weight of the feature vector type by using a hierarchical relationship between keywords, and provides an image database to be accessed. Since the search condition is converted and accessed in accordance with the above, image search can be performed using a common keyword for a plurality of image databases having different search conditions. Further, by providing a function of allowing the lower-level keywords to inherit the feature vector type and weight registered for the higher-level keyword, it is not necessary to manage the feature vector type and weight more than once.

In the image search technique according to claims 4, 14, and 24, a word that is the same or similar to a search keyword is associated with the search keyword, and a search target is set based on the associated keyword. Since image search conditions can be determined, even when a searcher does not know a search keyword, image data can be searched by entering a term having the same meaning or a similar term as the search keyword. . Therefore, the degree of freedom of keyword specification is expanded.

In the image search technique according to the fifth, fifteenth, and twenty-fifth aspects, a hierarchical relationship is set in advance between keywords for image search, and a higher rank of a keyword input by a searcher and / or Since the lower and lower keywords are presented to the searcher, more appropriate keywords can be specified by referring to the upper and lower keywords. Therefore, it is possible to efficiently and effectively perform an image search using a keyword.

The image retrieval technique according to claims 6, 16 and 26 integrally manages the feature vector type and weighting of the keyword, the feature vector and weighting for each database management system, and determines the feature vector type and weighting of the keyword. It can be converted into a feature vector and a weight for each database management system.

The image search technique according to claims 7, 17 and 27 is the same as the image search technique according to claims 6, 16 and 26, except that the terms managed by the keyword feature management table are synonymous with the terms of synonyms. A keyword alias management table for managing as a keyword, a searcher searches the keyword alias management table for a keyword on the concept keyword tree for a keyword specified as a search request, and searches the keyword on the concept keyword tree obtained by the search. Since it is used, it is possible to set a search condition of image data even by using a term synonymous with a keyword.

According to the image search technique according to claims 8, 18, and 28, by specifying a keyword, image data can be integratedly searched from a plurality of image databases having different search conditions.

The image retrieval technique according to the ninth, nineteenth, and twenty-ninth aspects converts a feature vector type and weight of a keyword into a feature vector type and weight for each database management system, and dynamically converts a search command sentence for each image database. And image data can be retrieved from each image database. Further, since a function is provided that allows the lower-level keywords to inherit the feature vector types and weights registered for the higher-order keywords, it is not necessary to manage the feature vector types and the weights more than once.

The image retrieval technology according to claims 10, 20, and 30 has a keyword alias management table in the image retrieval technology according to claims 9, 19, and 29. Can set image data search conditions, and can integrally search for image data from a plurality of image databases having different search conditions.

As described above, according to the present invention, at the time of image search using a keyword, a semantic vertical relationship between keywords is defined, and the feature vector type and weighting of the keyword are managed in a unified manner. Addition, deletion, and addition of databases are easy. In addition, the present invention eliminates differences in the management method of image data and feature vectors in each database management system, and dynamically generates a search statement corresponding to each database management system. It is not necessary to describe a unique search program for each image, and a conceptual image data search can be easily performed in a plurality of image database environments.

[Brief description of the drawings]

FIG. 1 is a block diagram of an image search device according to the present invention.

FIG. 2 is a flowchart showing an overall operation of the image search device according to the present invention.

FIG. 3 is an explanatory diagram showing a specific example of a concept keyword tree.

FIG. 4 is an explanatory diagram showing a specific example of the contents of a keyword feature management table.

FIG. 5 is an explanatory diagram showing a specific example of the contents of a keyword alias management table.

FIG. 6 is an explanatory diagram showing a specific example of the contents of a feature conversion function table.

FIG. 7 is an explanatory diagram showing a specific example of the contents of a key conversion function table.

FIG. 8 is an explanatory diagram showing a specific example of the contents of a local database management table.

FIG. 9 is an explanatory diagram showing a specific example of the contents of a search statement generation function table.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 image retrieval device 2 application program 3A, 3N image database device 4A, 4N database management unit 5A, 5N image database 10 syntax analysis unit 20 keyword conversion unit 21 local database management unit 22 keyword feature management unit 23 keyword feature extraction unit 24 feature conversion Function unit 25 Feature conversion unit 26 Key conversion function unit 27 Key conversion unit 30 Keyword conversion unit 31 Search command control unit 32 Search command sentence generation function unit 33 Search command sentence generation unit 34 Search result processing unit 40 Database access unit 51 Keyword feature management Table 52 Keyword alias management table 53 Feature conversion function table 54 Key conversion function table 55 Local database management table 56 Search statement generation function table

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Hiroki Machihara 3-19-2 Nishi-Shinjuku, Shinjuku-ku, Tokyo Inside Japan Telegraph and Telephone Corporation (72) Inventor Takashi Hoshino 3--19, Nishishinjuku, Shinjuku-ku, Tokyo No. 2 Nippon Telegraph and Telephone Corporation

Claims (30)

[Claims] 1. A search condition comprising a feature vector type of an image to be searched and a weight of the feature vector type is determined based on an image search keyword, and the search condition can be accessed to an image database to be accessed. An image search method characterized by performing image search using a common keyword with respect to a plurality of image databases having different search conditions by converting the image into search conditions and accessing the image database with the converted search conditions. 2. A search condition comprising a feature vector type of an image to be searched and weighting of the feature vector type is determined based on a keyword for image search, and an image database is accessed based on the search condition. In an image search method for performing a search, a vertical relationship is set in advance between the keywords for image search, and if the feature vector type and the weight of the feature vector type are not registered for the keyword, the keyword An image search method, wherein the search condition is set using a feature vector type registered for a high-order keyword and weighting of the feature vector type. 3. A search condition comprising a feature vector type of an image to be searched and weighting of the feature vector type is determined based on a keyword for image search, and an image database is accessed based on the search condition. In an image search method for performing a search, a vertical relationship is set in advance between the keywords for image search, and when the feature vector type and the weight of the feature vector type are not registered for the keyword, the keyword The search condition is set using the feature vector type registered for the high-order keyword and the weight of the feature vector type, and the search condition is converted into a search condition accessible to an image database to be accessed. By accessing the image database with the specified search conditions, multiple images with different search conditions An image search method, wherein an image search is performed on an image database using a common keyword. 4. An image search method for determining a search condition of an image to be searched based on a keyword for image search, and performing an image search by accessing an image database based on the search condition. An image search method, wherein a word that is the same as or similar to a keyword is associated with the search keyword, and a search condition for an image to be searched is determined based on the associated keyword. 5. A search condition comprising a feature vector type of an image to be searched and weighting of the feature vector type is determined based on a keyword for image search, and an image database is accessed based on the search condition. In an image search method for performing a search, an upper / lower relationship is previously set between the keywords for image search, and a keyword higher or lower and / or a keyword lower than a keyword input by the searcher is input to the searcher. A keyword for image search can be selected. 6. An image search method for searching image data based on a feature vector type of an image associated with a keyword and weighting of the feature vector type for a plurality of image databases having different search conditions. For a concept keyword tree to which a conceptual hierarchical relationship is applied, a keyword feature management table for managing feature vector types and weights for each keyword, and a database management of feature vector types associated with the keywords and feature vector type weights. A feature conversion function table for managing a conversion function for converting a feature vector type and a feature vector type weight for each system, and a key conversion for managing a conversion function for returning a key image name and a feature vector value of each image database for a keyword. Function table and concept key word A keyword feature extraction unit that applies an inheritance relationship to the definition of the feature vector type and the weighting of the feature vector type for each keyword of the tree, and extracts the feature vector of the keyword and the weight of the feature vector using the keyword feature management table; A feature conversion unit that converts the feature vector type and weight into a feature vector type and weight for each database management system using a conversion function that manages the feature vector in the feature conversion table; and a key conversion function that is managed in the key conversion function table. And a key conversion means for converting the keyword into a key image name and a feature vector value of each image database, and integrally managing the feature vector type and weight of the keyword and the feature vector and weight for each database management system. , Keyword features An image search method comprising converting a vector type and a weight into a feature vector and a weight for each database management system. 7. A keyword alias management table for managing a keyword managed in the keyword feature management table and a synonymous term as an alias keyword, wherein a keyword is specified on the concept keyword tree for a keyword specified as a search request by a searcher. 7. The image search method according to claim 6, wherein the keyword is searched from the keyword alias management table, and a keyword on the conceptual keyword tree obtained by the search is used. 8. A search condition including a feature vector type of an image to be searched and weighting of the feature vector type is determined based on an image search keyword, and an image database is accessed based on the search condition. In an image search method for performing a search, a local database management table for managing a database management system type and a data structure of each image database, a feature vector type and weight for each database management system, a key image name and a feature vector type, and the local database A search statement generation function table for managing a function for dynamically generating a search statement for each image database from a database management system type and a data structure managed by a management table; and a database management system from the local database management table. Local database information acquisition means for acquiring the program type and image database name; feature conversion means for respectively converting the feature vector type and weight of the keyword into the feature vector type and weight for each database management system; Key conversion means for converting to image names and feature vector values, and a search statement generation function managed in the search statement generation function table, using the feature vector type and weighting of the keyword obtained by the feature conversion means, Based on the key image name and the characteristic vector value obtained by the key conversion unit and the database management system type and data structure obtained by the local database information obtaining unit, a search statement for each image database is dynamically generated. Search statement generating means; and the search statement generating means Image data access means for searching for image data from each image database based on the search command sentence generated in the step, and searching for image data from each image database. 9. A search condition comprising a feature vector type of an image to be searched and weighting of the feature vector type is determined based on a keyword for image search, and an image database is accessed based on the search condition. In an image search method for performing a search, a feature conversion function table for managing a conversion function for converting a feature vector type and weight of a keyword into a feature vector type and weight for each database management system; a key image name of each image database for a keyword; A key conversion function table that manages a conversion function that returns a feature vector value, and a conversion function that manages the feature vector type and weight of a keyword in the feature conversion function table are used for the feature vector type and weight for each database management system Feature conversion means for converting A key conversion function for converting a keyword into a key image name or a feature vector value of each image database using a key conversion function managed in the key conversion function table; and a conceptual keyword tree in which a conceptual hierarchical relationship is applied to the keyword. In contrast, a keyword feature management table that manages the feature vector type and weighting for each keyword, and an inheritance relationship is applied to the definition of the feature vector type and weighting for each keyword in the concept keyword tree. Keyword feature extraction means for extracting the feature vector type and weight of each image database; a local database management table for managing the database management system type and data structure of each image database; feature vector type and weight for each database management system; key image name A search statement generation function table for managing a function for dynamically generating a search statement for each image database based on the feature vector value and the database management system type and data structure managed by the local database management table; A local database information acquisition unit that acquires a database management system type and an image database name from the local database management table; and a search statement generation function that is managed by the search statement generation function table. Based on the feature vector type and weight of the keyword, the key image name and the feature vector value obtained by the key conversion means, and the database management system type and data structure obtained by the local database information obtaining means. Search statement A search command sentence generation means for dynamically generating; and an image database access means for searching image data from each image database based on the search command sentence generated by the search command sentence generation means; An image search method characterized by converting the weight and the weight into a feature vector type and a weight for each database management system, dynamically generating a search statement for each image database, and searching for image data from each image database. 10. A keyword alias management table that manages a keyword and a synonymous term as an alias keyword managed in the keyword feature management table,
10. A keyword on the concept keyword tree for the keyword specified by the searcher as a search request is searched from the keyword alias management table, and the keyword on the concept keyword tree obtained by the search is used. Image search method described. 11. A search condition determining means for determining a search condition including a feature vector type of an image to be searched and weighting of the feature vector type based on an image search keyword, and the search condition is an access target. A search condition conversion unit for converting the search condition into a search condition accessible to the image database; and a database access unit for accessing the image database with the converted search condition, using a common keyword for a plurality of image databases having different search conditions. An image search device for performing an image search by using 12. A search condition comprising a feature vector type of an image to be searched and weighting of the feature vector type is determined based on an image search keyword, and an image database is accessed based on the search condition. In an image search device performing a search, a keyword hierarchical relationship storage unit that stores a hierarchical relationship between keywords set in advance, and when a feature vector type and a weight of a feature vector type are not registered for a keyword, An image search apparatus comprising: a search condition setting unit configured to set the search condition by using a feature vector type registered for a higher keyword of the keyword and weighting of the feature vector type. 13. A search condition including a feature vector type of an image to be searched and weighting of the feature vector type is determined based on a keyword for image search, and an image database is accessed based on the search condition. In an image search device performing a search, a keyword hierarchical relationship storage unit that stores a hierarchical relationship between keywords set in advance, and when a feature vector type and a weight of a feature vector type are not registered for a keyword, A search condition setting unit configured to set the search condition by using the feature vector type registered for the high-order keyword of the keyword and the weighting of the feature vector type; and enabling the search condition to access an image database to be accessed. A search condition conversion means for converting the search condition into search conditions; By accessing the image database, the image retrieval apparatus characterized by performing the image search by using the common keywords to a plurality of image database search conditions are different. 14. An image search apparatus which determines a search condition of an image to be searched based on a keyword for image search and accesses an image database based on the search condition to perform an image search. A term unifying unit for unifying terms by associating a word that is the same as or similar to a keyword with the keyword for search, and determining a search condition of an image to be searched based on the associated keyword. An image search device characterized by the following. 15. A search condition comprising a feature vector type of an image to be searched and weighting of the feature vector type is determined based on a keyword for image search, and an image database is accessed based on the search condition. An image search apparatus for performing a search, comprising: a keyword hierarchical relationship storage unit for storing a hierarchical relationship between keywords set in advance; and a keyword higher or lower than a keyword input with respect to a keyword input by a searcher. An image search apparatus comprising: a keyword selection assisting unit that is provided to a searcher to assist selection of an image search keyword. 16. An image search apparatus for searching image data based on a feature vector type of an image associated with a keyword and a weight of the feature vector type for a plurality of image databases having different search conditions. For a concept keyword tree to which a conceptual hierarchical relationship is applied, a keyword feature management table for managing feature vector types and weights for each keyword, and a database management of feature vector types associated with the keywords and feature vector type weights. A feature conversion function table for managing a conversion function for converting a feature vector type and a feature vector type weight for each system, and a key conversion for managing a conversion function for returning a key image name and a feature vector value of each image database for a keyword. Function tables and concept keys A keyword feature extraction unit that applies an inheritance relationship to the definition of the feature vector type and the weight of the feature vector type for each keyword of the keyword tree, and extracts the feature vector of the keyword and the weight of the feature vector using the keyword feature management table; A feature conversion unit that converts the feature vector type and weight into a feature vector type and weight for each database management system using a conversion function that manages the feature vector in the feature conversion table; and a key conversion function that is managed in the key conversion function table. And a key conversion means for converting the keyword into a key image name and a feature vector value of each image database, and integrally managing the feature vector type and weight of the keyword and the feature vector and weight for each database management system. , Keyword Features An image retrieval apparatus characterized in that a vector type and a weight are converted into a feature vector and a weight for each database management system. 17. A keyword alias management table for managing, as alias keywords, terms synonymous with keywords managed in the keyword feature management table,
17. The search method according to claim 16, wherein a keyword on the concept keyword tree for the keyword designated by the searcher as a search request is searched from the keyword alias management table, and the keyword on the concept keyword tree obtained by the search is used. The image search device according to the above. 18. A search condition including a feature vector type of an image to be searched and weighting of the feature vector type is determined based on an image search keyword, and an image database is accessed based on the search condition. In the image search apparatus for performing the search, a local database management table for managing a database management system type and a data structure of each image database, a feature vector type and weighting for each database management system, a key image name and a feature vector type, and the local database A search statement generation function table for managing a function for dynamically generating a search statement for each image database from the database management system type and data structure managed by the management table; and a database management system from the local database management table. Local database information acquisition means for acquiring a system type and an image database name; feature conversion means for converting a feature vector type and weight of a keyword into a feature vector type and weight for each database management system; and a keyword for a key of each image database. Key conversion means for converting to image names and feature vector values, and using a search statement generation function managed in the search statement generation function table, the feature vector type and weighting of the keyword obtained by the feature conversion means and Based on the key image name and the characteristic vector value obtained by the key conversion means and the database management system type and data structure obtained by the local database information obtaining means, a search statement for each image database is dynamically generated. Search statement generation means; and the search statement generation Image data access means for searching for image data from each image database based on the search command sentence generated by the means, and searching for image data from each image database. 19. A search condition comprising a feature vector type of an image to be searched and weighting of the feature vector type is determined based on a keyword for image search, and an image database is accessed based on the search condition. In an image search device for performing a search, a feature conversion function table that manages a conversion function for converting a feature vector type and weight of a keyword into a feature vector type and weight for each database management system; a key image name of each image database for a keyword; A key conversion function table that manages a conversion function that returns a feature vector value, and a conversion function that manages a feature vector type and weight of a keyword in the feature conversion function table are used for a feature vector type and weight for each database management system. Feature conversion means for conversion And key conversion means for converting a keyword into a key image name or a feature vector value of each image database using a key conversion function managed in the key conversion function table; and a conceptual keyword in which a conceptual hierarchical relationship is applied to the keyword. A keyword feature management table that manages the feature vector type and weight for each keyword for the tree, and an inheritance relationship is applied to the definition of the feature vector type and weight for each keyword in the concept keyword tree, and the keyword feature management table is used. Keyword feature extraction means for extracting the feature vector type and weight of the keyword; a local database management table for managing the database management system type and data structure of each image database; the feature vector type and weight for each database management system; Command generation function table that manages a function for dynamically generating a search command for each image database based on the database management system type and data structure managed by the local database management table and the feature vector value and the local database management table. A local database information acquisition unit that acquires a database management system type and an image database name from the local database management table; and a search command generation function that is managed by the search command generation function table. An image database based on the feature vector type and weight of the obtained keyword, the key image name and feature vector value obtained by the key conversion means, and the database management system type and data structure obtained by the local database information acquisition means. Search instructions for each A search statement generating means for dynamically generating a keyword, and an image database access means for searching image data from each image database based on the search statement generated by the search statement generating means. An image search apparatus for converting a type and a weight into a feature vector type and a weight for each database management system, dynamically generating a search statement for each image database, and searching for image data from each image database; . 20. A keyword alias management table that manages a keyword and a synonymous term that are managed in the keyword feature management table as alias keywords,
10. A keyword on the concept keyword tree for the keyword specified by the searcher as a search request is searched from the keyword alias management table, and the keyword on the concept keyword tree obtained by the search is used. Image search method described. 21. Search condition determining means for determining a search condition comprising a feature vector type of an image to be searched and weighting of the feature vector type based on a keyword for image search, and the search condition being an access target. A recording medium storing an image search program for causing a computer to execute a search condition conversion unit that converts the image database into search conditions that can be accessed, and a database access unit that accesses the image database with the converted search conditions. 22. A search condition comprising a feature vector type of an image to be searched and weighting of the feature vector type is determined based on a keyword for image search, and an image database is accessed based on the search condition. A computer-readable recording medium storing an image search program for executing a search, a keyword vertical relation information storage unit for storing predetermined vertical relation information between keywords, and a feature vector type for the keywords. And a search condition setting means for setting the search condition by using a feature vector type registered for a higher keyword of the keyword and a weight of the feature vector type when the weight of the feature vector type is not registered. Record an image search program to make a computer execute The recording medium. 23. A search condition comprising a feature vector type of an image to be searched and weighting of the feature vector type is determined based on an image search keyword, and an image database is accessed based on the search condition. A computer-readable recording medium storing an image search program for performing a search, a keyword vertical relation information storage means for storing predetermined vertical relation information between keywords, and a feature vector type for the keywords. When the weight of the feature vector type is not registered, a search condition setting unit configured to set the search condition using the feature vector type registered for the upper keyword of the keyword and the weight of the feature vector type. , The search condition can be accessed to the image database to be accessed A storage medium storing an image search program for causing a computer to execute a search condition conversion unit for converting the search condition into a search condition that can be converted into a search condition. 24. A computer storing an image search program for determining a search condition of an image to be searched based on a keyword for image search and accessing an image database to perform an image search based on the search condition. A readable recording medium, comprising: a term unifying unit for unifying terms by associating a word that is the same as or similar to the search keyword with the search keyword; and based on the associated keyword. A recording medium storing an image search program for causing a computer to execute a search condition determination unit for determining a search condition of an image to be searched. 25. A search condition including a feature vector type of an image to be searched and weighting of the feature vector type is determined based on a keyword for image search, and an image database is accessed based on the search condition. A computer readable recording medium storing an image search program for performing a search, a keyword vertical relation information storage means for storing predetermined vertical relation information between keywords, and An image search program for causing a computer to execute keyword selection assisting means for presenting a keyword higher or lower and / or lower than the input keyword to the searcher so as to assist selection of a keyword for image search. The recording medium on which it was recorded. 26. An image retrieval program for retrieving image data based on a feature vector type of an image associated with a keyword and a weight of the feature vector type for a plurality of image databases having different search conditions. A computer readable recording medium, comprising: a keyword feature management table for managing a feature vector type and a weight for each keyword for a conceptual keyword tree in which a conceptual hierarchical relationship is applied to the keyword; A feature conversion function table that manages a conversion function for converting a feature vector type and a weight of the feature vector type into a feature vector type and a weight of the feature vector type for each database management system; a key image name of each image database for a keyword; Return feature vector value A key conversion function table that manages the conversion functions to be performed, applying a succession relationship to the definition of the feature vector type and the weighting of the feature vector type for each keyword in the concept keyword tree, and Keyword feature extraction means for extracting the weight of the feature vector; and feature conversion means for converting the feature vector type and the weight of the keyword into the feature vector type and the weight for each database management system using a conversion function managing the feature conversion table. And a key conversion unit for converting a keyword into a key image name or a feature vector value of each image database using a key conversion function managed in the key conversion function table. Recording medium. 27. A keyword alias management table which manages a keyword and a synonymous term as an alias keyword in the keyword feature management table,
A searcher searching the keyword alias management table for a keyword on the concept keyword tree for a keyword specified as a search request, and causing a computer to use the keyword on the concept keyword tree obtained by the search. 27. A recording medium recording an image search program according to claim 26, wherein the image search program is recorded. 28. A search condition including a feature vector type of an image to be searched and weighting of the feature vector type is determined based on an image search keyword, and an image database is accessed based on the search condition. A computer-readable recording medium storing an image search program for performing a search, a local database management table for managing a database management system type and a data structure of each image database, and a feature vector type for each database management system. A search statement generation function for managing a function for dynamically generating a search statement for each image database from weights, key image names and feature vector values, and a database management system type and data structure managed in the local database management table. A table and Local database information acquisition means for acquiring a database management system type and an image database name from a local database management table; feature conversion means for converting a feature vector type and a weight of a keyword into a feature vector type and a weight for each database management system; A key conversion unit that converts a keyword into a key image name or a feature vector value of each image database; and a keyword obtained by the feature conversion unit using a search statement generation function managed by the search statement generation function table. Search for each image database based on the feature vector type and weighting, the key image name or feature vector value obtained by the key conversion means, and the database management system type and data structure obtained by the local database information obtaining means. Imperative sentence Image search means for causing a computer to execute: a search statement generation means for generating the image data; and an image data access means for searching for image data from each image database based on the search statement generated by the search statement generation means. A recording medium on which a program is recorded. 29. A search condition including a feature vector type of an image to be searched and weighting of the feature vector type is determined based on an image search keyword, and an image database is accessed based on the search condition. A computer readable recording medium storing an image search program for performing a search, wherein a feature conversion function for managing a conversion function for converting a feature vector type and weight of a keyword into a feature vector type and weight for each database management system A table, a key conversion function table for managing a conversion function for returning a key image name and a feature vector value of each image database for a keyword, and a conversion function for managing the feature vector type and weight of the keyword in the feature conversion function table. Per database management system A feature conversion unit that converts each of the feature vector types and weights, and a key conversion unit that converts a keyword into a key image name or a feature vector value of each image database using a key conversion function managed in the key conversion function table. A keyword feature management table that manages feature vector types and weights for each keyword for a conceptual keyword tree that applies a conceptual hierarchical relationship to keywords, and an inheritance relationship for the definition of feature vector types and weights for each keyword in the concept keyword tree A keyword feature extraction means for extracting a feature vector type and a weight of a keyword using the keyword feature management table; a local database management table for managing a database management system type and a data structure of each image database; Based on the feature vector type and weight for each image management system, the key image name and the feature vector value, and the database management system type and data structure managed in the local database management table. A search statement generation function table for managing a function to be generated in the database; a local database information acquisition unit for acquiring a database management system type and an image database name from the local database management table; and a search managed by the search statement generation function table. Using a command sentence generation function, the feature vector type and weight of the keyword obtained by the feature conversion means, the key image name and feature vector value obtained by the key conversion means, and the local database information acquisition means Database management system type Command statement generating means for dynamically generating a search command statement for each image database based on the search command statement generated based on the search command statement generated by the search command statement generating means. A recording medium on which an image search program for causing a computer to execute an image database access means for searching for an image is stored. 30. A keyword alias management table for managing a keyword and a synonymous term as an alias keyword in the keyword feature management table,
A search step that searches a keyword on the concept keyword tree for the keyword designated by the searcher as a search request from the keyword alias management table, and causes the computer to use the keyword on the concept keyword tree obtained by the search. 30. A recording medium recording an image search program according to claim 29, wherein the image search program is recorded.