JPH1115852A - Image database center device, image database registration / search method, and recording medium - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To provide an image database center apparatus which is versatile and does not specify a user and can be efficiently registered / retrieved, and a registration / retrieval method provided therein. An image database (12) can be accessed from paper that can be transmitted by fax terminals (16, 17) and a personal computer (18) or paper that can be decoded by a scanner (11). Furthermore, it is possible to enter the character string of the keyword data and the image data in the same page and the same page and to perform free format without limiting the description area thereof. Registration and search in the database 12 are enabled. In addition, use the respective dictionaries for kana and numbers, process voiced / semi-voiced characters for characters containing voiced / semi-voiced characters, and process separated characters for separated characters, and write in free format. A high character recognition rate and a high hit rate can be realized for a mixed character string of kana and numbers.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image database center apparatus and an image database registration / retrieval method, and more particularly, to accessing a remote database from a fax terminal.
The present invention relates to an image database center apparatus for searching and transmitting desired image data from an image database and an image database registration / search method provided for the apparatus.

[0002]

2. Description of the Related Art Conventionally, as this type of system, there is a facsimile guidance service system. In this service, when registering image data, a fax terminal is used to dial a service number for registration and a BOX number of a registration destination, and the image data is transmitted. On the other hand, when searching for registered image data, the service number for search and BO
When the X number is dialed, the service center searches the database for image data corresponding to the search request and transmits the image data to the search requester. However, in this system, BO
It takes time and effort to check the X number, and the image data registered in the BOX number is unclear.
It is necessary to search the menu screen once to check the image data registered in the X number, that the telephone number of the registrant required for the search is unclear, the operation at the time of registration or search (dial ) Cannot be remembered, so a manual is required.

[0003] On the other hand, without dialing a service number or a BOX number, a mark sheet transmitted from a fax terminal is limited to a dedicated mark sheet for each database center, and marks and characters described in the dedicated mark sheet are recognized. Some systems are configured to access a database.

In a system for constructing a database by character recognition, it is necessary to write a character on the dedicated mark sheet, and as a restriction in writing, a character is set in a predetermined frame at a predetermined position. It was necessary to write in the font of the size. Due to the necessity of writing on the special mark sheet and restrictions on the user side, the recognition rate of character recognition has been increased.

As a character recognition technique used for this, there is a means for simply combining individual candidate characters given as a result of character recognition to generate a candidate character string. Also,
Although there are database systems that use the generated candidate character strings as keywords, a service that generates one candidate character string for one image data is generally used.
In some systems, the generated candidate character strings are word-matched, and only those that match the result of the word matching are generated as candidate character strings.

[0006]

However, in the above-described conventional database system using character recognition, it is necessary to use a dedicated mark sheet and there is a restriction on the user side, so there is no versatility, and more users who can access the database are required. There was a problem that it was disadvantageous because it was specified.

Further, in the conventional character recognition technology used in this system, the recognition result of character recognition is bad, katakana may be erroneously recognized as a number, and a number may be erroneously recognized as katakana. Problems,
Matching is difficult because only one candidate character string is generated at the time of registration and search, respectively. Due to missing (overlooking) of voiced or semi-voiced characters, or character segmentation NG (failure) of separated characters, There was a problem that the number of characters did not match and matching was difficult.

An object of the present invention is to use a dedicated mark sheet for each database center in a conventional image database system, to restrict access to image data stored in a database, and to reduce character recognition accuracy. It solves problems such as difficulty in matching, is versatile, does not identify the user, and can perform high-precision character recognition. It is easy to match when searching, and as a result, the image desired by the user An object of the present invention is to provide an image database center device capable of efficiently transmitting data and an image database registration / search method provided for the image database center device.

[0009]

The present invention solves the above-mentioned problems by the following means (1) to (12).

(1) In an image database center for transmitting / receiving image data to / from a communication medium and registering / retrieving the image data, a registration requester sends an image from a communication medium via a communication network or from an image scanner. A first means for receiving data; and a second means for extracting a candidate character string from image data received from the registration requester and generating one or more candidate character strings with high accuracy including an uncertain character Means, the candidate character string generated by the second means and the image data from which the candidate character string is extracted are registered in an image database in association with each other, and the search requester receives the candidate character string from a communication medium via a communication network. Fourth means for receiving image data; extracting candidate character strings from the image data received from the search requester; Fifth means for generating one or more candidate character strings, and comparing the matching rate between the candidate character strings generated by the fifth means and the candidate character strings registered in the image database, Sixth means for searching the image database for image data corresponding to a candidate character string determined to match with the rate based on a preset matching rate;
And a seventh means for transmitting the searched image data to a communication medium of the search requester.

(2) The second and fifth means include a character recognizing means, an absolute reference value judging section, a fixed reference value judging section, and a relative reference value judging section. Generating a character string, wherein the character recognition means ranks a plurality of candidate characters as a character recognition result, and determines the plurality of candidate characters and a distance value corresponding to the plurality of candidate characters by the absolute reference value determination unit; Value determination unit, and returns to the relative reference value determination unit, the absolute reference value determination unit, the plurality of ranked candidate characters and the corresponding distance value value and a preset absolute reference value Compare the values,
Based on the magnitude of the value of the distance value and the value of the absolute reference value, it is determined whether or not the candidate character is to be combined with the candidate character string, and the final reference value determination unit,
The upper and lower candidate characters and the corresponding upper and lower distance values are subtracted, the subtracted value is compared with a preset fixed reference value, and the subtracted value and the fixed reference value are compared. Based on the magnitude of the value, it is determined whether only the upper candidate character is to be combined with the candidate character string, or whether the upper and lower candidate characters are to be combined with the candidate character string. The reference value determination unit subtracts the upper and lower candidate characters and the corresponding upper and lower distance values, compares the subtracted value with a preset relative reference value, and calculates the subtracted value. And depending on the value of the relative reference value, whether the upper and lower candidate characters are replaced with indeterminate characters and the candidate character strings are combined,
Alternatively, the image database center apparatus determines whether upper and lower candidate characters are to be combined with candidate character strings.

(6) The sixth means includes one or more candidate character strings generated from the image data received from the registration requester and one or more candidate character strings generated from the image data received from the search requester. When comparing the match rates with multiple candidate strings, the match rates between the individual strings of the candidate strings generated at the time of the registration request and the individual strings of the candidate strings generated at the An image database that is calculated for each character string and searches the image database for image data corresponding to a candidate character string including a character string determined to match with the match rate based on a preset match rate. Center equipment.

(4) The relative reference value determination unit includes means for replacing a candidate character with an uncertain character, and searching a candidate character string including the uncertain character for a character string from a dictionary database. Means for additionally generating a character string to be matched as a candidate character string.

(5) The second and fifth means are separated from a means for additionally generating a candidate character string from which a voiced point or a semi-voiced point is removed, to a candidate character string including a voiced point or a semi-voiced point. Means for additionally generating a candidate character string including a character and replacing the character with a character obtained by separating the character into two characters, an image database center apparatus.

(6) In a method for registering / searching an image database provided in an image database center for transmitting / receiving image data to / from a communication medium and registering / retrieving image data, image data from a registration requester is transmitted via a communication network. A first procedure of receiving from a communication medium or an image scanner, and extracting a candidate character string from image data received from the registration requester by using character recognition software, and selecting a candidate character string with high accuracy including an uncertain character A second procedure for generating one or more of the above, a third procedure for registering the candidate character string generated in the second procedure and the image data from which the candidate character string is extracted in the image database, A fourth step of receiving image data from a communication medium via a communication network; A fifth procedure of extracting one or more candidate character strings with high accuracy including an uncertain character by extracting candidate character strings from the character recognition software, and a candidate character string generated in the fifth procedure. Comparing a matching rate with the candidate character string registered in the image database, and searching the image database for image data corresponding to the candidate character string determined to match the matching rate with a preset matching rate. Image database registration / procedure, comprising: a sixth procedure; and a seventh procedure of transmitting the searched image data to a communication medium of the search requester.
retrieval method.

(7) When the character recognition software in the second and fifth procedures performs character recognition of a character string in which katakana and numbers are mixed, the character string in which katakana and numbers are mixed has katakana and Using a delimiter at the middle position of numbers, the characters before and after the delimiter are used separately with katakana and numbers, and the characters are recognized using a katakana dedicated dictionary for the katakana part and a numeric only dictionary for the number part. Database registration / search method to be performed.

(8) The second and fifth procedures include a step of ranking a plurality of candidate characters as a character recognition result of the character recognition software, and returning the plurality of candidate characters and a distance value corresponding thereto. The plurality of ranked candidate characters are compared with a value of a distance value corresponding thereto and a value of a preset absolute reference value, and the candidate character is determined as a candidate according to the magnitude of the value of the distance value and the value of the absolute reference value. An absolute reference value determination procedure for determining whether or not to be a combination of character strings;
The upper and lower candidate characters and the corresponding upper and lower distance values are subtracted, the subtracted value is compared with a preset fixed reference value, and the subtracted value and the fixed reference value are compared. Determining reference value determination procedure for determining whether only the upper candidate character is to be combined with the candidate character string or whether the upper and lower candidate characters are to be combined with the candidate character string according to the magnitude of the value. And subtracts the upper and lower candidate characters and the corresponding upper and lower distance values, compares the subtracted value with a predetermined relative reference value, and compares the subtracted value with the relative reference value. Depending on the magnitude of the value, the upper and lower candidate characters are replaced with indeterminate characters and the candidate character string is combined, or the upper and lower candidate characters are combined with the candidate character string, respectively. Or With a relative reference value determining procedure of a constant, the image database registration / search method characterized in that to generate the candidate character strings by combining the candidate characters on the basis of the determination.

(9) In the sixth procedure, one or more candidate character strings generated from the image data received from the registration requester, and one or more candidate character strings generated from the image data received from the search requester. When comparing the matching rates of a plurality of candidate character strings, the matching rate between the individual character strings of the candidate character group generated at the time of the registration request and the individual character strings of the candidate character group generated at the time of the search request is calculated. Image database registration / retrieval, characterized in that image data corresponding to a candidate character string including a character string that is calculated based on a character string and determined to be matched by a predetermined matching rate is searched from an image database. Method.

(10) The relative reference value determining step includes a step of replacing a candidate character with an uncertain character, and a step of searching a candidate character string including the uncertain character from a dictionary database. A method of additionally generating a character string to be matched as a candidate character string.

(11) The second and fifth procedures are separated from a procedure of additionally generating a candidate character string from which a voiced point or a semi-voiced point is removed for a candidate character string including a voiced point or a semi-voiced point. An additional step of generating a candidate character string including a character by replacing the character with a character obtained by separating the character into two characters, and a method of registering / retrieving an image database.

(12) A program for causing a computer to execute the procedure of any of the image database registration / retrieval methods described in (6) to (11) above is recorded on a computer-readable medium. Recording medium.

According to the present invention, an image database can be accessed from a sheet (access sheet) that can be transmitted by a fax terminal and a personal computer or a sheet that can be decoded by a scanner device. In this way, it is possible to enter a character string of keyword data and image data in a mixed manner, and to perform a free format that does not limit the description area of the character string of the keyword data and the image data. It is characterized in that registration and search in the image database are enabled based on a candidate character string.
This makes it possible to construct a system that is versatile and does not specify a user, and facilitates matching when accessing an image database. The conventional technology differs from the conventional technology in that a dedicated mark sheet required for each database center is used, and at the same time, a character string of keyword data and an area for describing image data are clarified. The difference is that a format described in an access sheet such as the mark sheet and a plurality of candidate character strings at the time of registration and search can be respectively generated.

Also, the present invention uses a katakana and a number and a dictionary for katakana and a number, respectively, and processes a character including a voiced point and a half-voiced voice for a voiced and a half-voiced voice. It is characterized in that a separated character process is performed on the character that is being performed. As a result, Katakana characters are not mistaken for numbers or numbers are mistaken for Katakana characters, and the character recognition rate is improved. Character recognition with a high recognition rate for character strings with mixed numbers and numbers, and the following features: omission of voiced dots (overlooked), omission of semi-voiced voices (overlooked), or character separation NG of separated characters The most important feature is to prevent the number of characters in the candidate character string from being inconsistent and to improve the search hit rate.

In the means for enabling access to a database according to the prior art, the mark sheet required for access is not versatile, and the user who uses the database is specified for that purpose. The means of using the access sheet that has been used classifies keyword data into character strings and image data by character recognition, so it is possible to mix keyword data character strings and image data in the same access sheet Thus, it is possible to efficiently transmit the image data desired by the user, which is the subject of the present invention.

According to the present invention, compared to the conventional facsimile guidance service system, a BOX number is not required, so that time and labor are not required. Further, since image data can be registered with a title name instead of a BOX number, registration is possible. It is possible to clarify the image data that has been entered, so that there is no need to search the menu screen, it is efficient because there is no need for the telephone number of the registrant required for the search, and the operation at the time of registration or search ( Dial) can be accessed with a single telephone number, and there is an advantage that a manual is not required.

[0026]

Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a system configuration diagram including an embodiment of the image database center apparatus of the present invention. The image database center device 10 includes a scanner 11, an image database 12, and an image database center main device 1.
3 and a line board 14. The line board 14 is connected to a communication network 15 via which the fax terminals 16 and 17 and personal computers (hereinafter, personal computers) 18 as communication media can access the image database center device 10. ing. Fax terminal 16
Is an example of a fax terminal that registers in the image database 12, and a fax terminal 17 is an example of a fax terminal that searches the image database 12.

FIG. 2 is a functional block diagram of the image database center apparatus 10. As shown in FIG. The image database center device 10 includes an image data transmitting / receiving unit that transmits / receives image data to / from the above-described communication media 16, 17, and 18 via the communication network 15, an image data receiving unit that receives image data input from the scanner 11, Character recognition software using a dictionary and a number dictionary, a delimiter deletion processing block, an absolute reference value determination unit, a fixed reference value determination unit,
It has a relative reference value determination unit, a candidate character string combination processing block, a voiced / semi-voiced voice processing block, a separated character processing block, a dictionary DB (database) comparison processing block, a dictionary DB, and an image data registration / search comparison block. Each function and operation will be described in the following description of the processing.

FIG. 3 is a diagram showing a service image by the image database center device. A description will be given of a service image in the case of registering / searching, for example, travel information and local detailed information on Hawaii as image data. An information provider such as travel information and local detailed information writes the information to be provided in a free format database access sheet and inputs the information to the image database center device 10 provided in the service center from the fax terminal 16 via a telephone line. Alternatively, the image database center device 1 is operated by using the scanner 11 provided in the service center.
Enter 0. The image database center apparatus 10 recognizes characters of the provided information and determines one keyword candidate.
Or, a plurality of extractions are made and registered in the image database corresponding to the image data of the extraction source. On the other hand, a general citizen who wants to obtain travel information, local detailed information, and the like regarding Hawaii writes keywords from a fax terminal 17 on a free format database access sheet and accesses the image database center apparatus 10 via a telephone line. Alternatively, the image database center device 10 is accessed from a personal computer through the Internet or the like. The image database center device 10 extracts one or more keyword candidates by character recognition from the input database access sheet, searches the image database with the keyword candidates, and searches for the keyword candidates registered in the image database. The travel information and the local detailed information on the corresponding Hawaii are fax-outputted to the fax terminal 17 accessed via the telephone line according to the coincidence rate of. Alternatively, output to a personal computer accessed through the Internet.

FIG. 4 shows a description example of a database access sheet. Reference numeral 20 denotes a sheet, reference numerals 21 and 23 denote examples of description of keyword data, reference numeral 22 denotes a black line (double line) for extracting a keyword, and reference numeral 24 denotes an area where image data can be written. FIGS. 4A and 4B are examples in which the keyword 21 is indicated by a double line 22, and FIG. 4C is an example in which the keyword 23 is described in an image data writeable area 24 for extracting the keyword. .

FIG. 5 shows, as examples of character types that can be described in the database access sheet, (a) numerals,
(B), (c) indicate katakana characters.

FIGS. 6 to 12 are flowcharts showing an embodiment of the program processing in the image database center apparatus having the above-mentioned configuration. Hereinafter, an embodiment of the processing of the program will be described.

[Embodiment 1] In this embodiment, a basic processing example up to registration of an image database will be described (processing flows 101 to 152).

In the following embodiment, an example will be described in which the medium used for registration is a facsimile terminal, and the recognition character type selection preset in the system is set to the "kana only" mode. The character string of the keyword data registered from the fax terminal is "multi-media" as an example.

101: Branch depending on the media used for image database registration. In the present embodiment, a fax terminal (see FIG. 1) is used. When a scanner is used as a medium, the external input / output function of 201,
The image data input from the scanner is received by the scanner registration 202 and the utility software 203.

102: FAX transmission / reception software receives image data transmitted from a facsimile terminal, a portable information terminal, or a personal computer. In the present embodiment, the image data transmitted from the fax terminal is received.

At step 103, an xbm (x bitmap) file is created from the image data received by the fax transmission / reception software at 102 or the utility software at 203.

104: System setting (FIG. 1) set by GUI (Graphical User Interface)
3. Referring to the state of the system setting screen example of FIG. 14 (see FIG. 14 is a continuation screen of FIG. 13), the process branches at 105 and 106 depending on the system setting state.

105: Branch depending on whether or not there is a delimiter "#". In the system setting of 104, "Kana only" is selected in the recognition character type selection (see FIG. 13).
Alternatively, if “only numbers”, “kana + numeric”, or “numeric + kana” is selected, processing is performed in the kana character presence / absence mode of 106. If “kana + (#) + numerical” or “numeric + (#) + kana” is selected in the recognition character type selection, processing is performed by extracting keyword data and separating image data in 107-1. . In the present embodiment, the “kana only” mode is set and there is no “#”. In this case, processing is performed in the kana character presence / absence mode of 106.

106: Branch depending on whether there is a kana character or not. In the system setting of 104, if “kana only”, “kana + numeral” or “numeric + kana” is selected in the recognition character type selection (see FIG. 13), the keyword data extraction and image data extraction of 107-1 are performed. Processed in separation. In this embodiment, “only kana”
Since the mode is set, in this case, processing is performed by extracting keyword data and separating image data in 107-1.

107-1: Extract keyword data and separate image data. First, only the keyword data is extracted from the image data. Only the extracted keyword data is processed by the character recognition AP 107-2. Also,
The image data other than the extracted keyword data is determined to be image data, and the image data is directly registered in the image database corresponding to the keyword data without being processed by the character recognition AP 107-2.

Here, a specific method for extracting keyword data and separating image data will be described.

First, a sheet that can be used as a database access sheet is not specified, but is set in a free format. It is possible to describe target keyword data and image data on a sheet in a mixed manner on the same page. As a description method, (1) a method of drawing a double underline below keyword data as illustrated in FIG. 15A, and (2) a description of keyword data as illustrated in FIG. There are two ways to specify the location (range).

Next, the target keyword data described on the sheet is extracted by one of the following two extraction methods: (1) a double underline extraction method, and (2) an entry place designation extraction method. Extract.

(1) The double underline extraction method is as follows.
As shown in FIGS. 16A, 16B, and 16C, this is a method of extracting keyword data based on a double underline drawn below target keyword data. More specifically, a double underline in one page is searched for, a character string described above the double underline is extracted as target keyword data, and the character string is recognized. The result of character recognition is registered as keyword data in the image database as a candidate character string. Further, image data other than the target keyword data is determined as image data, and linked to the keyword data and registered in the image database.

(2) The entry location designation extraction method is described in FIG.
As shown in (c), this is a method of extracting target keyword data only within a predetermined range. Specifically, only within a predetermined range of a page, a character string described in the range is extracted as keyword data, and the character string is extracted as a character string. recognize. For the result of character recognition,
It is registered as keyword data in the image database as a candidate character string. In addition, image data other than the target keyword data is determined as image data,
Link to the keyword data and register it in the image database.

107-2: In the character recognition AP (application program), the keyword data extracted in 107-1 is converted into a file (xb
m files), and character recognition of the file (xbm file) returns the candidate characters and the corresponding distance values for each character. FIG. 17 shows an example of a character recognition AP output result. Here, in the character recognition AP, ten characters from the first place to the tenth place are listed as candidates, and
A distance value is assigned to each candidate character in units of one character. The distance value indicates a value that quantitatively expresses the likelihood of the candidate character recognized and returned by the character recognition AP. That is, the likelihood (accuracy) of each candidate character is represented by a numerical value, and the smaller the value, the higher the accuracy. Note that "#" is used in 105.
If it is determined that the mode is the presence mode, the delimiter character deletion processing of 301 deletes the delimiter “#” in 302 and then processes in 108.

108: In the absolute reference value determination unit,
Based on the candidate character returned from the second character recognition AP and its distance value (see FIG. 17), the candidate character for which the candidate character string is to be created is closely examined. Specifically, the first to tenth candidate characters are compared with the corresponding distance value and the absolute reference value set in the system, and the distance value and the absolute reference value are compared. It is determined whether or not the candidate character is to be combined with the candidate character string based on the magnitude. In the present embodiment, since the “kana only” mode is set, in this case, in the system setting of 104 (see FIG. 13),
While referring to "19000" of the absolute reference value (Kana),
The following 109 to 111 are processed. Note that the absolute reference value refers to a reference value for determining whether or not a candidate character is to be combined with a candidate character string.

109: Character recognition AP of 107-2 from the first candidate character in the first column to the first candidate character in the last column
Is compared with the absolute reference value set in the system setting of 104, and the process branches depending on the magnitude of the value. In the present embodiment, the first candidate character in the first column is "ma" and its distance value is "-6162" (see FIG. 17), and the first candidate character in the second column is "ru". It is referred to that "-14814" (see FIG. 17) is returned as the distance value. In the same way, the reference is continued up to the first candidate character “ta” in the last column. As a result, since the distance values from the first candidate character “MA” in the first column to the first candidate character “TA” in the last column are all smaller than “19000”, the process is continued at 110. In addition, 1
If the distance value is larger than the absolute reference value even for characters, 40
At 1, a candidate character string is created by combining only the first candidate character, and then processed at 133.

110: Distance values returned from the character recognition AP 107-2 for the second to tenth candidate characters for the characters in the first column, and the absolute reference set in the system settings 104 Compare values and branch depending on the magnitude of the value. In the present embodiment, the second-place candidate character in the first column is “A” and its distance value is “8500” (see FIG. 17).
Is returned, and the third-place candidate character in the same column is “U”, and “12500” is returned as the distance value. Similarly, since the fourth to tenth candidate characters have a value smaller than "19000" set by the absolute reference value (kana), they are processed at 111.

As a result of comparing the distance value returned from the character recognition AP 111: 107-2 with the absolute reference value set in the system setting 104, the first candidate character in the first column to the tenth candidate character The final reference value determination unit 112 processes ten characters up to the character.

112: In the final reference value determination section,
Based on the candidate characters returned from the character recognition AP of No. 2 and their distance values (see FIG. 17), candidate characters for which a candidate character string is to be created are closely examined. Specifically, the upper and lower candidate characters and the corresponding upper and lower distance values are subtracted, and the subtracted value is compared with the value of the final reference value set in the system. Based on the value of the reference value, it is determined whether only the upper candidate character is to be combined with the candidate character string, or whether the upper and lower candidate characters are to be combined with the candidate character string. In the present embodiment, the “kana only” mode is set. In this case, in the system setting of 104 (see FIG. 13), the following 113 to 113 are set while referring to “6000” of the fixed reference value (kana). Process 116. In addition, the fixed reference value is
It indicates a criterion value for determining whether only candidate characters with high accuracy are to be combined with candidate character strings.

113: The distance value of the first candidate character in the first column and the distance value of the second candidate character are subtracted, and the difference value is processed at 114 as a calculated value. In the present embodiment, the first candidate character in the first column is “ma” and its distance value is “−6”.
Since “162” is the second-place candidate character and “8500” is returned as its distance value, the calculated value becomes “14662” as a result of the subtraction process.

114: The calculated value at 113 is compared with the fixed reference value (see FIG. 13) set by the system setting at 104, and the process branches depending on the magnitude of the value. In the present embodiment,
Since the calculated value “14662” is a value larger than “6000” set as the fixed reference value (Kana), it is processed at 115.

115: At 114, the calculated value is larger than the fixed reference value, so the comparison between the calculated value at 114 and the fixed reference value is terminated, and the process is continued at 116.

The candidate character determined at 116: 115 is set as a candidate for combining candidate character strings. In the present embodiment, since the candidate character "ma" is determined by the determination reference value, the candidate character "ma" is set as a candidate for combining candidate character strings.

At 117: 116, the candidate character is branched depending on whether the candidate character is the last column. In the present embodiment, since the candidate character "MA" is not the last column, the second reference character in the next column of 118 is processed by the absolute reference value determination unit.

118: The second reference candidate character in the next column is processed by the absolute reference value determination unit. In the present embodiment, the second reference character “e” in the second column is processed by the absolute reference value determination unit 108.

108: In this embodiment, since the “kana only” mode is set, in this case, in the system setting of 104 (see FIG. 13), referring to “19000” of the absolute reference value (kana). , 109-111 below
Process.

109: This process is for the first candidate character in the first column to the first candidate character in the last column, so the second candidate character in the second column is processed in 110.

110: For the second column of characters, the distance value returned from the character recognition AP 107-2 for the second to tenth candidate characters and the absolute value set in the system setting 104 The reference value is compared, and branching is performed depending on the value. In the present embodiment, the second candidate character in the second column is “e” and its distance value is “16921” (see FIG. 17), and the third candidate character in the second column is “f”. It is assumed that “28323” is returned as the distance value.
At this point, the third-place candidate character "K" is the absolute reference value (Kana)
Since the value is larger than "19000" set in step 11,
9 is processed.

At 119: 110, the third-place candidate character "" in the second column was a distance value "28323" which was larger than "19000" set by the absolute reference value (Kana). The comparison of the reference values is terminated, and the process is continued at 120.

As a result of 120: 110, only the candidate characters whose distance value is within the absolute reference value are subjected to the determination of the fixed reference value. In the present embodiment, since the second-place candidate character "e" is a value smaller than "19000" set by the absolute reference value (kana) with the distance value "16921" for the character in the second column, "e" Is the target of the determination of the reference value. Also, since the third-place candidate character "K" is a distance value "28323" and a value larger than "19000" set by the absolute reference value (Kana), "K" is excluded from the determination of the fixed reference value. As a result, the second column is processed by the 112 fixed reference value determination units for one character of the second candidate character.

112: The fixed reference value determination unit 107-
Based on the candidate characters returned from the character recognition AP of No. 2 and their distance values (see FIG. 17), candidate characters for which a candidate character string is to be created are closely examined. In the present embodiment, the “kana only” mode is set. In this case, in the system setting of 104 (see FIG. 13), “60” of the fixed reference value (kana) is set.
The following 113 to 116 are processed with reference to “00”.

113: The distance value of the first candidate character in the second column is subtracted from the distance value of the second candidate character, and the difference value is processed at 114 as the calculated value. In the present embodiment, the first candidate character in the second column is “ル” and its distance value is “−1”.
4814 ", the second-place candidate character is" e ", and" 16921 "is returned as its distance value. As a result of the subtraction processing, the calculated value becomes" 31735 ".

114: The calculated value at 113 is compared with the fixed reference value set in the system setting at 104, and the process branches depending on the magnitude of the value. In the present embodiment, the calculated value “3173
Since “5” is a value larger than “6000” set as the fixed reference value (Kana), it is processed at 115.

At 115: 114, the calculated value is larger than the fixed reference value, so the comparison between the calculated value at 114 and the fixed reference value is terminated, and the process is continued at 116.

The candidate character determined at 116: 115 is set as a candidate for combining candidate character strings.

In the present embodiment, since the candidate character "R" is determined by the determination reference value, the candidate character "R" is set as a candidate for combining candidate character strings.

The candidate character determined at 117: 116 branches depending on whether it is the last column. In the present embodiment, since the candidate character “ル” is not the last column, the second reference candidate character in the next column of 118 is processed by the absolute reference value determination unit.

Step 118: The second reference candidate character in the next column is processed by the absolute reference value determination unit. In the present embodiment, the second reference character "" in the third column is processed by the absolute reference value determination unit 108.

108: In this embodiment, the "kana only" mode is set. In this case, in the system setting of 104 (see FIG. 13), referring to "19000" of the absolute reference value (kana). , 109-111 below
Process.

109: Since this process is a process from the first candidate character in the first column to the first candidate character in the last column, the second candidate character in the third column is processed in 110.

110: For the third column of characters, the distance value returned from the character recognition AP 107-2 for the second to tenth candidate characters and the absolute value set in the system setting 104 The reference value is compared, and branching is performed depending on the value. In the present embodiment, the second candidate character in the third column is “K” and its distance value is “3000” (see FIG. 17).
Is returned, and the third-place candidate character in the third column is “te”, and “19355” is returned as the distance value. At this point, the third-place candidate character "te" is larger than "19000" set by the absolute reference value (kana), and is therefore processed at 119.

At 119: 110, the third-order candidate character "te" in the third column was a distance value "19355", which was larger than "19000" set by the absolute reference value (kana). The comparison of the reference values is terminated, and the process is continued at 120.

As a result of 120: 110, only the candidate characters whose distance value is within the absolute reference value are subjected to the determination of the fixed reference value. In the present embodiment, since the second-place candidate character "" is a value smaller than "19000" set by the absolute reference value (Kana) with the distance value "3000" for the character in the third column, "" Is the target of the determination of the reference value. Since the third-place candidate character "te" is a distance value "19355" and a value larger than "19000" set by the absolute reference value (kana), "te" is excluded from the determination of the final reference value. As a result, the third column is processed by the 112 fixed reference value determining units for one character of the second candidate character.

112: The final reference value judgment unit 107-
Based on the candidate characters returned from the character recognition AP of No. 2 and their distance values (see FIG. 17), candidate characters for which a candidate character string is to be created are closely examined. In the present embodiment, the “kana only” mode is set. In this case, in the system setting of 104 (see FIG. 13), “60” of the fixed reference value (kana) is set.
The following 113 to 116 are processed with reference to “00”.

113: Subtract the distance value of the first candidate character and the distance value of the second candidate character in the third column, and process the difference value as a calculated value at 114. In the present embodiment, the first candidate character in the third column is “H” and its distance value is “31”.
Since “6” is the second candidate character and “3000” is returned as the distance value, the calculated value becomes “2684” as a result of the subtraction process.

114: The calculated value at 113 is compared with the fixed reference value set in the system setting at 104, and the process branches depending on the magnitude of the value. In the present embodiment, the calculated value “268
Since "4" is a value smaller than "6000" set as the fixed reference value (Kana), it is processed at 121.

At 121: 114, since the calculated value was smaller than the fixed reference value, the comparison between the calculated value of 114 and the fixed reference value was terminated, and the upper candidate character and the lower candidate character were determined as relative reference value determination targets. I do. In the present embodiment, the relative reference value determination unit 122 processes two characters of the first candidate character “H” and the second candidate character “K” in the third column.

122: In the relative reference value determination section,
Based on the candidate characters returned from the character recognition AP of No. 2 and their distance values (see FIG. 17), candidate characters for which a candidate character string is to be created are closely examined. Specifically, the upper and lower candidate characters and the corresponding upper and lower distance values are subtracted, and the subtracted value is compared with the value of the relative reference value set in the system. Depending on the value of the reference value, the upper and lower candidate characters may be replaced with uncertain characters and combined with candidate character strings, or the upper and lower candidate characters may be combined with candidate character strings, respectively. Is determined. In this embodiment, the “kana only” mode is set. In this case, in the system setting of 104 (see FIG. 13), “35” of the relative reference value (kana) is set.
The following 123 to 128 are processed with reference to “00”. Note that the relative reference value indicates a reference value for determining whether or not a candidate character having low accuracy is replaced with an uncertain character and is set as a candidate for combining candidate character strings.

123: The distance value of the first candidate character in the third column is subtracted from the distance value of the second candidate character, and the difference value is processed as a calculated value at 124. In the present embodiment, the first candidate character in the third column is “H” and its distance value is “31”.
Since “6” is the second candidate character and “3000” is returned as the distance value, the calculated value becomes “2684” as a result of the subtraction process.

124: The calculated value of 123 is compared with the relative reference value set in the system setting of 104, and the process branches depending on the magnitude of the value. In the present embodiment, the calculated value “268
Since "4" is a value smaller than "3500" set by the relative reference value (kana), it is processed at 125.

Since the calculated value is smaller than the relative reference value at 125: 124, the comparison between the calculated value of 124 and the relative reference value is terminated, and the processing is continued at 126.

At 126: 125, the candidate character is replaced with the uncertain character "@" and added to the candidate character string combination target. In the present embodiment example, the candidate characters of the first candidate character “H” and the second candidate character “G” in the third column are replaced with the candidate character Add to

At 127: 126, the candidate character is branched depending on whether the candidate character is the last column. In this embodiment, since the candidate character “＠” is not the last column, the second reference character in the next column of 128 is processed by the absolute reference value determination unit.

128: The second reference candidate character in the next column is processed by the absolute reference value determination unit. In the present embodiment, the second reference character “me” in the fourth column is processed by the absolute reference value determination unit 108. That is, the same processing as above is repeated for the second candidate character “me” in the fourth column in order from the absolute reference value of 108. As a result, the calculated value of the first candidate character “me” and the second candidate character “nu” in the fourth column is “20514” by the subtraction processing of the distance value, and the final reference value (kana)
Since the value is larger than “6000” set in the above, only the first candidate character “me” is targeted for combination of candidate character strings. Similarly, for the first candidate character "te" and the second candidate character "$" in the fifth column, only one character of the first candidate character "te" is subjected to the combination of candidate character strings according to the determined reference value. In the same manner, the sixth and seventh columns are also subject to the combination of candidate character strings, with the first candidate character “゛” in the sixth column and “A” in the seventh column each being one character. The 108th absolute reference value determination unit processes the second-place candidate character “ta” in the eighth column (final column).

108: In the present embodiment, the "kana only" mode is set. In this case, in the system setting of 104 (see FIG. 13), referring to "19000" of the absolute reference value (kana). , 109-111 below
Process.

109: Since this processing is for the first candidate character in the first column to the first candidate character in the last column, the second candidate character in the eighth column is processed in 110.

110: For the characters in the eighth column, the distance value returned from the character recognition AP 107-2 for the second to tenth candidate characters and the absolute value set in the system setting 104 The reference value is compared, and branching is performed depending on the value. In the present embodiment, the second candidate character in the eighth column is "A" and its distance value is "11200" (see FIG. 17), and the third candidate character in the eighth column is "MA". Assume that "19275" is returned as the distance value.
At this point, the third candidate character "ma" is the absolute reference value (kana)
Since the value is larger than "19000" set in step 11,
9 is processed.

At 119: 110, the third-place candidate character “ma” in the eighth column was a distance value “19275”, which was larger than “19000” set by the absolute reference value (kana). The comparison of the reference values is terminated, and the process is continued at 120.

As a result of 120: 110, only the candidate characters whose distance value is within the absolute reference value are subjected to the fixed reference value determination. In the present embodiment, the second-place candidate character “A” is smaller than “19000” set by the absolute reference value (Kana) with the distance value “11200” for the character in the eighth column. Is the target of the determination of the reference value. Further, since the third-place candidate character "ma" is a distance value "19275" and is a value larger than "19000" set by the absolute reference value (kana), "ma" is excluded from the determination of the fixed reference value. As a result, the eighth column is processed by the 112 fixed reference value determination units for one character of the second candidate character.

112: In the final reference value determination section,
Based on the candidate characters returned from the character recognition AP of No. 2 and their distance values (see FIG. 17), candidate characters for which a candidate character string is to be created are closely examined. In the present embodiment, the “kana only” mode is set. In this case, in the system setting of 104 (see FIG. 13), “60” of the fixed reference value (kana) is set.
The following 113 to 116 are processed with reference to “00”.

113: The distance value of the first candidate character in the eighth column is subtracted from the distance value of the second candidate character, and the difference value is processed as a calculated value in 114. In the present embodiment, the first candidate character in the eighth column is “ta” and its distance value is “70”.
Since “00” is the second candidate character and “11200” is returned as its distance value, the calculated value becomes “4200” as a result of the subtraction processing.

114: The calculated value of 113 is compared with the fixed reference value set in the system setting of 104, and the process branches depending on the magnitude of the value. In the present embodiment, the calculated value "420
Since “0” is a value smaller than “6000” set as the fixed reference value (Kana), it is processed at 121.

At 121: 114, since the calculated value was smaller than the fixed reference value, the comparison between the calculated value of 114 and the fixed reference value was terminated, and the upper candidate character and the lower candidate character were set as targets of the relative reference value determination. I do. In the present embodiment, the relative reference value determination unit 122 processes two characters of the first candidate character “ta” and the second candidate character “a” in the eighth column.

122: In the relative reference value determination section,
Based on the candidate characters returned from the character recognition AP of No. 2 and their distance values (see FIG. 17), candidate characters for which a candidate character string is to be created are closely examined. In the present embodiment, the “kana only” mode is set. In this case, in the system setting of 104 (see FIG. 2), “350” of the relative reference value (kana) is set.
The following 123 to 128 are processed with reference to “0”.

123: The distance value of the first candidate character in the eighth column is subtracted from the distance value of the second candidate character, and the difference value is processed as a calculated value at 124. In the present embodiment, the first candidate character in the eighth column is “ta” and its distance value is “70”.
Since “00” is the second candidate character and “11200” is returned as its distance value, the calculated value becomes “4200” as a result of the subtraction processing.

124: The calculated value of 123 is compared with the relative reference value set in the system setting of 104, and the process branches depending on the value. In the present embodiment, the calculated value "420
Since "0" is a value larger than "3500" set by the relative reference value (kana), it is processed in 129.

Since the calculated value was larger than the relative reference value at 129: 124, the upper and lower candidate characters are set as a candidate for combining candidate character strings. In the present embodiment, the first candidate character "T" and the second candidate character "A" are determined based on the relative reference value in the eighth column. Are added to the candidate for combining candidate character strings.

130: Branch depending on whether or not there is a candidate character for which the relative reference value is determined below. In the present embodiment, for the second candidate “A” in the eighth column and thereafter,
Since it is not a target, it is processed in 127.

At 127: 130, the candidate character branches depending on whether it is the last column. In the present embodiment, since two candidate characters of the first candidate character “ta” and the second candidate character “a” in the eighth column are the last column, the process is performed by the 131 candidate character string combination process. I do.

131: 108 absolute reference value determination unit, 11
A candidate character string is created by combining candidate characters scrutinized by the final reference value determination unit 2 and the relative reference value determination unit 122.

132: A candidate character string is created from candidate characters for which the candidate character strings have been combined. In the present embodiment, the candidate character in the first column is one character “ma”, the candidate character in the second column is one character “ru”, the candidate character in the third column is two characters “chi” and “ , The candidate character in the fifth column is one character, the candidate character in the sixth column is one character, the candidate character in the seventh column is one character, and the candidate character in the last column is The two characters “TA” and “A” are created as candidate character strings. The candidate character strings created in the present embodiment are of four types: "multi-media,""multi-media,""multi-media," and "multi-media."

133: The allowable uncertain character reference value determination unit calculates the ratio of the uncertain character "@" to the candidate character string created in 132, and scrutinizes the candidate character string according to the allowable range. In this embodiment, since the “kana only” mode is set, in this case, the system setting of 104 (FIG. 14)
), The process is performed in the following 134 with reference to the conversion allowable ratio (kana) “84”.

134: Branch depending on whether or not there are a plurality of candidate character strings. In the present embodiment, since there are a plurality of candidate character strings (four), processing is performed at 135. If there is not a plurality, processing is performed at 138.

135: Branch by comparing the uncertain character ratio with the allowable uncertain character reference value. In this embodiment, the “kana only” mode is set. In this case, in the system setting of 104 (see FIG. 14), “8” set with the minimum number of characters (kana) within the maximum uncertain character reference value is set. Then, the number of “$” that can be permitted in one candidate character string is determined from “84” set by the conversion allowable ratio (Kana). According to the calculation method, 84% does not allow “$” for 8 characters in one candidate character string. Conversely, for an 8-character candidate string, 1
6% accepts "＠". That is, in this case, two characters in the eight character candidate character string allow "@". As a result, 13
Since the second-place candidate character string “multi-meta-ita” of No. 2 does not include the uncertain character “＠”, the value becomes smaller than the allowable uncertain character reference value, and is processed at 136.

As a result of 136: 135, the candidate character string "multi-me- iterator" is set as a candidate for the candidate character string, and is processed at 137.

137: Branch depending on whether the candidate character string is the last character string. In the present embodiment, since the candidate character string “multi-meta-data” is not the final character string, it is processed at 134.

134: Branch depending on whether or not there are a plurality of candidate character strings. In the present embodiment, since there are a plurality of candidate character strings (four), processing is performed at 135.

135: Branch by comparing the uncertain character ratio with the allowable uncertain character reference value.

In this embodiment, the "kana only" mode is set. In this case, in the system setting 104 (see FIG. 14), the minimum number of characters (kana) within the maximum uncertain character reference value is set. "8" and conversion allowable ratio (Kana)
The number of “$” that can be allowed in one candidate character string is determined by “84” set in the above. According to the calculation method, 84% does not allow “$” for 8 characters in one candidate character string. Conversely, 16% of the eight character candidate character strings allow "$". That is, in this case, two characters in the eight character candidate character string allow "@". As a result, for the second-place candidate character string of “132”, the uncertain character “
Contains only one character, the value is smaller than the allowable uncertain character reference value, and is processed at 136.

As a result of 136: 135, the candidate character string "multi-me- iterator" is set as a candidate for the candidate character string, and is processed at 137. Similarly, the uncertain character ratio and the allowable uncertain character reference value of 135 are compared for the 132nd third candidate character string “Multimedia” and the fourth candidate character string “Multimedia”. As a result, in the present embodiment,
Four types of “multi-medium,” “multi-medium,” “multi-medium,” and “multi-medium” from the first candidate character string to the fourth candidate character string are all candidate character strings and are processed at 137. Is done.

137: Branch depending on whether the candidate character string is the last character string. Since four types of 136 candidate character strings are targeted for the candidate character strings and become the final character strings, they are processed in 138.

138: Branch depending on the mode. In the present embodiment, the “kana only” mode is set. In this case, in the system setting of 104 (see FIG. 13), the following 139 to 139 are used while referring to “kana only” as the recognition character type determination. 146 is processed.

139: In the cloud point / semi-voice point processing unit, 104
In the system setting (see FIG. 13), the processing differs depending on the setting of “ON / OFF” as the turbid point / semi-turbid point.

In the system setting at 140: 104 (see FIG. 13), the process branches depending on whether the turbid point / semi-turbid point is ON or OFF. In the present embodiment, since it is set to “ON” as a cloud point / semi-voice point in the system setting of 104 (see FIG. 13), the processing is performed in 141. Note that "OF
If it is set to "F", processing is performed in 143.

141: Branch depending on whether or not the first candidate character string has a clouded point “゛” or a semi-voiced point “゜”. In the present embodiment, since the first candidate character string “multi-meta-data” includes the turbid point “゛”, it is processed in 142.

142: One additional candidate character string from which the voiced point "@" and the semi-voiced point "@" have been removed is created. In the present embodiment, one additional candidate character string “multi-meta” obtained by removing the turbid point “゛” from the first-place candidate character string “multi-meta-ita” is created and processed in 143.

143: In the separation character processing unit, the processing differs depending on the setting of "ON / OFF" as a separation character in the system setting of 104 (see FIG. 13).

In the system setting of 144: 104 (see FIG. 13), the process branches depending on whether the separation character is ON. In the present embodiment, since “ON” is set as the separation character in the system setting 104 (see FIG. 13), the processing is performed at 145.

145: Branch depending on whether or not there is a separation character "ru", "no", or "re" in the first place candidate character string. In the present embodiment, since the first candidate character string “multi-meta-data” includes the separating character “ru”, 146
To process. If no separator character is included,
Processing is performed at 147.

146: Separation characters “ru” are replaced by “no” and “re”
One additional candidate character string is created. In the present embodiment example, one additional candidate character string "Manorechimeitaita" in which one character separated from the first-place candidate character string "Multimeter" is separated into two characters "No" and "Le", Processing is performed at 147.

The candidate character strings created at the time of completion of the processing of 146 are the first candidate character string “multi-meta-data”, the second candidate character string “multi-meta-data”, and the third candidate character string “multi-method”. It is assumed that there are six types, namely, "medium", the fourth candidate character string "multimedia", the fifth candidate character string "multimeter", and the sixth candidate character string "manorechimeitaita".

147: In the dictionary DB (database) comparison processing, the processing differs depending on the setting of “used / unused” as the keyword dictionary in the system setting of 104 (see FIG. 14).

In the system setting at 148: 104 (see FIG. 14), the process branches depending on whether or not to use a dictionary DB for matching keyword data. In the present embodiment, since "dictionary A" is set to "in use" as the dictionary DB for matching the keyword data in the system setting of 104 (see FIG. 14), the processing is performed at 149. In addition, when it is set to “unused”, 1
Process at 51.

149: Branch depending on whether or not the candidate character string contains the uncertain character "@". In the present embodiment, since the second-place candidate character string “multi-meta-data” and the fourth-place candidate character string “multi-meta-data” include the uncertain character “＠”, the processing is performed at 150. If it does not include the uncertain character "@", it is processed at 151.

150: Using "dictionary A" set to "in use" from the dictionary DB, search for an uncertain character "@" from a character string registered in the dictionary DB (dictionary A) Then, the matched character string is added as a candidate character string. In the present embodiment, a character string is searched from the dictionary DB (dictionary A) for the second-place candidate character string “Maru-me-data” and the fourth-place candidate character string “Maru-me-time” including the uncertain character “＠”. I do. In the dictionary DB (dictionary A), as shown in the example of the dictionary DB in FIG. 18, “multi-media” is a noun that is assumed to be registered in the image database in advance.
Is assumed to be registered. As for the second-place candidate character string "Maru-me-taitaita", there is no matching character string for the character "ta" in the eighth column other than the uncertain character "@". Next, with regard to the fourth candidate character string "multimedia", there is a character string in which all characters other than the uncertain character "＠" match, so in this case, one "multimedia" is added as a candidate character string. I do. A specific example of a method of using the dictionary DB and an example of processing will be described in a third embodiment below.

151: The candidate character string branches depending on whether it is registered or searched. In the present embodiment, since the purpose is registration in the image database, the processing is performed in step 152.

[0151] 152: Register the candidate character string in the image database. In the present embodiment, the finally created candidate character strings are the first candidate character string “multi-medium”, the second candidate character string “multi-measure”, and the third candidate character string “multi-measure”. , The fourth candidate character string "multi-medium", the fifth candidate character string "multi-meta", the sixth candidate character string "manorechi-me-taita", and the seventh candidate character string "multi-media" The character string is linked to the extraction source image data as shown in FIG. 19 and registered in the image database.

[Embodiment 2] In this embodiment, a basic processing example up to image database search will be described (processing flows 153 to 155).

In the following embodiment, an example will be described in which the medium used for the search is a facsimile terminal, and the recognition character type selection preset in the system is set to the “kana only” mode. In addition, "Multimedia" is used as a character string of keyword data searched from the fax terminal.
Is taken as an example. In this embodiment, the same processes as 101 to 151 described in the first embodiment are executed, but the description is omitted here. However, 151
In the following, an example will be described in which the candidate character string created at the point of time when the above processing is completed is the first candidate character string “Altmedium” and only one of the above.

151: The candidate character string branches depending on whether it is registered or searched. In the present embodiment, since the purpose is to search the image database, the processing is performed in 153.

153: Search the image database for the candidate character string. In the present embodiment, character strings registered in advance in the image database are set as seven types of candidate character strings registered in 152 (see FIG. 19). The candidate character string of the keyword to be searched is one type of “Altmedium”.

154: Branch by comparing the correct answer rate of the candidate character string with the search match rate. In the present embodiment, since the “kana only” mode is set, in this case, the system setting 104 (see FIG. 14) matches “8” set by the allowable value (kana) in the DB search relation. The hit condition of the candidate character string registered in the image database and the candidate character string for searching the image database is determined based on “68” set by the rate (Kana). The calculation method requires that 68% of the characters match the number of characters in one candidate character string. Conversely, 32% may not match for an eight character candidate string. That is, in this case, two characters in the candidate character string of eight characters are hit even if the characters do not match. As a result, 152
Since two characters do not match the character string "Altometia" which is searched for the second candidate character string "Malmetaitita" and the fourth candidate character string "Malmetaia", the two candidate character strings Is hit, so it is processed at 155. Here, when the correct answer rate of the candidate character string is smaller than the search matching rate set in the system, the process returns to the search source medium in 801 by notifying that there is no corresponding data in the image database, and ends.

The above-described image database search method and a specific method of comparing the matching rate (correct answer rate) of candidate character strings with the matching rate set in the system in advance are described in the fourth section.
The embodiment will be described.

155: The data matched by the character string registered in the image database is returned to the search source medium, and the processing ends. In the present embodiment, matching image data is returned to the search source fax terminal with respect to the second-place candidate character string “multi-media” and the fourth-place candidate character string “multi-media” registered in the image database. . However, in the present embodiment, two candidate character strings are hit for the same image data, but since they are the same image data, only one image data is returned to the search source fax terminal.

[Embodiment 3] This embodiment is similar to FIG.
9 is a specific example of a method for using and processing the dictionary DB shown in the flowchart of FIG. FIG. 20 is an explanatory diagram of the present method, wherein (1) to (8) indicate keyword data (candidate character strings).

[0139] As described above, the noun and the number of characters that are assumed to be registered as keyword data in the image database are registered in the dictionary DB. When the keyword data is registered in the image database, a noun registered in the dictionary DB is referred to as necessary. More specifically, if the keyword data to be registered in the image database contains an uncertain character “＠”, the dictionary D
B, if keyword data similar to the dictionary DB is registered in advance, the word is quoted and additionally registered as keyword data in the image database. Hereinafter, description will be made in accordance with the operation flows (a) to (e) with reference to FIG.

(A) It is determined whether or not the keyword data after character recognition includes a character string including the uncertain character "@". If there is no keyword data including the uncertain character "@" (in the case of the keyword data of (7) and (8)), the keyword data after character recognition is used as it is without referring to the dictionary DB. Register in the database. If there is keyword data (3) including the uncertain character "@", the dictionary DB is referred to.

(B) The reference to the dictionary DB is performed by searching from the number of characters of the keyword data, and only the keyword data having the same number of characters is to be compared. In the present embodiment, since the keyword data (3) after character recognition is "Ultimate Window" and includes the uncertain character "@", the dictionary DB is referred to.

(C) In the dictionary DB, two keyword data of 8 characters, “multi-media” of (4) and “computer” of (5) are registered.
B is to be collated.

(D) First, "multi-media" in the keyword data (3) after character recognition is compared with "multi-media" in the dictionary DB (4). The collation is performed on a character-by-character basis, and all characters except the uncertain character “@” are targeted. In the case of "multimedia" in the dictionary DB (4), all characters other than the uncertain character "@" match, so
(4) "Multimedia" as keyword data 1
And register it in the image database. Next, (3) the "multimedia" is collated with the "computer" in the dictionary DB (5). As a result of the collation, there is a character that does not match any character other than the uncertain character “@”, and therefore “computer” in (5) is not additionally registered as keyword data in the image database.

(E) As a final result of the above processes (a) to (d), the keyword data to be registered in the image database is character-recognized "Ultimate" and (2) "Ultimate". , And (3) “Multimedia” and (4) “Multimedia” as a result of collation with the dictionary DB.

[Embodiment 4] This embodiment is different from the embodiment shown in FIG.
9 is a specific example of the embodiment of the image database search method and the process of comparing the matching rate of the candidate character string with the matching rate set in the system in advance. FIG. 21 is an explanatory diagram of this processing, and (1) to (11) show keyword data (candidate character strings).

This process searches the image database using the keyword data registered in advance in the image database and the keyword data requested to be searched. In the present embodiment, three types of image data registered in advance in the image database are defined as "multi-media" (image data A), "PC" (image data B), and "computer" (image data C). It is assumed that five keyword data are created for “multi-media”, two keyword data are created for “computer”, and two keyword data are created for “consumer” for each type of image data. Also, in the search request, two keyword data, "multi-media" and "multi-media", are created for one image data (image data D), and the search is performed using the two keyword data. Less than,
The process will be described with reference to FIG. 21 in accordance with the processing flows (a) to (h).

(A) First, (1) "Multimedia"
Search the image database with the keyword data.

(B) The “multi-medium” in (1) has eight characters in terms of the number of characters (the dull point is also one character). If the search requesting side has eight characters, only the data registered in the image database and the keyword data to be searched this time correspond to only eight characters. That is, in the keyword data search method, first, the number of characters in the keyword data is referred to, and only data having the same number of characters is searched.

(C) As a result of a search request for “multi-media” in (1), the keyword data that was compared with the match rate calculation this time is “multi-media” in (3) and “ulti-media” in (4). , "(5)" Yaruchimetima "and (10)" Consumer ". The coincidence rate is defined as a value obtained by collating the keyword data requested to be retrieved with the retrieved keyword data on a character-by-character basis, and calculating the collation result in percentage. The formula for calculating the match rate is: match rate = number of matched characters (by comparing the keyword data in the search request with the keyword data in the image database) ÷ the number of characters in the keyword data,
Becomes For example, if the keyword data for which a search request is made is “multi-media” and the searched keyword data is also “multi-media”, eight characters out of eight characters match, so the matching rate is 100 [%]. Becomes

(D) Similar to the above (c), the keyword data requested to be searched and the four keyword data to be compared in the above (c) for calculating the matching rate are collated one by one. The result is as follows. In the case of (3) “Multi-metadia”, 8 characters out of 8 characters matched, so the matching rate was 100.
[%], (4) "Ultimate timer" matches 7 characters out of 8 characters, so the match rate is 87.5 [%], and (5) "Alarm data" matches 6 out of 8 characters, so match "Consumer" with a rate of 75% and (10)
Matches 0 characters out of 8 characters, so the matching rate is 0 [%].

(E) Also, a match rate calculated by collating the keyword data requested for search described in (d) above with keyword data registered in advance on a character-by-character basis. There is a match rate to set. It is necessary to input a numerical value from the GUI on the system setting screen of FIG. 14, and is used to determine whether to return the search result to the search request source. More specifically, the matching rate is further compared with the matching rate. If the matching rate is higher than the matching rate set in the system settings, it is determined that a hit has occurred, and the image data is returned to the search request source. Also,
If the matching rate is smaller, a notification message stating that "the corresponding data does not exist" is returned. However, even if the coincidence rate is smaller, if at least one other keyword data is hit, the notification message to the effect that "there is no corresponding data" is not returned.

In this embodiment, it is assumed that “the coincidence rate set in the system setting is 80”. In this case, since the match rate of “multi-media” of (3) is 100 [%], it is determined that a hit has occurred. Since the match rate of “ultimate wear” in (4) is 87.5 [%], it is determined that a hit has occurred. (5)
Is not hit because the matching rate is 75 [%]. The "computer" in (10) has a matching rate of 0 [%], so it is determined that no hit was made. As described above, it is determined that (5) and (10) have not been hit, but since (3) and (4) have been hit, a notification message to the effect that "corresponding data does not exist" is not returned. Also, since (3) and (4) are the same image data (image data A), in this case, the same screen is not returned two times. As a result, the image data returned to the search request source is one image data A.

(F) Similar to the above (a) to (e), the image database is searched by the "multimedia" of (2).

(G) Since the "multi-media" in (2) has 7 characters, the keyword data to be compared in the calculation of the coincidence rate are "multi-media" in (6) and "commuter" in (11). The result of calculating the match rate is
In the case of “(6)”, the matching rate was 100 [%] because 7 out of 7 characters matched, and in “(11)” the matching rate was 0 since 0 out of 7 characters matched.
[%], (8) Regarding the comparison between the calculated coincidence rate and the coincidence rate set in the system, also in this embodiment, the coincidence rate set in the system setting is set to “80”. in this case,
It is determined that a hit has occurred in the “multi-media” of (6) because the matching rate is 100 [%]. Since the match rate of “Consumer” in (11) is 0 [%], it is determined that no hit was made. As described above, (6) was hit, but since the same image data (image data A) as (3) in (5) above, in this case, the same two screens are not returned. Also, (1
It is determined that no hit has occurred in 1), but since (3) has hit, a notification message stating that "corresponding data does not exist" is not returned.

(H) As the final result of (a) to (g), the image data returned to the search request source is one of the image data A.

In the program processing described in the first to fourth embodiments described above, the processing procedure can be executed by a computer, and a program for causing the computer to execute the processing procedure is provided. The program can be recorded on a computer-readable medium such as a floppy disk or a CD-ROM and distributed.

[0157]

According to the present invention, since the database access sheet can be sent to the image database center as image data, the existing fax terminal can access the image database center. Further, the database access sheet can be read as image data by a scanner, and the scanner can also access the image database center. In addition, by character recognition, keyword data and image data can be mixed and described on the database access sheet on the same page and on the same page. The description area is free format, so it has versatility and can identify the user. Therefore, an effect of facilitating access to the image database can be provided. In the case where katakana and numbers are used, high-precision character recognition can be realized by separating the two and performing character recognition using dedicated dictionaries. In addition, by making it possible to generate a plurality of candidate character strings, or by additionally generating a candidate character string from which a clouded point and a semi-voiced point are deleted, matching is facilitated at the time of search. Therefore, according to the present invention, the processing result at the image database center desired by the search requester can be efficiently transmitted to the fax terminal of the search requester.

[Brief description of the drawings]

FIG. 1 is a system configuration diagram including an image database center device according to a first embodiment of the present invention.

FIG. 2 is a functional block configuration diagram of an image database center device in the embodiment.

FIG. 3 is a diagram showing a service image by the image database center device in the embodiment.

FIG. 4 is a diagram showing a description example of a keyword in a database access sheet according to the embodiment;
(A), (b) shows a description example in which the keyword is indicated by a double line, and (c) shows an example in which the keyword is described in a region where the image data can be extracted for extracting the keyword.

FIGS. 5A and 5B are diagrams illustrating examples of character types that can be described in the database access sheet according to the embodiment, wherein FIG. 5A illustrates numbers, and FIGS. 5B and 5C illustrate katakana characters.

FIGS. 6A and 6B are flowcharts (part 1) illustrating an embodiment of a program process in the image database center device in the embodiment.

FIG. 7 is a flowchart (part 2) illustrating an embodiment of a program process in the image database center device in the embodiment.

FIG. 8 is a flowchart (part 3) illustrating an embodiment of a program process in the image database center device in the embodiment.

FIG. 9 is a flowchart (part 4) illustrating an embodiment of a program process in the image database center device in the embodiment.

FIG. 10 is a flowchart (part 5) showing one embodiment of the program processing in the image database center device in the above embodiment.

FIG. 11 is a flowchart (part 6) illustrating an embodiment of the program processing in the image database center device in the embodiment.

FIG. 12 is a flowchart (part 7) illustrating an embodiment of a program process in the image database center device in the embodiment.

FIG. 13 is a diagram showing an example of a system setting screen in the embodiment.

FIG. 14 is a diagram illustrating a continuation of the example of the system setting screen in the embodiment.

FIGS. 15A and 15B are diagrams (part 1) for explaining a method for extracting a keyword described in a database access sheet in the embodiment.

FIGS. 16 (a), (b), (c) and (d) are diagrams (part 2) for explaining a method of extracting a keyword described in a database access sheet in the embodiment.

FIG. 17 is a diagram showing an example of an output result of a character recognition AP in the embodiment.

FIG. 18 is a diagram showing a registration example of a dictionary DB in the embodiment.

FIG. 19 is an explanatory diagram of an image database for explaining a second embodiment of the present invention.

FIG. 20 is an explanatory diagram showing a usage example of a dictionary DB for explaining a third embodiment of the present invention.

FIG. 21 is a diagram illustrating a keyword search method for explaining a fourth embodiment of the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 10 ... Image database center apparatus 11 ... Scanner 12 ... Image database 13 ... Image database center main apparatus 14 ... Line board 15 ... Communication network 16, 17 ... Fax terminal 18 ... Personal computer (PC) 20 ... Paper 21 ... Keyword data description example 22 black line for extracting keyword data (double line) 23 keyword data description example 24 image data description area

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Takahiro Odo 3-19-2 Nishi Shinjuku, Shinjuku-ku, Tokyo Nippon Telegraph and Telephone Corporation

Claims (12)

[Claims] 1. An image database center for transmitting / receiving image data to / from a communication medium and registering / retrieving the image data, wherein a registration requester transmits the image data from a communication medium via a communication network or from an image scanner. A second means for extracting a candidate character string from the image data received from the registration requester and generating one or more candidate character strings with high accuracy including an uncertain character Means for registering the candidate character string generated by the second means with the image data from which the candidate character string is extracted in the image database.
Means for receiving image data from a search requester via a communication medium via a communication network; extracting candidate character strings from the image data received from the search requester; A fifth means for generating one or more candidate character strings having high accuracy including: a matching rate between the candidate character strings generated by the fifth means and the candidate character strings registered in the image database A sixth means for searching the image database for image data corresponding to a candidate character string determined to match the match rate with the preset match rate from the image database; and And a seventh means for transmitting the image data. 2. The apparatus according to claim 1, wherein the second and fifth units include a character recognizing unit, an absolute reference value determining unit, a fixed reference value determining unit, and a relative reference value determining unit. Generating a sequence, the character recognition means ranks a plurality of candidate characters as a character recognition result, the plurality of candidate characters and the corresponding distance value the absolute reference value determination unit, the fixed reference value A determination unit, and a relative reference value determination unit, wherein the absolute reference value determination unit includes the plurality of ranked candidate characters, a value of a distance value corresponding thereto, and a value of a preset absolute reference value. Comparing the value of the distance value and the value of the absolute reference value to determine whether or not the candidate character is to be combined with a candidate character string.The fixed reference value determination unit includes: The upper and lower candidate sentences The value of the upper and lower distance value corresponding to the character and the corresponding value is subtracted, the subtracted value is compared with a value of a preset reference value, and the value of the subtracted value and the value of the reference value are determined according to the magnitude of the value. It is to determine whether only the upper candidate characters are to be combined with the candidate character strings, or whether the upper and lower candidate characters are to be combined with the candidate character strings, respectively. The upper and lower candidate characters and the corresponding upper and lower distance values are subtracted, the subtracted value is compared with a preset relative reference value, and the subtracted value and the relative reference value are compared. Depending on the magnitude of the value, whether the upper and lower candidate characters are replaced with uncertain characters and the candidate character string is combined or whether the upper and lower candidate characters are combined with the candidate character string respectively What to judge The image database center device according to claim 1, wherein: 3. The method according to claim 1, wherein the sixth means includes one or more candidate character strings generated from the image data received from the registration requester, and one or more candidate character strings generated from the image data received from the search requester. When comparing the match rates with the candidate strings of the characters, the match rate between the individual strings of the candidate strings generated at the time of the registration request and the individual strings of the candidate strings generated at the The image data corresponding to a candidate character string including a character string that is calculated for each column and is determined to match with the matching rate based on a predetermined matching rate is searched from an image database. The image database center device according to claim 1 or 2. 4. A method according to claim 1, wherein the relative reference value determining unit replaces a candidate character with an uncertain character, and searches the dictionary database for a character string for the candidate character string including the uncertain character, and performs matching from the dictionary database. 4. An image database center apparatus according to claim 1, further comprising: means for additionally generating a character string to be executed as a candidate character string. 5. The method according to claim 2, wherein the candidate character string including a cloud point or a semi-voice point includes a cloud point or
Means for additionally generating a candidate character string from which the semi-voiced voice has been removed; and means for additionally generating a candidate character string in which the character is replaced with a character obtained by separating the character into two characters, for a candidate character string including the separated character. The image database center device according to any one of claims 1 to 4, wherein the image database center device is provided. 6. Sending and receiving image data to and from a communication medium,
An image database registration / search method provided in an image database center for registering / searching image data, comprising: a first procedure of receiving image data from a registration requester from a communication medium via a communication network or an image scanner; A second procedure of extracting a candidate character string from the image data received from the registration requestor and generating one or more candidate character strings having high accuracy including an uncertain character; and Register the candidate character string generated in the procedure 2 in the image database in association with the image data from which the candidate character string is extracted.
And a fourth step of receiving image data from a communication medium of the search requester via a communication network; and extracting a candidate character string from the image data received from the search requester by character recognition software. A fifth procedure of generating one or more candidate character strings having high accuracy including an uncertain character; and a candidate character string generated in the fifth procedure and a candidate character string registered in the image database. A sixth procedure of comparing image data corresponding to a candidate character string determined to match with the coincidence rate with a preset coincidence rate from the image database; and a communication medium of the search requester. A seventh procedure of transmitting the searched image data to the image database registration / search method. 7. The character recognition software according to the second and fifth procedures, when performing character recognition of a character string in which katakana and numbers are mixed, the character string in which katakana and numbers are mixed includes katakana and numerals. Characters are distinguished using katakana and numbers before and after the delimiter, using katakana-only dictionaries for the katakana part and numeric-only dictionaries for the number part. The image database registration / search method according to claim 6. 8. The second and fifth procedures include: ranking a plurality of candidate characters as a character recognition result of character recognition software, and returning the plurality of candidate characters and a distance value corresponding to the plurality of candidate characters; A plurality of ranked candidate characters, a value of a distance value corresponding thereto and a value of a preset absolute reference value are compared, and the candidate character is determined as a candidate character according to the magnitude of the distance value and the value of the absolute reference value. An absolute reference value determination procedure for determining whether or not to combine the columns; and subtracting the upper and lower candidate characters and the corresponding upper and lower distance values, and setting the subtracted value to a preset value. The value of the fixed reference value is compared, and the value of the subtracted value and the value of the fixed reference value are compared to determine whether only the upper candidate character is to be combined with the candidate character string or the upper and lower candidate characters. Each A fixed reference value determination procedure for determining whether or not the candidate character string is to be combined; subtracting the upper and lower candidate characters and the corresponding upper and lower distance values, and setting the subtracted value to a preset value By comparing the value of the relative reference value, by the value of the subtracted value and the value of the relative reference value, the upper and lower candidate characters are replaced with uncertain characters and the combination of candidate character strings, or A relative reference value determination step of determining whether the upper and lower candidate characters are to be combined with the candidate character strings, respectively, comprising: generating the candidate character string by combining the candidate characters based on the determination. The image database registration / retrieval method according to claim 6 or 7, wherein: 9. The method according to claim 6, wherein one or more candidate character strings generated from the image data received from the registration requester and one or more candidate character strings generated from the image data received from the search requester are included in the sixth procedure. When comparing the match rates with the candidate character strings of the above, the match rate between the individual character strings of the candidate character group generated at the time of the registration request and the individual character strings of the candidate character group generated at the 7. The image data corresponding to a candidate character string including a character string that is calculated for each column and determined to match with the matching rate based on a predetermined matching rate is searched from an image database. Item 9. The image database registration / search method according to any one of Items 8 to 8. 10. The relative reference value determining step includes: replacing a candidate character with an uncertain character; searching a dictionary database for a candidate character string including the uncertain character, and performing matching from the dictionary database. The image database registration / retrieval method according to claim 8, comprising: a step of additionally generating a character string to be executed as a candidate character string. 11. The second and fifth procedures are as follows: a candidate character string including a cloud point or a semi-voice point includes a cloud point or
A step of additionally generating a candidate character string from which the semi-voiced voice is removed; and a step of additionally generating a candidate character string in which the character is replaced with a character obtained by separating the character into two characters for the candidate character string including the separated character. The image database registration / search method according to any one of claims 6 to 10, wherein the method is provided. 12. A program for causing a computer to execute the procedure of the image database registration / retrieval method according to any one of claims 6 to 11, wherein the program is recorded on a computer-readable medium. Recording medium.