JPH06103404A - Business card recognition device - Google Patents

Abstract

PURPOSE:To more correctly correct the error of a dwelling place read out of a business card. CONSTITUTION:A storage means to store the dwelling place and a telephone number at that dwelling place as correlating them with each other, an image scanner 5 as a reading means to read the image of the business card, and a character recognizing device 6 as a detecting means to detect the dwelling place and the telephone number out of the image of the business card read by the image scanner 5 are provided, and the telephone number coincident with the telephone number detected by the character recognizing device 6 is retrieved, and the dwelling place detected by the character recognizing device 6 is corrected on the basis of the dwelling place coordinated with that telephone number. Further, it is made more correct by a postal code.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a business card recognition device suitable for use in correcting an error in an address read from a business card.

[0002]

2. Description of the Related Art In a conventional business card recognition device, an image of a business card is read, the image is recognized, and a character string corresponding to, for example, a name, an address, a telephone number, or a postal code is cut out from the business card. Has become.

The name, address, telephone number, zip code, etc. as a character string are recorded in a recording medium such as a disk and made into a database, so that
If necessary, for example, a telephone number can be searched for.

[0004]

By the way, from the image of a business card, there are many characters, such as kanji, hiragana, katakana,
Alternatively, it is difficult to accurately recognize an address containing a mixture of numbers, as compared with the case of recognizing a telephone number or a postal code that is composed almost entirely of numbers.

Therefore, there is a method of correcting an address using a postal code associated with the address. That is, there is a method of comparing an address which is limited to a certain extent by the postal code with an address where the character is recognized and correcting the address where the character is recognized so as to match any of the addresses limited to the postal code. .

However, this method can correct the address error to some extent, but it is difficult to correct the error more accurately.

The present invention has been made in view of such a situation, and it is possible to more accurately correct an error in an address read from a business card.

[0008]

According to another aspect of the present invention, there is provided a business card recognition device including: a RAM 13 as a storage means for storing an address and a telephone number at the address in association with each other; and a reading means for reading an image of the business card. And a character recognition device 6 as detection means for detecting an address and a telephone number from an image of a business card read by the image scanner 5, and a telephone number detected by the character recognition device 6 from the RAM 13. Processing steps S2 to S9 of the program as a correction unit that searches for a matching telephone number and corrects the address detected by the character recognition device 6 based on the address associated with the telephone number.
And is provided.

The business card recognition device according to claim 2 is a RAM
The address and the postal code at the address are stored in association with each other in 13, and the character recognition device 6 is caused to further detect the postal code from the image of the business card. From the RAM 13 to the processing steps S2 to S9 of the program. , A postal code that matches the postal code detected by the character recognition device 6 is searched, and the address detected by the character recognition device 6 is further corrected based on the address associated with the postal code. .

The business card recognition device according to claim 3 is a RAM
13, the address, the telephone number at the address, or the zip code at the address is hierarchically stored.

[0011]

In the business card recognition device having the above structure, the RAM 1
The address and the telephone number or zip code at the address are stored in association with each other in 3. Then, the image of the business card is read, and the address, telephone number, or zip code is detected from the image. Further, the RAM 13 is searched for a telephone number or a zip code that matches the telephone number or zip code detected by the character recognition device 6, and based on the address associated with the telephone number or the mail, the character recognition device 6 Correct the detected address. Therefore, an error in the address read from the business card can be corrected more accurately.

[0012]

1 is a block diagram showing the configuration of an embodiment of a business card recognition device of the present invention, and FIG. 2 is an overall view thereof. WS (workstation) 1 has CPU 11 and R
It is composed of an OM 12 and a RAM 13, and an interface 14 and an image board 15 are inserted into its expansion slot (not shown). The CPU 11 controls the entire apparatus and reads and executes the system program stored in the ROM 12 or the user program stored in the RAM 13.

A system program is stored in the ROM 12. The RAM 13 stores the user program and also data necessary for the operation of the device. Further, the RAM 13 stores a dictionary in which an address as shown in FIG. 3, a postal code at the address, and a telephone number, for example, an area code are associated with each other.

In the figure, the symbol "*" added to the end of the zip code indicates the keyword of the zip code, and the symbol "+" added to the end of the area code indicates the keyword of the area code. Show.

Further, in the RAM 13, for example, "city", "road", "prefecture", "prefecture", etc.
A keyword table including “ward”, “county”, “city”, “village”, or “town”, a zip code keyword “*”, and a telephone number keyword “+” is stored (FIG. 5). ).

The interface 14 manages input / output with the character recognition device 6 (interface 22) in place of the CPU 11. The image board 15 manages input / output with the image scanner 5 or the printer 7 instead of the CPU 11.

The display 2 displays character data and graphic data supplied from WS1. The keyboard 3 or the mouse 4 is operated, for example, when a command or data necessary for operating the device is input.

The image scanner 5 reads the image of the business card set there and outputs the image data to WS1.
(Interface 14).

The character recognition device 6 uses an OCR (Optical Char
acter Reader) board 21 and interface 22
Composed of. The OCR board 21 is WS1 (RAM
13) The image data (character data) of the business card read by the image scanner 5, which is supplied from 13), is recognized.
The interface 22 manages the input of data to the character recognition device 6 or the output of data therefrom.

The printer 7 prints out characters or graphics according to the print data supplied from WS1.

Next, referring to the flowchart of FIG.
The operation will be described. First, the image scanner 5
A business card is set in, the image is read, and image data is generated. The image data of the business card is stored in the RAM 13 of the WS 1 via the image board 15 and then supplied to the character recognition device 6 via the interface 14.

In the character recognition device 6 (OCR board 21), the character pattern stored in the built-in memory (not shown) is referred to, and the image data of the business card supplied from WS1 (RAM 13) is recognized. It Further, in the character recognition device 6, character strings corresponding to a postal code, a telephone number, or an address are extracted from the image data subjected to character recognition.

Then, in step S1 of FIG.
The dictionary (FIG. 3) stored in the AM 13 is expanded as shown in FIG. That is, in the dictionary, first, the address keywords "city", "road", "fu", or "prefecture" are special keywords belonging to the highest hierarchy, and are classified into four categories together with the prefecture names attached to them. . Then, in the lower hierarchy of each special keyword “city”, “road”, “prefecture”, or “prefecture”, there is a place name (for example, a municipality name, a ward name, a county name, etc.) under the prefecture name. , Hierarchically (on the tree structure) along with keywords attached to the place name (eg "city", "town", "village", "ward", or "county")
Be expanded.

Further, the zip code or area code is expanded in the hierarchy below the corresponding address, together with the keywords "*" or "+", respectively.

In one category, keywords or place names belonging to the same hierarchy are connected by one chain (for example, in the figure, keywords "ward", "village", and "village" belonging to the lower hierarchy of Tokyo. "City" is connected by one chain). In addition, the same keyword in different categories (for example, the keyword "city" that belongs to the lower hierarchy of Tokyo and the keyword "city" that belongs to the lower hierarchy of Saitama prefecture) can be searched so that you can search for keywords or place names across categories. They are tied together in one chain.

The address (pointer) of the keyword in the dictionary expanded as described above is stored together with the keyword in the keyword table in the RAM 13 (FIG. 5).

After the dictionary is developed in step S1, the area code extraction processing, the keyword extraction processing of the address, or the postal code extraction processing in steps S2 to S4 is performed in parallel.

That is, in step S2, first, from the character string corresponding to the telephone number of the business card output from the character recognition device 6, a keyword peculiar to the telephone number (for example, the character strings "TEL", "Tel", "telephone" , “Telephone number”, phone mark, etc.) are removed, and if they contain Kanji numbers, they are converted to numbers (alphanumeric characters).

When the first character of the character string subjected to the above-mentioned processing is the numeral "0", the character immediately before the delimiter "-" or "(" is deleted from the first character of the character string. Up to the character, or when there is no delimiter "-" or "(", the first to sixth characters of the character string are extracted.

The first character of the character string subjected to the above-mentioned processing is "(", and the second character is the numeral "0".
Is from the second character of the string (the number "0") to the character before the delimiter ")", or when there is no delimiter "(", the first character of the string Characters up to the sixth character are extracted.

As described above, the area code is extracted from the telephone number, the keyword "+" is added to the end of the area code, and the process proceeds to step S5.

If the first character of the character string subjected to the above-mentioned processing is neither the numeral "0" nor the character "(", the area code is not extracted from the telephone number.

On the other hand, in step S3, first, each character forming the character string corresponding to the address of the business card output from the character recognition device 6 is collated with the keyword stored in the keyword table to obtain the address. From the corresponding character string, the keyword of the address and its position (what character) are detected.

Here, if the keyword of the address cannot be detected from the character string corresponding to the address, for example, only 10 prepared beforehand, candidate characters for each character constituting the character string (each character Character similar to)
And the keywords stored in the keyword table are compared with each other. If no keyword is detected from the candidate characters, the process is stopped.

After the address keyword and its position are detected from the character string corresponding to the address, a character string for address search is cut out from the character string.

That is, from the beginning of the character string corresponding to the address,
Up to the keyword of the address located at the front of the character string is cut out, and the character string obtained by removing the first character from the cut out character string is repeated until the number of characters of the address keyword becomes one. It is cut out.

The above processing is performed for the character string 2 corresponding to the address.
The keyword of the address located after the th is also sequentially performed, and the character string for address search is cut out, and step S6 is performed.
Proceed to.

That is, the character string corresponding to the address output from the character recognition device 6 is, for example, "3-1 Shibamata, Katsushika-ku, Kyoto.
9-3 "(the exact address is" 3-19-3 Shibamata, Katsushika-ku, Tokyo ", but in the character recognition device 6,
It is assumed that the first character "east" is erroneously recognized as a character "bundle"). In step S3, first, the keywords "tou" and "ward" of the address are detected from this character string and correspond to the address. From the character string, based on this keyword, the character string for address search (search character string) "Bundle Kyoto""Kyoto""Bundle Kyoto Katsushika Ward""Kyoto Katsushika Ward""Matsu Katsushika Ward""KatsushikaWard""Decoration" The ward will be cut out.

At the same time, in step S4, the postal code keyword "*" is added to the end of the character string corresponding to the postal code of the business card output from the character recognition device 6,
Go to step S7.

As described above, steps S2 to S4
After the processing in step S5 is performed in parallel, steps S5 to S
7, the address is selected from the dictionary developed in step S1 based on the area code, search character string, or zip code.

That is, in step S5, first, the keyword "+" added to the end of the area code extracted in step S2 is detected from the keyword table, and from there, the dictionary of that keyword "+" is added. (Fig. 5)
The address stored in is read. Then, the area code stored together with the keyword "+" is read out from the dictionary based on the address and collated with the area code extracted in step S2.

As a result of the collation, if the area code that matches (matches) the area code extracted in step S2 is stored in the dictionary, it is stored in the upper hierarchy connected to the hierarchy of the area code. The belonging address (place name) is read out, stored as a candidate address (correct address candidate) in a predetermined area secured in the RAM 13, and the process proceeds to step S8.

When the area code that matches (matches) the area code extracted in step S2 is not stored in the dictionary, for example, only 10 prepared area codes are prepared in advance and the area extracted in step S2 is stored. Candidate characters (characters similar to each character) for each character (each number) as an area code and each character that constitutes the area code stored in the dictionary are collated and extracted in step S2. The area code and the area code stored in the dictionary are matched.

On the other hand, in step S6, first, in step S6
The last character of the search character string cut out in 3, that is, the keyword of the address is detected from the keyword table, and the address of the keyword stored in the dictionary (FIG. 5) is read therefrom. And from the dictionary
Based on the address, the place name (including the prefecture name) stored with the address keyword is read out,
The search character string cut out in step S3 is collated with the character string excluding the last character (address keyword).

As a result of the collation, when a place name that matches (matches) the character string excluding the last character (address keyword) of the search character string cut out in step S3 is stored in the dictionary, the place name is stored. All the place names belonging to the upper and lower layers connected to the hierarchy are read out, and an address obtained by combining these place names is stored in a predetermined area secured in the RAM 13 as a candidate address (correct address candidate). Then, the process proceeds to step S8.

That is, for example, when the search character string is "Katsushika-ku", a place name that matches (matches) with the character string "Katsushika" excluding the last character (keyword of the address) "ku" of this character string All the place names (addresses) (the shaded portions in FIG. 6) belonging to the upper and lower layers connected to the "Katsushika" layer are read out, and the address obtained by combining these place names is the candidate address. It is stored in a predetermined area secured in the RAM 13 as (correct address candidate).

As in the case of step S5, also in step S6, the last character (keyword of address) of the search character string cut out in step S3.
If there is no place name stored in the dictionary that matches (matches) with the character string excluding, the last character (keyword of the address) of the search character string cut out in step S3 prepared in advance, for example, 10 characters The candidate characters (characters similar to each character) for each character forming the character string excluding the character string and each character forming the place name stored in the dictionary are collated with each other, and the search cut out in step S3 is performed. The last character of the string (address keyword)
Matching is performed on the character string excluding "" and the place name stored in the dictionary.

At the same time, in step S7, the keyword "*" added to the end of the postal code extracted in step S4 is detected from the keyword table.
Further from that, the dictionary of the keyword "*" (Fig. 5)
The address stored in is read. Then, the postal code stored together with the keyword "*" is read out from the dictionary based on the address, and collated with the postal code extracted in step S4.

As a result of the collation, when the postal code that matches (matches) the postal code extracted in step S4 is stored in the dictionary, the address (that belongs to the upper hierarchy connected to the hierarchy of the postal code ( The place name is read out and stored as a candidate address (correct address candidate) in a predetermined area secured in the RAM 13, and the process proceeds to step S8.

As in the case of step S5, also in this step S7, when the postal code matching (matching) with the postal code extracted in step S4 is not stored in the dictionary, for example, only 10 postal codes are stored. Candidate characters (characters similar to each character) for each character (each number) as the postal code extracted in step S4 prepared in advance, and each character that constitutes the postal code stored in the dictionary, respectively. By collating, the zip code extracted in step S4 and the zip code stored in the dictionary are matched.

As described above, steps S5 to S7
Then, the candidate address is selected. Note that step S5
If the candidate addresses are redundantly selected in S7 to S7, the candidate addresses are deleted (ignored) except for the first selected address, for example.

Then, the process proceeds to step S8 and the RAM 13
All of the candidate addresses stored in and the character string corresponding to the address output from the character recognition device 6 are respectively collated,
From among them, the one corresponding to the address output from the character recognition device 6 is determined.

That is, in step S8, RAM1
Whether each character forming the candidate address stored in No. 3 and each character forming the character string corresponding to the address output from the character recognition device 6 are sequentially collated from the beginning and match (match). Is determined. Then, for each candidate address stored in the RAM 13, a correct answer rate p p = (the number of characters determined to match) / (the number of collated characters) defined by the following equation is calculated.

Here, in the character string corresponding to the address output from the character recognizing device 6, a character or a character string irrelevant to the true (correct) address is erroneously detected at the beginning of the character string in the processing process. It is possible that the character or character string corresponding to the true (correct) address is accidentally deleted at the beginning of the message.

Therefore, when the character string corresponding to the address output from the character recognition device 6 and the candidate address are sequentially collated from the first character, the correct accuracy rate p may not be obtained.

Therefore, the character string of the candidate address stored in the RAM 13 is collated with the character string corresponding to the address output from the character recognition device 6 while deleting the character string from the leading character one by one. , The character string corresponding to the address output from the character recognition device 6 is collated with the character string of the candidate address stored in the RAM 13 while deleting the character from the leading character one by one, and the correct answer rate p is It is calculated.

[0057] Then, when the maximum value of the accuracy rate p was p _max, (p _max -C) or better rate p is among the candidate address that is calculated, as the largest number of characters is detected, step S9 Proceed to.

C is a predetermined coefficient, such as 0.135, which is obtained in advance by a statistical method.

In step S9, the address output from the character recognition device 6 is corrected by the candidate address detected in step S8, and the process ends.

By the above processing, when the character string corresponding to the address output from the character recognition device 6 is, for example, "3-10-3 Shibamata, Shibamata, Katsushika-ku, Kyoto", the first character "bun" is The characters are corrected to "east", and the correct address "3-19-3 Shibamata, Katsushika-ku, Tokyo" is output.

The corrected address is recorded on, for example, a magnetic disk (not shown) and used as a database.

In this embodiment, the area code and the address of the telephone number are associated with each other. However, not only the area code but also the local area code is associated with the address. You can

[0063]

As described above, according to the business card recognition device of the present invention, the address and the telephone number or the postal code at the address are associated and stored in the storage means. Then, read the image of the business card, and from that image,
Find addresses, phone numbers, or zip codes. further,
The storage means is searched for a telephone number or postal code, respectively, which matches the telephone number or postal code detected by the detection means, and the address detected by the detection means is determined based on the address associated with the telephone number or postal mail. correct. Therefore, an error in the address read from the business card can be corrected more accurately.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of a business card recognition device of the present invention.

FIG. 2 is an overall view of the embodiment shown in FIG.

FIG. 3 is a diagram showing a dictionary in which an address, a zip code at the address, and an area code are associated with each other.

FIG. 4 is a flow chart for explaining the operation of the embodiment of FIG.

5 is a diagram showing how the dictionary of FIG. 3 is expanded in a hierarchical structure.

FIG. 6 is a diagram for explaining selection of an address in step S6 of the flowchart of FIG.

[Explanation of symbols]

 1 WS (Workstation) 2 Display 3 Keyboard 4 Mouse 5 Image Scanner 6 Character Recognition Device 7 Printer 11 CPU 12 ROM 13 RAM 14 Interface 15 Image Board 21 OCR (Optical Character Reader) Board 22 Interface

Claims (3)

[Claims] 1. A storage unit for storing an address and a telephone number in the address in association with each other, a reading unit for reading an image of a business card, and an image of the business card read by the reading unit,
Detecting means for detecting an address and a telephone number, and searching the storage means for a telephone number that matches the telephone number detected by the detecting means, and based on the address associated with the telephone number, the detecting means. A business card recognition device, comprising: a correction unit that corrects the address detected by. 2. The storage means further stores the address and a zip code in the address in association with each other, and the detecting means further detects a zip code from the image of the business card, and the correcting means. Searching the storage means for a postal code that matches the postal code detected by the detection means, and further correcting the address detected by the detection means based on the address associated with the postal code. The business card recognition device according to claim 1, wherein: 3. The business card recognition device according to claim 2, wherein the storage unit hierarchically stores the address, the telephone number at the address, or the postal code at the address.