JPH02181287A - Address recognizing device - Google Patents

Abstract

PURPOSE:To attain an address recognizing system with high flexibility and high allowance by recognizing the attribute and the position of each of plural words out of a read character string and recognizing an address in accordance with the word position. CONSTITUTION:The output of a character recognizing part 13 is supplied to a word recognizing part 14. The part 14 recognizes the characters supplied from the part 13 and to be recognized based on the words read out of a word dictionary 15 and at the same time reads the attribute information corresponding to the corresponding candidate word out of the dictionary 15. The output received from the part 14, i.e., a candidate word and its attribute information and the position information on each word are supplied to an address recognizing part 16. The part 16 decides a ZIP code, a state name, a town name, a street name, an address, etc., based on the candidate word and its attribute information. Thus the part 16 recognizes an address. Then the section designating data is outputted to the recognized address.

Description

[Detailed Description of the Invention] [Objective of the Invention] (Industrial Application Field) This invention relates to an address system that recognizes the address written on the mail from the mail being conveyed, for example in an automatic address reading and sorting machine. Relating to a recognition device.

(Prior art) Postal sorting operations can be roughly divided into dispatch sorting operations, which are the sorting operations for sending the collected mail to the destination post office, and sorting operations, which are the sorting operations for delivering the arrived mail. There is a certain delivery classification business.

In the past, dispatch sorting operations have been largely automated using automatic postal code reading and sorting machines, but delivery sorting operations have lagged behind in automation.

However, in recent years, there has been a growing movement among postal ministries around the world to automate the sorting of mail down to the delivery person level (automation of delivery sorting operations) by reading addresses.

The technology for reading addresses using OCR (optical character reader) differs from conventional postal code automatic reading technology because it requires the machine to have address knowledge and can only read numbers or a mixture of alphanumeric characters. This is extremely difficult compared to postal code reading technology.

For example, a character string that constitutes an address is optically read from a piece of mail, and words are recognized from this read character string.

Then, using the hierarchical structure of addresses and the basic knowledge of address writing, from the bottom line to the bottom line, the foreign name, postal code (ZIP), state name, city name, street name, building name, It searches in the order of company names and recognizes addresses.

Therefore, if there is even a slight difference from the basic knowledge regarding the hierarchical structure of addresses and address description, the address cannot be recognized, resulting in inflexible address recognition and low tolerance.

(Problems to be Solved by the Invention) As mentioned above, if there is even a slight difference from the basic knowledge regarding the hierarchical structure of addresses and address description, it will be impossible to recognize addresses, there will be no flexibility in address recognition, and there will be limited tolerance. This eliminates the drawback of low address recognition, and allows address recognition even if there is even a slight difference from the basic knowledge regarding the hierarchical structure of addresses and address descriptions, and has high flexibility in address recognition. An object of the present invention is to provide an address recognition device capable of highly accurate address recognition.

[Structure of the Invention] (Means for Solving the Problems) The address recognition device of the present invention recognizes an address by detecting character information, and includes a character reading means for reading a character string constituting an address; word recognition means for recognizing a plurality of words, attributes of each word, and position of each word from a character string read by the means; It consists of an address recognition means that recognizes the address according to the address.

(Operation) The present invention detects character information to recognize an address, reads a character string constituting an address, and from this read character string, a plurality of words, the attributes of each word, and the position of each word. This system recognizes addresses based on the recognized words, the attributes of each word, and the position of each word.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings.

FIG. 2 shows an example of an object to be read, such as a mail P, according to the present invention, on the surface of which address information 1 and the like are written in the horizontal direction. The address information 1 is each composed of multiple lines of character information. In such mail P, address information 1 generally includes foreign name, postal code (ZIP), state name, city name, street name,
Bostal station name, building name, company name, street address, etc. are listed. This mail P is conveyed in the direction of arrow a in the figure.

FIG. 1 schematically shows an optical character reading device used in an automatic address reading and sorting machine. That is, reference numeral 11 denotes a scanning section (photoelectric conversion section), which is provided along a transport path (not shown) for mail P that is transported in the direction of arrow a.

This scanning unit 11 obtains a pattern signal by optically scanning the postal item P and photoelectrically converting it. It is composed of a self-scanning COD image sensor that converts into signals.

The output of the scanning section 11 is supplied to an address area detection section 12. This address area detecting unit 12 is a well-known device that detects an area where address information 1 is written based on the shape, position, etc. of the area from among each piece of information written on the postal item P.

The determination result of the address area detection section 12, for example, the coordinate values of the determined area, is output to the selection circuit 21 in the character recognition section 13.

Further, the output of the scanning section 11 is output to a binarization circuit 20 within the character recognition section 13.

” The bipedal character recognition unit 13 includes a binarization circuit 20 that binarizes the read signal from the scanning unit 11, and the address area detection unit 1.
A selection circuit 21 that selectively outputs only the binary signal from the binary conversion circuit 20, that is, the read signal from the scanning unit 11, corresponding to the coordinate information supplied from the selection circuit 21; As shown in Figure 3(a), the address information 1 corresponding to the address area to be searched is detected and extracted line by line, and based on the detection and extraction results for that line, the size of the extracted character cluster is calculated. Then, based on position information, etc., as shown in Figure (b), we detect and cut out each character one by one, and as shown in Figure (c), we calculate the inter-character gap and calculate the character gap in advance. A part larger than a predetermined value is determined to be a break between words, and the character blocks are grouped as words. A word is determined using the group of characters grouped in this manner and the recognition results of each character. A line and character detection cutting circuit 22, a normalization circuit 23 that normalizes and samples the character information supplied from this line and character detection and cutting circuit 22, and which is supplied from the normalization circuit 23. Character information, for example, alphabetical characters in the dictionary 25,
It is constituted by a recognition circuit 24 that recognizes characters such as alphabets and numbers by a method of matching numbers with a reference pattern (decoding similarity method).

The output of the character recognition section 13, ie, the recognition circuit 24, is supplied to the word recognition section 14. This word recognition unit 14 is -1-
Words are recognized based on the words read out from the word dictionary 15 for the recognized characters supplied from the written character recognition unit 13, and attribute information corresponding to the candidate word is also read out from the word dictionary 15 and output at the same time. . The word recognition unit 14 also outputs position information of each word.

The word recognition unit 14 is configured to perform word recognition using a word recognition method called trigram, for example.

That is, a word is determined using the group of characters grouped in the line/character detection/cutout circuit 22 and the recognition results of each character. At this time, the word is divided into trigrams of three characters, and candidate words are extracted by matching the divided trigrams with the word dictionary 15. Next, the frequency of appearance of the extracted candidate words is checked for each trigram, and the word with the highest frequency of appearance is output as a candidate word.

In addition, the word recognition unit 14 stores attribute information corresponding to the candidate words along with the state name, state name, etc. for each word.
It reads out and outputs attribute information such as city name, street name, major name, person's name, etc. from the attribute information table 15a stored therein.

For example, for the words rWASHI NGTONJ and IU, WAS as shown in FIG.

It is divided into trigrams of three characters each: ASH, SHIl, -TON, ONW, and NWA, and words whose positions and characters match each trigram are extracted from the word dictionary 15, and the appearance frequency is calculated. As a result, the word rWASHI NGTONJ with the highest frequency of appearance is output as a candidate word. At this time, the attribute information for the word rWASHINGTONJ includes state name, city name, street name, large name,
A person's name is read from the attribute information table 15a and output.

Additionally, if rNWJ is recognized as a word, the direction name is output as attribute information, and rAVE is recognized as a word.
If J is recognized, the street type is output as the attribute information, and rMASSACHUSE is output as the word.
If TTSJ is recognized, the state name,
Street name and major name are output.

The attribute information stored in the attribute information table 15a is as follows:
The address tape on the mask (for example, address information managed by the Ministry of Posts and Telecommunications) is analyzed and created at the same time as the dictionary 1f15 for the address reader.

The word recognition unit 14 also takes into consideration the evaluation score (calculated from the degree of similarity) of each character of the recognition circuit 24 to determine a word.

The output from the word recognition section 14, that is, the candidate words, their attribute information, and the position information of each word are supplied to the address recognition section 16. This address recognition unit 16 is similar to the word recognition unit 14.
Using candidate words and their attribute information supplied from zi
The address is recognized by determining the p (zip code), state name, city name, street name, street number, etc., and the classification designation data for the address, that is, the classification designation data indicating the classification pocket in the subsequent classification section (not shown). This is what is output.

The address recognition section 16 includes a zip search section 31, a state name search section 32, a city name search section 33, a street name search section 34, an address search section 35, a state name table 36, and a city name table 37.
, a street name table 38, a major name table 39, an address determination section 40, and the like.

The zip search unit 31 uses the candidate words searched by the word recognition unit 14, the attribute information corresponding to each candidate word, and the word position information to search a 5-digit or 9-digit number string from the lower right corner to the left. Search for each word and if there is a number string, zi
p (zip code) candidates, and the search results are supplied to the state name search section 32, city name search section 33, street name search section 34, and address determination section 40.

If there is a zip candidate supplied from the zip search unit 31, and if there is no word on the left side, the state name search unit 32 sets a pointer indicating the word at the lower right end in the PTI, and performs a state name search. Word attribute information has 2 bytes for each word,
Each bit indicates whether or not it exists as a state name, city name, etc. If the state name bit of the attribute of the word indicated by the pointer PTI is on (1), the official name of the state name is stored. A search is performed using the state name table 36, and the words of the official name are used as state name candidates, and the search results are supplied to the city name search section 33, street name search section 34, and address determination section 40. The state name search is performed, for example, using the bottom two lines as the limit, and when the pointer PTI reaches 1, which indicates the last word, the state name search ends.

” If the state name search unit 32 has supplied the state name candidate for the city name search, the city name search unit 33 sets a pointer to the left side of the state name candidate, if not, sets a pointer indicating the word at the lower right corner to PT2, If the city name bit of the attribute of the word indicated by pointer PT2 is on (1), search the city name table 37 in which official city names are stored (40,000 city names) and find the word of the official name. Candidate city name. At that time, if there is a candidate state name, only the city names within that state are searched.

For example, if there is a state name candidate, the city name search is performed within two lines above the line containing the state name candidate, and if there is not, the city name search is performed within two lines below, and the pointer PT2 moves to the pointer PT2 pointing to the last word.
When reaching , the city name search ends. Therefore, the city names that have come up as candidates are evaluated based on their relative positions with the state name and zip, and, for example, a maximum of four candidate city names are output to the street name search section 34 and address determination section 40 as city name search results.

The above-mentioned street name search section 34 is connected to the above-mentioned city name search section 32.
If a city name candidate for a street name search is supplied from the state name search section 32, then the state name candidate for a city name search is supplied from the state name search section 32, then both must be supplied. For example, if the pointer indicating the word at the bottom right corner is set in PT3, and the street name bit of the attribute of the word indicated by pointer PT3 is on (1), the official name of the street name is stored (1st child case). The street name table 38 is searched, and the words of the official name are used as street candidates. At that time, if there is a city name candidate, only the street names within that city will be searched, and if there is a state name candidate,
Search only street names within that state. For example, if there is a city name candidate or a state name candidate, the street name search is performed within the upper two lines from the line containing the city name candidate or state name candidate, and if there is not, the street name search is performed within the two lines below. When the pointer indicating the final word reaches 3, the street name search ends. Therefore, street names that have come up as candidates are evaluated based on the city name, state name, and relative position with ZIP, and, for example, a maximum of four street names are output as street name search results to the street number search unit 35 and address determination unit 40. do.

When a street name candidate is output from the street name search unit 34, the street name search unit 35 selects the word on the left side of the street name candidate as a street number candidate if there is a numeric string, and uses the search result as an address. This is output to the determination section 40.

” The address determination unit 40 includes a zip search unit 31, a state name search unit 32, a city name search unit 33, a street name search unit 34,
In accordance with the search results supplied from the address search section 35, the address is determined, classification designation data corresponding to this address is searched from a table (not shown), and a subsequent classification section (not shown)
This is what is output to. Further, the address determination unit 40
When street name candidates are not supplied from the street name search unit 34, the state name candidates from the zip mil code name search unit 32 from the zip search unit 31 and the city name search unit 3
According to the city name candidate from 3, the large customer table 39 in which the large customer corresponding to the zip mill code name candidate and the city name candidate is stored is searched, and if the corresponding large customer is searched, the large customer It outputs classification designation data corresponding to a customer to a subsequent classification section (not shown).

Next, the operation in such a configuration will be explained.

For example, now, a mail P shown in FIG. 2 is conveyed in the direction of arrow a and reaches the scanning section 11. Then, this scanning section 11
optically scans the mail P and performs photoelectric conversion, and the pattern signal is supplied to the address area detection section 12. Then, the address area detection unit 12 detects the address area in which the address information 1 is written from among the pieces of information written in the mail P.

Further, the output of the scanning section 11 is supplied to the binarization circuit 20 in the character recognition section 13, and after being binarized, the output is sent to the selection circuit 20.
is output to.

Thereby, the selection circuit 21 selects the address area detection section 12.
The binarization circuit 20 corresponds to the coordinate information supplied from
The output from the character detection cutout circuit 22, that is, only the read signal is selectively outputted to the character detection cutout circuit 22.

As a result, the line/character detection/cutout circuit 22 detects and cuts out the address information 1 corresponding to the address area supplied from the selection circuit 21 line by line, as shown in FIG. 3(a). Based on the line detection and extraction results, that is, based on the size and position information of the extracted character clusters, detection and extraction is performed one character at a time as shown in (b) of the same figure. ), the inter-character gap is calculated, the part where the inter-character gap is larger than a predetermined value is cut off as the end of words, and the character blocks are grooved as words. A word is determined using the group of characters grouped in this manner and the recognition results of each character. This line,
The character information detected and extracted by the character detection and extraction circuit 22 is normalized and sampled by the normalization circuit 23, and then supplied to the recognition circuit 24. This recognition circuit 24 converts the supplied character information into, for example, ``!''. Characters such as alphabets and numbers in 25 are recognized by a matching method (decoding similarity method) with the basic pattern for alphabets and numbers.

The output of this recognition circuit 24 is supplied to the word recognition section 14. This word recognition unit 14 uses the above-mentioned tridalum to recognize words based on the words read out from the word dictionary 15 for the recognized characters supplied from the above-mentioned recognition circuit 24, and outputs this recognition result and its position information. At the same time, the attribute information corresponding to the candidate word is also stored in the attribute information table 15.
Reading from a, each search section 31 to 3 in the second address verification section 16
Output to 5.

As a result, the address recognition unit 16 recognizes the address using the supplied recognition results of each word, position information of each word, and attribute information of each word, and uses the classification designation data corresponding to this address to be classified in the subsequent stage. output to the department.

Regarding the address recognition by the address recognition unit 16, '<+
This will be explained with reference to the flowcharts shown in 5d(a) and (b). That is, the word recognition unit 14
The candidate words searched for, attribute information corresponding to each candidate word, and word position information are sent to the zip search section 31, state name search section 32, city name search section 33, street name search section 34, and location search section 35. Supplied. As a result, the zip search unit 31 uses the candidate words searched by the word recognition unit 14,
Using attribute information and word position information corresponding to each candidate word, a 5-digit or 9-digit number string is assigned to the lower right word (200
36) to the left, and if there is a number string, it is considered as a ZIP (postal code) candidate, and the search results are searched in the state name search section 32, city name search section 33, and street name search section 34. , is supplied to the address determination section 40.

Further, if there is a zip candidate supplied from the zip search unit 31, and if there is no word on the left side of the zip candidate, the state name search unit 32 searches for state names by examining word attribute information in order from the word at the lower right end. That is, when the bit corresponding to the state name of a predetermined word is on, it is checked using the state name table 36 whether or not the word is within the range of the above zip, and if it is within the range,
The word is set as a state name candidate, and this state name candidate is output to the city name search section 33, street name search section 34, and address determination section 40.

Further, if the city name search unit 33 is supplied with a state name candidate for a city name search from the state name search unit 32, the city name search unit 33 examines the word attribute information in order from the word on the left side of the candidate name, and if not, the word at the bottom right corner. Perform a city name search. That is, when the bit corresponding to the city name of a given word is on, the city name table 37 is used to determine whether or not the word is within the range of the zip mentioned above and whether it is within the range of the state name. If the word is within the range, the word is selected as a city name candidate, and this city name candidate is output to the street name search unit 34 and address determination unit 40.

Furthermore, when the street name search unit 34 is supplied with a city name candidate for a street name search from the city name search unit 32, on the left side, a state name candidate for a city name search is supplied from the state name search unit 32. If supplied, to the left of
If both are not supplied, the street name search is performed by examining the word attribute information in order from the word at the bottom right corner. That is, if the bit corresponding to the street name of a predetermined word is on, it is determined whether the word is within the range of the above zip.
Use the street name table 38 to check whether it is within the range of the above state name or city name, and if it is within the range, use that word as a street name candidate, and select this word as a street name candidate. The name candidates are output to the address search unit 35 and address determination unit 40.

Furthermore, when a street name candidate is output from the street name search unit 34, the street name search unit 35 selects the word on the left side of the street name candidate as a street name candidate if there is a numeric string, and uses the search result as a street name candidate. It is output to the address determination section 40.

Therefore, the address determination section 40 determines the address according to the search results supplied from the zip search section 31, the state name search section 32, the city name search section 33, the street name search section 34, and the house number search section 35, The classification designation data corresponding to this address is output to the subsequent classification section. In addition, the address determination unit 40
When the street name candidate is not supplied from the street name search section 34, the state name candidate from the zip mill code name search section 32 from the zip search section 31 and the city name candidate from the city name search section 33 are used. te, large mouth core table 39
is searched, and if a corresponding large-mouth group is found, the classification designation data corresponding to the large-mouth group is output to the subsequent classification section.

For example, a case will be described in which the address of mail P as shown in FIG. 2 is recognized. The candidate words searched by the word recognition unit 14 are r20038 J, rD, c, J,
r WASHINGTONJ, r lnJ, rAVHJ
, r MASSAC) ltlsETTs J, r 17
17J, r NEVSJ, r ENVIRONMENT
J, r ATLANTIC J, rcENTRAL J
, attribute information corresponding to each candidate word, for example r
For D, C, J, state name, r WASHINGTONJ
In the case of , state name, city name, street name, Oguchi group, person's name,
In the case of r NVJ, the direction name, in the case of rAVEJ, the street type, in the case of rMAssAcHUsETTs J, the state name, street name, Oguchi base, and location information for each candidate word are zip search section 31, state name search section 32, city. The information is supplied to a name search section 33, a street name search section 34, and an address search section 35.

As a result, the zip search section 31 can perform the word recognition section 14.
Using the candidate words searched for, the attribute information corresponding to each candidate word, and the word position information, search for each word in a 5-digit or 9-digit number string from the word r20036 J at the lower right corner to the left, and If there is a column, it is a zip (postal code) candidate, and in this case, r20036 J is 5
State name search unit 32 and city name search unit 3 as digit zip candidates
3. Output to the street name search unit 34 and address determination unit 40.

Next, since there is a zip candidate supplied from the zLp search unit 31, the state name search unit 32 searches for the word r on the left side of the zip candidate.
The word attribute information is checked in order from D, C, and J to perform a state name search. In this case, since the bit corresponding to the state name in the attribute information of the words rD, C, and J is on, the words rD, c,
State name table 3 to check whether J is within the range of the zip above.
6, and if it is within the range, the word r D, C
, J are output as state name candidates to the city name search unit 33, street name search unit 34, and address determination unit 40.

Furthermore, the city name search unit 33 searches the above state name candidates r D, C.
, J is supplied, so the word r WAS on the left side
City name search is performed by examining word attribute information in order from the old NGTONJ. In this case, the word “νASHI NGTONJ
Since the bit corresponding to the city name is on, the word r WASHI NGTONJ is
20038 Whether or not it is within the scope of J, the above state name r D
, C, J using the city name table 37, and if it is within the range, the word r WASHI
NGTONJ is output as a city name candidate to the street name search unit 34 and address determination unit 40.

Furthermore, the street name search unit 34 searches for the city name candidate r.
Since WASII NGTONJ is supplied, the street name search is performed by checking the word attribute information in order from the word r NWJ on the left side. In this case, the word r MA
Since the bit corresponding to the street name of SSACIIUSETTS J is on, that word is
20036 Whether or not it is within the scope of J, the above state name r D
, whether or not it is within the range of C, J, and the above city name r
Check whether it is within the range of WASIIIINGTONJ using the street name table 38, and if it is within the range,
The word rMASSACIIUSETTSJ is output to the address search unit 35 and address determination unit 40 as a street name candidate.

Next, the address search unit 35 searches the street name candidate r.
Since MASSACIIUSIETTSJ is supplied, the address search is performed in order from the word on the left side. In this case, the address determination unit 4 uses the word r 1717J as an address candidate.
Output to 0.

As a result, the address determination unit 40 has zip candidate r2003.
6 J State name candidate r D, C, J City name candidate r
WASHINGTONJ X treat name candidate r MA
SSACIIUSETTS J address candidate r 171
By supplying 7J, the address is determined and r
1717 MASSACIIUSETTS AYE,
, NV WAS old NGTOND, C, 20038J corresponding classification designation data (for example, 7-digit zip mill code, output to the subsequent classification section.

As mentioned above, devices that detect character information and recognize addresses read the character strings that make up the address, and recognize multiple words, the attributes of each word, and the position of each word from this read character string. However, since the address is recognized according to the recognized words, the attributes of each word, and the position of each word, the address can be recognized even if it differs from the basic knowledge regarding the hierarchical structure of addresses and address descriptions. It is possible to perform address recognition with high flexibility and tolerance.

In addition, by combining character arrangement information, it is possible to evaluate addresses with high reliability and improve address reading accuracy.

[Effects of the Invention] As detailed above, according to the present invention, an address can be recognized even if there is even a slight difference from the basic knowledge regarding the hierarchical structure of addresses and address description, and the flexibility of address recognition is improved. It is possible to provide an address recognition device capable of recognizing addresses with high tolerance.

[Brief explanation of the drawing]

The drawings show an embodiment of the present invention; FIG. 1 is a diagram schematically showing the overall configuration, FIG. 2 is a plan view showing an example of a mail item, and FIG. 3 is an explanation of the detection and cutting operation. FIG. 4 is a diagram for explaining the word recognition method, and FIG. 5 is a flowchart for explaining the address recognition operation. P... Mail, 1... Address information, 11... Scanning unit, 12... Address area detection unit, 13... Character recognition unit,
14... Word recognition unit, 15... Word dictionary, 15a.
...Attribute information table, 16...Address recognition section, 31.
...zip search section, 32...state name search section, 33...
City name search section, 34...Street name search section, 35.
...Street address search section, 36...State name table, 37...
City name table, 38...Street name table, 3
9... Oguchi base table, 40... Address determination section. Applicant's agent Patent attorney Takehiko Suzue (a)! YliNIS日IS 0m0fSJ ra 口. mouth. 2003 WASHINGTON D, C,
20036(C) Figure 3 WAS-WASHINGTON WASTEWATE
RASH-CASHDOLLAR, WASHINGTO
NHE WASHINGTON Figure

Claims (1)

[Scope of Claims] An address recognition device that recognizes an address by detecting character information, comprising: character reading means for reading a character string constituting an address; A word recognition means for recognizing the attributes of words and the position of each word; and an address recognition means for recognizing an address according to the words recognized by the word recognition means, the attributes of each word, and the position of each word. An address recognition device characterized by: