JP3425257B2 - Mail reading sorter - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a mail automatic reading / sorting device for automatically sorting mails in detail, which reads the detailed postal code and address on the mail being conveyed. Regarding the device. The present invention also relates to a manual input device for sorting information of the mail reading system.

[0002]

2. Description of the Related Art Conventionally, as a method for automatically reading and sorting postal items, a first method is known in which a postal code on a postal item is read as sorting information and sorting is performed according to the postal code. For example, Japanese Laid-Open Patent Publication No. 1-161587 describes that when a zip code is read, the city name is also read to improve the reliability of the zip code used for classification. In addition, the prefecture name of the address stated on the mail,
Various proposals have been made for reading a place name such as a city / ward name, a town name, etc. or a house number to determine a classification destination. For example, JP-A-1-316888, JP-A-2-308384, and JP-A-4 In Japanese Patent Laid-Open No. 205685/1985, the place name portion of an address is read as an address, and in Japanese Patent Laid-Open No. 6-444.
A second method is described in which the house number is read as in Japanese Patent No. 01. Further, Japanese Patent Application Laid-Open No. 61-234481 discloses a third method of reading a recipient name, which is the only address in a certain area and can specify the destination.

[0003]

However, in the first method described above, in order to make finer divisions according to the zip code and to assemble the route for delivery, it is necessary to increase the number of zip codes to 10 or more. However, there is a problem that the burden on the user increases. Further, in the second method, the place name portion of the address is read, but it is necessary to read the chome address, building number, etc. in order to perform more detailed division and route assembly. There are problems that the street address and the building number are different in the way they are entered, which makes it difficult to recognize, and there are many misrecognitions and the reliability is low. Also, it does not consider fixing non-reading. Further, in the third method, the recipient name that can specify the destination is used. However, when multiple recipients with the same surname and the same name live in the same area, or when there are multiple business establishments with the same name. Has the problem that the destination cannot be specified.

Therefore, the present invention addresses the mailing, especially
To improve the reliability of recognition of chome address and building number,
The purpose is to reduce manual work when correcting unread.

[0005]

The mail reading / sorting device of the present invention is a mail reading / sorting device for recognizing an address by detecting character information and collating it with information in a dictionary. A means for reading the postal code, a means for reading the recipient's first and last name, a means for holding a detailed address such as the recipient's street address, building number, room number, etc. as a dictionary, and a dictionary for the detailed postal code and first and last name. A means to access and generate detailed classification information including chome address, building name, room number, and a means to read the chome address and generate the detailed classification information when a detailed address cannot be determined by the address area and surname. It has.

Further, in the correction unit for correcting the read result of the mail, a means for key-inputting a part of the family name and a detailed classification information including a chome, an address, a building name and a room number from the key input character. And means for generating.

[0007]

The present invention holds a detailed address such as the recipient's street address, building number, and room number as a dictionary. By reading the detailed postal code corresponding to the town area of the address and the recipient's first and last name, Since the division information including the detailed address of the recipient can be generated, it is not necessary to read the chome address and the building number, and the division information can be generated with high reliability. Further, since the detailed dictionary is provided, highly accurate recognition can be realized when reading the chome, street address, building name, and room number. Further, in the correction unit that corrects the read result of the mail matter, detailed classification information can be generated by only keying in a part of the family name, which has the effect of reducing the correction work of the operator.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to FIG. The postal matter 140 is sequentially sent to the supply unit 100. The postal matter passes through a predetermined position in the supply unit 100, and an image of the surface of the postal matter is captured by the optical system 101 and the image input unit 102 during the passage. In the address recognition unit 103,
The address and the first and last name of the recipient are printed on the surface of the mail and the classification information is generated. On the other hand, the postal matter whose front surface image has been captured is sent to the delay conveyance path 109. In the delay conveyance path 109, the mail moves along the conveyance path 109 for a predetermined time for generating the sorting information from the surface image. The sorting unit 110 sorts the mail items according to the sorting information. In addition, for the mail matter that the address recognition unit 103 has determined that the acquisition of the classification information is unreadable, that is, unsuccessful, the unread correction unit 111 manually adds the classification information. Optical system 10
Reference numeral 1 includes an illumination optical system and an imaging optical system. The image input unit 102 digitizes an image signal from a photoelectric conversion element such as a line sensor. Address recognition unit 103
Is composed of a control unit 104, a user identification code conversion unit 105, and a knowledge processing unit 106. An image processing unit 107 and a character recognition unit 108 are connected to the address recognition unit 103 and controlled by the control unit 104. In the knowledge processing unit 106,
Access the word dictionary group 120 and the connected dictionary group 130,
An error or the like in the recognition result of the character recognition unit 108 is automatically corrected by using the dictionary group. Word dictionary group 120
Is a place name dictionary 121, an address display number dictionary 122, a building name, a building name dictionary 123, a floor number, a room number dictionary 124, an institution name dictionary 125, a title dictionary 126, a family name dictionary 12.
It consists of 7. The connection dictionary group 130 includes an address area connection dictionary 131, an address display number connection dictionary 132, a building name, a building name connection dictionary 133, a floor number, and a room number connection dictionary 134.
It consists of an affiliated institution name connection dictionary 135 and a family name connection dictionary 136.

FIG. 2 is a flow chart of a process for reading a mail item. In step 200, an image of the surface of the mail is input. This process is performed by the image input unit 102.
Step 210 is a process of image processing, and is executed by the image processing unit 107. The step includes a step 211 of extracting a destination area from the entire surface image and a step 212 of extracting a character line from the destination area. After extracting the character line, the character is cut out in step 220 and cut out 1
Recognize each character and convert it to a character code. Here, a plurality of recognition results are output for each cut out character. The process is performed by the character recognition unit 108. Then, the process proceeds to step 230 of knowledge processing. Here, first, in step 231, the address area is extracted, and then in step 23
2. Extract the address display number. Further, step 233
The building name and the building name are extracted with, and the floor number and room number are extracted in step 234. In addition, the institution name is extracted in step 235, the title is extracted in step 236, and the family name is extracted in step 237. Based on these extraction results, a verification process is performed in step 238 to correct the recognition results for errors and unread. In step 240, the read result is converted into a user identification code.

FIG. 3 is a diagram for explaining a process of reading a mail item in which an address and a family name are written. The line image 320 extracted in the character line extraction processing 300 is sent to the character recognition unit 301. The line image 320
May be composed of a plurality of lines such as a line in which an address area is described, a line in which an address display number is described, and a line in which the recipient's first and last names are described. In the character recognition processing 301, a character pattern is cut out and character recognition is performed for each character. The result of character recognition is shown at 321.
The recognition result 322 of the rank, the recognition result 323 of the second rank, and the recognition result 324 of the third rank. For example, in the fourth digit "thousand" of the described address, the first rank of the recognition result is "ten", the second rank of the recognition result is "thousand", and the third rank of the recognition result is "me".
In the address area extraction process 302, the address area portion 325 is extracted from the recognition result 321 using the place name dictionary 303 and the address area connection dictionary 304. In this example, “Kanda Surugadai, Chiyoda-ku, Tokyo” is extracted as a candidate for the address area portion. On the other hand, in the address display number extraction processing 305,
Using the address display number dictionary 306, the address display numbers 327 such as chome and address are extracted from the recognition result 321.
At this time, in the previous address area extraction process 302, the last digit of the address area character string is output. Based on this last digit position, the first digit of the address display number is set and the address display number is extracted. ing. In this example, the extraction result 32 of the process is
As for 7, there are multiple candidates for "4" and "9",
Plural candidates of “8th address” and “6th address” are extracted as addresses. Furthermore, in the family name extraction process 307, the family name dictionary 3
08 is used to extract the full name character string 326. Here, the last digit of the address display number is output by the above address display number extraction processing 305, and the first digit of the family name is set by using this digit position as a clue, and the family name extraction processing 307 is performed. As a result 326 of the family name extraction processing, in this example, a plurality of candidates of “Kawata” and “Yamada” as the family name and a plurality of candidates of “Taro” and “Taro” as the family name are obtained. Then, the process proceeds to the verification process 309, the above-mentioned results 325, 327, 326 are inputted, and the validity is judged whether or not there is a contradiction in the connectivity between these results. At this time, the address display number connection dictionary 310 and the family name connection dictionary 311 are accessed to judge the validity of the connection of the plurality of results. In this example, as the result 328 of the verification process 309, the chome “4 chome”, the address “6”, and the family names “Yamada” and “Taro” are required.
In this verification process 309, the chome address connected to the address area extraction result 325 is obtained from the address display number extraction result 327. In addition, the extraction result of the first and last name 326 is included in the extraction result of the address area 32
5. It is determined whether the address display number extraction result 327 is connected. Thus, for example, even if there are multiple candidates for "Kawata" and "Yamada" as surnames, it is determined whether the address area or address display number connected to the surname is valid. You can select "Yamada" as your family name because you have decided that you do not live in. Furthermore, when "Yamada" is not included in the surname and first name extraction result 326, it may be unreadable, or from the extraction result of the address area and the address display number, the corresponding surname "Yamada" from the surname and first name connection dictionary 311. Can be selected and output to the result 328.

FIG. 4 is a diagram for explaining the process of word matching. This word matching is a basic process for realizing the knowledge process 230, and each extraction process 231, 232, 233,
This is processing inside 234, 235, 236, and 237.
The word dictionary 400 is each word dictionary 12 of the word dictionary group 120.
1, 122, 123, 124, 125, 126, 127
Equivalent to. Input of word matching processing is recognition candidate lattice 4
04, which corresponds to the recognition result 321. The output of the process is the corresponding word, the digit position of the word, and the cost. In the process of the process, first, at the start of the word matching process, the matching start digit is set 408 in the recognition candidate lattice 404. Here, the items 410 to 415 in the horizontal direction of the recognition candidate lattice 404 correspond to the digits of the character string, and characters are compared in the direction of the arrow indicated by the digit direction scan 407. The vertical items 420 to 427 of the recognition candidate lattice 404 correspond to the candidate rank direction, and the character of the recognition result is accessed from the lattice 404 according to the candidate rank direction scan 406. The word dictionary 400 is accessed using the character code of the recognition candidate lattice 404 as a key. At the set start digit, for example, the digit 411, the character codes 1 to 8 of the recognition result are sequentially fetched, and the words whose character code at the leading digit of the word is 1 to 8 are sequentially fetched from the word dictionary 400. Take it out. These words 401 are collated in the comparison unit 422 with the character code of the corresponding digit of the recognition candidate lattice 404 for each character according to the scanning 405 in the digit direction. Cost calculator 4
In 03, the digit-direction scanning 405 is performed with the degree of matching as the cost.
The cost is output for each word.
Note that a method other than the above can be used to access a word from the word dictionary 400. For example, there is a method in which not only the leading digit of the word but also the nth digit character code can be accessed.

FIG. 5 is a diagram for explaining the structure of the connection dictionary relating to the addressee of the recipient. The address area connection dictionary 500 has connection relationships between prefecture level words, city / ward county words, and town level words. For example, there are prefecture level words such as Saitama prefecture 510, Tokyo 511 and Kanagawa prefecture 512, and Tokyo 511 is connected to city / ward level words, Minato 520, Chiyoda 521, and Shinjuku 522. . Further, for example, Chiyoda Ward 521 is a town level word, Marunouchi 530, Kanda Surugadai 531, Otemachi 53.
It is connected to 2. The connection relation between these prefecture level words, city / ward county level words, and town level words is stored in the address area connection dictionary 500. The address display number connection dictionary 501 has a connection relation between a town-level word in the address area connection dictionary and the lower-order chome and address. For example, below the Kanda Surugadai 531, as the chome level, 1 chome 540, 2 chome 541, 3 chome 542,
There are 4 chomes 543 and 5 chomes 544. Further, as the address level, for example, at the lower order of the 4 chome 543, the 1 st 55
0, 2 address 551, 3 address 552, 4 address 553, 5 address 554, 6 address 555, 7 address 556, 8 address 557
Is present in the dictionary 501.
The first and last name connection dictionary 502 holds the connection relation between the address display number and the first and last name of the resident of the residence. For example, in Tokyo 511, Chiyoda-ku 521, Kanda Sugawadai 531, 4 chome 543, 5 address 554, Jiro Suzuki 560 resides, in 6 place 555 of that place Taro Yamada is located, and in that place of 7 556 place. Jiro Takahashi 562 and Goro Yamamoto 563 are in the same place 8
Saburo Saburo 564 will live in each house,
It is held in the dictionary 502.

FIG. 6 is a diagram for explaining the construction of a connection dictionary relating to the addressee of the recipient in which the building name, floor number, and institution name are described. In addition to the address area connection dictionary 600 and the address display number connection dictionary 601 described in FIG. 5, a building name and building name connection dictionary 602 holding building names and building names corresponding to each address display number are provided. Further, each building or building has a floor number holding a building number or room number, and a room number connection dictionary 603. Further, an institution name connection dictionary 60 that hierarchically holds the institution name to which the recipient belongs
4 is equipped. Further, it is provided with a surname / first name connection dictionary 605 which holds the surname / first name of the recipient who belongs to each department in association with the institution. For example, Tokyo 610, Chiyoda 620, Marunouchi 630, 1 chome 640, 5th 650,
For the address of No. 1 660, Shinmaru Building 670 is connected to No. 1 660 of the address display number. Further, in the floor number / room number connection dictionary 603, the basement floor 680, the first floor 68
Shinmaru Building 670 where the 1st, 2nd floor 682 and 7th floor 683 are higher
It is connected to the. In the affiliated institution name connection dictionary 604, the organization of the institution is hierarchically held, and the Japan Manufacturing 690 is held in the lower part of the 7th floor 683, and the industrial department 691 and the sales department 692 are held in the lower part. The family name connection dictionary 605 holds information that Taro Yamada 693 and Ichiro Suzuki 694 belong to the sales department 692.

FIG. 7 is a diagram for explaining a process of selecting a plurality of candidates in the verification process. For the image of Description Example 700, multiple candidates 710 are assumed as inputs to the verification process. In the example, multiple candidates of 711, 712, 713 are assumed. An address whose address is erroneous as "8", such as 711, and an address whose maiden name is erroneous as "river", such as 712, 7
Each of the correct addresses, such as No. 13, is input to the verification process 720 as a candidate. In the verification process, the first and last name connection dictionary 502, the address display number connection dictionary 501, and the address area connection dictionary are used, and it is determined that Taro Yamada 561 resides at the 6th address 555. Therefore, the processing result is 730
The address shown in is output.

FIG. 8 shows an example of a detailed procedure of the verification process.
In step 800, surname and surname candidates are generated by combining surname and surname candidates. For example, the surname “Kawata” “Yamada” and the surnames “Taro” “Taro” in the surname extraction result 326 are combined to generate four surname surname candidates “Taro Kawada” “Taro Kawada” “Taro Yamada” “Taro Yamada”. In step 801, an address display number is generated. Here, a plurality of candidates for chome and address are combined and generated. For example, the address display number extraction result 327 chome “4 chome” “9 chome”, address “8 address” “6 address”
4 address display number candidates “4-8th address”, “4th-6th address”, “9th-8th address” and “9th-6th address” are generated. Next, in step 802, a combination of the address area candidate and the address display number is generated. For example, “Kanda Surugadai, Chiyoda-ku, Tokyo” of the address area candidate 325,
The four address display number candidates “4 generated in step 801
"8-8", "4-6", "8-8", "6-6", "4-6-8 Kanda Surugadai, Chiyoda-ku, Tokyo""4-6-6 Kanda Surugadai, Chiyoda-ku, Tokyo" Generate candidates for "9-8 Surugadai Kanda, Chiyoda-ku, Tokyo" and "9-6 Surugadai, Kanda-Surugadai, Chiyoda-ku, Tokyo". Next, in step 803, the processes from step 804 to step 806 are performed until the above combinations are exhausted. First, in step 804, it is determined whether the connection between the address area and the address display number is appropriate. If it is determined that the connection is valid, the validity of the first and last name connection is determined in step 805, and if the connection is determined to be valid, the result is registered in step 806. In step 804, for example, by accessing the address display number connection dictionary 501 and the family name connection dictionary 502, the “Kanda Surugadai” 5
Since it can be seen that there are 1-chome 540 to 5-chome 544 in 31 and that "Kanda Surugadai 9-chome" does not exist, it is possible to narrow down the candidates. In addition, step 8
In 05, "Taro Yamada" is "4-6, Kanda Surugadai"
It is determined from the 561 of the surname / first name connection dictionary 502 that the user lives in the city, and among the multiple candidates, "4 Kanda Surugadai, Chiyoda-ku, Tokyo"
"Chome 6" is registered in step 806.

FIG. 9 shows an example of the screen structure of the non-reading correction section 111. For postal items for which it is difficult to determine the correct answer due to multiple address candidates, or for postal items for which it is difficult to determine the correct answer even if there is only one address candidate, use the screen. Manually enter the correct address. Screen 9
In 00, the image of the address part is displayed in 901,
The candidates of the recognition result of the part are displayed in 902, 903, and 904. Reference numeral 905 is for scrolling and displays a large number of recognition result candidates by scrolling. A correct answer candidate is selected from the displayed recognition result candidates by observing the image 901 with the naked eye. Also, the screen 91
No. 0 is another embodiment, in which an image 911 of the address portion is displayed, and the recognition result 912 is displayed below the image 911. Here, it is also possible to input a part or all of the family name using the keyboard or the like in the family name input unit 913 and extract the correct answer candidate 914 based on the input character. Here, the first and last name connection dictionary 502, the address display number connection dictionary 501, and the address area connection dictionary 500 are used.
A resident who is under the Kanda Sugawadai 531 and who has a "mountain" keyed in the family name input unit 913 is searched for and a correct answer candidate is extracted.

FIG. 10 is a flow chart showing another embodiment of the present invention. Here, it is assumed that the detailed postal code, which is obtained by dividing the area in detail, is described in the postal matter. The address area can be uniquely determined by reading the detailed postal code. Therefore, the address including the street address is determined based on the read detailed postal code and first and last name. Step 1000
Enter the image of the surface of the mail with. This process is performed by the image input unit 102. Step 1010 is a process of image processing, and is executed by the image processing unit 107. The step is a step 1011 of extracting a detailed postal code line and a step 10 of extracting an address area from the entire surface image.
12 and step 1 of extracting a character line from the addressed area
013. After extracting the character line, step 1020
The character is cut out with and the character is recognized for each character cut out and converted into a character code. Here, a plurality of recognition results are output for each cut out character. The process is performed by the character recognition unit 1.
At 08. Then, step 10 of knowledge processing
Move to 30. Here, first, the detailed postal code is extracted in step 1031. The detailed postal code is written in a character box in a predetermined area of the postal matter, or in the address area. Recognizing a postal code is described in, for example, Japanese Patent Laid-Open No. 1-316888.
It is possible to easily realize the detailed postal code. Step 1
The full name is extracted at 032. The method of extracting the family name is, for example,
The place where the surname is written may be determined based on the layout of the character lines extracted from the mail, or the surname and first name may be extracted from the character string by collating with the surname and surname dictionary 127. In step 1040, the address display number is searched based on the extraction result of the family name. Here, step 10
Using the detailed zip code extracted in 31 as a key, the detailed zip code connection dictionary 1103 is accessed to find the corresponding address area, and the address display number that matches the first and last name of the recipient living in the address area is selected. There may be a plurality of recipients having the same surname and the same name in the address area, and at this time, a plurality of address display numbers are retrieved. Therefore, step 105
When 0, the address display number is determined from a plurality of address display number candidates. Here, the address display number is extracted from the recognition result, and the number that matches the previously retrieved address display number is selected. Then, in step 1060, the detailed postal code as the read result and the address display number obtained in the first and last name extraction step 1032 are used to convert into a user identification code.

FIG. 11 is a diagram for explaining the structure of the detailed zip code connection dictionary. The detailed postal code connection dictionary 1103 is composed of, for example, seven-digit numbers 1170, 1171, 1172, etc., and can access the corresponding town area name in the address area connection dictionary 1100, for example, 1130, 1131, 1132. As such, the address of the town area name is stored in the detailed zip code connection dictionary. Here, "Marunouchi" 1130 by detailed postal code "1001234" 1170
Can be accessed. In addition, detailed postal code "100123
5 ”1171 to access“ Kanda Surugadai ”1131. With the address display number connection dictionary 1101 and the surname / first name connection dictionary 1102, it is possible to set the range of surnames and first names belonging to the relevant town area. For example, the detailed postal code “1001235” 1171 of the detailed postal code connection dictionary 1103 passes through “Kanda Surugadai” 1131 in the address area connection dictionary 1131 and then “1 chome” 114 of the address display number connection dictionary 1101.
It is possible to access "5 chome" 1144 from 0 to retrieve the corresponding surname and surname from the surname and surname connection dictionary 1102. As a result, “Taro Yamada” 1161 who lives in “Surugadai Kanda” 1140 can be searched in step 1040, and at this time, the corresponding address display number is “4th-6th address”.
From “4 chome” 1143 and “6 address” 1155 of the address display number connection dictionary 1101, step 10
It can be decided at 50. When a plurality of people with the same surname and the same name live in the same town area, the address display numbers may be extracted from the character recognition result, and the surname and first name connection dictionary 1102 may be used to test the validity of the selection.

FIG. 12 is a flowchart of the address display number searching process shown in step 1040. In step 1200, the town name of the address is accessed by the read detailed zip code. Then, in step 1201, a search range is set for the first and last names of the recipient who lives in the town name, based on the first and last name connection dictionary, and scanning is performed. In step 1202, the designated surname and surname are searched, and in step 1203, the address display number in which the recipient having the corresponding surname and surname resides is registered. It should be noted that the present invention is not limited to registration of the address display number, and for example, the building name, the building name connection dictionary 602, the floor number, and the room number connection dictionary 603 are used, and the corresponding building name and floor number are also combined. You may register.

FIG. 13 is a flow chart of the address display number determination processing shown in step 1050. In step 1300, it is determined whether or not there are a plurality of address display number candidates. If there is a single address display number, the validity of the address display number is verified in step 1304, and the address display number is registered in step 1305. To do. On the other hand, if it is determined in step 1300 that there are a plurality of address display number candidates, that is, if residents with the same surname and the same name exist in the same address area, the character recognition result is input in step 1301 and the address is input in step 1302. Extract the display number. Then, in step 1303, the address display number is collated, in step 1304 the validity is verified by the collation result, a single address display number is selected, and in step 1305 the number is registered.

[0021]

As described in detail above, according to the present invention, in the mail reading / sorting device for detecting the character information and collating it with the information in the dictionary to recognize the address, the detailed postal code corresponding to the address area can be obtained. To read the recipient's first and last name, a means to hold a detailed address such as the recipient's street address, building number, room number, etc. as a dictionary, and access the dictionary from the detailed postal code and first and last name. Chome address, building name,
It is provided with a means for generating detailed division information including a room number and a means for reading the chome address to generate the detailed division information when the detailed address cannot be determined by the address area and the family name. It holds a detailed address such as the recipient's street address, building number, room number, etc. as a dictionary. Therefore, by reading the detailed postal code corresponding to the town area of the address and the surname of the recipient, It is possible to generate the division information including a detailed address, it is not necessary to read the chome address and the building number, and the division information can be generated with high reliability. Also, when the detailed address cannot be determined by the address area and surname, the chome, street address, building name, and room number are read, but at this time, a detailed address dictionary is held, and high-accuracy recognition can be realized. it can.

Further, according to the present invention, in the correction unit for correcting the reading result of the postal matter, the means for inputting a part of the family name and the key, the chome, the address, the building name, and the room number from the key input character. It is possible to generate detailed classification information by simply inputting a part of the family name in the correction unit that corrects the read result of the mail. This has the effect of reducing the correction work by the operator.

[0023]

[Brief description of drawings]

FIG. 1 is a configuration diagram of a mail piece reading / sorting method according to an embodiment of the present invention.

FIG. 2 is a flowchart of a process for reading a mail item.

FIG. 3 is an explanatory diagram of a process of reading a mail item in which an address and a family name are written.

FIG. 4 is an explanatory diagram of a word matching process.

FIG. 5 is an explanatory diagram of a configuration of a connection dictionary regarding a recipient address.

FIG. 6 is an explanatory diagram of a configuration of a connection dictionary relating to a recipient address in which a building name, a floor number, and an institution name are described.

FIG. 7 is an explanatory diagram of a process of selecting a plurality of candidates in the verification process.

FIG. 8 is a flowchart showing an example of a detailed procedure of a verification process.

FIG. 9 is a configuration diagram of a screen of an unread correction unit.

FIG. 10 is a flow chart of a mail piece reading section according to another embodiment of the present invention.

FIG. 11 is an explanatory diagram of a configuration of a detailed postal code connection dictionary.

FIG. 12 is a flow chart of an address display number search process.

FIG. 13 is a flowchart of determination processing of an address display number.

[Explanation of symbols]

103 ... Address recognition unit, 106 ... Knowledge processing unit, 120 ... Word dictionary group, 231 ... Address area extraction step, 232 ... Address display number extraction step, 237 ... Surname and surname extraction step,
238 ... Verification process, 321 ... Character recognition result, 404 ... Recognition candidate lattice, 500 ... Address area connection dictionary, 501 ...
Address display number connection dictionary, 502 ... Last name connection dictionary, 602
... building name, building name connection dictionary, 603 ... floor number, room number connection dictionary, 604 ... affiliation institution name connection dictionary, 805 ... address area / address display number connection determination step, 902 ... recognition result candidate display window, 913 … First and last name input window, 1040… Search for address display number, 1050… Determine address display number, 11
03 ... Detailed zip code connection dictionary, 1200 ... Steps for accessing the address town name by the detailed zip code.

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Hisao Ogata               1-280, Higashikoigokubo, Kokubunji, Tokyo                 Central Research Laboratory, Hitachi, Ltd. (72) Inventor Tatsuhiko Kagehiro               1-280, Higashikoigokubo, Kokubunji, Tokyo                 Central Research Laboratory, Hitachi, Ltd. (72) Inventor Katsumi Marukawa               1-280, Higashikoigokubo, Kokubunji, Tokyo                 Central Research Laboratory, Hitachi, Ltd. (72) Inventor Hiroshi Shinjo               1-280, Higashikoigokubo, Kokubunji, Tokyo                 Central Research Laboratory, Hitachi, Ltd. (72) Inventor Kazuki Nakajima               1-280, Higashikoigokubo, Kokubunji, Tokyo                 Central Research Laboratory, Hitachi, Ltd.                (56) References JP-A-61-243481 (JP, A)

Claims (3)

(57) [Claims] 1. A postal mail reading / sorting device for detecting character information and collating it with information in a dictionary to recognize an address, and holds a postal code corresponding to an address area name as a postal code dictionary for performing detailed classification. Means, recognize the zip code on the mail, recognize the surname on the mail, address area name and address display number, building name, room number, institution name, surname and means for holding address information as address dictionary connected to, on the basis of the corresponding zip code and last name recognized, location number from the zip code dictionary and the address dictionary, building name, room number,
And a means for generating at least one of the affiliated institution names, and a means for outputting classification information of the postal matter based on the generated address display number, building name, room number, and / or institution name and the postal code. A postal matter reading / sorting device comprising: 2. The mail piece reading / sorting device according to claim 1, when the recognition of the family name described in the mail piece is determined to be unsuccessful, or when there are a plurality of addresses having the same family name in the address area. If determined, a means for recognizing at least one of the address display number, building name, room number, and institution name, and at least one of the recognized address display number, building name, room number, and institution name. A postal matter reading / sorting device comprising: means for outputting postal matter classification information based on a postal code; 3. The surname and first name written on the mail above is the recipient's name
Either claim 1 or claim 2 characterized in that
The mail reading and sorting device described in.