JP3162552B2 - Mail address recognition device and address recognition method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition apparatus, and more particularly to a mail address recognition apparatus using a postal code.

[0002]

2. Description of the Related Art In the current state of mail flow and automatic mail machines, in recent years, mails have been equipped with automatic address reading and sorting machines for sorting mails into delivery offices and delivery divisions. FIG. 3 is a diagram showing the flow of mail and the current state of a mail automation machine. In this figure, the postal matter collected in the post 31 is collected in the dispatch station 33 by the transportation means 32. Here, after the postal matter is automatically sorted and sorted and stamped by the stamping machine 36, sorting is performed for each delivery station 34 by the postal code automatic reading sorting machine 37.

Further, an automatic mail address reading / sorting machine 38 installed at the delivery office 34 is provided with a mail address recognition device. The mail address is read by using the mail address recognition device, and the mail is delivered. It is based on the function of classifying the delivery person 35 according to the delivery area.

[0004] Therefore, if the delivery station has a kanji dictionary memory included in the name of the town in the area served by the delivery office or the name of the company that is the major recipient, it can be handled. Conversely, it is difficult to read at once the address of the entire country of Japan, which is freely handwritten or printed on mail, at the current level of character recognition technology. There is a technical limitation that the address of the area under the jurisdiction of the bureau and the name of the large receiving company) and the mail address automatic reading sorter must be the mail of one delivery bureau.

[0005] However, as automatic mail reading and sorting machines are increasingly deployed and utilized, their own office (one delivery office)
In addition, there has been a strong demand to be able to process postal items from neighboring stations as well. However, to simultaneously process the postal matter of the own station and the postal matter of the adjacent station, it is necessary to enhance the address recognition dictionary memory (increase the number of character types to be read). However, as the number of characters increases for multiple delivery stations, the number of characters that need to be compared and examined increases accordingly, which places a greater burden on character recognition processing, and the recognition performance (processing speed, reliability Etc.) decrease.

[0006] Further, there is a potential demand to centralize the delivery division processing. In this case, not only two stations, that is, a local station and a neighboring station, but also a plurality of stations are required. The problem of the reduction in recognition performance comes out.

[0007]

SUMMARY OF THE INVENTION The present invention is directed to a mail address recognition apparatus capable of recognizing a mail address of a plurality of delivery stations without deteriorating the recognition performance (processing speed, certainty, etc.). The purpose is to provide.

[0008]

Means for Solving the Problems To achieve the above object,
The postal mail address recognition device of the present invention
In the mail address recognition device for posted mail, multiple mail
Dictionary information for address recognition is stored in advance according to the number.
Detecting multiple storage means and character information of mail
Detecting means, and the character information detected by the detecting means.
Identification means for identifying the postal code from the
The storage means corresponding to the separated postal code is stored in the plural
Selecting means for selecting from storage means, and selecting means for selecting by the selecting means.
Using the selected storage means, before the detection means detects
Recognition means for recognizing the address of the postal matter based on written information
And a step. Also the mail of the present invention
The destination address recognition method is based on the postal code
In the name recognition method, detection of character information of mail
Out of the character information detected in the output step and the detection step
Identification process for identifying postal codes from
A dictionary in which dictionary information for address recognition is stored in advance.
The postal code identified in the identification step from the number storage means
Selecting a storage means corresponding to a signal,
Using the storage means selected in the step,
Identify the address of the postal matter based on the character information detected in
And a recognition step for recognizing.

Further, the mail address recognition device of the present invention
In the mail address recognition device with the number written,
Dictionary information for address recognition according to the number of postal codes
Multiple dictionary memories and character information of mail
A photoelectric conversion element for detecting the
Identification means for identifying a postal code in said character information,
Dictionary memo corresponding to the postal code identified by the identification means
Selecting means for selecting a directory from the plurality of dictionary memories;
Using the dictionary memory selected by the selection means,
According to the character information detected by the photoelectric conversion element, the mail
A first recognizing means for recognizing an address of an object;
When the postal code could not be identified,
According to the station with the largest number of processings among multiple dictionary memories
Select a dictionary memory and use it to address the mail
And a second recognizing means for recognizing.
You. In the address recognition method of the present invention, the postal code is described.
Character information of postal items
A detecting step of detecting the sentence, and the sentence detected in the detecting step
Identification process to identify the postal code in the
Prestores dictionary information for address recognition according to flight number
From the plurality of storage means that have been identified in the identification step.
A selection step of selecting a storage means according to the postal code,
Using the storage means selected in the selection step,
The character information detected in the detection step allows the
A first recognition step of recognizing a name,
Based on the inability to identify the flight number,
According to the station with the largest number of processings among multiple storage means
Select the storage means and use it to identify the address of the mail.
And a second recognition step of
You.

[0010]

The address recognition apparatus of the present invention has the above-described configuration.
The mail address information of a single station as in the past.
Instead of recognizing the address, address mail from a nearby office.
Prepare multiple dictionary memories for multiple stations to perform name recognition
And selects only the dictionary memory corresponding to the recognized postal code.
Select only the required dictionary memory.
Can be. Therefore, this postal code is used for the dictionary memory.
Recognize mail from many nearby offices
Despite being able to perform, the collation work for recognition only needs to be performed for one station.
This will increase the awareness of addressing mail in a wide area.
It is possible to perform the reading quickly with the reading accuracy. The present invention
The address recognition method is based on the above-mentioned features.
For the same purpose as post office, postal items only in the conventional unit
Instead of performing address recognition based on address information,
Only the dictionary memory corresponding to the postal code
By selecting from the memory, the dictionary can be narrowed down.
No, high readability of address recognition for wide area mail
It can be performed quickly with high accuracy.

[0011] Also, the address recognition device described later.
, The photoelectric conversion element can identify the postal code on the postal matter.
If not, simply declare this post as unprocessable
Instead of handling errors and releasing them,
Selection is made by assuming a dictionary memory corresponding to the number of stations. So
Then, according to the selected dictionary memory, address recognition is performed.
Process, which requires human intervention.
Minimize error mail that must be done
Is made possible. Also in address recognition method
However, for the same purpose, the photoelectric conversion element
Most processed in the past when the issue could not be identified
Assuming the dictionary memory according to the station, select
The address recognition process is performed. Therefore, human processing
As much error mail as possible.
To be able to

[0012]

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 2 shows an outline of a hardware configuration applied to the address recognition device. The character recognition unit 101 includes a photoelectric conversion unit 102, an area detection unit 104, a line detection unit 106, a character detection unit 108, and a character recognition unit 101, and is referred to when the character recognition unit 101 performs character recognition. A character dictionary memory 130 including all basic characters, and an address recognition unit 120 that recognizes an address from a combination of read characters recognized by the character recognition unit 101 with reference to the character type table 140 and the address dictionary memory 150. . Character recognition unit 110 and address recognition unit 12
0 includes a microcomputer (CPU), and the functions of these units are realized by software running on the CPU. Note that the character recognition unit 101 itself can be configured by a character recognition technique employed in a conventional optical character reader (OCR) or the like.

First, the contents of the system of this embodiment will be briefly described. The photoelectric conversion unit 102 captures a surface image of the mail including the character to be read, binarizes the image, and outputs original image data including the character image.

The area detection unit 104 extracts an area in which an address is described from a window signal and a fine scanning signal obtained by manipulating the surface of a mail. That is, first, image information (original image data) of the entire postal matter is two-dimensionally compressed, and the processed image is used to block a global image, thereby performing global area detection that is not restricted to details. Next, projection data for each block is generated, the complexity / direction of the line text is determined, and the determination result is edited by using a previously assigned address area editing knowledge to determine an address area.

From the destination area, binarization threshold candidates obtained from the image density histogram inside the area,
A description direction candidate obtained from the detected destination area position and a font candidate such as handwritten / printed type are output. In the destination area editing knowledge, data based on a statistical survey for the destination position / area extracted from a large number of mails is taken in as destination area image knowledge.

On the surface of the mail, there are a sender name, a sender address, a communication portion, a fee printing surface such as a stamp, and various kinds of advertisements, which are described in the same manner as the address, and the location, area, complexity, The printing direction and the like are used to construct the destination area image knowledge.

The line detection unit 106 receives the output from the area detection unit 104, removes noise components such as a frame line surrounding the line of the address character or a ruled line on the address name, and separates and extracts the character line unit. In addition, a print density histogram is obtained for each detected character line, and an optimum binarization threshold value is determined for each detected character line.

The algorithm for character line unit detection here is basically the same as the algorithm for address region detection. However, the character line unit detection does not take a general view, but excludes the underlined region, the blank region, and the like by performing a more detailed image analysis than in the case of the destination region detection, and selectively extracts the image to be read (line image).

The character detection unit 108 separates an image into characters in accordance with the line image / binarization threshold value extracted / determined by the line detection unit 106. That is, it is assumed that a value equal to or less than a threshold value determined in advance by the character line width data is a name assigned to a print type, and the others are assumed to be names assigned to a handwriting. Then, character separation in the line direction is detected from the projection information. The detection logic is space detection caused by blanks between characters, and contact between characters and separation within a character are determined by predicting fluctuation of a detected character circumscribed frame on the basis of a square. If the evaluation decision is ambiguous, multiple detection candidates are accepted.

The character recognition unit 110 performs a character recognition process similar to that of a known optical character reader (OCR).
The identification target character types prepared in advance for the character recognition processing are all characters of Arabic numerals / kanji numerals and katakana / hiragana and approximately 200 kanji characters used for addressing. These characters (400 characters or less in total) are prepared for print and handwritten characters (up to 800 characters).
Characters).

It is not realistic to achieve 100% correct character identification for an input character image containing ambiguity.
Therefore, in the embodiment of the present application, on the assumption that there is a post-processing using the address knowledge (address dictionary memory 150) later,
Ten candidates are output in the order of higher similarity in the form of identification candidates for each input image. By doing so, it is possible to obtain a high identification rate that cannot be obtained by using a single identification candidate. In addition, regarding the classification of the writing direction (vertical writing / horizontal writing) left as ambiguity in writing,
For input, the self-portrait is rotated by 90 ° and 180 ° to perform identification processing, and a series of character candidate matrices is output for each rotation unit.

The address identification unit 120 determines a destination from a character identification candidate matrix using a knowledge database (part of the dictionary memory 150) such as a ward name, a town name, and a name of a major recipient given to the delivery office. For example, taking the Adachi Post Office in Adachi-ku, Tokyo as an example, there are 31 types of standard notation town names prepared as addresses, and 10 types of large recipients such as Adachi City Office. However, since the notation of these names includes various deformations, learning is advanced in accordance with the frequency of the deformations, and the address knowledge data is increased as long as it does not conflict with other town names / large recipient names.

For example, taking the name of a small town in Adachi Ward as "Kodai" as an example, the standard knowledge is "Odai", but this town name is "Kodai", "Otai", "Kotaimachi",
There are deformations such as "Odaicho". These modified notations are stored in the address knowledge database (address dictionary memory 150) as pointing to the "small stand" of Adachi-ku.

From the knowledge based on the address / major recipient name including the above-described deformation, a word formed by a group of two or more consecutive characters is created, and the character identification candidate matrix is evaluated based on the word. For example, if the created word is “Tokyo” and the corresponding character identification candidate matrix is as follows, “Tokyo group”, “Toto city”, “Tokyo group”, “Kyoto city”, From the word groups such as “Kyoto” and “Kyoto”, the place name registered as the address knowledge “Tokyo” in the address dictionary memory 150 is finally selected.

After that, when all the words have been integrated, knowledge processing is performed again on some word candidates. This is a process of evaluating whether or not a word arrangement as an address or a major recipient name can be obtained while maintaining the adjacent relationship between words at the same time as word matching. For example, it is evaluated whether the word "Tokyo" is followed by the word "Adachi-ku", followed by the town name, or the word "Adachi-ku" is followed by the name of a major recipient such as a ward office. In this evaluation, a predetermined score is given to each word (a person who is closer to the actual address and place name notation has a higher score), and the address having the highest score obtained by accumulating the scores can be output.

Except for postal items addressed to the major recipients, in order to identify the streets and streets (numbers / numbers, etc.) after the street name, the position of the detected street name on the character image and the destination name format code (vertical writing / Information such as horizontal writing / rotation) is passed to the next processing as dependent information.

After the recognition of the street name, it becomes clear that the characters following the street name represent chome / block. The address recognition unit 120 identifies the chome / block and determines a final address, converts it to an address classification code, and outputs it.

In summary, the embodiment of FIG. 2 functions as follows. That is, the character recognizing unit 101 refers to the character dictionary memory 130 and performs character reading / recognition of handwritten characters or print characters. Address recognition unit 1
In 20, the character table 140 is referred to, and the character recognition unit 1 is referred to.
The combination of one or more read characters recognized in 01 is 1
Created above. Next, the address dictionary memory 150 is retrieved by the created combination of read characters, and a correct address display word corresponding to the read character combination is read from the dictionary memory 150. Character codes of one or more address display words read from the dictionary memory 150 and address display numerical values (chome, number, number, etc.) following these address display words
Are collected and output as an address reading result (address classification code).

FIG. 1 shows further features of the address recognition apparatus of the present invention. FIG. 1 is a flowchart for address recognition according to an embodiment of the present invention. Hereinafter, features of the present invention and FIG. 1 will be described.

In this figure, almost in the same manner as in the case described above with reference to FIG. 2, the mail information is fetched by the photoelectric conversion 1 into the mail address recognition device by the photoelectric conversion 1. Next, it is detected 2 to which of the surface areas of the mail the name is distributed. Further, row 3 such as a destination name is specified,
Finally, each character is detected 4 (detection means, detection
Process, photoelectric conversion element) .

Here, the postal matter address recognition device of the present invention recognizes the postal code 5 (identification means, identification).
Another process) specifies the location of the delivery office of this postal matter. Next, the corresponding dictionary memory 11 corresponding to the postal code
Only one (storage means, storage step, dictionary memory) is selected.

Therefore, a character recognition operation 6, which will be described below,
The character collation operation such as the street name / large name identification 7 and the chome / block identification 8 is performed using the dictionary memory 10 for one station (recognition means and recognition step) . It is possible to read the address of the mail of the neighboring office at the same level as in the case of only one office.

In the conventional device, in order to read addresses not only for a single station but also for a plurality of stations, it is necessary to cover addresses in the jurisdiction for the number of stations, and the character / word dictionary memory is simply enlarged. If it is performed only, the dictionary memory pattern to be compared is enlarged, and the reading accuracy and the processing speed are reduced.

According to the present invention, a zip code is read before performing word / address recognition for address reading, and the post office corresponding to the destination of the postal matter is specified by the zip code. That is, a handwritten postal code in a red frame and a printed postal code in a window / label for a destination are read.

More specifically, an embodiment of the present invention will be described with reference to FIGS. 4 and 5. FIG. FIG. 4 is a table showing addresses of a dictionary memory according to an embodiment of the present invention, and FIG. 5 is an external view of an example of a mail to be recognized by the address recognition device of the present invention.

In this figure, for example, when the present invention is applied to the postal matter shown in FIG. 5, the upper left character string in the enclosing frame can be read as the postal code "140". The dictionary memory selection database is searched to determine that the mail is addressed to the Shinagawa Bureau. As a result, the address of the dictionary memory of the characters that make up the address within the Shinagawa Bureau,
An address in an address dictionary memory in which words constituting a street name / a large name and its connection condition are registered is obtained, and characters or addresses are recognized.

In the part indicated by the address in the address dictionary memory of the Shinagawa Bureau, words such as the name of a town belonging to the Shinagawa Bureau and the name of a large company within the Shinagawa Bureau are registered in Shinagawa Wards such as Kita-Shinagawa, Minami-Oi, and Yashio. In the area indicated by the character dictionary memory address, each character constituting the word is registered.

Therefore, in the proposed method, by selecting a dictionary memory in a range corresponding to one station specified by the postal code, the address reading processing capability for one station can be maintained and the address can be addressed to a plurality of stations. Can read the mailing address. However, the dictionary memory used for recognition must be switched
Since the switching is performed by switching the address of the dictionary memory as described above, the dictionary memories for a plurality of stations need to be held in a memory such as a semiconductor. However, a circuit for collating the input pattern with the dictionary memory pattern and a circuit for searching for an address word are sufficient for one station. If the dictionary memory is not switched by the postal code, the dictionary memory pattern to be compared and the number of words to be searched increase, and the time required for address reading increases unless a circuit proportional to the dictionary memory is prepared. As the types of target characters and words increase, the performance of address reading itself decreases.

Further, an embodiment of the present invention will be described with reference to FIG. FIG. 6 is a flowchart of a name recognition apparatus according to another embodiment of the present invention. In other words, in this flowchart, the process is the same as that of the above-described embodiment up to selecting the dictionary of the station corresponding to the number 22 when the number can be read 22 as a result of the number recognition 21 of the postal matter (first).
Recognition means, first recognition step) . However, if the postal code is unreadable22, assuming that the postal code of this post is the one of the plurality of dictionary memories that has the highest number of processings, this post is selected 26 and Address character recognition 24
(Second recognition means, second recognition step) . This reduces the number of cases that are ejected from the machine as being unreadable and reduces the number of cases that require human processing. (If this processing is not performed, it becomes impossible to read immediately when the postal code cannot be read. Discharged from 27
Will be done). The station with the largest number of processing cases may be pointed out in advance when the machine is installed, or the frequency of appearance of the postal code may be stored and automatically determined.

Further, in the above-described embodiment, the method of selecting the plural dictionary memories by the postal code is adopted. However, this method is applied to a more detailed address division (for example, XX town).
The same effect of the present invention can be obtained by providing a plurality of dictionary memories corresponding to the above, and recognizing the address section, and switching the dictionary memory according to the recognition result to perform processing. .

As described above, according to the present invention, it is possible to read the address of a plurality of stations without deteriorating the performance only by expanding the capacity of the dictionary memory composed of relatively inexpensive semiconductor memory. Becomes possible.

[0043]

As described above in detail, according to the present invention, by using the result of reading the postal code, the used part of the dictionary memory is switched, so that the mailing address for a plurality of post offices can be addressed. Can be read. This method can greatly reduce the contents of the dictionary memory to be compared with the input pattern compared to the case where the dictionary memory is not switched, and has the same properties (recognition accuracy and processing speed) as when reading the address of one postal mail. ) Is realized only by increasing the dictionary memory capacity.

[Brief description of the drawings]

FIG. 1 is a flowchart for address recognition according to an embodiment of the present invention.

FIG. 2 is a block diagram of a general address recognition device.

FIG. 3 is a conceptual diagram showing the flow of mail in recent years and the current state of mail automation machines.

FIG. 4 is a table showing addresses of a dictionary memory according to one embodiment of the present invention.

FIG. 5 is an external view of an example of a mail to be recognized by the address recognition device of the present invention.

FIG. 6 is a flowchart of a name recognition apparatus according to another embodiment of the present invention.

[Explanation of symbols]

1 ... photoelectric conversion, 2 ... address area detection, 3 ... row detection, 4 ...
Character detection, 5: Postal code recognition, 6: Character recognition, 7: Street name / large name identification, 8: Chome / block identification, 9: Recognition result, 1
0: One station corresponding dictionary memory, 11: One station corresponding dictionary memory,
33: dispatch station, 34: delivery office, 35: deliverer, 36: stamping machine, 37, 38: postal code / address automatic reading / sorting machine, 41: postal matter, 101: character recognition system, 102
... Photoelectric conversion unit, 104 ... Area detection unit, 10
6 ... line detection unit, 108 ... character detection unit, 11
0: character recognition unit, 120: address recognition unit, 1
30: character dictionary memory, 140: character type table, 15
0 ... Address dictionary memory

Claims (6)

(57) [Claims] 1. Recognition of a mailing address in which a postal code is described
In the apparatus, the character information and a plurality of storage means previously stores dictionary information for address recognition according to a plurality of zip codes, a detecting means for detecting character information of mail, the detection means detects Zip code from
And identifying means for identifying, storing hand the identification means corresponding to the zip code that identifies
Selecting means for selecting a stage from the plurality of storage means, and using the storage means selected by the selection means, recognizing means for recognizing the address of the postal matter based on the character information detected by the detecting means, A mail address recognition device comprising: 2. The method according to claim 1, wherein the detecting means detects character information of the mail.
A photoelectric conversion element, wherein the storage means is assigned to an address according to the plurality of postal codes.
Includes multiple dictionary memories that pre-store information about
2. The mail address recognition device according to claim 1, wherein
Place. 3. A method for recognizing a mailing address in which a postal code is described.
In the method, a detecting step of detecting character information of a postal matter, and a zip code from the character information detected in the detecting step.
And dictionary information for address recognition according to a plurality of postal codes.
From the plurality of storage means stored in advance,
Select a storage means corresponding to the identified postal code.
And-option step, by using a storage means selected by said selecting step, said
The character information detected in the detection step allows the
And a recognition step of recognizing the name of the mail. 4. The detecting step detects character information of a postal matter.
4. A photoelectric conversion device according to claim 3, wherein
Name recognition method for mail items described. 5. A method for recognizing a mailing address in which a postal code is described.
In the device, dictionary information for address recognition according to a plurality of postal codes is stored.
A plurality of storage means stored in advance , a detection means for detecting character information of a mail, and a postal code from the character information detected by the detection means
Identification means for identifying, a storage means corresponding to the postal code identified by the identification means, a selection means for selecting from the plurality of storage means, using the storage means selected by the selection means,
The mail information is detected based on the character information detected by the detecting means.
First recognizing means for recognizing the first name and the recognizing means cannot identify the postal code
Of the plurality of storage means,
Select the storage means according to the large number of stations and use it to
And a second recognizing means for recognizing the address of the mail. 6. Recognition of a mailing address in which a postal code is described.
In the method, a detection step of detecting the character information of mail, the postal code in the character information detected by the detecting step
Identification information and dictionary information for address recognition according to a plurality of postal codes
From the plurality of storage means stored in advance,
Selection to select storage means according to the identified postal code
Step and using the storage means selected in the selection step,
The character information detected in the detection step allows the
Whether the postal code cannot be identified in the first recognition step of recognizing the postal name and the identification step
Of the plurality of storage means,
Select the storage means in accordance with the number of large stations before using this
A second recognition step of recognizing a destination name of the postal matter.