JP2685862B2 - Address recognition method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Field of Industrial Application) The present invention relates to an address recognition method for reading an address of a mail or the like.

(Prior Art) For example, regarding the reading of an address written in a US mail, in the conventional art, candidates are narrowed down in order of a state name and a city name from the read word group, and finally the street is searched. I used to find an address by searching for a keyword word such as my name, POBOX (equivalent to a POBOX in Japan).

However, in such a conventional technique, if the characters of the word that is the keyword are faint or the notation of the address on the postal matter is ambiguous, if the characters cannot be read correctly, the address candidates are naturally narrowed down. I can't.

Further, depending on the character recognition method adopted, it may be easy to make an error in distinguishing between a number and an alphabet. In such a case, a word presumed to have the street attribute and a word presumed to have the pe-o-box attribute may be listed as recognition candidates, and the normal rule for mail items alone is: I could not narrow down the recognition candidates any further.

(Problems to be Solved by the Invention) As described above, according to the conventional technique, it may not be possible to narrow down the recognition candidates of the address.

Therefore, the present invention further reduces the cases where it is not possible to narrow down the recognition candidates of the address, and in particular, even if the address notation on the mail is defective or the character recognition is defective, the recognition is performed. It is an object to provide an address recognition method that can specify candidates.

[Means for Solving the Problems] (Means for Solving the Problems) In order to solve the above problems, the present invention reads an zip code and an address having a hierarchical structure written in a letter to recognize an address. In the first step of reading the postal code and the address, the second step of recognizing an upper-layer address and the postal code of the address read in the first step, and the second step A third step of narrowing down the range of the postal code from the address recognized in step 3, and determining whether or not the postal code recognized in the second step is within the range;
The fourth step of searching the attribute corresponding to the postal code determined to be within the range by the third step, and the address of the lower hierarchy of the address according to the attribute searched in the fourth step. And a fifth step for recognizing the address.

(Operation) According to the present invention, it is determined whether or not the read zip code is within the range indicated by the read address, and the ambiguity with respect to the read address is compensated by the attribute assigned to the zip code. , More accurate address search becomes possible.

(Embodiment) Next, an embodiment of the present invention will be described in detail with reference to the drawings. This embodiment relates to an address recognition device for US mail. This address recognition device includes an input device 11 for reading an address on a mail piece, as shown in FIG.
The input device 11 is composed of a photoelectric conversion device such as a CCD and converts an image on a mail item into image information of "0" and "1". The image information obtained by the input device 11 is supplied to the arithmetic device 13. This arithmetic unit 13 performs recognition processing using a dictionary. Further, in this embodiment, the recognition process is performed more efficiently by using a 5-digit zip (ZIP) table 15 described later. The result of recognition by the arithmetic unit 13 is displayed on the output unit 17, etc.

Next, the 5-digit zip table 15 will be described. As shown in FIG. 2, this table 15 is constructed in the order of the size of the 5-digit zip code. The content describes the type, state name ID, and city name ID for each of the 5-digit zip codes. The zip code is a type of postal code used in the United States.

The above-mentioned type is 7-bit information, and the attribute of the address for the zip code is expressed.
Specifically, as shown in Fig. 3, non-unique zip (NON-UNIQE ZIP), P-box (POBO)
X), unique zip (UNIQE ZIP), AP / FP (APO / FPO), etc. Unique zip is a code attached to a specific building or organization, and only one of the codes on the left corresponds to that code. On the other hand, the non-unique zip means that a plurality of houses and buildings correspond to one code. P-box corresponds to a post office box in Japan. AP / FP is a code mainly given to the military.

The state name ID is a code of the state name,
As shown in FIG. 4, it is expressed here as 7-bit information. A city name ID is a code of a city name, and is composed of a state name ID and how.

That is, if the 5-digit zip table 15 configured as described above is pulled, the corresponding address attribute, state name ID, city name ID
Is obtained.

Next, a recognition process using the 5-digit zip table 15 configured as described above will be described.

In this process, as shown in FIG. 5, character inspection is performed on the image information obtained by the input device. Since this is a well-known technique, it will be briefly described that distribution statistics are taken in two axis directions (horizontal and vertical) with respect to image information. At this time, the distribution is taken by the pixels having a coarser density than that used in the normal recognition processing. Therefore, if the pixels are distributed along the same direction as the character arrangement (called the first direction), the lines can be cut out.

When the line is cut out, the pixels are distributed along the second direction perpendicular to the first direction with respect to the image of the area forming the line. As a result, the pixels are counted in the area with the character, and the pixels are not counted in the area without the character. As a result, characters can be cut out from the line, and character inspection can be achieved.

Following the character inspection, character recognition is performed. In this embodiment, character recognition is performed using, for example, the composite similarity method.
After this character recognition, word recognition, which is a combination of characters, is executed. For both character recognition and word recognition, a recognition result is not simply obtained, but a plurality of recognition candidates are obtained in descending order of likelihood. In the usual case, a reading on the mail piece will recognize several words. For example, a state name, a city name, a street name, and a peobox are obtained as a result of recognition. To be precise, some restrictions presume the recognized words to be those mentioned above.

When word recognition is completed in this way, the state name and city name are searched. For example, it is assumed that the word "aaa" is recognized by the recognition processing. As explained below, if the rule infers this word to be a city name, the address dictionary will say "aa
Search for a city (or state) named "a". In this embodiment, not only the presence / absence of the address but also the code of the corresponding address is obtained. That is, after recognition, if there is a corresponding address, read the code,
If there is no address, the code corresponding to "none" is read.

Also, as mentioned above, some empirical rules are used when searching for addresses. For example, in the area recognized as an address, the rightmost alphabet in the lowest row is a state name and the left is a city name. The number to the right of the state name is the five-digit zip code. There is a street name in the line above the city name (which may be the left of the city name in rare cases), and there may be a PEO Box to the left of this street name. Using such a rule, a dictionary (area) to be searched is specified and searched for the result of word recognition.
As described in the (Prior art) section, if the written state of the characters is bad or if the reading is bad,
Recognition and search cannot be performed well. For example, if there is the above defect, there is a possibility that the recognition process cannot correctly recognize where the state name, the city name, the street name, the peer box, or the like is.

However, the probability of making a mistake in the name of a state or city is lower than that of making a mistake in PEO Box, etc. That is, it can be said that the recognition of the state name and the city name is more reliable in reading and recognizing the address. Therefore, in this embodiment, as will be described below, a zip code is obtained from the recognition result of the state name and the city name, and the inspection is satisfactorily performed using the zip code as auxiliary information.

Specifically, when the state and city name search is completed,
Get the zip code from the city name. The zip code is usually 5 digits, and when the state name and the city name are estimated, the zip code corresponding to the zip code is obtained in most cases. Then, the attribute of the corresponding address is checked from this zip code. The information of the type may be read out using the 5-digit zip code table. The attribute of the corresponding address is known from this information. For example, in type as above,
If "1" is set at the 6th bit, it can be seen that there is a "Peo Box". Also, if "1" is set at the 5th bit, it can be seen that it is a "unique zip".

Next, it is checked whether or not the read word has a keyword. Specifically, the attributes such as "Peo Box" and "Unique Zip" in the read word can be understood.

Therefore, based on the attribute of the read address, the one having the corresponding attribute is searched from the 5-digit zip code table. When there is a zip code having the corresponding attribute, the priority of the address corresponding to the zip code is increased. This priority is an evaluation measure of the degree of similarity in pattern recognition.

After such processing, the one having the highest overall similarity is output as the address recognition result as in the case of the normal pattern recognition.

Next, the above processing will be described for a specific mail item. Now, it is assumed that the recognition target on the postal matter is configured as shown by 61 in FIG. Here, "1111" on the second line is
It is a "Peo Box" and is an attribute in this address. "CINCINNATI" on the third line is the city name. The left "OH" in the 4th line is the state name, and the right "00000" is the zip code.

For such an address, "OH" is obtained as a candidate state name (Ohio) after word recognition. Similarly, "CINCIN
NATI "is obtained as a city name candidate.

A zip code is obtained from such recognition candidates. In this example, the zip code corresponding to the recognition candidate Cincinnati, Ohio is “45200 to 45900”.

On the other hand, the information on the second line in the address is recognized as the word "1111", but it is presumed to be "Peo Box" from the rules of mail. A 5-digit zip table is searched based on this attribute information. That is, of the types of the 5-digit zip table, the one having "1" at the 6th bit is searched. In this example, three zip codes “45201” and “452”
"50" and "45258" correspond to this.

So, the state names corresponding to these three zip codes,
Take out the city name with priority. That is, only the corresponding state name and city name are searched, and matching with the state name candidate and city name candidate is performed. By this processing, the state name and the city name having the highest degree of coincidence are set as the recognition result. When there are a plurality of state name candidates and city name candidates, the above process is performed for all the candidates, and the one with the highest degree of coincidence is comprehensively determined and is set as the final recognition result.

Next, another example of postal matter will be described. This example is the 7th
As shown in the figure, the first is the destination (group name), the second
It is assumed that the line is the city name, the left "NY" in the third line is the state name, and the right "10149" is the zip code.

For such an address, "NY" is obtained as a state name candidate and "NEW YORK" is obtained as a city name candidate. The range of corresponding zip codes from these state name candidates and city name candidates is limited to “9001 to 10292”.

On the other hand, the information “10149” on the right of the third line is read and treated as a zip code. Then, since this zip code is included in the above range, it is determined that it is accurate. Moreover, if you pull the 5-digit zip table with "10149", you can see that it is a "unique zip". This unique zip shows a specific company or organization.
Moreover, in this example of address, there are no candidates for PEO Box or street name. Therefore, the information on the first line is the specific company or organization name, and the read state name or city name is output as the recognition candidate with the highest degree of coincidence.

Although the embodiments of the present invention have been described above, the present invention is not limited to the above embodiments, and it goes without saying that various modifications can be made without departing from the spirit of the present invention.

[Effects of the Invention] As described above, according to the present invention, it is possible to further improve the accuracy in address search. Further, even if the notation of the address is ambiguous, more accurate address search can be performed.

[Brief description of the drawings]

1 to 7 are diagrams for explaining an embodiment, and FIG. 1 is a block diagram showing an outline of an apparatus according to one embodiment,
2 to 4 are diagrams showing the structure of the 5-digit zip table shown in FIG. 1, FIG. 5 is a flow chart showing the recognition procedure in FIG. 1, and FIGS. 6 and 7 are postal items. 5A and 5B are diagrams showing a concrete example of FIG. 11 …… input device 13 …… computing device 15 …… recognition dictionary 17 …… 5-digit zip table 19 …… output device

Claims (1)

(57) [Claims] 1. An address recognition method for recognizing an address by reading a postal code and an address having a hierarchical structure written on a letter, comprising: a first step of reading the postal code and the address; A second step of recognizing an upper-layer address and a postal code of the addresses read in the step, and a range of the postal code from the address recognized in the second step is narrowed down to the second step. The third step of determining whether or not the zip code recognized in the step is within the range, and the attribute corresponding to the zip code determined to be within the range by the third step is searched. Address recognition, which comprises a fourth step of recognizing, and a fifth step of recognizing an address in a lower hierarchy of the address based on the attribute retrieved in the fourth step. Law.