KR100286163B1 - Address recognition method, address recognition device and paper sheet automatic processing system - Google Patents

Abstract

[Purpose] Read addresses quickly and accurately without registering a large number of neighborhood names.

[Configuration] Image extraction of input strings indicating addresses including neighborhood name codes, prefectural prefecture neighborhood names, and residential display numbers corresponding to neighborhood names listed on paper sheets, and sliced image extracted input strings by 1 character The character recognition section 104, which inquires each character in advance of the contents of the character dictionary 105 in which the character information is registered and the character recognition section, and the character recognition section of the input string based on the character recognition result in the character recognition section 104 A character type determination unit 106 that determines the character type of each character, and checks the character type sequentially from the head of the input string, detects the neighborhood name code and the residential display number, and detects the detected neighborhood name code and the residential display number. Has an address recognition unit 107 that recognizes as an address together.

Description

Address recognition method, address recognition device and paper sheet automatic processing system

1 is a block diagram showing a configuration of a first embodiment (Embodiment 1) of the present invention.

2 is a diagram showing an example of an address described in a paper sheet.

3 is a diagram showing an example of a text dictionary.

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

5 shows the relation between input string and character position P (P), dictionary NO. A diagram showing a relationship between D (j), a dictionary, a category, and a character type, a category of an inquiry result stored in r (iii), and a character recognition result output from the character recognition unit.

6 is a diagram showing a relationship between a character type and a category, a category of the final result stored in f (1) to f (m), and a character type determination result output from the character type determination unit.

7 is a block diagram showing a configuration of a second embodiment (Embodiment 2) of the present invention.

8 is a diagram showing an example of the neighborhood name character table.

9 is a diagram illustrating an example of a city name table.

10 is a flowchart for explaining the operation of the second embodiment.

11 is a block diagram showing the configuration of a third embodiment (Embodiment 3) of the present invention.

12 is a diagram illustrating an example of a city name table.

13 is a view showing an example of the neighborhood name table.

14 is a flow chart for explaining the operation of the third embodiment.

FIG. 15 is a diagram illustrating an outline of an automatic mail processing system that converts a recognized address into a barcode code, prints on paper sheets, and performs processing; FIG.

FIG. 16 is a diagram showing examples of mail items in China and Korea.

17 is a block diagram showing the structure of a conventional address recognition apparatus.

<Explanation of symbols for main parts of drawing>

101: image extraction unit 102: address position detection unit

103: character slice portion 104: character recognition portion

105: character dictionary 106: character type determination unit

107: address recognition unit 201: neighborhood name character search unit

202: the number of characters in the neighborhood name table 203: city name search unit

204: City name table 205: Address recognition unit

301: City name search unit 302: City name table

303: neighborhood name search unit 304: neighborhood name table

305: character name search unit neighborhood name 306: address recognition unit

1 A: Neighborhood name code 1 B: City name / neighborhood name

1 C: Residential labeling number

The present invention relates to an address recognition method for generating an address written on a paper sheet at a high speed and high accuracy in a paper sheet classifier, an address recognition device and the read address, and converts it into a barcode to print and process on a paper sheet. It relates to a paper sheet automatic processing system.

A conventional address recognition apparatus will be described with reference to FIG. In this address recognition apparatus, first, in the image extraction unit 121, the character pattern described on the paper sheet is input, converted into an image signal, and the character pattern of the part corresponding to the address in the image signal is extracted. Then, the character slice processing is performed on the character pattern extracted by the image extraction unit 121 in the character recognition unit 122. Then, in the character recognition unit 122, the characters previously stored in the dictionary 123 and the characters sliced for each character are compared, and character recognition is performed.

Next, in the word recognition unit 124, a word is read from the word dictionary 125 in which the word is stored in advance with respect to the character string recognized by the character recognition unit 122, and the read word is compared with the recognized character string. By inquiring, word recognition is performed. Then, the address recognition unit 126 compares the address read out from the address dictionary 127 in which the address is stored in advance with the word recognized by the word recognition unit 124 to perform address recognition.

Then, register the address name to be read in advance in the address dictionary, compare the result of reading the string on the paper sheet with the address name read from the address dictionary, and refer to e. For example, Japanese Patent Application Laid-Open No. 2-181287 or USPS Receiving Address Automatic Reading Separator (NEC Publication No. 187 Vol. 38 No. 4 P41 to 47 USPS Receiving Address Automatic Reading Separator (1985)) is disclosed.

In addition, word extraction means for scanning characters written on paper sheets and extracting scanning patterns as word units, storage means for storing a preset dictionary pattern indicating an address, and words extracted from the word extraction means on paper sheets. A method for recognizing an address by providing a comparison means for comparing with a prepattern while inquiring according to the described procedure, and selecting the combination having the highest similarity from the comparison result in the comparison means, for example, JP-A-62 62388 is disclosed.

However, in such a conventional address recognition apparatus, all address names to be read must be registered in advance of the address dictionary. For example, in Japan, about 3,000 prefectures and up to 120,000 address names must be registered, and the contents of address dictionaries become enormous. As a result, there is a problem that the apparatus becomes large. In addition, there is a problem that the processing speed is very slow because the number of comparison queries is very large.

SUMMARY OF THE INVENTION The present invention has been made to solve such a problem, and an object thereof is to provide an address recognition method and an address recognition device capable of reading an address at high speed and with high accuracy without registering a large number of neighborhood names. have. Another object is to provide a paper sheet automatic processing system which converts the read address into a barcode and prints and processes the paper sheets.

In order to achieve this object, the first and second inventions (invention according to claim 1) and the sixth and seventh inventions (invention according to claim 6) correspond to each neighborhood name described on the paper sheet. An image is extracted from an input string representing an address including a neighborhood name code, a district name (provincial prefecture's neighborhood name), and a housing display number, and the image extracted input string is sliced by one character, and each character is pre-text information. Character recognition is compared with the contents of the registered text dictionary, and the neighborhood name code and residential indication number are detected based on this character recognition result, and the detected neighborhood name code and residential indication number are recognized together as addresses, and Based on the recognition result, the character type of each character of the input string is determined, and the character string is sequentially checked from the head of the input string to determine the character type. It will have to detect a code name and number display housing.

For example, based on the character recognition result, the character type (a distinction of Chinese characters, numbers, symbols, etc.) of each character of the input string is determined, and the character string is sequentially checked from the head of the input string, and the number of characters is firstly counted. When the character type is changed from a kanji to a number or symbol in the input string, for example, the area name code is regarded as a neighborhood name code. Detect by number.

In the third invention (invention according to claim 3) and the eighth invention (invention according to claim 8), in the second invention and the seventh invention, the number of characters of the neighborhood name corresponding to the neighborhood name code is registered in advance. A first table and a second table in which the city name is registered in advance are installed, the neighborhood name code is detected based on the character recognition result, and the number of characters of the neighborhood name corresponding to the neighborhood name code is retrieved from the first table. Based on the character recognition result, the city name constituting the input string is searched from the second table, and the residential display number is detected based on the number of characters and the city name of the searched neighborhood name.

For example, in the third invention and the eighth invention, based on the character recognition result, the portion where the numbers or symbols are continuous is regarded as the neighborhood name code, and the number of characters of the neighborhood name corresponding to the neighborhood name code is shown in the first table. Search from. Further, based on the character recognition result, the name of the city constituting the input string is retrieved from the second table. Then, the starting position of the housing display number is detected based on the number of characters of the neighborhood name retrieved from the first table and the municipality name retrieved from the second table, and the character string after the starting position is regarded as the residential display number.

In the third invention and the eighth invention, the fourth invention (invention according to claim 4) and the ninth invention (invention according to claim 9) are used in the specific area designated in advance among the city names registered in the second table. If the neighborhood name flag is added to the city name and the searched city name is added, the neighborhood corresponding to the city name to which the neighborhood name flag is added based on the character recognition result and the searched city name. The name of the neighborhood and the neighborhood name code corresponding to the neighborhood name are searched from the third table in which the name and the neighborhood name code corresponding to the neighborhood name are pre-registered, and the neighborhood name code corresponding to the neighborhood name. Based on this, the house number is detected.

In the fifth invention (invention according to claim 5) and the tenth invention (invention according to claim 10), in the fourth invention and the ninth invention, a specific region to which a neighborhood name flag is added is handled for each region. Based on the number of paper sheets to be specified, they are to be designated in advance.

The eleventh invention (invention according to claim 11) recognizes an address by the address recognition method of the second invention from an input string representing an address written on a conveyed paper sheet, and converts the recognized address into a barcode. The paper sheet is converted and printed on the paper sheets, and the paper sheets printed on the barcode are processed according to the printed barcode.

EMBODIMENT OF THE INVENTION Hereinafter, this invention is demonstrated in detail based on embodiment.

For example, in the case of Japan, an address written on a paper sheet is composed of a postal code corresponding to a city name, a prefectural prefecture's neighborhood name, and a residential indication. The housing number is indicated by symbols such as numbers and hyphens. Here, in the embodiment described below, it is assumed that a code consisting of a number corresponding to the neighborhood name (hereinafter referred to as neighborhood name code) is added to the head of the address. Hereinafter, this neighborhood name code, the prefecture prefectural municipal district name, and the residential indication number will be called an address.

Here, an example of an address is shown in FIG. In the figure, 1 A is a neighborhood name code, 1 B is a city name / village name, and 1 C is a residential display number. In the example of this address, the upper three digits of the neighborhood name code 1 A, namely, "183", correspond to a currently defined postal code, and in this case, "Chu Zhong City". The lower three digits of the neighborhood name code 1 A, `` 123, '' correspond to the neighborhood name 日新 町.

Embodiment 1

Next, 1st Embodiment (1st Embodiment) of this invention is described with reference to drawings.

In this embodiment, an input string corresponding to an address is character-recognized for each character, and a character type (a distinction of Chinese characters, numbers, symbols, etc.) is determined based on the recognition result. Then, the character type is sequentially checked from the head of the input string. First, the portion where the number or symbol is continuous is viewed as a neighborhood name code. In addition, the position where the type of character changes from a Chinese character to a number or a symbol is detected, and the subsequent part is regarded as a residential display number. This neighborhood name code and residential identification number are recognized together as an address.

FIG. 1 is a block diagram showing the structure of Embodiment 1, and the image extraction unit 101 converts letters, symbols, and the like on paper sheets into electrical image signals.

The address position detection unit 102 detects information such as the length, the direction and the number of columns of the character string from the image signal converted by the image extraction unit 101, and corresponds to the address described on the paper sheet based on the detected information. Find the string to As the address position detection technique in this case, for example, the technique described in Japanese Patent Application Laid-Open No. 59-39065 by the present applicant can be considered.

In Japanese Unexamined Patent Publication No. 59-39065, a certain range of the area of a mailing address traveling at a constant speed is scanned, and a character string obtained by this is compressed for a certain period, and a fixed period of time is obtained. The character string start and end positions of the compressed character string of the compressed character string are detected, and the group of address names written on the mail, its line, position, size of the character name character, and the slope of the line are determined from the combination of the compressed strings obtained thereby. .

The character slice unit 103 slices the character string detected as an address for each character. As the character slicing technique in this case, for example, the adoption of the technique described in Japanese Patent Application Laid-Open No. 63-57831 by the present applicant can be considered.

In Japanese Patent Application Laid-Open No. 63-57831, the input operation signal is updated while being compressed at right angles to the direction of the scanning line, and the compression width is sequentially stored for each character block of the compressed scanning signal. The scanning signal is compared, and if the input scanning signal corresponding to the continuous character block is compressed and there is no portion representing the character, the character slice is detected. The compression width stored at the timing of the character slice is read. I'm making it.

The character recognition unit 104 compares the sliced character with the previously registered character dictionary 105 and recognizes the sliced character. In the character dictionary 105, about 3000 Chinese characters, numbers, symbols, and the like are registered in advance, and the character recognition results in the character recognition unit 104 are output as JIS codes, for example. That is, when the sliced character is "1" or "府 (part)", the character recognition result in the character recognition part 105 is output as "2331" and "495C" of JIS code, respectively. As the character recognition technology in this case, for example, the application of the technique shown in Japanese Patent Laid-Open No. 1-94490 or Japanese Patent Laid-Open No. 2-137080 by the present applicant can be considered.

In Japanese Patent Laid-Open No. 1-94990, the standard dictionary of the same category is blocked in the recognition dictionary, and the set value for stopping the inquiry in the middle of the category is sequentially updated to smaller ones, and the inconsistency is 0 days. At that time, the inquiry of the category is stopped, and one minimum non-uniformity per category is output as a result.

In Japanese Patent Laid-Open No. 2-137080, a feature pattern signal is read from a feature pattern storage unit, a reference pattern for each character to be read is read from the reference pattern storage unit, and a corresponding pattern on each partial region of the feature pattern and the reference pattern When calculating the degree of difference between the subregions, a sequence obtained by adding a feature sequence in each direction describing the feature of each subregion of the feature pattern and a portion of the adjacent feature sequence on the adjacent subregion with respect to each feature sequence A plurality of types of feature sequences are created by a plurality of methods, nonlinear matching is performed between each of the original feature sequences and the created plurality of feature sequences and the feature sequences of reference patterns corresponding to the original feature sequences in each direction. The character pattern and the true value of each character type are determined by determining the sum of the minimum of the nonlinear matching in each direction as the difference between the subregions. Determining a top view of a pattern, and subjected to classification according to the high-road, and outputs the separated results.

The character type determination unit 106 determines the character type (a Chinese character, a numeric symbol, etc.) on the basis of the recognition result of the character used for the address output from the character recognition unit 104. That is, when the character recognition result in the character recognition part 104 is output as a JIS code, if the JIS code is 2300th (three digits of the code 233), the character is used for the calculation of "0"-"9". If the JIS code is 3000 or more (the first digit of the code is 3 or more), the character is determined to be a Chinese character. The same determination is made for symbols such as hyphens. 3 shows an example of the result of the character type determination.

First, in the character type determination unit 106, the address recognition unit 107 checks the character type sequentially from the head of the input string corresponding to the address based on the determination result, and the number or symbol is continuous. Look at the neighborhood name code, and remember it temporarily. In addition, the point where the character type changes from Chinese characters to numbers in the input character string, that is, the start position of the residential display number is detected, and the numbers and symbols after the start position are regarded as the residential display number. Then, the temporarily stored neighborhood name and residential display number are recognized as addresses.

Next, the address recognition device will be described in more detail.

On the paper sheet, for example, the character string "183-123 Nippon-shi Shinjuku 1-10" corresponding to the address shown in FIG. 2 is described. The character string is composed of the neighborhood name code 1 A, the prefecture prefectural city neighborhood name 1 B, and the residential display number 1C described at the beginning of the address.

First, the image extracting unit 101 converts this paper sheet character, code, etc. into an electric image signal (step 401 shown in FIG. 4). The extracted image signal is sent to the address position detection unit 102. The address position detecting unit 102 detects the address position (step 402), and inputs an input string corresponding to the address, that is, an image signal of 1183-123, Nishi-jin, Japan. Extract.

The input string extracted by the address position detection unit 102 is sliced by one character in the character slice unit 103 (403). The sliced characters are then inquired by the character recognition unit 104 with the contents of the character dictionary 105. That is, the character recognition unit 104 initializes the character position and simultaneously (i = 1: step 404) and the dictionary NO. Is initialized (j = 1: step 405), and the character at the character position P (i) and the dictionary NO. The characters of D (j) are compared and queried (step 406). Fig. 5A shows the relationship between the input string and the character position P (i). In Fig. 5 (b), the dictionary NO. Shows the relationship between D (j), dictionary, category (JIS code), and character type.

The character recognition unit 104 includes a character at the character position P (i) and a dictionary NO. If the characters of D (j) do not match, after setting j = j + 1 (step 408), the process returns to step 406. Character at character position P (i) and dictionary NO. If the characters of D (j) match, the process proceeds to step 409. In step 409, the category (JIS code) of the search result is stored in r (i) (see FIG. 5 (c)), and it is checked whether i = n (step 410), and i = n Otherwise, after setting i = i + 1 (step 411), the flow returns to step 405. If i = n in step 410, the inquiry results r (1) to r (n) are output from the character recognition unit 104 to the character type determination unit 106. That is, the inquiry results r (1) to r (n) are output as the character recognition results in the JIS code as shown in FIG. 5 (d).

The character type determination unit 106 obtains the inquiry results r (1) to r (n) from the character recognition unit 104 and initializes the storing position of the inquiry result (i = 1: step 412). The character type is determined based on the category of the inquiry result r (i) (step 413), and the determined character type is stored in k (i) (step 414). Fig. 6A shows the relationship between the character type and category (JIS code). That is, if "2330" ≤ r (i) ≤ "2339", it is determined as a number, if r (i) = "213D", it is determined as a symbol, and if r (i) ≥ "3000", it is determined as a Chinese character, and the judgment is made. The result is stored in k (i). Then, it is checked whether i = n (step 415), and if i = n, i = i + 1 (step 416), and the flow returns to step 413. When i = n in step 415, the determination results k (1) to k (n) are output from the character type determination unit 106 to the address recognition unit 107. That is, the determination results k (1) to k (n) are output as the character determination results as shown in Fig. 6C.

The address recognition unit 107 obtains the determination results k (1) to k (n) from the character type determination unit 106, initializes the storing position of the inquiry result and the determination result (i = 1: step). (417), the storage position of the final result is initialized (j = 1: step 418). Then, the result of the determination k (i) is checked (step 419), and if it is a number or a symbol, the category of the inquiry result r (i) is stored in the final result f (i) (step 420), and j = It becomes j + 1 (step 421), and it progresses to step 422. In step 419, if the determination result k (i) is a Chinese character, the flow proceeds directly to step 422. In step (422), it is checked whether i = n, and if i = n, i = i + 1 (step 423), and the process returns to step 419. When i = n in step 422, the final results f (1) to f (m) obtained in step 420 are recognized as addresses (see FIG. 6 (b)).

That is, the address recognizing unit 107 checks the character type in sequence from the head of the input string corresponding to the address, firstly sees the area &quot; 183-123 &quot; where the numbers or symbols are consecutive as the neighborhood name code, Remember temporarily. Further, in the input string, the point at which the character type changes from Chinese characters to a number, that is, the starting position of the housing display number is detected, and the numbers and symbols "1-10" after the starting position are regarded as the housing display number. Then, the temporarily stored neighborhood name code "183-123" and the residential display number "1-10" are recognized together as an address.

Embodiment 2

Next, a second embodiment (second embodiment) of the present invention will be described with reference to the drawings.

In this embodiment, by character recognition of an input string corresponding to an address for each character and first recognizing the city name in the input string, the boundary position between the city name and the neighborhood name is detected, and the neighborhood name code is detected. Detects the number of characters in the neighborhood name corresponding to the neighborhood name code. Then, the starting position of description of the housing display number is detected based on the boundary of the city name, the neighborhood name, and the number of characters of the neighborhood name, and the input string after the start position is detected as the housing display number. The residential identification number and the neighborhood name code are the address recognition results.

7 is a block diagram showing the configuration of Embodiment 2, wherein the image extraction unit 101, the address position detection unit 102, the character slice unit 103, the character recognition unit 104, and the character dictionary 105 are Since it is the same as the structure of Embodiment 1 mentioned above, description of a duplicate part is abbreviate | omitted. In this embodiment, the character recognition results in the character recognition unit 104 are sent to the neighborhood name character number retrieval unit 201, the municipality name retrieval unit 203, and the address recognition unit 205.

The neighborhood name character number search unit 201 sequentially checks the input string corresponding to the address which is the result of the character recognition in the character recognition unit 104 from the beginning, detects and temporarily stores the neighborhood name code. Then, the number of characters of the neighborhood name corresponding to the neighborhood name code, the neighborhood name code as shown in FIG. 8 and the number of characters of the neighborhood name corresponding to the neighborhood name code are registered in advance. Search from.

The city name search unit 203 searches the city name by comparing the text recognition result in the character recognition unit 104 with the contents of the city name table 204 in which the city name is registered in advance. Then, by searching for the name of the city, the boundary position of the name of the city in the input string and the neighborhood name described subsequently is detected. Here, in the city name table 204, as shown in FIG. 9, the JIS code of the city name of the whole country and the corresponding character is registered. And the capacity of this municipality table 204 is very small compared with the capacity of the table in which the neighborhood name is registered, since it is limited to about 3000 municipal names as a phenomenon.

The address recognizing unit 205 is based on the number of characters of the neighborhood name corresponding to the neighborhood name code retrieved by the neighborhood name character number retrieval unit 201 and the boundary position of the city name and the neighborhood name detected by the municipality name retrieval unit 203. To detect the starting position of the residential display number. Then, the residential display number and the neighborhood name code temporarily stored in the neighborhood name character number search unit 201 are used as the recognition result of the address.

Next, the address recognition device will be described in more detail.

Also in this embodiment, it is assumed that on the paper sheet, the character string "183-123 Nippon City Shinjuku 1-10" corresponding to the address shown in FIG. 2 is described.

First, the image extraction unit 101 converts the characters, symbols, and the like on this paper sheet into electrical image signals (step 401 shown in FIG. 4), and the extracted image signals are sent to the address position detection unit 102. FIG. Lose. The address position detecting unit 102 detects the address position (step 402), and inputs an input string corresponding to the address, that is, an image signal of 1183-123, Nishi-jin, Japan. Extract.

The input string extracted by the address position detection unit 102 is sliced by one character in the character slice unit 103 (step 403). The sliced characters are then inquired by the character recognition unit 104 with the contents of the character dictionary 105. Hereinafter, the same processing as described in Embodiment 1 is performed (steps 404 to 411), and the character recognition unit 104 uses the inquiry results r (1) to r (n) as character recognition results. It outputs with JIS code as shown to (d).

The neighborhood name character number search unit 201 obtains the inquiry results r (1) to r (n) from the character recognition unit 104, and sequentially orders the inquiry results r (1) to r (n) from the head part. It checks and extracts the category corresponding to a neighborhood name code (step 501 shown in FIG. 10). That is, the character string corresponding to the address which is the character recognition result in the character recognition unit 104 is checked from the beginning, and the neighborhood name code "183-123" is detected and stored temporarily. The character number "3" of the neighborhood name "日新 町" corresponding to this neighborhood name code "183-123" is searched from the neighborhood name number table 202 shown in FIG. (CNO = "3": step 502).

Simultaneously with the processing in the neighborhood name character number search unit 201, the city name search unit 203 performs the character recognition result and the name of the city name table in the character recognition unit 104. We compare contents and search for city name "府中 市". Then, by searching for the name of the city, (中 市, the name of the city in the input string `` 中 中 日 日 '' was written. ) And the neighborhood name "日新 町".

That is, the city name search unit 203 sets the head position of the city name to i = 8 (step 503). Then, the table position of the city name is initialized (j = 1: step 504), and the result after the character code r (8) of C (j) of the city name table 204 is inquired. (Step 505). If the character code of C (j) of the city name table 204 and the result after r (8) coincide, the flow proceeds to step 507 in response to NO in step 506, and j = j + 1. The flow returns to step 505. If the result after the character code r (8) of C (j) of the city name table 204 matches, it progresses to step 508 according to the example in step 506. FIG. In step 508, the boundary between the city name and the neighborhood name, that is, the head position of the neighborhood name, is stored in the CST. In this case, since the city name is obtained as "middle city", CST = "11".

The address recognition unit 205 is detected by the neighborhood name character number "3" (CNO = "3") and the city name search unit 203 corresponding to the neighborhood name code retrieved by the neighborhood name character search unit 201. Based on the boundary position (CST = “11”) of the city name 府中 市 and the neighborhood name 日新 町, the starting position JST of the residential display number is calculated (JST = CNO + CST). In step 509), the character &quot; 1-10 &quot; described after the start position JST of this residential display number is regarded as the residential display number. The residential display number &quot; 1-10 &quot; and the neighborhood name code &quot; 183-123 &quot; temporarily stored in the neighborhood name character number search unit 201 are set together as the recognition result &quot; 183-123 1-10 &quot;.

That is, the address recognition unit 205 stores the search results r (1) to r (7) corresponding to the neighborhood name codes in the final results f (1) to f (7) (step 510), After the inquiry result r (JST) corresponding to the display number is stored after the final result f (8) (step 511). As a result, as shown in FIG. 6 (b), final results f (1) to f (m) are obtained.

Embodiment 3

Next, Embodiment (Embodiment 3) of this invention is described with reference to drawings.

In this embodiment, for example, a specific area having a large amount of paper sheets to be handled is designated in advance, and in this designated specific area, even if the neighborhood name code is not described in the input string corresponding to the address, Is to be able to recognize. That is, in this specific area, a dictionary in which a neighborhood name and a neighborhood name code corresponding to the neighborhood name is registered is prepared, and a neighborhood name code is found from the neighborhood name described in the input string.

11 is a block diagram showing the configuration of Embodiment 3, wherein the image extraction unit 101, the address position detection unit 102, the character slice unit 103, the character recognition unit 104, and the character dictionary 105 are Since it is the same as the structure of embodiment (1) mentioned above, description of a duplicate part is abbreviate | omitted. In this embodiment, the character recognition results in the character recognition unit 104 include the city name search unit 301, the neighborhood name character number search unit 303, the neighborhood name character number search unit 305, and the address recognition unit. 306 is sent to.

The city name search unit 301 searches the city name by comparing the character recognition result in the character recognition unit 104 with the contents of the city name table 302 in which the character name is registered in advance. Then, by searching for the name of the municipality, the boundary position with the name of the neighborhood of the name of the municipality is detected in the character string of the part where the prefectures of the prefecture municipality is written, and the detected result is the neighborhood name searching unit 303 and the address recognition unit 306. )

In the city name table 302, as shown in FIG. 12, the neighborhood name flag "Yes" is included in the city name corresponding to the specific area designated in advance among the registered city names, and the town name is in the city name not applicable. It is registered with the flag "none". And if the neighborhood name flag "Yes" corresponding to the municipal name searched by the city name search unit 301, the neighborhood name flag "Yes" is output to the neighborhood name search unit 303. If the searched neighborhood name flag is "none", the neighborhood name flag "none" is outputted to the neighborhood name character number search unit 305. Here, as for the city name where the neighborhood name flag is "is", a very large area is designated, for example compared with the area | region where papers are handled.

When the neighborhood name search unit 303 receives the output of the neighborhood name flag "Yes" from the street name search unit 301, the boundary position and character recognition of the street name and neighborhood name output from the street name search unit 301 Based on the character recognition result outputted from the unit 104, the character recognition result of the character string after the name of the municipality is compared with the neighborhood name table in which the neighborhood name of the specific region is registered in advance, and the neighborhood name is searched. As shown in FIG. 13, the neighborhood name and the neighborhood name code corresponding to the neighborhood name are registered in this neighborhood name table. The neighborhood name search unit 303 outputs a neighborhood name code corresponding to the searched neighborhood name and neighborhood name. In addition, the neighborhood names registered in the neighborhood name table are only neighborhood names in a specific area designated in advance, and their use capacity is very small as compared with the case of registering neighborhood names throughout Japan.

When the neighborhood name character number search unit 305 receives the output of the neighborhood name flag "none" from the city name search unit 301, the corresponding input string which is an address which is the character recognition result in the character recognition unit 104 is received. Are checked sequentially from the beginning, and the neighborhood name code is detected and stored temporarily. Then, the number of characters of the neighborhood name corresponding to this code is retrieved from the neighborhood name code table 202 in which the neighborhood name code as shown in FIG. 8 and the number of characters of the neighborhood name corresponding to the neighborhood name code are registered in advance. do.

The address recognition unit 306, when the neighborhood name flag is searched for in the city name search unit 301 with "Yes," the character recognition result of the input string output from the character recognition unit 104, the city name search unit ( The housing display number is detected based on the boundary name of the city name and the neighborhood name output from 301 and the neighborhood name output from the neighborhood name search unit 303, and the neighborhood from the neighborhood display number and the neighborhood name search unit 303. Recognize an address with a name code.

In addition, when the neighborhood name flag is searched for in the neighborhood name search section 301 as "none", the character section of the input character string output from the character recognition section 104 results in the city section output from the street name search section 301. Based on the boundary position between the name and the neighborhood name and the number of characters of the neighborhood name output from the neighborhood name character number search unit 305, the starting position of the housing display number is detected, and the characters after the starting position are detected as the housing display number. The residential display number and the neighborhood name character number search unit 305 together recognize the temporarily stored neighborhood name code as an address.

Next, the address recognition apparatus will be described in more detail.

Also in this embodiment, it is assumed that on the paper sheets, the character string "183-123 Nippon-shi Shinjuku 1-10" corresponding to the address shown in FIG. 2 is described.

First, the image extraction unit 101 converts the letters, symbols, etc. on the paper sheets into electrical image signals (step 401 shown in FIG. 4). The extracted image signal is sent to the address position detection unit 102. The address position detecting unit 102 detects the address position (step 402), and inputs an input string corresponding to the address, that is, an image signal of "183-123 Nippon City Shinjuku 1-10". Extract.

The input string extracted by the address position detection unit 102 is sliced by one character in the character slice unit 103 (step 403). The sliced characters are then inquired by the character recognition unit 104 with the contents of the character dictionary 105. Hereinafter, the same processing as described in Embodiment 1 is performed (steps 404 to 411), and the character recognition unit 104 uses the inquiry results r (1) to (n) as the character recognition results. It outputs with JIS code as shown to d).

The city name search unit 301 compares the character recognition result in the character recognition unit 104 with the contents of the city name table 302 in which the city name is registered in advance, and selects the city name "市 中 市". Search. Then, by searching for the name of the city, the boundary position of the city name "府中 市" and the neighborhood name "日新 町" in the input string of the part where the prefecture prefecture city name is written is detected. The detection result is output to the neighborhood name search unit 303 and the address recognition unit 306. In addition, as shown in FIG. 12, the searched "city of city" corresponds to a specific area designated in advance, so that the neighborhood name flag "is present" is detected. In this way, the neighborhood name flag "Yes" is output to the neighborhood name search unit 303. In this case, the neighborhood name character count search unit 305 does not operate because the neighborhood name flag "none" is not sent from the city name search unit 301.

The neighborhood name search unit 303 receives the output of the neighborhood name flag "Yes" from the municipal name search unit 301, and outputs the neighborhood name "府中 市 (Bujung City)" and the neighborhood output from the municipal name search unit 301. Based on the boundary position of the name "日新 町" and the character recognition result output from the character recognition unit 104, a neighborhood name "日新 일" is searched from the neighborhood name table 304, The neighborhood name code corresponding to the searched neighborhood name and neighborhood name is output to the address recognition unit 306.

The address recognition unit 306 is a character recognition result of the input string output from the character recognition unit 104, the city name "府中 市 (Bujung City)" output from the city name search unit 301 and the neighborhood name "日新 町Based on the boundary position information of &quot; (New Decision) &quot; and the neighborhood name output from the neighborhood name search unit 303, a housing display number &quot; 1-10 &quot; is detected, and this residential display number &quot; 1-10 &quot; The neighborhood name code "183-123" from the search unit 303 is also recognized as the addresses "183-123, 1-10".

That is, the city name search unit 301 receives the inquiry results r (1) to (n) from the character recognition unit 104 and initializes the table position of the city name (j = 1: step shown in FIG. 10). (504)). Then, the result after the character code r (8) of the city name table C (j) is inquired (step 505), and if there is a match, the corresponding j is stored in the CFL (CFL = j: shown in FIG. 14). Step 601). Then, the presence or absence of the neighborhood name flag g (CFL) is checked (step 602). If the neighborhood name flag g (CFL) is not present, the neighborhood name flag "none" is returned in response to a no at step 603. Send to the neighborhood name character search unit (305). In this way, the process after step 508 shown in FIG. 10 is executed by the neighborhood name search unit 305 and the address recognition unit 306. If there is a neighborhood name flag g (CFL), in response to the example in step 603, after storing the leading location of the neighborhood name in the CST (step 604), the leading location (CST) and neighborhood name of the neighborhood name The flag "Yes" is sent to the neighborhood name search unit 303.

The neighborhood name search unit 303 receives the neighborhood name flag "Yes" from the city name search unit 301, initializes the table position of the neighborhood name (j = 1: step 605), and the neighborhood name table t. The character code of (j) and the result after r (CST) are inquired (step 606). If the character code of the neighborhood name table t (j) and the result after r (CST) do not match, the process proceeds to step 608 in response to NO in step 607, where j = j + 1, and step ( Return to 606). If the character code of the neighborhood name table t (j) and the result after r (CST) coincide, the process proceeds to step 609 in accordance with the example in step 607. In step 609, the neighborhood name and the neighborhood name code indicated by the neighborhood name table t (j) corresponding to the result after r (CST) are output to the address recognition unit 306. At this time, for the neighborhood name code, the category of each character is stored in r '(1) to r' (7) from the beginning, and this is referred to as the address recognition unit (r '(1) to r' (7). 306).

The address recognition unit 306 is a boundary between the city name and the neighborhood name from the city name search unit 301, namely, the head position of the neighborhood name (CST) and the first position of the residential display number from the neighborhood name search unit 303. The JST is found (step 610). The search results r '(1) to r' (7) from the neighborhood name search unit 303 are stored in the final results f (1) to f (7) (step 611). In addition, the inquiry result (JST) after the first position JST of the residence indication number is regarded as the residence indication number, and after inquiry result r (JST) corresponding to this residence indication number is stored after the final result f (8). Thereby, final results f (1) to f (m) similar to those shown in FIG. 6 (b) can be obtained.

[Mail Automatic Processing System]

Then, by converting "183-123 1-10" recognized as an address in each of the above embodiments into a barcode code, for example, and printing the barcode code on paper sheets, the processing of the automatic processing system becomes very efficient. can do. In this case, the processing of the automatic mail processing system can be made very efficient. An outline of the automatic mail processing system in this case is illustrated in FIG. In the figure, (2-1) is a paper sheet supply unit, (2-2) is an optical character recognition unit, (2-3) is a barcode factor unit, (2-4) (2-4l to 2-4n) The silver integrated part, (2-5) is a gate controller, (2-6) is a conveyance path, and (2-7) (2-71-2-7n) is a gate. (3-1) is a paper sheet supply part, 3-2 is a barcode lead part, (3-3) (3-31 to 3-3n) is an integrated part, (3-4) is a gate controller, and (3- 5) is a conveyance path, and (3-6) (3-61 to 3-6n) is a gate.

In this postal matter processing system, the paper sheets 4 from the paper sheet supply unit 2-1 are supplied to the conveying path 2-6. When the paper sheet 4 reaches the optical character recognition unit 2-2, the optical character recognition unit 2-2 detects a character string corresponding to the address described on the paper sheet 4 and according to the present invention. It is recognized as an address. In other words, if the paper sheet 4 is described as "183-123, Nippon-Dongjin 1-1", "183-123 1-10" is recognized as an address. And the optical character recognition part 2-2 sends this recognized address "183-123 1-10" to the barcode printing part 2-3. The barcode printing unit 2-3 converts the address "183-123 1-10" from the optical character recognition unit 2-2 into a barcode, which is sent through the optical character recognition unit 2-2. It prints on the predetermined part of the paper sheet 4.

The paper sheet 4 on which this barcode has been printed is sent to the accumulation section 2-4. Here, the address "183-123 1-10" of the paper sheet 4 recognized by the optical character recognition unit 2-2 is also sent to the gate controller 2-5, and the gate controller 2-5 is The opening / closing of the gates 2-71 to 2-7n is controlled based on the address “183-123 1-10”. Thereby, the paper sheet 4 is integrated in the accumulation part 2-4 corresponding to the address "183-123 1-10" recognized by the optical character recognition part 2-2.

The paper sheets 4 from the paper sheet supply unit 3-1, that is, the paper sheets 4 with the barcode printed by the above-described method, are supplied to the conveying path 3-5. When the paper sheet 4 reaches the barcode reading section 3-2, the barcode reading section 3-2 reads the barcode printed on the paper sheet 4. As shown in FIG. Then, the contents of the read barcode are sent to the gate controller 3-4. The gate controller 3-4 controls the opening and closing of the gates 3-61 to 3-6n based on the contents of the bar code sent. In this way, the paper sheets 4 are accumulated in the accumulation section 3-3 corresponding to the address indicated by the barcode read by the barcode reading section 3-2.

In each of the above-described embodiments, the address form is described by a combination of Chinese characters, symbols, and numbers for the domestic market. However, the same method can be applied to paper sheets such as China and Korea. An example of a postal matter in China is shown in FIG. 16 (a), and an example of a Korean postal matter is shown in FIG. 16 (b). Even in the case of China's mail, the portion corresponding to the prefecture's prefecture's name is written in Chinese characters, and the housing number is marked with numbers. In the Korean postal matters, portions corresponding to the prefectures of the prefectures of the prefecture are written in Korean, and the residential display numbers are indicated by symbols such as numbers and hyphens.

As will be apparent from the above description, according to the present invention, since the neighborhood name code and the residential display number are recognized together as addresses in the first and sixth inventions, a large number of neighborhood name codes consisting of numbers and the like are matched in advance. Without registering the name of the neighborhood, the address can be read at high speed and with high accuracy.

In the second invention and the seventh invention, in the first invention and the sixth invention, the neighborhood name code and the residential display number are detected by checking the character type sequentially from the head of the input string. In addition to the effects of the invention, it is possible to ensure the detection of the neighborhood name code and the residential display number.

In the third invention and the eighth invention, in the first invention and the sixth invention, the number of characters of the neighborhood name corresponding to the neighborhood name code is retrieved from the first table, and the city name constituting the input string is determined. 2 The table was searched, and the residential display number was detected based on the number of characters and the city name of the searched neighborhood name. Therefore, in addition to the effects of the first and sixth inventions, The effect of ensuring the detection is obtained.

In the fourth and ninth inventions, in the third and eighth inventions, the neighborhood name flag is added to the municipal name of a specific region previously designated among the municipal names registered in the second table, and the neighborhood is searched for. When the name flag is attached, the neighborhood name described in the input string and the neighborhood name code corresponding to the neighborhood name are searched from the third table based on the character recognition result and the searched city name. Since the residential display number is detected based on the name and the neighborhood name, in addition to the effects of the third and eighth inventions, the address can be reliably recognized even if the neighborhood name code is not described.

In the fifth invention and the tenth invention, in the fourth invention and the ninth invention, the specific area to which the neighborhood name flag is added is specified in advance based on the number of paper sheets handled for each area. In addition to the effects of the invention, at least registration of the neighborhood name can achieve a very large effect.

In the eleventh invention, an address is recognized by the address recognition method of the first invention rather than an input string indicating an address written on a paper sheet to be returned, and the recognized address is converted into a barcode to be printed on the paper sheet. Since the printed paper sheets are processed according to the printed bar codes, in addition to the effect of the first invention, the paper sheets can be automatically sorted and accumulated according to the barcodes or recognized addresses. have.

Claims (7)

A neighborhood name code corresponding to a series of numbers or numbers and symbols corresponding to each neighborhood name listed on a paper sheet, a region name consisting of a series of Korean, Chinese, or English characters, and a residential designation number consisting of a series of numbers or both numbers and symbols. Image extracting an input string indicating an address to be included, and extracting the image extracted input string by one character, and each character is previously registered with character information limited to the character type of the neighborhood name code and the residential display number. Compare and search the contents of, but only character recognition of the neighborhood name code and the residential display number consisting of a series of numbers or both numbers and symbols, and detects only the neighborhood name code and the residential display number based on the character recognition result. To recognize the detected neighborhood name code and residential identification number as addresses. The character type of each character of the input string is determined based on the character recognition result, and the neighborhood name code and the residential display number are detected by checking the character type sequentially from the head of the input string. Install a first table that pre-registers the number of characters of the neighborhood name corresponding to the neighborhood name code and a second table that pre-registers the city name, and detect the neighborhood name code based on the character recognition result; The number of characters of the neighborhood name corresponding to the neighborhood name code is searched from the first table, and the city name constituting the input string is searched from the second table based on the character recognition result. Thus, the address characterized in that to detect the residential display number based on the number of characters and the city name of the searched neighborhood name Recognition method. The character recognition result according to claim 1, wherein the town name flag is added to a preliminary name of a specific area among the precinct names registered in the second table, and the town name flag is added to the searched prefecture name. And the neighborhood name and its neighborhood name described in the input string from the third table in which the neighborhood name corresponding to the neighborhood name to which the neighborhood name flag has been added and the neighborhood name code corresponding to the neighborhood name are pre-registered based on the retrieved city name. And searching for a neighborhood name code corresponding to the neighborhood name code, and detecting a residential display number based on the searched district name and neighborhood name. The address recognition method according to claim 2, wherein the specific area to which the neighborhood name flag is added is specified in advance based on the number of paper sheets handled for each area. A neighborhood name code corresponding to a series of numbers or numbers and symbols corresponding to each neighborhood name listed on a paper sheet, a region name consisting of a series of Korean, Chinese, or English characters, and a residential designation number consisting of a series of numbers or both numbers and symbols. The image extracting means and the input string extracted by the image extracting means are sliced by one character, and each character is previously limited to the character type of the neighborhood name code and the residential display number. Character information means for searching and comparing the text information with the contents of the registered text dictionary, a character recognition means for character recognition only the neighborhood name code and the residential display number consisting of a series of numbers or both numbers and symbols, and the character recognition in the character recognition means Detecting means for detecting only the neighborhood name code and the residential display number based on a result; Address recognition means for recognizing the neighborhood name code and the residential display number detected by the detection means together as an address, wherein the detection means is based on the character recognition result of the character recognition means for each character of the input string. A character type determination means for determining a character type, and checking the character type sequentially from the head of the input string to detect a neighborhood name code and a residential display number, and the detection means is a neighborhood corresponding to the neighborhood name code The neighborhood name code is detected based on the first table in which the number of characters of the name is registered in advance, the second table in which the city name is registered in advance, and the text recognition means in the text recognition means, and the name of the neighborhood corresponding to the neighborhood name code is detected. Neighborhood name character number search means for searching the number of characters from the first table, and character recognition result in the character recognition means; And a county name search means for searching for a county name constituting the input string from the second table, wherein the number of characters of the neighborhood name searched by the neighborhood name search means and the county searched by the county name search means are provided. Address recognition device characterized in that for detecting the residential display number based on the name. 5. The method of claim 4, wherein a neighborhood name flag is added to a preliminary name of a specific region among the precinct names registered in the second table, and a neighborhood name flag is added to the precinct name retrieved by the prefecture name searching means. Based on the character recognition result of the character recognition means and the city name searched by the city name search means, the neighborhood name corresponding to the neighborhood name to which the neighborhood name flag is added and the neighborhood name code corresponding to the neighborhood name are previously displayed. Neighborhood name search means for searching for a neighborhood name described in the input string and a neighborhood name code corresponding to the neighborhood name from a registered third table, a neighborhood name searched by the city name search means and the neighborhood name search means Address recognition device characterized in that for detecting the residential display number based on the name. The address recognition device according to claim 5, wherein the specific area to which the neighborhood name flag is added is specified in advance based on the number of paper sheets handled for each area. An address recognition means for recognizing an address by the address recognition method of claim 1 from an input string representing an address written on the returned paper sheet, and converting the address recognized by the address recognition means into a barcode to print on the paper sheet. And a bar code printing means, and processing means for processing paper sheets on which the bar code has been printed by the bar code printing means in accordance with the bar code.