JP2671311B2 - Address reader - Google Patents

Description

The present invention relates to an address reading device for reading an address written as a set of consecutive words on a paper sheet such as a mail piece. [Prior Art] As a conventional address reading device, as shown in JP-A-57-146380, a set of consecutive words forming an address is sequentially collated with an address dictionary to extract the best matching address. Method is used. However, in the address notation,
A large number of destination information such as country name, prefecture name or state name, city name, town name, street name, street address or street number, company name, building name, etc. are generally included. In this address reading, a means for recognizing the final address while dividing and recognizing these pieces of information is taken. [Problems to be Solved by the Invention] As described above, in the conventional device, an address is constructed in order to detect destination components such as a country name, a prefecture name or a state name, a city name, etc. from a large number of these destination information. Since a method of extracting all combinations of one or more consecutive words from the consecutive consecutive words and collating them with the address dictionary is used, it takes a lot of time to read and the processing capability of the device. Was a factor that lowers. FIGS. 2A to 2C are address layout diagrams showing an example of the address notation. FIG. 2 (a) shows the case where the address notation spans three lines, the city name and zip code are on the first line from the bottom, the building name and destination floor are on the second line from the bottom, and the third line from the bottom. The street name and area number (lot number) are described in. Second
2B and 2C are diagrams in which the address notation of FIG. 2A is decomposed into word units, and in address reading, the address configuration is made from the word array of FIG. 2C. It detects elements such as city names, street names, and building names. As shown in the above-mentioned Japanese Patent Laid-Open No. 57-146380, this detection process detects and separates the address constituent elements from the bottom to the first line of the address notation shown in FIG. 2 (a). Also,
Detection is performed in the order of major to minor, that is, city name to street name, lot number, and building name. As a result, as shown in FIG. 5, in the matching with the address dictionary executed in each line, there are three combinations of words to be matched in the first line, 15 in the second line, and three in the third line. 28 with the eyes
It becomes a street. Here, the number of collations of the conventional device in the address dictionary with 500 registered country names and city names, 2000 registered street names, and 500 registered building names is roughly calculated as follows. The address dictionary of this conventional device uses the number of characters (word length) of the registered name.
The method of classifying address dictionaries is adopted. Country name, city name classification …… 3 word length to 20 word length 18 classification Street name classification …… 6 word length to 30 word length 25 classification Building name classification …… 6 word length to 30 word length 25 classification As a method, a dynamic programming (DP) matching method is adopted, and an input word having a word length n is collated with a dictionary of three classifications of n-1, n, n + 1. Therefore, the average number of collations is (1) 1st line… Country name / city name detection 3 ways × 500 cases ÷ 18 × 3 = 250 times (2) 2nd line… street name, building name detection 28 ways × (2000 (+500 cases) ÷ 25 × 3 = 8,400 times (3) 3rd line… Building name detection 15 streets × 500 cases ÷ 25 × 3 = 900 times In order to detect the address in Fig. 2 (a), Will be executed 9,550 times in total. Therefore,
If the conventional device requires 20 μs for one verification, it takes about 190 ms to execute the verification completely, which is 1 hour / hour.
This is equivalent to a processing speed of 9,000. On the other hand, in order to satisfy the device specifications of 30,000 / hour
This resulted in the need for a collation processing unit for the machine. An object of the present invention is to solve such a problem, detect a notational characteristic of an address written on a paper sheet such as a mail, that is, detect a keyword of an address notation, and detect a continuous word constituting an address. An object of the present invention is to provide an address reading device in which the number of collations with the address dictionary is reduced and the reading processing time is significantly reduced by limiting the words to be subjected to the address discrimination from the position of the keyword therein. [Means for Solving Problems] An address reading device of the present invention is a character reading unit that reads a character string relating to an address notation written on a paper sheet, and a character string output of the character reading unit is edited into a word. A word extraction means for
Address dictionary storage means for storing an address character string corresponding to the address notation, and address discrimination means for collating the word output from the word extraction means and the character string stored in the address dictionary storage means to discriminate the address. In an address reading device comprising: a keyword dictionary storage unit that stores a predetermined keyword corresponding to the address notation, a keyword detection unit that detects a keyword included in the address notation, based on the keyword detection result, And means for limiting combinations of the words to be collated with the character strings stored in the address dictionary storage means in the address determination means. [Operation] According to the configuration of the present invention, the keyword representing the feature of the word forming the address notation is registered in the keyword dictionary storage means, and the output character string from the character reading means is divided into words by the word extracting means. The keyword detection means first compares the registered keyword with the registered keyword to detect the keyword. The matching combination of the address constituent words to be detected next from this detected keyword is determined by the word limiting means, and only the limited word is sent to the address determining means, and the address corresponding to the read character string is determined. That is, since the matching range is limited by the keyword, the matching time can be shortened. EXAMPLES The present invention will be described in detail below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention. This embodiment is also configured to read the city name, street name, lot number, and building name in the address notation as shown in FIGS. 2 (a) to (c). This detection process is
The first from the bottom of the address notation shown in FIG.
The address components are detected and separated from the first line to the upper line, and detection is performed in the order of the large section to the small section, that is, the city name to the street name, the lot number, and the building name. In this embodiment, a character reading unit 1, a word extracting unit 2, a keyword detecting unit 3, a keyword dictionary storing unit 4,
It is composed of an address discrimination unit 5 and an address dictionary storage unit 6.
Further, the keyword detection unit 3 receives an output word from the word extraction unit 2, an input word selection unit 31, a maximum match detection unit 32 that compares the input word with a keyword dictionary, a keyword determination unit 33 that determines a keyword, and a determination. It comprises an input word limiting unit 34 for selecting a word to be used for address detection in the address discriminating unit 5 from the keywords, and a word buffer 35 for storing output word information. It is composed of a dictionary index unit 42 and a matching keyword buffer 43. The output character string 100 of the character reading unit 1 is connected to the input of the word extracting unit 2, and the output word string 101 of this word extracting unit 2 is input to the input word selecting unit 31. The word 102, which is the first output of the input word selection unit, is connected to the first input of the maximum match detection unit 32. The word length data 103 of the word which is the second output is input to the dictionary index unit 42. The dictionary index unit 42 receives corresponding word length registration data 104 from the keyword dictionary storage unit 41 based on the input word length data 103. Dictionary index section 42
The dictionary registration data 105 which is the output of is input to the matching keyword buffer 43, and the dictionary registration data 106 of the output of this matching keyword buffer 43 is connected to the second input of the maximum match detecting unit 32, The matching result 107 is input to the keyword determination unit 33. The keyword determination end signal 108 that is the first output of the keyword determination unit 33 is connected to the input word selection unit 31, and the keyword determination result 109 that is the second output is input to the input word limiting unit 34. The output 110 of the input word limiting unit 34 is input to the word buffer 35, the output 111 of the word buffer 35 is connected to the first input of the address discriminating unit 5, and the dictionary registration data 112 of the output of the address dictionary storage unit 6 is the address. It is connected to the second input of the discriminator 5. Next, the operation of this embodiment will be described with reference to FIGS. 1, 2, and 6 to 8. From the character reading unit 1, a character reading result 100 shown in FIG. 2A is sequentially output from the left end or the right end for each line. The word extraction unit 2 separates words (spaces, etc.)
Is detected and separated into words as shown in FIG. 2 (b), and the word numbers are (1), in order from the rightmost word on the bottom line to the left.
They are given as (2), (3) ... (14). These word strings
101 is sent to the input word selection unit 31. The data structure of the output word string 101 of the word extracting unit 2 is shown in FIG. It consists of the total number of detected words, the number of words in each line, the number of characters contained in each word, and a pointer indicating the memory address storing the read result of the constituent characters of each word. Next, the operation of the keyword detecting means will be described with reference to the flowchart of FIG. The input word selection unit 31 uses the input word string information 101 shown in FIG.
Then, the total word number data is input and set in the counter WC which controls the keyword detection processing loop 108. Next, the register WN that stores the number of the word for which the keyword is detected is initialized to 1, and the keyword detection is started. First, the word length data of the word No. set in the register WN, that is, the number of characters included in the word, is shown in FIG.
It is input from 101, transferred to the keyword dictionary indexing section 42, and the dictionary is read. Next, the character read output of the same word is transferred to the maximum match detection unit 32. Then, it waits until the reading from the keyword dictionary is completed. The end of the keyword reading is detected when the collation keyword buffer 43 becomes "Ready". On the other hand, the keyword dictionary indexing section 42 operates as shown in FIG. 6 (b). That is, when the word length data n of a word is input, the word length is checked and the range of the keyword dictionary group to be referred to is determined. In the embodiment, when the word length n is 5 characters or less, a keyword dictionary group having the same word length as the input word is read and set in the matching keyword buffer 43. When the word length n is 6 characters or more and 16 characters or less, three keyword dictionary groups having word lengths (n-1), n, (n + 1) are referred to. Word length n is 17
If there are more than characters, (n-2), (n-1), n, (n +
Reference is made to five keyword dictionary groups 1) and (n + 2). When all dictionary groups have been read and set in the collation keyword dictionary buffer 43, the buffer “Read
The flag "y" is set. Then, the maximum match detection unit 32 is activated. The maximum match detection unit 32 performs matching between the word 102 given from the input word selection unit 31 and the keyword 106 given from the matching keyword buffer 43. The input word selection unit 31 inputs the words of Fig. 2 (b) in order from the word (1) on the first line to the maximum match detection unit 32. The maximum match detection unit 32 uses the "DP matching method". Collate. The result is output to the keyword determination unit 33. For this keyword, for example, as shown in Table 1, the keyword and its attribute code are registered in the keyword dictionary storage unit 41. From the word (1) on the first line to the word (14) on the third line in FIG. 2B, the keyword determination unit 33 outputs a matching request 108 with the next word to the input word selection unit 31, and this matching is performed. repeat. The keyword dictionary in Table 1 is composed of word length data for indexing, keywords, and their attribute codes. The attribute code represents the property of the keyword. For example, as shown in the layout diagram of FIG.
Keyword flag used for street name in "0" bit,
The "1" bit is provided with a keyword flag used for a building name, and the "2" bit is provided with a keyword flag accompanied by a number for distinction. Also, information that is placed at the beginning of "14" bits (prefix), that that is placed at the end of "15" bits (LSB) (suffix), and information that indicates whether the position is undefined at "13" bits ( Flag) is included. For example, "AVEN
The attribute code of the keyword "UE" is used in the street name in (8001) ₁₆ and indicates that the keyword is placed at the end, and the attribute code of the keyword "BLK" or "BLOCK" is accompanied by a number in (2002) _16. It indicates that it is a keyword and has a number after the keyword.As for the keyword, most frequently appearing in address notation, for example, Singapore mail such as "SINGAPORE" Registering a keyword as described is effective for speeding up the address detection processing.The keyword determination unit 33, as shown in FIG. The high keyword candidates are stored in the temporary memory, and the same keyword detection as above is repeated for all input word strings 101. All input word strings
When the keyword detection ends in 101, the presence or absence of the keyword is checked for each line of the address. The result is set in the row keyword flag register as shown in FIG. In matching the word (2) on the first line of FIG. 2 (b) with the keyword, "SINGAPORE" in the keyword dictionary of Table 1 is used.
And the keyword determination result 109 matches the input word limiting unit 34
Sent to The first line keyword flag register contains (30
00) The code ₁₆ is set. In the second line, word (3) matches the keyword “AVENUE” word (9) matches the keyword “BLOCK”, and the determination result 109 is the input word limiting unit 3
Sent to 4. The second line keyword flag register contains (A
000) ₁₆ code is set. In the third line, the word (10) and the keyword “FLOOR” match, and the word (12) and the keyword “BLDG” match, and the determination result 109 is sent to the input word limiting unit 34. The code (6000) ₁₆ is set in the third line keyword flag register. The input word limiting unit 34 determines a word combination for address determination and sets it in the word buffer 35 by referring to the keyword flag register of each line and the keyword attribute code. FIGS. 3 (a) to 3 (c) are explanatory views showing the detection states of the keywords in Table 1. FIG. 3 (a) shows the keyword detection result on the first line. The input word limiting unit 34 learns from the attribute code (2002) ₁₆ of the keyword “SINGAPORE” that the first line word (2) is a keyword accompanied by a number after it. As a result, the city name has already been detected in the first line, and the subsequent address discrimination unit 5 does not need to handle the first line. In the second line, word (3) and word (9) are detected as keywords, as shown in FIG. 3 (b). The word (3) is the keyword “AVENUE”, and it can be understood from the attribute code (8001) ₁₆ that there is a keyword at the end of the street name. Further, it can be seen that the word (9) is the keyword "BLOCK", and is a keyword accompanied by a number after the attribute code (2002) ₁₆ . As a result,
In the second line, it is understood that the street name may be determined by a combination of a series of words including the word (3) at the end. Therefore, it suffices to match only the four combinations of words from the steps of FIG. 3 (b) with the street name dictionary in the address dictionary 6, and the combinations of words when the keyword is not detected (see FIG. Compared to 28 in b)), 1/7 combination is required. In the 3rd line, the word (1
0) and the word (12) are detected as keywords.
It can be seen from the attribute code (2004) ₁₆ that the word (10) is the keyword "FLOOR", and that a number is attached either before or after the keyword. The word (12) is the keyword "BLDG", and it can be seen from its attribute code (4001) ₁₆ that it is used as the building name and is the keyword placed at the end. As a result, in the third line, it is understood that the building name may be determined by a combination of a series of words including the word (12) at the end. Therefore, it suffices to check the building name dictionary in the address dictionary storage unit 6 only with the combination of two words in the step of FIG. 3 (c), and the combination of words when no keyword is detected (FIG. 5 (c)). ) Is 1 / 7.5 compared to 15 ways. The input word limiting unit 34 inputs the limited combination of input words to the word buffer 35. Then, the address discrimination unit 5 collates the input terminal 111 with the address dictionary stored in the address dictionary storage unit 6 as in the conventional case, and outputs the address discrimination result 113. The approximate number of collations with the address dictionary in this embodiment is as follows. (1) Keyword detection 14 ways × 15 (keywords) = 210 times (2) First line Country name City name detection 0 times (already detected city name by keyword detection) (3) Second line Street name detection 4 ways × 2000 cases ÷ 25 × 3 = 960 times (4) 3rd line Building name detection 2 ways × 500 cases ÷ 25 × 3 = 120 times Therefore, the matching statistics with the dictionary is 1290 times. Therefore,
This means that the number of collations was reduced to 1 / 7.5, which is 1 / 7.5 of the conventional collation count of 9550 without keyword detection. [Advantages of the Invention] As described above, the present invention detects a keyword in the process of performing address discrimination from a series of word strings written on a paper sheet, so that it is possible to limit the words to be subjected to address discrimination, This has the effect of significantly reducing the number of collations with the address dictionary and greatly reducing the address reading processing time.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention, FIGS. 2 (a) to (c) are address layout diagrams showing an example of address notation and word configuration, and FIG. FIGS. 5A to 5C are address arrangement diagrams for explaining the operation of this embodiment, FIG. 4 is a flag array diagram showing the configuration of the keyword dictionary of this embodiment, and FIGS. FIG. 6 is a diagram for explaining the operation of the keyword detecting means, FIG. 7 is a diagram for constructing word string information data, and FIG. 8 is a diagram for constructing a row keyword flag register. . 1 ... Character reading unit, 2 ... Word extraction unit, 3 ... Keyword detection unit, 4 ... Keyword dictionary storage unit, 5 ... Address discrimination unit, 6 ... Address dictionary storage unit, 31 ... Input word selection Part, 32 ... maximum match detection part, 33 ... keyword determination part,
34 …… Input word limiter, 35 …… Word buffer, 41 …… Keyword dictionary storage, 42 …… Keyword dictionary index, 43
…… Collation keyword buffer, 100 …… Output character string, 101
…… Output word string, 102 …… Word, 103 …… Word length data, 10
4 …… Word length registration data, 105,112 …… Dictionary registration data, 10
6 ... Output data, 107 ... Collation output, 108 ... Judgment end signal, 109, 113 ... Judgment output.

   ────────────────────────────────────────────────── ─── Continuation of front page    (56) References JP-A-57-137976 (JP, A)                 JP-A-60-233782 (JP, A)                 JP-A-57-146380 (JP, A)

Claims (1)

(57) [Claims] Character reading means for reading a character string relating to the address notation written on the paper sheet, word extracting means for editing the character string output of the character reading means into words, and storing an address character string corresponding to the address notation. An address reading device comprising: an address dictionary storage means; and an address discrimination means for discriminating an address by collating a word output from the word extraction means with a character string stored in the address dictionary storage means. A keyword dictionary storage means for storing a predetermined keyword corresponding to, a keyword detection means for detecting a keyword included in the address notation, and stored in the address dictionary storage means in the address determination means based on the keyword detection result. An address reading device, comprising means for limiting a combination of the words to be collated with the character string.