JP2001014311A - Place name representing method, and method and device for place name character string recognition - Google Patents

Abstract

PROBLEM TO BE SOLVED: To permit a high-speed collating process by generating a place name dictionary covering all place name character strings even when place names has many different notations, with less man-power. SOLUTION: Different notations of a place name are represented by extending a BNF notation method known as one representing method for the generation rule of context free grammar to match to the notation of a place name. According to the generation rule, the pattern of a typical different notation, e.g. small 'ke' and large 'ke' in the square form of Japanese syllabary and 'ga' in the cursive form of Japanese syllabary are defined as one syntax category, so that a set of the different notations of the place name can be simply represented. Further, the different notations of the place name can be simply represented by using select symbols adopted by the BNF notation method. Namely, the different notations of the place name are represented according to the generation rule (102) and recognition (104) is performed by using a network obtained from the generation rule. Consequently, the dictionary into which the set of various different notations is entered can be easily generated without any omission.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

TECHNICAL FIELD The present invention relates to a method for expressing a place name group,
BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a place name character string recognition method and apparatus, and in particular, a place name group expression method and a place name character string recognition method suitable for a place name character string storage means and a collation means used in an apparatus for reading a place name written on a document. And an apparatus.

[0002]

2. Description of the Related Art In general, a character recognition device that reads a character string composed of a sequence of place name words such as a prefecture name, a municipal name, a character name, and the like (hereinafter, referred to as a place name character string) from an image has the following characteristics. (Character extraction), (2) identifying the character type (character code) of each character pattern (character identification), (3) collating the character identification result with a prestored string of place name words (character string collation) ) And three functions.

As a conventional technique relating to a character string collation method, for example, a method by Marukawa et al. (Information Processing Society of Japan, Vol. 35, No. 6, "Error Correction Algorithm for Handwritten Kanji Address Recognition") is known. . In addition, as a conventional technology related to a method that integrates character extraction, recognition, and collation, a method based on a hidden Markov model (OE Agazzi,
et al., "Connected And Degraded Text Recognition
using Planar Hidden Markov Models, "Proceedings o
f International Conference on Acoustics, Speech,
and Signal Processing), a method for exploratory recognition of character strings (Koga, et al., “Lexical Search Approach fo
r Character-String Recognition ”Third Internation
al Association for Pattern Recognition Workshop on
Document Analysis Systems 1998) is known.

The above-mentioned prior art requires a means for storing a place name character string that can appear in advance and a place name character string dictionary for the character string collation processing. There are three types of place name character string dictionaries as shown below. (1) “Dictionary source file” stored in the file This is a “place name notation rule file” described later, and the like.
Editing must be possible for new creation or modification. (2) “Dictionary table” stored on the memory This is a “place name notation network” described later, and the like.
The contents of the dictionary file are expanded on the memory in a format suitable for the matching process. (3) "Dictionary binary file" at an intermediate stage between the above (1) and (2) This is to store in a file the result of performing a part of the expansion processing in advance in order to facilitate expansion on the memory. It was done.

[0005] In many cases, the format of a dictionary source file used in the prior art has not been clarified. However, all of the conventional techniques are based on the premise that all possible place name character strings are stored in a dictionary table in advance, so that a text file listing all possible place name character strings in advance is stored in a dictionary. It is considered to be used as a source file.

[0006]

The prior art described above is
It is necessary to prepare a dictionary for the string matching process,
In Japanese, there are many different notations that represent the same area with different character strings, and it is difficult to register possible place name character strings in the dictionary, so it is necessary to manually create a complete dictionary for this Has the problem that it is virtually impossible.

[0007] The different notations of Japanese place names include different notations due to differences in characters used, different notations due to omission of words, different notations due to additional character strings, different notations due to notation of street names, and the like. Hereinafter, examples of these different notations will be described.

(1) Different notations due to differences in characters used There are “Ozawa” and “Ozawa”, “Ichigaya”, “Ichigaya”, “Ichigaya”, and the like.

(2) Different notation by omitting words There are different notations by omitting the names of prefectures, and different notations by omitting “large characters” and “characters”. Omitting the abbreviation of a prefecture name is often seen in the case of a mailing address of a mail, for example, "Ogaya, Kawagoe-shi, Saitama Prefecture" and "Ogaya, Kawagoe-shi, Ogaya". In addition, as an example of omitting “large characters” and “characters”, for example,
"Ogaya, Kawagoe City, Saitama Prefecture" and "Ogaya, Kawagoe City, Saitama Prefecture"
Etc.

(3) Different notation by additional character strings Character strings that are not necessary for identification of an address such as a small letter name are added and are different notations. For example, "Ogaya Ogaya, Kawagoe-shi, Saitama""Saitama prefecture Kawagoe city Ogaya character Higashiseki" and so on.

[0011] (4) Different notation of street name Notation that is often found in Kyoto and the like, for example, "Omasamachi-cho, Shimogyo-ku, Kyoto" and "Karasuma-dori Bukkoji-jiru".

As described above, there are various types of different names of place names. For example, when a place name of "Ogaya, Kawagoe-shi, Saitama" is taken as an example, the corresponding different notation is examined. It can be seen that there are a large number of different notations, as listed below. `` Ogaya, Kawagoe City, Saitama Prefecture '' `` Ogaya, Kawagoe City, Saitama Prefecture '' `` Ogaya, Kawagoe City, Saitama Prefecture '' `` Ogaya, Kawagoe City, Saitama Prefecture '' `` Ogaya, Kawagoe City, Saitama Prefecture '' `` Saitama Prefecture Ogaya, Kawagoe-shi Ogaya, Kawagoe-shi Ogaya, Kawagoe-shi Ogaya, Kawagoe-shi Ogaya, Kawagoe-shi Ogaya, Kawagoe-shi Ogaya, Kawagoe-shi Ogaya, Kawagoe-shi valley"

[0013] In the above example, small names such as "Ogaya Higashida, Kawagoe City, Saitama Prefecture", "Ogaya Higashiseki, Kawagoe City, Saitama Prefecture", and "Ogaya Nishiseki, Kawagoe City, Saitama Prefecture" are also used. In consideration of the combination with the twelve different notations listed above, there are a total of 84 different notations.

In the case of the prior art, it is necessary to comprehensively register all combinations of the various different notations described above in the dictionary file manually, and therefore, it takes a lot of labor to create the dictionary file. There was a problem. Moreover,
In the case of Kyoto City, where there are many different notations, the total name of street names and street names in the city is hundreds of thousands, and it was virtually impossible to create a complete dictionary by hand .

A first object of the present invention is to solve the above-mentioned problems and to provide a place name notation method capable of easily registering various different notations without exception in a character string collation dictionary. is there.

As described above, in the case where there are many different notations in place names, even if it is possible to write all the different notations in the dictionary, the conventional technology increases the storage capacity of the dictionary and increases the processing time. Also increases according to the number of different notations.

As a technique capable of solving the above-mentioned problem, a data format called Trie is used.
A technology that reduced the storage capacity of dictionaries and made the matching process faster was developed (Koga, et al.,
“Lexical Search Approach for Character-String Rec
ognition ”Third International Association for Pat
tern Recognition Workshop on Document Analysis Sys
tems 1998). This technology describes place names as tree-like data, such as branching only in parts with various notations, and facilitates automatic generation of Tries from a set of character strings. .

[0018] The above-mentioned technology is based on, for example, three notations of "Ogaya Higashida, Kawagoe-shi, Saitama", "Higashiseki, Ogaya, Kawagoe-shi, Saitama", and "Ogaya-Nishiseki, Kawagoe-shi, Saitama". Trie like
Can be easily generated. Hereinafter, a network that represents the concatenation relationship of the characters in the place name character string will be referred to as a place name notation network.

However, in the Trie type place name notation network, when there is a difference in a part of a character string, it is necessary to treat the character string as a completely different character string and generate another branch. Therefore, for example, Trie corresponding to the different notation group of “Ogaya, Kawagoe-shi, Saitama” becomes a large one as shown below.

[0020] (Abbreviated below)

As described above, even if an attempt is made to express a different notation by a method using a Trie type place name notation network, there arises a problem that both the dictionary capacity and the processing time are greatly increased.

Accordingly, an object of the present invention is to provide a place name expression method having a storage format with a small storage capacity and capable of high-speed collation processing for use in a place name dictionary for recognizing various different notations. And a place name character string recognition method and apparatus.

[0023]

According to the present invention, it is an object of the present invention to provide a method in which a place name representing an area has a plurality of different notations represented by an array of words having different character strings but meaning the same area. In the place name expression method of representing a set of character strings, an array of characters or syntax categories is defined for each substring constituting a part or all of the place name character string, and the array of characters or defined syntax categories is formed. Achieved by representing place name strings by syntax category.

In addition, the object is that the place name character string is replaced with a substitution symbol indicating what kind of character or syntax category string the syntax category is replaced with, and that a certain syntax category represents a specific region. This is achieved by expressing using the place name symbols shown.

[0025] The object is to define an array of character or syntax categories for each partial character string that constitutes part or all of a place name character string, wherein the partial character string of the input character string is provided. This is achieved by matching a place name from the input string by determining whether it matches one of the place name strings represented by a syntax category consisting of a character or an array of defined syntax categories.

The object is to define an array of characters or syntax categories for each partial character string that constitutes a part or all of a place name character string, and to define an array of characters or defined syntax categories. Storage means for storing the represented place name character string; input means for inputting the character string; means for comparing whether the input character string is the place name character string stored in the storage means; This is achieved by providing means for outputting a result.

[0027] Further, the object is to provide a character reading means for reading a character written on a document by using an image obtained by converting the density of the surface of the document into an electric signal as an input,
This is achieved by the input means inputting a character string from the character reading means.

Specifically, in order to achieve the above object, the present invention expresses a different notation of a place name using a generation rule of a context-free grammar. In the context-free grammar, generation rules indicate which elements of a certain sentence (syntax category) are replaced by what other syntactic categories ("Introduction to Natural Language Processing", Modern Science, ISBN4-7649-0143 -9). The present invention uses the BNF notation (Back
us-Naur-Form) (Nakada “Compiler” ISBN4-7828-5057-
Use extended BNF notation that extends 3) to be suitable for the representation of place names.

According to the above-described generation rules, typical patterns of different notations, for example, “ga”, “ke”, and “ga” can be defined as one syntax category, and a set of different notations of place names can be simplified. Can be expressed as Further, by using the selection symbols adopted in the BNF notation, it is possible to express the notation of the place name more simply. Therefore, according to the present invention, it is possible to easily create a dictionary in which a set of various different notations is completely described.

The BNF notation is a notation for expressing the generation rule of the context-free grammar by using symbols such as replacement, option, and selection. The following symbols are used. :: = substitution. This means that the left-hand syntax category can be replaced with the right-hand syntax category or character array. [] Optional. It means that the description in [] is optional. | Select. It means either the right side or the left side.

As an example, an example in which the above-described generation rule of the different notation of “Ogaya, Kawagoe-shi, Saitama” is expressed by BNF notation is shown below. <W ka> :: = ka | ke | is <Ogaya, Kawagoe-shi, Saitama> ::: = [Saitama Prefecture] Kawagoe-shi [larger characters]
Small <w> valley [[character] Higashida |

Further, by using the notation format as described above, it is possible to reduce the size of the place name notation network. In the above-mentioned notation format, the difference between the partial character strings is explicitly expressed using the symbols “[]” and “|”. For this reason, in the case where the difference between the partial character strings may be different, it is possible to easily set a path on the network that bypasses the partial character string. For example, the BNF notation described above can be replaced with a compact network as shown below. It has been difficult to generate such a compact network from a sequence of character strings as in the related art.

[0033]

[0034]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of a place name notation method and a place name character string recognition method according to the present invention will be described in detail with reference to the drawings.

FIG. 1 is a flow chart for explaining a processing example of place name character string recognition according to the embodiment of the present invention.
This flow will be described. In addition, the flowchart used in the following description was expressed in accordance with the Gain-Thurson notation. For this notation, see "J. Martin" Software Structuring Technique "Modern Science Company", "ISBN4-7
649-0124-2 C3050 P5562E ".

(1) First, prior to the recognition of a place name, a place name character string generation rule editing process (step 101) creates a place name character string generation rule based on a case of a different notation of a place name, It is stored in the character string generation rule file 102. The place name character string generation rule editing process in step 101 can be realized by a human editing operation via a computer.

(2) Next, the place name notation network generation processing (step 103) reads the place name character string generation rule file 102 and generates a place name notation network which is a dictionary for the place name recognition 104. The place name notation network generation processing in step 103 can be realized as a program on a computer.

(3) Next, place name recognition processing (step 10)
4) reads the place name character string from the input image with reference to the place name notation network. The place name recognition processing 104 in step 104 can be realized as a program on a computer.

The place name character string generation rule file 102 uses the "extended BNF notation" according to the present invention to represent a group of different names of place names according to a generation rule of a context-free grammar. Extended BNF
The notation is an extension of a symbol such as a combination to the BNF notation, and the notation described below is used. :: = substitution. This means that the left-hand syntax category can be replaced with the right-hand syntax category or an array of characters. [] Optional. This means that the description in [] may be omitted. | Select. Means either right or left of this symbol. () Join. Evaluate the parentheses before the surrounding variables. <W character string> Syntax category. <N number> Syntax category representing a group of different notations of a place name character string indicating a specific area. The number is the identifier of the place name. Use an integer greater than 0.

The above-mentioned symbols are evaluated according to the following priorities. (1) Definition of variable names of <W character string> and <N number> (2) Parentheses of [] and (). When parentheses are used in double or more nests, evaluation takes precedence over inner parentheses. (3) | (4) ::: =

FIG. 2 is a diagram showing an example of notation of place names expressed by the place name character string generation rules edited by the editing processing in step 101 and an example of listing different notations without using the generation rules.

The notation example of the place name represented by the place name character string generation rule shown in FIG. 2A is, as an example, "Ogaya Ogaya, Kawagoe-shi, Saitama"("Higashida""Higashiseki""Nishiseki") This is an example of using the extended BNF notation according to the present invention to express the different notations of "Kazahata, Kawagoe City, Saitama Prefecture"("Kubo" and "Kanan" are small letters) and "Shimohiroya, Kawagoe City, Saitama Prefecture". As described above, a place name including many different notations can be expressed very easily by using the symbol introduced according to the present invention. On the other hand, in the example in which different notations are listed without using the generation rules shown in FIG. 2B, only a large number of different notations are listed, so the four lines shown in FIG. As many as 106 different notations are generated. FIG. 2B shows a part thereof.

The place name character string generation rule file 102 is a normal text file, and a general text editor can be applied as a means for implementing step 101 of the place name character string generation rule editing process.

FIG. 3 is a diagram schematically showing a place name notation network created from the example of the generation rule of FIG. 2A, which will be described below.

The place name notation network is a directed graph in which each side corresponds to a partial character string and each vertex corresponds to a boundary of the partial character string. The direction of each side matches the order of the characters in the character string. In FIG. 3, the side marked NULL indicates a null transition, that is, that there is no need for any character string at that location. A circle 301 with a line at the lower right in FIG. 3 indicates the start position of the place name character string. Circles 302 to 304, which are hatched in the center, indicate the end positions of the place name character strings.

FIG. 4 is a diagram for explaining a data format when the place name notation network is mounted on a computer. This will be described below. When a place name notation network is implemented on a computer, the place name notation network is configured as shown in FIG.
Data format as shown in (left-child.right-sibling
representation, T. Colmen et al. "Algorithm Introduction" Modern Science, pp. 201-202). In this data format, the concatenation of characters is represented by a child pointer, and the branch of the place name notation network is represented by a sibling pointer.

FIG. 4A shows the components of each data record. Each data record has a data item c4.
01, b402, and d403. Data item c is a character code, and data item b is a sibling pointer. The data item d is a child pointer. The branch from a certain data record is represented by a sibling pointer, and the character string is represented by a list connected by a child pointer. For example, if the place name notation network shown in FIG. 3 is represented in a list format by the above-described data record, it becomes as shown in FIG. 4B.

In the place name notation network expressed in the list format shown in FIG.
(Corresponding to the character code "small"), data record 4
Although the process branches from 04 to 406, data records 404 'to 404 are linked by a child pointer, and data records 404, 405, and 406 are linked by sibling pointers. The character string "Saitama" is represented by data records 407, 408, and 409 connected by a child pointer. When the data record corresponds to the NULL transition, the character code c401 of the data record contains NU.
This means that a data record storing an LL character and branching from the data code storing the NULL character may be omitted. Further, after the data record corresponding to the last character of the place name character string, the data record 410
, One extra data record is provided, and the child pointer d of this data record 410 has NU
The LL pointer is stored to indicate that the network is at the end, and the identifier of the place name is stored in the sibling pointer b.

The place name notation network in the form of a list shown in FIG. 4B, which is expressed as described above, can be regarded as a graph in which each data record corresponds to a node.
Each side in the place name notation network schematically shown in FIG. 3 is represented here by nodes for the number of characters.

FIG. 5 is a flowchart for explaining the process of generating a place name notation network from the place name character string generation rule in step 103 of FIG. 1, and FIG. 6 is a view for explaining an example of a generated syntax tree. Will be described.

First, the generation rule of each place name character string in the place name notation generation rule file 102, for example, for each line starting with <N number> from the second line from the top in FIG. Process one by one. For each row,
First, in step 502, a syntax analysis of a character string in a line is performed to create a syntax tree as shown in FIG. Next, at step 503, to produce a terminal node t _i place name notation network corresponding to the place name different notation group. In the following, unless otherwise specified, "nodes on the place name notation network"
Indicates a data record in the format of FIG. NULL is set for the character code c in t _i and the child pointer d
Is stored as NULL, and the sibling pointer b is stored as the place name identifier of the place name variant notation group. Next, in step 504, a place name notation network corresponding to the place name variant notation group is generated using a function proc described later. After all the place name character string generation rules have been processed, in step 505, the redundant parts of the generated place name notation network are integrated.

The process of generating a syntax tree from the place name character string generation rules is described in, for example, “Introduction to Natural Language Processing” (Kindai Kagakusha,
ISBN4-7649-0143-9, pp. A method of generating a transition network according to a generation rule as described in 19-30) can be used. The example of the syntax tree generated in the process of step 502 illustrated in FIG. 6 is an example of the syntax tree generated from the second line in FIG. In FIG.
Circles marked with “+” indicate concatenation of character strings, circles marked with “[]” indicate options, circles marked with “|” indicate selection, and squares indicate character strings. In the extended BNF notation, parentheses “(”, “)” are also used. However, the syntax tree used in the embodiment of the present invention does not include a node corresponding to the parenthesis, and determines the order of operations determined by the parentheses. Is reflected in the structure itself.

The function proc is a function used to generate a place name notation network from a syntax tree, and takes two arguments, p and a. The argument p designates a value taken by the child pointer d of the node at the end of the generated place name notation network. The argument a indicates the top node of the syntax tree to be processed. When an argument a is specified for a certain node,
All nodes below the argument a are processed recursively.

FIG. 7 is a flowchart for explaining the processing operation by the function proc. FIGS. 8 and 9 are diagrams for explaining the process of generating a place name notation network by the function proc.
Is a diagram showing a group of place name notation networks generated from the place name notation generation rules, which will be described below.
In FIG. 7, p, q, and r shown in FIG.
This is a variable representing the address of the data record in the format shown in FIG. 4, and the symbol "->" represents a data item in the data record. The processing of the flow shown in FIG. 7 is executed in four cases according to the type of the node a of the syntax tree.

(1) Determine the type of node a in the syntax tree,
It is determined whether the type is “+”, “|”, “[]”, or “character string” (step 701).

(2) If the type of the node a of the syntax tree is “+”, that is, a join in the judgment in step 701,
First, the argument p is copied to the variable q. That is, the address of the terminal node of the partial network generated in this process is copied (step 702).

(3) Next, the child nodes n _i (1 ≦ 1) of the syntax tree
(i ≦ the number of child nodes) are sequentially processed from the right by the function proc () to generate a partial network of the place name notation network. At this time, an argument is passed so that the end point of the partial network generated by the function proc () becomes q. The pointer of the start point of the partial network generated as a result is re-substituted into q, which is set as the end point of the partial network to be generated next.
By repeatedly calling the function proc () in this way, the partial networks of the place name notation networks generated from the child nodes of the “+” of the syntax tree are successively connected (steps 703 and 704).

(4) When all child nodes have been processed, q at that time, that is, the head of the partial network is returned as a return value (step 705).

(5) If the type of the node a in the syntax tree is "|"
First, a partial network is generated from one of the child nodes n _1, and the head address of the obtained partial network is set as a variable q
(Step 706).

(6) Next, the value of q is substituted for the variable r, and partial networks are generated in order from other child nodes n _i (2 ≦ i ≦ number of child nodes). The head address of the generated partial network is stored in the sibling pointer b of r. Further, the head address of the generated partial network is substituted for r, and the same processing is repeated thereafter (step 7).
07-710).

(7) When all child nodes have been processed, q, that is, the top address q of the partial network generated first is returned as a return value (step 71).
1).

(8) If the type of the node a in the syntax tree is “[]”, that is, the option in the determination in step 701, first, a partial network corresponding to the child node of the syntax tree is generated. Store the address in variable q. At this time, a parameter is specified so that the end of the generated partial network becomes p (step 712).

(9) Next, a node corresponding to the NULL transition is generated using the function newNd (), and its address is stored in the sibling pointer b of q. Note that newNd () is a function for newly allocating one storage area for the data record in the format shown in FIG. 4, and a NULL pointer is set in the data item b of the allocated data record (step 71).
3).

(10) Next, NULL is assigned to the character code c of the node corresponding to the NULL transition, and p is set to the child node pointer d of the node corresponding to the NULL transition (steps 714 and 715).

(11) The first address q of the partial network generated last is returned as a return value (step 71).
6).

(12) If it is determined in step 701 that the type of the node a in the syntax tree is a character string, first, the variable q
Is substituted for the value of p (step 717).

(13) Next, the following processing is repeated for each character C _i (1 ≦ i ≦ character string length) in the character string in order from the end of the character string, and the node corresponding to each character is set to 1 Generate one after another. Here, first, one storage area of the node is secured by the function newNd (). Then, the character code c of the newly generated node substitutes C _i. Next, the value of q is assigned to the child node d of the newly generated node. Then, q
Is replaced with the address of the newly generated node (steps 718 to 722).

(14) After the above processing is performed for each character C _i , the address q of the newly generated partial network is returned as a return value (step 723).

8 and 9 showing a process of generating a place name notation network by the function proc, reference numeral 801 denotes
The terminal node is generated in the process of step 503 of the flow shown in FIG. 5 and the identifier “3501104” is stored. Thereafter, by the processing of each step of the flow shown in FIG. 7, a place name notation network is generated as shown in order from top to bottom in the diagrams shown in FIGS. Then, when the node 603 of the syntax tree shown in FIG. 6 is processed by the function proc,
First, the partial network corresponding to the node 602 is a function
A partial network generated by the proc and corresponding to the node 602 as shown by 802 is generated. Next, node 6
A partial network corresponding to 04 is generated by a function proc. In this case, p stores the address of the node 804, and the generated partial network is connected to the node 804 as indicated by 803.

The control loop 501 of the flow shown in FIG. 5 generates a separate place name notation network for each place name character string generation rule. As a result, the place name notation network group generated from the place name notation generation rule of FIG. 2 is generated as shown in FIG.
By the process of 5, the redundant part of these network groups, for example, the part of Kawagoe-shi, Saitama, is integrated, and the place name notation network described with reference to FIG. 3 is generated.

FIG. 11 is a flowchart for explaining a processing procedure for generating a place name notation network using a conventional technique.
FIG. 12 is a diagram for explaining an example of a generation process of a place name notation network generated according to the related art, and FIG. 13 is a diagram illustrating an example of a place name notation network generated according to the related art. Next, a method of generating a place name notation network when no generation rule is used will be described.

Here, the reason why the prior art is explained is that only the tree-structured place name notation network called Trie can be generated from the conventional place name character string notation method, and the place name generated from the notation method of the present invention. This is because the notation network is superior in both storage capacity and processing time required for collation. A method of expressing the notation of a place name according to the related art is a list of place name character strings as shown in FIG.
The flow described with reference to FIG. 11 is a procedure for generating a place name notation network from such a list of words. Here, the k-th character string in FIG. 2B is S _k , its length is L _k , and the i-th character of each character string is C _i . Also,
It is assumed that an identifier corresponding to each character string is separately stored. The generated place name notation network is shown in FIG.
This is realized in the data format shown in

(1) First, a node rr serving as a temporary root of the place name notation network is generated. NULL is set to the child node pointer d of this node (step 1101,
1102).

(2) All character strings S _k are processed one by one by a loop 1103.

(3) First, the root address is substituted for the variable p. Next, for each character in the character string, the subroutine Sr
Call chNxt. The subroutine SrchNxt is a processing procedure for determining whether or not a node corresponding to each character has already been generated, and if not, adding a new node. This procedure will be described later (steps 1104 to 1106). ).

(4) When the subroutine SrchNxt has finished processing the characters in the character string, a new child node is added to the function ne
wNd (), the identifier of the character string is stored in the area of the pointer b, and the address of this new child node is
Is assigned to the child node pointer d. The child node of rr at the end of the loop 1103 becomes the root of the place name notation network (steps 1107 to 1110).

Next, the processing of the subroutine SrchNxt will be described.

(1) First, the value of the child node d of p is substituted for the variable q, and then the loop processing is performed to scan all the child nodes of p with the variable q, and the corresponding character code, that is, the data item Check whether c is equal to C _i . If they are equal, it is considered that a node corresponding to C _i has already been generated, and the pointer p is advanced to that node q, and the processing ends (steps 1111, 1113 to 1115, loop 11).
12).

(2) In the check in step 1113,
If the data item c is not equal to C _i, and substitutes the value of the sibling pointer of q to q, and repeats a loop process until q is NULL. (Step 1116).

(3) If the node corresponding to C _i cannot be found even after the loop processing ends, a new node is added to the function ne
Generated by wNd (), and the character code c of the new node is C _i
, NULL for child node pointer d, p for sibling pointer b
Is assigned to the value of the child node pointer d, the address of the new node is assigned to the child node pointer d of p, the address of the new child node is assigned to the pointer p, and the processing of this subroutine is completed. (Step 11
17 to 1122).

FIG. 12 shows a process of generating a place name notation network according to the processing procedure of FIG. The example given here is a process of processing the top three rows in FIG. First, a place name notation network corresponding to "Ogaya, Kawagoe-shi" is generated (120).
1). Next, "Kawagoe City Kasabata" is processed.
Is already generated in 1201, no new node is generated. However, if the pointer p is 120
When the character reaches the position shown in FIG. 2 and the character "shade" is processed, the node corresponding to "shade" is not a child node of "city". Therefore, a node corresponding to “kasa” is newly generated as a sibling node of “small”. Hereinafter, the node corresponding to the remaining character “Hata” is connected as a child node of the newly generated node (1203). The same applies to the case of "Kawagoe Shimohiroya", where the nodes for "below" are "small"
A new node is generated as a sibling of "Kasa" (1204), and nodes corresponding to subsequent characters are linked (1205).

FIG. 13 schematically shows a part of a place name notation network generated from the different notation group shown in FIG. 2B. This example differs from the case shown in FIG. The place name notation network generated from the notation method is in the form of a tree, that is, a form that, once branched, does not merge again. This is a form of representation of data known as Trie. Compared to FIG. 3, it can be seen that there are many redundant parts. For example, "Higashida", "Higashiseki",
The partial network corresponding to "Nishiguan" is repeated six times. This means that the required storage capacity is increased. In the case of a computer having a hierarchical memory configuration, if the memory space to be accessed becomes large, the access becomes slow due to a cache mishit or the like, which will be described later. The string matching process itself will be slow.

The fact that the present invention can generate a place name notation network with little redundancy as shown in FIG. 3 is an essential advantage of the place name word notation according to the generation rules. By using this generation rule, a redundant portion can be clearly expressed. For example, in the case of the example shown in FIG. 2 (A), there are three different notations for “ga” of “Ogaya”,
It is shown by the extended BNF notation that the character strings after “ka” are the same. For this reason, as shown in FIG.
Only between "small" and "valley", a network having three paths is generated.

On the other hand, the conventional notation method of the place name character string as shown in FIG. 2 (B) cannot detect whether or not the different notation groups after “「 ”are equivalent. Only networks as shown can be created.

FIG. 14 is a flowchart for explaining the processing operation in the place name recognition processing 104 shown in FIG. 1. This will be described below.

(1) First, an image of a character line portion is cut out from the input image by a character line cutout process (step 14).
01).

(2) Next, a pattern which seems to be a character, that is, a candidate pattern, is extracted from the character line image by a character extracting process. If character boundaries cannot be uniquely determined at this stage, character pattern extraction is attempted based on a plurality of boundary hypotheses, and candidate patterns corresponding to the respective hypotheses are output (step 1402).

(3) Next, the character recognition process recognizes what character each of the extracted candidate patterns is, and outputs it as a candidate character string. When the method of extracting characters is based on a plurality of hypotheses,
When a plurality of candidate characters are output for one pattern, the character recognition process outputs a plurality of candidate character strings corresponding to the respective extraction methods and combinations of the candidate characters (step 1403).

(4) Finally, in the character string collation processing, whether or not each candidate character string is a correct place name character string is collated with reference to the place name notation network. The candidate character string received by the collation is set as a place name recognition result (step 1404).

FIG. 15 is a flowchart for explaining the processing operation in the above-described character string collation processing 1404, which will be described below. In this process, one character string is input, and it is determined whether at least a part of the input character string can be accepted as a place name character string, and if acceptable, a corresponding identifier of the place name notation is obtained. Here, the length of the input string L, and the i-th character in the string and C _i.

(1) First, by changing the starting point s of the collation from 1 to L by the loop 1501,
2. Repeat 1503.

(2) The root address of the place name notation network is set in the variable p indicating the node. Next, the function srch is called with the arguments p and s. Function srch
Is a function that finds a route that matches the input character string from the place name notation network and returns the address of the terminal node at the end. If the return value of srch is not a NULL pointer, it is assumed that the matching is successful, and the identifier stored in the node indicated by the return value of the function srch is output (step 150).
2 to 1504).

(3) If collation does not succeed even if s reaches L, it is determined that the character string collation processing has failed and the processing is terminated (step 1505).

In the above-described processing, the function srch is recursively called by itself, and searches the place name notation network for a path that matches the input character string in a depth-first manner.
The function srch takes two arguments, p and i. The argument p points to the node where the search starts. The argument i is an integer, and indicates the number of the character in the input character string that is focused on in the current processing. The function srch returns the address of the node at the end of the string if an acceptable string is found, and returns a NULL pointer if no acceptable string is found.

FIG. 16 is a flowchart for explaining the processing operation of the function srch in the above-described processing, which will be described below.

(1) First, it is checked whether or not the argument p points to the character string end node. If it points to the character string end node, it is considered that the input character string has been accepted, and p is returned as a return value, and the processing is terminated (step 1601).

(2) Next, it is determined whether or not all characters have already been processed. If i is larger than L and all characters have been processed, but p has not reached the end of the place name notation network, NULL is returned (step 1602).

[0098] (3) Next, determine whether the data item c of p matches the i-th character C _i of the string. If they match, the child node p-> d of p is set as the starting point of the search, and i +
Call the function srch recursively to process the string from the first. If the return value r is not NULL, it is determined that the character string has been accepted, and the process ends with r as the return value (step 1603).

(4) Next, it is checked whether p is a node corresponding to a NULL transition. If so, the child nodes p-> d of p
Is used as the starting point of the search, and the function srch is recursively called so as to process the character string from the ith position. This return value r is NULL
If not, it is determined that the character string has been accepted, and the process ends with r as a return value (step 1604).

(5) Next, it is checked whether or not a sibling node p-> b is connected to p. If connected
The function srch is recursively called so that the character string is processed from the i-th node, with the sibling node p-> b of p as the starting point of the search, and this return value is returned to the higher order (step 1605).

(6) If the input character string is not accepted in any of the above-described processes, further search cannot be performed, and the process is terminated with NULL as a return value (step 1).
606).

The embodiment of the present invention described above is
Although it has been described that character extraction, character recognition, and character string collation are sequentially performed, the present invention relates to Koga et al., "Address reading device, mail sorting machine, and character string recognition method" (Japanese Patent Application No. 10-2).
No. 8077), it is possible to easily extend the method to feed back the character string collation result to character extraction.

FIG. 17 is a block diagram showing a configuration example of a system to which the place name character string recognition processing according to the embodiment of the present invention is applied, and FIG. 18 is a block diagram showing a configuration of a place name character string generation rule editing device. This system example is an example in which the present invention is applied to a postal sorting system. 17 and 18, 1701 is a mail sorting machine, 1702 is a scanner, 1
703 is a delay line, 1704 is a sorter, 170
5 is a place name recognition device, 1706 is an input interface,
1707 is an arithmetic processing unit, 1708 is an output processing unit,
1710 is a memory, 1711 is a network interface, 1712 is a hard disk, 1713 is a removable medium storage device, 1714 is a place name character string generation rule editing device, 1718 is a network, 1801 is a mouse,
802 is a keyboard, 1803 is a display, 180
4 is a place name character string generation rule editing program, 1805 is a character string collation program, 1806 is a place name notation network display program, 1807 is a place name character string generation rule file, 1808 is a place name notation network generation program,
Reference numeral 1809 denotes a place name notation network data, reference numeral 1810 denotes a communication device, reference numeral 1811 denotes a removable medium storage device, reference numeral 181.
2 is a computer.

In the system shown in FIG. 17, one or a plurality of mail sorting machines 1701 and one or a plurality of place name character string generation rule editing devices 1714 are connected to a network 171.
8 to be connected. The mail sorting machine 1701 includes a scanner 1702, a delay line 1703, a sorter 17
04, a place name recognition device 1705. In addition, the place name recognition device 1705 includes an input interface 170.
6, arithmetic processing unit 1707, output processing unit 1708,
Memory 1710, network interface 171
1, a hard disk 1712, and a medium removable storage device 1713. The bold line in the figure is
Shows the flow of mail.

In the system shown in FIG. 17, the image information of the place name described in the mail sent from the scanner 1702 is transferred to the place name recognizing device 1705. Then, while the mail is conveyed on the delay line 1703, the place name recognition device 1705 recognizes the place name described in the mail and transfers the recognition result to the sorter 1704.
Sorter 1704 sorts the mail according to the recognition result.

As a preparation stage for sorting mail, the place name recognizing device 1705 stores the place name notation network generation program file from the hard disk 1712 into the memory 1710.
And is started by the arithmetic unit 1707. Under the control of the place name notation network generation program, the place name recognition device 1705 transmits the place name character string generation rule from the place name character string generation rule editing device 1714 to the network interface 17.
11, a place name notation network file is created and stored in the hard disk 1712.

Note that the place name character string generation rule may be read from a removable medium storage device 1713 such as a floppy disk drive instead of being input from the place name character string generation rule editing device 1714 via a network.

The place name recognizing device 1705 stores the recognition program file and the place name notation network file from the hard disk 1712 in the memory 1710 when sorting mail.
And executed by the arithmetic unit 1807. Then, the place name recognition device 1705 inputs an image from the input interface 1706 under the control of the recognition program,
The place name described in the mail is recognized, and the recognition result is output via the output interface 1708.

The place name character string generation rule editing device 1714
As shown in FIG. 18, a computer 1802 has a mouse 1801, a keyboard 1802, a display 180
3. A disk device for storing the place name character string generation rule file 1807, a communication device 1810, and a removable media storage device 1811 are connected to each other. The editing work is executed by editing the place name character string generation rule file 1807 via the place name character string generation rule editing program 1804 operating on the computer 1812. The place name character string generation rule file 1807 is a text file, and an ordinary text editor can be used for editing. Also, the place name notation network generation program 1808 is executed on the computer 1812, and the place name notation network 18
09 can be generated.

The place name character string generation rule editing device 1714
With the above-described function, it is possible to confirm whether or not the place name word generation rule being edited is grammatically correct, and further, a program 1805 equivalent to the character string collation 1404 in the recognition processing.
Is executed, it is possible to confirm whether or not the test character string input from the keyboard 1803 is accepted.

Also, since the computer 1812 executes the place name notation network display program 1806 for displaying the place name notation network 1809 in, for example, the format shown in FIG. 3, the operator visually recognizes the editing result. You can check. The edited place name character string generation rule file 1807 is transferred to the place name recognition device 1705 via the communication device 1810, or copied to a removable storage medium such as a floppy disk by the media removable storage device 1811 and stored. It may be transported to the mail sorter 1701 by media.

FIG. 19 is a block diagram showing the configuration of another embodiment of the present invention, and FIG. 20 is a diagram for explaining an example of a screen displayed on a display. This example is a gazetteer for retrieving information relating to a place name from a character string representing a place name, using a place name character string notation method and a place name collation method according to the present invention. In FIG. 19, reference numeral 1901 denotes a mouse;
2 is a keyboard, 1903 is a display, 1904 is a printer, 1905 is an input file, 1906 is an output file, 1907 is a gazetteer program, 1908 is a place name additional information file, 1909 is a place name character string generation rule file, 1910 is a communication module, 1911 Is an interface module, 1912 is place name list data, 19
13 is a place name list sort module, 1914 is a place name information search module, 1915 is a place name list generation module, 1916 is a character string collation module, 1917 is a place name notation expansion module, 1918 is a place name notation network generation program, and 1919 is a place name notation network data. is there.

The device shown in FIG. 19 provides the following services. (1) Display or print the standard form of the place name character string input from the keyboard. (2) Display or print the different notation of the place name character string input from the keyboard. (3) Display or print area information (postal code, etc.) corresponding to the place name character string input from the keyboard. (4) The place name character string input from the file is converted into information unique to the corresponding area, such as a standard form or a zip code, and output to a file. (5) The place name character string input from the network is converted into information specific to the corresponding area, such as a standard form or a zip code, and output to the network. In the above description, the standard form is, for example, a formal character string representing a certain area defined by an administrative division.

In the embodiment shown in FIG. 19, a place name book program 1907 executed on a computer has a mouse 1901,
Keyboard 1902, display 1903, printer 1904, input file 1905, output file 190
6, a place name additional information file 1908 and a place name character string generation rule file 1909 are connected and configured. Display and input / output are performed via the interface module 1911. When a character string to be searched is input, the place name information search module 1914 causes the character string collation module 191
Call 6. The character string collation module 1916 is configured as shown in FIG.
4 is a module that performs processing equivalent to the character string collation processing 1404 in FIG.
, Refer to the place name notation network data 1919 generated by the place name notation network generation program 1918 to check which identifier the input character string corresponds to.

The place name information search module 1914 uses the obtained identifier as a clue to add the place name additional information file 190.
From 8, a standard form and additional information such as a postal code are retrieved. The place name notation development module 1917 enumerates all possible notations from the place name notation network data 1919. The obtained different notation group is stored in the place name list data 1912, and output via the interface module 1911 as necessary. In addition, the place name list sort module 1913 rearranges and outputs the order of the different notation groups according to the instruction of the operator. Inputs for such processing include a keyboard 1901, an input file 1905,
This may be performed via any of the communication modules 1910. The output may be performed via any of the display 1904, the output file 1906, and the communication module 1910.

In the example of the screen displayed on the display 1903 of the embodiment of FIG. 19 shown in FIG.
In the example shown in (A), when the operator inputs a character string “Ogaya, Kawagoe-shi” and executes a search, the display 1 is displayed.
903 is a screen example displayed on the screen. The input character string is input to field 2005, and a search is executed by clicking button 2006 with a mouse. As a result of the search, the character string found to correspond to the input character string is displayed in window 2007. The "standard" item in each line indicates whether the character string is in a standard form. The item "place name" displays the character string. In the item “postal code”, a postal code corresponding to the character string is displayed, but other additional information of the area may be displayed.

The frames of “standard”, “place name”, and “zip code” arranged in the area 2004 are buttons, and when each button is clicked with a mouse, the rows are rearranged based on the respective items. To instruct. Window 2
008 is for designating a search option. Here, whether to display only the standard form, to display a group of different notations based on characters, capital letters, etc.
Specify whether to display a group of different notations based on. Button 20
Reference numeral 02 denotes a button for instructing printing of display contents, and a button 2001 is a button for switching to a file input / output mode instead of the keyboard and the display. A button 2003 is a button for instructing termination of the program.

The window 200 opened in FIG.
Reference numeral 9 denotes a window for displaying detailed information such as how to read place names, small letters, and postal codes obtained as a result of the comparison. This window 2009 is activated by clicking a search result displayed on the window 2007 with a mouse.

The place name character string described by the notation method according to the embodiment of the present invention can be provided as a place name dictionary stored in a storage medium such as FD, MO, and DVD.

[0120]

As described above, according to the present invention, a place name dictionary covering all possible place name character strings can be created with a small number of hands, even when there are many different notations in place name notation. . Further, it is possible to easily create a network-type place name dictionary that enables high-speed collation processing.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating a processing example of place name character string recognition according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating a description example of a place name represented by an edited place name character string generation rule and an example in which different notations are listed without using the generation rule.

FIG. 3 is a diagram schematically illustrating a place name notation network created from an example of a generation rule.

FIG. 4 is a diagram illustrating a data format when a place name notation network is implemented on a computer.

FIG. 5 is a flowchart illustrating a process of generating a place name notation network from a place name character string generation rule.

FIG. 6 is a diagram illustrating an example of a generated syntax tree.

FIG. 7 is a flowchart illustrating a processing operation by a function proc that generates a place name notation network from a place name notation generation rule.

FIG. 8 is a diagram (part 1) illustrating a process of generating a place name notation network by a function proc.

FIG. 9 is a diagram (part 2) illustrating a process of generating a place name notation network by a function proc.

FIG. 10 is a diagram showing a group of place name notation networks generated from a place name notation generation rule.

FIG. 11 is a flowchart illustrating a processing procedure for generating a place name notation network using a conventional technique.

FIG. 12 is a diagram illustrating an example of a generation process of a place name notation network generated according to the related art.

FIG. 13 is a diagram illustrating an example of a place name notation network generated according to the related art.

FIG. 14 is a flowchart illustrating a processing operation in the place name recognition processing illustrated in FIG. 1;

FIG. 15 is a flowchart illustrating a processing operation in the character string collation processing illustrated in FIG. 14;

FIG. 16 is a flowchart illustrating a processing operation of a function srch.

FIG. 17 is a block diagram illustrating a configuration example of a system to which a place name character string recognition process according to an embodiment of the present invention is applied.

FIG. 18 is a block diagram illustrating a configuration of a place name character string generation rule editing device.

FIG. 19 is a block diagram showing a configuration of another embodiment of the present invention.

FIG. 20 is a diagram illustrating an example of a screen displayed on a display.

[Explanation of symbols]

 101 place name character string generation rule editing processing 102 place name character string generation rule file 103 place name notation network generation processing 104 place name recognition processing 1404 character string collation processing 1701 postal sorter 1702 scanner 1703 delay line 1704 sorter 1705 place name recognition device 1706 input interface 1707 Arithmetic processing unit 1708 Output processing unit 1710 Memory 1711 Network interface 1712 Hard disk 1713 Removable media storage device 1714 Place name character string generation rule editing device 1718 Network 1801 Mouse 1802 Keyboard 1803 Display 1804 Place name character string generation rule editing program 1805 Character string collation program 1806 Place name notation network display program 1807 Place name sentence Column creation rule file 1808 place name notation network generation program 1809 place name notation network data 1810 communication device 1811 removable media storage device 1812 computer 1901 mouse 1902 keyboard 1903 display 1904 printer 1905 input file 1906 output file 1907 gazetteer program 1908 geolocation additional information file 1909 Place name character string generation rule file 1910 Communication module 1911 Interface module 1912 Place name list data 1913 Place name list sort module 1914 Place name information search module 1915 Place name list generation module 1916 Character string collation module 1917 Place name notation expansion module 1918 Place name notation network generation program 19 19 Place Name Notation Network Data

Continued on the front page (72) Inventor Shoji Ikeda 1-280 Higashi-Koikekubo, Kokubunji-shi, Tokyo Inside the Central Research Laboratory, Hitachi, Ltd. (72) Inventor Hisao Ogata 1-280 Higashi-Koikekubo, Kokubunji-shi, Tokyo Hitachi, Ltd. In the laboratory (72) Inventor Hiroshi Saka 1-280 Higashi Koikekubo, Kokubunji-shi, Tokyo Inside the Hitachi, Ltd. Central Research Laboratory (72) Inventor Hiromichi Fujisawa 1-280 Higashi Koikekubo, Kokubunji-shi, Tokyo F-term in the Hitachi, Ltd. (Reference) 5B009 LA03 ME02 ME17 ME23 QA16 VA02

Claims (9)

[Claims] 1. A place name expression method for representing a set of place name character strings having a plurality of different notations in which place names representing regions are different character strings but are represented by an array of words meaning the same region. An array of characters or syntax categories is defined for each substring constituting part or all of the character string, and the place name character string is represented by a syntax category consisting of an array of characters or defined syntax categories. Place name expression method. 2. The method according to claim 1, wherein the place name character string is replaced with a replacement symbol indicating what kind of character or syntax category string the syntax category is replaced with, and a place name symbol indicating that a certain syntax category represents a specific region. The place name expression method according to claim 1, wherein the place name is represented by using the following. 3. A partial character string of an input character string defines an array of characters or syntax categories for each partial character string that constitutes a part or all of a place name character string, and the character or defined character string is defined. A place name character string collating method, wherein a place name is collated from an input character string by determining whether or not it matches one of the place name character strings represented by a syntax category composed of an array of syntax categories. . 4. An array of characters or syntax categories is defined for each partial character string constituting a part or all of a place name character string, and a place name represented by a syntax category composed of an array of characters or defined syntax categories. Storage means for storing a character string, input means for inputting the character string, means for checking whether the input character string is the place name character string stored in the storage means, and outputting the result of the check And a place name character string collating device. 5. A character reading means for reading a character described in a document by inputting an image obtained by converting the density of a surface of the document into an electric signal, and a place name character string collating means according to claim 4. Wherein the input means in the place name character string collating means inputs a character string from the character reading means. 6. A method for recognizing a place name character string in an address of a postal matter by using the place name character string recognizing device according to claim 5, classifying the postal matter, or printing a recognition result on the postal matter. A mail sorting system characterized by the following. 7. A place name character string having a plurality of different notations in which a place name representing an area is a different character string but represented by an array of words meaning the same area, is converted into a part of the place name character string or A place name character string recording medium characterized in that an array of characters or syntax categories is defined for each of the partial character strings constituting the entirety, and is stored by being represented by a syntax category consisting of an array of characters or defined syntax categories. 8. A character reading device for inputting an image obtained by converting the density of a surface of a document into an electric signal and reading characters described on the document, using the place name expression method according to claim 1. Means for storing a place name character string, and an array of partial images in the input image, wherein each partial image is one of the place name character strings represented by the place name expression.
A place name character string recognition device comprising means for recognizing a place name by finding a character similar to each character included in the place name. 9. A place name expression method in which a place name representing an area is a different character string but expresses a set of place name character strings having a plurality of different notations represented by an array of words meaning the same area. Alternatively, a place name expression method characterized by expression according to a partial character string generation rule composed of word parts.