JPH06274701A - Word collating device - Google Patents

Abstract

PURPOSE:To provide a word collating device which can retrieve at a high speed such a word that includes the correct characters in a character string of combination of recognized candidate characters in number more than a fixed rate in regard of those recognized candidate characters supplied from a character recognizing device in particular. CONSTITUTION:In regard of each of recognized candidate characters which are supplied to a word retrieval processor 1, the word numbers of these candidate characters are read out the corresponding 1st-N-th character word tables and then registered in a word register table 7 under the control of a table registration processing part 2. Then the word numbers registered in the table 7 are sorted and the emerging frequency of each word is calculated. Then the umber of characters of a word having a large emerging frequency is read out of a character number table 9. This character number is compared with the word emerging frequency by an output word decision processing part 4. If the coincidence between these number and frequency exceeds a prescribed degree, the part 4 decides that those component characters are recognized.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a word collating device, and more particularly to a recognition candidate character output from a character recognizing device.
The present invention relates to a word matching device that matches each character of a recognition candidate character and searches for a correct word.

[0002]

2. Description of the Related Art When a character recognition device performs recognition processing on a character string read by a sensor such as a CCD, a plurality of recognition candidate characters similar to one character string are output. The character string is collated with the word dictionary to determine the correct character. For example, as a conventional word matching device, when a recognition candidate character is read,
A table in which a large number of words are registered is searched in advance and the matching word is read out. However, not all the characters forming the recognition candidate character match the word to be read. For this reason, for example, if there is a match with one less character, it is searched as the target word.

[0003]

In the above-mentioned conventional word collating device, the number of words to be searched from the table for one character string is huge, and it takes time to search.

The present invention has been made in view of the above conventional circumstances, and an object of the present invention is to provide a word collation device capable of performing word search at high speed.

[0005]

As one configuration example of the word collating apparatus of the present invention, a word number is set for a predetermined word, and the word number is registered for each character forming the word. In the case of a word having a four-character structure, there is a table for each character from the first character to the fourth character, and the word table means for each character position in which a word number including the character is registered in each table, For example, the input means for inputting a recognition candidate character string similar to the target word output from the character recognition device, and the recognition candidate character string input from the input means, search the character position by word table means,
The word number search means for reading out the word number of each character of the recognition candidate character string and the number of appearances of the word number searched by the word number search means are obtained, and the number of appearances is the total of the word numbers searched for each character. Is a number, for example 4
When it is a recognition candidate character having a character structure, the number of appearances is calculated on the basis of the number of appearances of word numbers for obtaining the total number of word numbers searched for four characters, and the number of appearances of word numbers calculated by the above-mentioned number of appearances calculation means. The number of constituent characters of a word having a large number of words and the number of appearances of the word are compared, and the output word determining means determines the word based on the degree of coincidence.

Another example of the structure of the word collating device of the present invention is to have a word position-based word table means similar to the above and an input means, and to recognize a recognition candidate character string input from the input means. , Searching the word table for each character position, determining the table address using a hash function, and comparing the word occurrence count and the number of word constituent characters with table registration processing means for registering the searched word and its appearance count in the table And an output word determining means for determining a word according to the degree of coincidence obtained.

[0007]

The word collating device of the present invention registers the word number of each character forming the recognition candidate character in the table each time a word including the character appears, and finally sorts the table by the word number. , A word collating device that obtains the number of word appearances, compares the number of constituent characters of a target word with the number of word appearances, and determines a word based on a certain degree of matching, which enables high-speed word search processing. it can.

Further, by performing the word appearance count using a hash function and writing it to the address designated by the table registration processing means, the word search processing can be performed at a higher speed.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a system configuration diagram illustrating a word matching device according to the present embodiment. In the figure, character data output from a character recognition device (not shown) is input to a word search processing unit 1 as an input unit and a word number search unit. The character recognition device reads, for example, handwritten characters by a sensor, analyzes the read characters, and outputs similar characters. The word search processing unit 1 cuts out a predetermined character string from such a character string output from the character recognition device, and sets this character string as a recognition candidate character string. The word search processing unit 1 is connected to the memory unit 5 and refers to a character position-based word table 6 in the memory unit 5 described later to search a word number for each character of the above candidate character string. . That is, a word search is performed for each character from the first character to the Nth character in the candidate character string.

The table registration processing unit 2 registers the word numbers output from the word search processing unit 1 in the word registration table 7 in the memory unit 5 in the order in which they are searched. The registration of the word number in the word registration table 7 performed by the table registration processing unit 2 is performed for each character as described later.

The word appearance number calculation unit 3 as the word number appearance number calculation means is connected to the above-mentioned word registration table 7, reads the word number registered in the word registration table 7, and appears the corresponding word number. Calculate the number of times. This calculation process is a calculation process for the word numbers obtained by all the search processes from the first character to the Nth character of the recognition candidate character string. The calculation result of the word appearance frequency calculation unit 3 is output to the output word determination processing unit 4.

The output word determination processing unit 4 as output word determination means is connected to the notation table 8, the character number table 9, and the minimum matching character number table 10 provided in the memory unit 5, and is output from the word appearance frequency calculation unit 3. Which word the recognition candidate character corresponds to is determined from the calculation result.

FIG. 2 is a diagram for explaining a specific memory configuration of the word position-specific word table 6 provided in the memory section 5 described above. The character position-specific word table 6 as the character position-specific word table means has N words from the first character word table shown in (a) of the figure to the Nth character word table shown in (b) of the figure. Each word table is composed of a key character information section 6a and a word number information section 6b. The key character information part 6a is (a) in FIG.
As shown in (b), it is composed of a key character area, a word count area, and a pointer area for word number information. In this key character area, the first of the words registered in advance is
The letters of the letters are stored, for example, the word "walking tired",
"Step" corresponding to the first character such as "rewrite",
Characters such as "calligraphy" are registered. Also, in the word count area, the character "step" stored in the key character area described above is used.
Among words registered in the character position-specific word table 6, such as “walking tired”, “walking too much”, and “walking”, the number of words including “step” as the first character is registered. Has been done. That is, “78” shown in FIG.
Means that 78 words including the character “step” as the first character are registered. In addition, for example, when the character registered in this key character area is "writing", "rewrite",
This means that 45 words including “writing” as the first character such as “wrong writing” are registered in the character position-based word table 6.

The above-mentioned character strings "beginning of walking" and "rewriting"
A specific word number is given in advance to the character string such as, and this word number is specifically stored in the above-mentioned word number information section 6b. For example, the word number “100” is given to the character string “rewriting”, and the word number “35” is given to the character string “beginning of walking”. The address for searching the word number information section 6b is designated by the pointer data stored in the above-mentioned pointer area for word number information. For example, in the case of the character "Ayu",
Point data for corresponding word number information “1904
8 "is data designating the uppermost address of the word number information portion 6b shown in (a) of the same figure. Similarly, point data" 2 "for word number information corresponding to the character" calligraphy "
"0684" is data designating, for example, the second address of the word number information section 6b.

The above-mentioned configuration explains the table configuration of the first character word table, but the same applies to the word table of the second character to the Nth character. On the other hand, FIG. 3 is a diagram showing the configuration of the notation table 8 described above. The notation table 8 stores the character information of the character strings stored in the character position-specific word table 6, and for example, “Aun”, “Asa”, “Ano”, “Are”, ... , ..., “Rewrite” and other word information are registered. As described above, each word is given a word number, and the word number, the number of constituent characters of the word, and the address to the notation table 8 are registered in the character number table 9 shown in FIG. For example, as described above, the character string “rewrite” is the word number “100”, and if the word number “100” in the character number table 9 is known,
Along with the number of characters, by referring to the notation table 8 by the point data to the notation table 8, the reading "rewrite" can be understood.

On the other hand, FIG. 5 shows the minimum matching character number table 10
Is a diagram for explaining the table configuration of FIG. 3, in which information on the minimum number of matching characters for each word length (the number of constituent characters of words) is registered. The information on the minimum number of matching characters is registered with respect to the word length of the registered word (the number of constituent characters of the word), even if the candidate character is recognized as the word, it can be determined that there is no mistake. For example, in a word having a word length (the number of constituent characters of the word) of “4” such as the words “rewriting” and “walking fatigue”, the minimum number of matching characters is “3” and the word length is “3”.
In the case of, the minimum number of matching characters is “2”, and in the case of word lengths of “9” and “10”, the minimum number of matching characters is “6”.

A specific operation of word matching in the word matching device having the above configuration will be described below. Figure 6
For example, for the word "rewrite", character recognition is performed,
It is an example of a recognition candidate character output from a character recognition device (not shown). That is, "calligraphy", "ki", enclosed in □ in the figure,
Although the character "Kana" is a character that has been correctly recognized, the character recognition device uses other characters shown in the figure, such as "Ayumu", "Look", and "Sa".
Is also recognized as a candidate character and is output to the word search processing unit 1. When the recognition candidate character string is input, the word search processing unit 1 first performs a search process for the character at the first position of the first character in the recognition candidate character string. That is, the first
The word number of the word having the first character "Ayu" of the first character as the first character is searched from the above-mentioned first character word table.

FIG. 7 is a diagram showing the flow of this search processing and the subsequent word matching processing. First of the first character above
The character "Ayu" of the rank is the first in the word table 6 for each character position.
By searching the character word table, the word number including the corresponding character "step" is read. That is,
As shown in FIG. 2A, when the word "step" is searched, the word number including the word "step" is read from the word number information section 6b. The word numbers read at this time are "35" and "5" as shown in FIG.
9 ”,“ 83 ”, etc. The word numbers thus read are output to the table registration processing unit 2 and stored in the word registration table 7 under the control of the table registration processing unit 2. A indicates that the word number “35” (this word number “35” corresponds to the word “beginning to walk”) is stored in the word registration table 7 by the above-described processing. It should be noted that, although not shown, word numbers such as "53" and "83" read from the above-mentioned word number information section 6b are written below the word number "35" in FIG.

Next, the word search processing unit 1 makes the word number of the word having the second character "calli" of the first character the first character the first word number.
It is read from the word number information section 6b of the character word table. The word number read at this time is, as shown in the figure,
The word numbers such as “100”, “104”, and “920” are read out in this way, and the word numbers thus read out are written in the word registration table 7 under the control of the table registration processing unit 2. In FIG. 6, b indicates the word numbers “920”, “100”, etc. by the processing at this time (the word number “920” is, for example, the word “writing”, word number “1
00 ”corresponds to the word“ rewrite ”) is stored in the word registration table 7.

Thereafter, similarly, a search process is performed up to the last N-th character "to" of the first character, and the read word number is registered in the table. This registration may be performed sequentially in the order of retrieval. As a result of the above processing, the word registration table 7 has "35", "920",
Word numbers such as "100", "78", "1230" are registered.

As described above, the first character of the first character
When the word search process and the table registration process are completed up to the Nth position, the same process is performed from the first character "ki" of the second character. At this time, even when the word that has already appeared in the first character appears again, the registration process in the word registration table 7 is performed. For example, the word search processing unit 1 reads the word number of the word having the second character "ki" in the second character as the second character from the second character word table, and as shown in c of FIG.
The word numbers such as “100”, “700”, “35” including the word numbers “35” and “100” retrieved by the first character are written in the word registration table 7. At this time, the word number "100" corresponds to the word "rewrite" as described above, the word number "700" corresponds to the word "write down", and the word number "35" corresponds to the word "start walking".

Similarly, the search process is performed up to the last Nth character of the second character, and the search process is similarly performed from the third character to the Nth character. Similar to the above, the result of this search is registered in the word registration table 7. For example, as shown in FIG. 7, the third character is “100”, “46”, “70”.
A word number such as "0" is registered, and similarly, a word number such as "65", "10", "1054", etc. is stored as the fourth character. Note that the above-mentioned word "rewrite" is an example. Is a four-character word, the fourth character is the Nth character, and the word search process ends in the above example.

Next, the word appearance number calculation unit 3 reads out the word numbers stored in the word registration table 7 and performs a sorting process. This sorting process is performed by the word registration table 7
This is performed for all the word numbers registered in the word registration table 7 regardless of the first character, the second character, etc. stored in. Therefore, all the registered word numbers are rearranged in the order of word numbers as shown in d of FIG. When the word numbers are rearranged in this way, the word numbers read a plurality of times are always continuous, and the number of appearances can be easily obtained. For example, the word number "100" has three appearances, and the word numbers "35" and "700" have two appearances.
The data of the number of appearances is output to the output word determination processing unit 4.

When the above-mentioned appearance count data is input, the output word determination processing unit 4 outputs the word number having the maximum appearance count to the character count table 9 to obtain the data of the character count of the corresponding word number. For example, in the above example, the word number "10"
0 "is the maximum number of appearances 3, and this word number" 10 "
The character number table 9 is searched for "0". By this processing, the number of characters of the word number "100" is 4, and the notation table 8 is searched according to the pointer to the corresponding notation table 8 (point data "108464" at this time). The word "rewrite" corresponding to the word number "100" is retrieved and read.

Next, the output word determination processing unit 4 outputs the character number data (for example, 4) read from the character number table 9 to the minimum matching character number table 10 and reads the data of the minimum matching character number. At this time, in the above example, the word number "1"
Since the character count data of "00" is "4", the minimum matching character count defined in the minimum matching character count table 10 is "3", and the data of this minimum matching character count "3" is sent to the output word determination processing unit 4. The output word determination processing unit 4 compares the data of the minimum matching character number "3" with the number of appearances of the word number "100" (word "rewriting"), and performs word matching. Here, the output word determination processing unit 4
Indicates that the data read from the minimum matching character number table 10 is “3” and the number of appearances output from the word appearance number calculation unit 3 is also “3”, so that both data match. That is, the number of appearances “3” is “rewriting” (word “10”, which is considered to be the correct word (target word) in the recognition candidate characters.
0 ") means that three characters match, and if the minimum number of matching characters" 3 "matches up to three characters in the case of a word consisting of four characters, it can be judged that there is no error even if it is judged as the correct word. Therefore, the output word determination processing unit 4 determines that the word "rewrite" with the word number "100" is obtained as a result of matching the recognition candidate characters read from the character recognition device based on the coincidence of the above two data. The word "rewrite" is output as the correct word.

By processing as described above, word matching can be easily performed. FIG. 8 is a block diagram of a word matching device for explaining another embodiment of the present invention.

This embodiment uses a hash function for word search, and FIG. 8 is a system configuration diagram thereof.
That is, the word search processing unit 11 and the table registration processing unit 1
2. The output word determination processing unit 13 and the memory unit 15. FIG. 8 does not include the configuration of the word appearance frequency calculation unit 3 as compared with the configuration of FIG. 1 described in the above embodiment.
The memory unit 15 is composed of a character position-based word table 16, a word registration table 17, a notation table 18, a character number table 19, and a minimum matching character number table 20, and is similar to the memory unit 5 shown in FIG. However, the internal structure of the word registration table 17 is different. That is, in this word registration table 17, word numbers are registered in the area addressed by the hash function. Therefore, the table registration processing unit 12 determines a table address for registering the word number from the searched word number using a hash function, and registers the word number and the number of word appearances at the corresponding address in the word registration table 17.

FIG. 9 is a diagram for explaining the above-mentioned word search processing, table registration processing, and subsequent processing. That is, the word search processing unit 1 performs the search process from the first character of the recognition candidate characters output from the character recognition device (not shown), and has the first step "Ayu" in the first character. The word number of the word is obtained from the first letter word table. A word registration table 17 is created by a hash function based on this word number.
Obtain the registration destination address for and register the word number and the number of appearances. The number of appearances at this time is once. Similarly, the first
The word search processing is performed for the second character “calli” to the Nth character “to” to obtain the registration destination address in the word registration table 17, and the registration processing is performed in the corresponding area of the word registration table 17. To do. By the above-described processing, the word number such as "35" is registered as in the above-mentioned embodiment, and when the second character word table is searched for the second-rank "calligraphy", the word number such as "100" is registered. . Therefore,
When the above process is repeated up to the Nth position, the above “35”,
Word numbers such as “78”, “100”, and “920” are searched, and the word numbers searched at this time are all the first read word numbers. This word number is registered in the word registration table 17 as shown in FIG. 9 under the control of the table registration processing unit 12. That is, the word registration table 17
The read word number and the number of appearances “1” of the word number are registered in.

Next, when the search process is repeated in the same manner from the first character to the Nth character of the second character, the word number that has already appeared in the first character may appear again. However, in this case, since the table registration processing unit 12 of the present embodiment performs the registration processing based on the hash function, the word registration table 17
The number of registrations of the corresponding word number in the (hash table) is incremented. That is, the number of appearances "2" is set for the word numbers "35" and "100" read again as shown in FIG.

In the same manner, the word search processing is executed for the third and fourth characters. For the fourth character, for example, the word numbers "35", "920" shown in FIG.
"100", "10", "78", "1230", "4"
Word numbers such as "6" and "700" are registered. In this, "3" is set to "100" which is the word number searched three times, and the word number "35" searched twice. ,
“2” is set in “700”.

The output word determination processing unit 13 performs the character position-based word table 16 after the above-described search processing for the second and subsequent characters.
Each time a word is read from the table, the character number table 19 and the minimum matching character number table 20 are referred to, and the word registration table 1
It is checked whether or not the number of word appearances obtained from No. 7 is equal to or more than the minimum number of matching characters of the number of word constituent characters. Then, as in the above-described embodiment, if the number of matching characters is the minimum or more, the corresponding word is output. This process is repeated until the fourth character "n" at the end of the recognition candidate character string. For example, in the case of the above example, when the word number including the first character “ka” of the third character is registered in the word registration table 17, the word number “100” included therein is read three times. This means that "3" is set in the area corresponding to the word number "100". Therefore, at this point, the output word determination processing unit 13
Searches the character number table 19 and reads the data "4" of the number of constituent characters of the word number "100", and reads the data of the minimum number of matching characters "3" from the minimum number of matching characters table 20 based on the data "4" of the number of constituent characters. . The output word determination processing unit 4 determines from the above data that the correct word is the "rewrite" of the word number "100" in the same manner as in the above embodiment.

In the above-described other embodiment, the words selected as candidate words by the word search processing unit 11 are registered in the table by using the hash function, but are also registered in the word table for each word constituent character. It may be configured. By configuring in this way, from the word table for each number of characters constituting the word, the minimum matching character number table, the character position currently being processed, and the number of word occurrences, for words of a certain word length or more, the character position It may be found that the collation result cannot be obtained even if the process is advanced to. That is, even if it is assumed that the word numbers appear at all the character positions after the current processing position for all words, the number of characters constituting the word may not exceed the minimum matching character number. As a concrete example of this case, let W1 be the number of constituent characters of a word, Wh be the number of appearances of that word, and Min (Wl) be the minimum number of matching characters in the case of that word length, and let the current character position be m.
Then, if Min (Wl)-(Wl-m) ≤Wh, the collation result cannot be obtained. In this case, the processing is not continued for words whose constituent characters are Wl or more. As a result, the processing time for word matching can be shortened.

For example, considering the word "beginning of walking" with the above-mentioned word number "35", if the number of appearances is two at the time when the search processing for the fourth character is completed,
The minimum number of matching characters in the case where the number of constituent characters Wl of the word is “4”, the number of appearances of the word Wh is “2”, and the word length (Wl) is Min (Wl) is the minimum matching character number table 20 described above. Since it is "3" and the current character position m is "4", the above expression Min (Wl)-(Wl
-M) ≤ Wh is 3- (4-4) ≤ 2, which does not satisfy this expression, so that the subsequent processing is terminated at this point.

[0034]

As described above in detail, according to the present invention, for each character of the recognition candidate character string, the word containing the character is searched, the searched word is registered in the table, and this table is stored. Since the word matching is performed based on the matching number obtained by comparing the number of word occurrences that is easily obtained by sorting and the number of word constituent characters, the table that records the number of appearances each time a word is read as in the conventional case This eliminates the need for the word search process within, and makes it possible to increase the processing speed.

Further, by registering the number of word appearances in the table address determined by using the hash function, the word and the number of appearances thereof can be obtained at high speed by using the hash function, so that the processing speed is further increased. Can be realized.

Furthermore, when registering words in the table,
By registering in the table according to the number of word constituent characters,
The word matching process after the predetermined character position can be omitted, and the processing speed can be increased from this point as well.

[Brief description of drawings]

FIG. 1 is a system configuration diagram of a word search system according to an embodiment.

FIG. 2 is a diagram showing a table structure of a character position-based word table.

FIG. 3 is a diagram showing a configuration of a notation table.

FIG. 4 is a diagram showing a configuration of a character number table.

FIG. 5 is a diagram showing a structure of a minimum matching character number table.

FIG. 6 is a diagram showing a specific example of a recognition candidate character string.

FIG. 7 is a diagram illustrating an operation of a word search method according to an embodiment.

FIG. 8 is a system configuration diagram of a word search system of another embodiment.

FIG. 9 is a diagram for explaining the operation of the word search method of another embodiment.

[Explanation of symbols]

 1, 11 word search processing unit 2, 12 table registration processing unit 3 word appearance number calculation processing unit 4, 14 output word determination processing unit 5, 15 memory unit 6, 16 character position-based word table 7, 17 word registration table 8, 18 notation table 9, 19 character number table 10, 20 minimum matching character number table

Claims (3)

[Claims] 1. A word position-specific word table means in which a word number is set for a predetermined word and the word number is registered for each character forming the word, and an input means for inputting a recognition candidate character string, A word number search means for searching the word table means for each character position for the recognition candidate character string input from the input means and reading a word number for each character of the recognition candidate character string; and the word number search means. A number-of-words-number-of-occurrences calculation unit that obtains the number of occurrences of the searched word number, and the number of constituent characters of a word of a word number with a large number of occurrences and its number of occurrences are compared based on the number of occurrences of word numbers calculated by the number of occurrences of word number And an output word determining unit that determines a word based on the degree of coincidence. 2. A word position-specific word table means in which a word number is set for a predetermined word, and the word number is registered for each character forming the word, and an input means for inputting a recognition candidate character string. A table registration process for searching the word table means for each character position for the recognition candidate character string input from the input means, determining a table address using a hash function, and registering the searched word and the number of appearances thereof in a table. A word collating apparatus comprising: a means for determining a word based on a degree of coincidence obtained by comparing the number of appearances of the word and the number of characters constituting the word. 3. The address of the word number obtained by searching the character position-based word table is determined using a hash function, and the address is registered in the table.
Registration means for registering in a word table by the number of constituent characters of the word corresponding to the word number, and using the word table, word matching processing continuation for deciding whether or not to continue word matching at a predetermined character position 3. The word matching device according to claim 2, further comprising: determining means.