JPH0196779A - Word reader - Google Patents

Abstract

PURPOSE:To realize the high-speed collation of words without deteriorating the accuracy of word reading, by providing a reading object word deciding means. CONSTITUTION:A reading object word deciding means 6 is provided to decide the word where the character of a specific position in a word dictionary 7 is included in recognition candidate characters as a reading object word. The word decided by the means 6 is collated with said candidate characters. While the dictionary 7 is entirely collated with the candidate characters only in case no word is decided. Therefore, it is possible to perform the collation of words at a high speed without deteriorating the word reading accuracy since the collation is carried out first among a small number of words having high possi bility for satisfactory collation and then performing the collation among all words of the dictionary 7 only when the first collation fails.

Description

[Detailed Description of the Invention] [Field of Industrial Application] This invention relates to a word reading device that reads and recognizes words such as addresses and names, and in particular, recognizes each character that constitutes a word, and uses the recognition results. This invention relates to shortening the query time of a word reading device that corrects words using .

[Conventional technology]

In recent years, large amounts of high-speed data transmission to computers, etc.) I
Word reading devices are attracting attention as a means of J-1.Since Japanese has a complex mixture of kana characters, alphanumeric characters, etc., there is hope for the development of word reading devices that can read words with high accuracy. It is rare. This word reading device is known as an optical character reading device (OCR), and as shown in Fig. 2, for example, “Word information in handwritten kanji recognition” "Use of", etc. The configuration of a conventional word reading device is shown.

In the figure, (the opening is a form with letters written on it.

(2) is a scanning means for causing a reading device to read the form (original part); (3) is a character recognition means for cutting out and recognizing input characters character by character, and outputting recognition candidate characters.

141 is a word dictionary that stores words to be entered in the form (1).

(51 is a word determining means that determines a word by comparing the recognition candidate character outputted from the character recognition means (3) with a plurality of words in the word dictionary (41).) Conventional word reading devices With the above structure, according to the judgment conditions shown below, words that are ranked high among the recognition candidate characters and have been entered in the word dictionary (4-) are preferentially selected for word reading. This is what is output as a result.

Judgment conditions> When m recognition candidate characters are output from the character recognition means (3) and a word consisting of n characters is input, the jth word Aj of a last name with n characters in length is determined by the zero-order formula. Calculate Pj.

and.

pk= [nin (Pj IAj is n characters single gJ<n
If there exists a word k (Ak is an n-character word) such as XT (T is a threshold), the second item in the n-character dictionary of the last name is output as the matching result.

The third factor is an explanatory diagram showing an example of a form (1) read by a word reading device. In the figure, (81 is the input character "Minakami", and (9) is the input character 1 "Iceberg").

Examples of recognition candidate characters output from the character recognition means (31) for each input character are shown in FIGS. 4 and 5.

Figure 4 shows the recognition candidate characters selected by the 0 character recognition means (3:) for the input character 1 "Minakami" (8). For the first character ``1 water'', there are two recognition candidate characters ``water'' and ``ice'', or for the second input character 1 upper'', there are ``earth''. 1. There are three recognition candidate characters: ``up'' and @stop.

Figure 5 shows the recognition candidate characters selected by the 0 character recognition means (3) for the input character 1 Nagayama'' (9), and g
As shown in (Ill) of srs, for the first input character ``1ei'', there are two recognition candidate characters @ice'' and @water'', or the second input character There are three recognition candidate characters for ``1 mountain'', 0 m mountain, 1.1 small'', and 1J1 bi.

Further, FIG. 6 is an explanatory diagram showing an example of words with word length 2 in the word dictionary of surnames in the word reading device. In the diagram, LLz is the word 1 Nagai 1, (1 m is 1 Mizuyama)
(14 is the word 1 water, aji is the word ``ice island,'' and ae is the word 1 water.)

Next, the operation of the conventional device is determined based on the determination condition m=3.
n=2. O=10. The case where T=8 will be explained. The word determination means (51 First, the recognition candidate character QI is checked at each character position of the word 1 Nagai" (12) with word length 2 in the word dictionary.

In other words, the first character "1 ice" is compared with the first input character Issei candidate character 7. However, the first input character
Since there is no ``1ei'' in the mR sleeve letter, p11=
t.

becomes. Next, input the second character ``1i'' and the second input character.
Match the recognition candidate character with the first character. Here too, since ``1i'' does not exist in the recognition candidate characters for the second input character, P21=10, resulting in PIT-20. Next, the word 1 Mizuyama''aj in the word dictionary is compared. The matching operation is performed in the same manner as described above. At each character position, the first character "1 water" does not exist in the wt* candidate characters, and the second character "1 mountain" does not exist in the verification candidate characters, so P2
=P12+P22=10+10=20.

Next, the word "minakami" I in the word dictionary is compared. At each character position, the first character "1 water" is in the first position among the recognition candidate characters, and the second character "1 top" is in the second position among the recognition candidate characters, so pg = p + 5 + P25 = 1 + 2 = It becomes 3.

Then, a query is made for the word 1 Mizushima'19 (!-) in the word dictionary. At each character position, the first character 1 mizu m exists in the first place among the B recognition candidate characters, and the second character 1 mizushima m does not exist in the recognition candidate characters, so P4 ” Pj4 + P24=
1+10=11.

Next, a match is made with the word "Ibara-tte" in the word dictionary. At each character position, the first character "Water" is in the first place among the recognition candidate characters, and the second character "1hara" is a recognition candidate. Since it does not exist in the characters, P5 ”pts + P25=1+
10=11. As a result, among Pj (j=1 to 5), i & small value is P5=3 ((nXT (・=111))
As a result of the matching, the correct answer ``1 Mizukami'' is obtained.

In addition, for the recognition candidate character 0 old ζ of 1 Nagai' (9) written on the form in Figure 5, the word length 2 in the word dictionary shown in prisoner 6, 'Nagai' (12, ″″Nagayama (13゜1
Water #1. 'Ice Island' Q9. Perform the same matching operation as above with "Ice Field" ae.

As a result, p, = 2Q for 'Nagai' asHc.

“Iceberg” (p2=ll for 13, P5=12 for @Mizukami “I” P4=12゜1 Suwon for “Water M” tte Ps = 12 t!:, 'Q,
Among Pj (j=1 to 5), the minimum value is P2=1
1 ((nXT(=16)), and the correct answer "Iceberg" is obtained as a matching result.

[The problem that the invention attempts to solve]

However, in the conventional matching method described above, it is necessary to match the recognition candidate characters of the entered characters with all the words that have the same number of characters as the number of entered characters, which has the disadvantage of requiring a large amount of time. .

The present invention was made to solve this problem, and an object of the present invention is to provide a word reading device that can perform word inquiries at high speed without reducing word reading accuracy.

[Means for solving problems]

The word reading device according to the present invention is provided with a reading target word determining means for determining a character at a specific character position in a word dictionary or a word included in a recognition candidate character as a reading target word, and the reading target word determining means is provided. The word selected by the means is compared with the recognition candidate characters, and only when the word cannot be determined, the entire word dictionary and S! ! It is designed to match against weak candidate characters.

[Effect]

The word reading device in this invention first compares recognition candidate characters with words in the word dictionary.

Matches a small number of words with a high probability of matching.

Only when this matching is unsuccessful, matching against all words in the word dictionary is performed, so word matching can be performed at high speed without reducing word reading accuracy.

[Embodiments of the invention]

Hereinafter, an embodiment of the present invention will be described based on the drawings.

FIG. 1 shows the components of an example of the word reading device according to the present invention, and (l) to (3) and (51) are the same components as the conventional word reading device shown in FIG. In the figure, (6) is a word that is included in a character at a predetermined specific character position or a recognition candidate character obtained from the single character recognition means (3) among the words in the word dictionary.
The reading target word 7 is a reading target word determining means that reads out the reading target word from a word dictionary.

Further, ()) is a word dictionary in which an index is added to the word dictionary (4) so that words satisfying predetermined conditions can be extracted using characters at predetermined specific character positions as keys.

FIG. 7 is an explanation showing an example of a word with a word length of 2 in the surname word dictionary (7) in the word reading device. -, aX8 is the word 1 Nagai, (13G21 is the word Nagai 1, (141 is the word 1 Minakami), and 1 is the word 1.
The word ``Mizushima'', aS is the word ``Suwon''.

Now, αη is an index whose key is the first letter of a single line, and U is the key 1 in the in-book system η.

α9 is the index (key 1 in 17+ Z cl
Q is the key 1 in the index αη.

Qυ is from the key of 1 ice #al in Inbooksumi n.

A pointer pointing to the word ``1 Nagai'' az, c13 is a pointer pointing to the word ``Iceberg'' from the key 1 Nagai#- in the index αη, a pointer pointing to the word Q3 ``Iceberg'', @ is a key pointing to the word ``1 water'' aS in the index ul, Pointer pointing to the word ``water'' I, c! 4 is 1 water "f1" in index QD
From the key 9, a pointer points to the word ``water bird'', and (c) points from the key σ9, ``1 water'' in the index Qη, to the word ``@'' water freeze field 1, as.

Note that (e) is a pointer to the key 1 ice even in the index fin, and indicates that there is no word for the key 1 ice even.

Next, the operation will be explained. The recognition candidate character ``1 Minakami'' (8) written on the middle form in Section 4 is compared with a word with a word length of 2 in the word dictionary (7) shown in Figure 7. Target word determining means (6) and word determining means (5)
).

First, the reading target word determining means (6) selects the first character of the recognition candidate character 1 water of the input character 1 water m (8) in FIG.
Referring to the first recognition candidate character 1 ', is it the first character?
The word "water" is extracted from the word dictionary (7).In this extraction, the index aTI of the word dictionary shown in Figure 0m7 is
Please refer to the 1 water "0" key.

Extract the words indicated by pointer (2), (goods), @, ``Water'' I, @ice island #similarity, @ice field'' (LLl).

Next, the reading target word determining means (al 41 , the first character in the recognition candidate character al of the input character 1 Mizukami m (8) in the fourth factor)
The first character or the word "ice" is extracted from the single candy word @ (?) by referring to the second-ranked recognition candidate character "1 ice".

When extracting, if you refer to the false key in the dictionary in the dictionary shown in Figure 1, the pointer (2) indicates that the word does not exist.

Words whose first character is "1 ice" cannot be extracted.

Thus, the words to be read are @Minakami"1.@Suzuki"aS
, 1 Mizuhara"ae is determined.Now that the word to be read has been determined, the zero-order word determination means (5) performs a verification.This verification operation is performed in the same manner as in the conventional device.However, The constants for the judgment conditions of the conventional fct are m=3.n=2
．． C=10. Let T==3. Similar to the comparison with the conventional device, when the recognition candidate character is compared with the target word for each character position, for 'Mizukami' I, pH = 3. - for Hyojima' Kasumi, P2 = '11. ``Ice field'' (1 [Pjc vs. P5 = 11 Tona Q, Ko's result Pj = (j =
1 to 3), the minimum value is P1=3 ((nXT
(=8)), and the correct answer is 1 Mizukami" as a matching result. In this way, in the conventional system, matching was performed with words in the entire word dictionary, but in this invention, the correct answer is "1 Minakami". The word determination means (6) selects 3 words (1.
, @″Mizuhyo Island.

You can get the correct answer just by narrowing down your search to ``Suwon'').

Next, for the recognition candidate character a old of ``1 iceberg'' (9) written on the form in Figure 5, the word length 2 in the word dictionary (7) shown in Figure 7'll is compared. Verification 8. A case in which the reading target word determining means (6) and the word determining means (5) are used will be described.

As above, first, the reading target word determining means (6).

Recognition candidate character for input character 1 Nagayama 1 (9) in Figure 4 +1
1) Referring to the first character, the first recognition candidate character "Ice", or the word "Ice" as the first character does not exist, as described above.

Next, the first character in the recognition candidate character aD of the input character 1 Nagayama m (9) in Figure 4, and the second recognition candidate character 1 water'''
Referring to ``1 Minakami'' al., as above.

Three words, “waterfowl” aS and “ice field” ae, are extracted.Furthermore, in the same way as above, the constants of the judgment conditions of the conventional device are set as m=3.n=2.O=10.T=3 When collated as follows, P1=12°1 for 'Mizukami' Q4, p2=i2 for Mizushima 15. @Iceland” (Ps=12 for Lll
Therefore, both P, 1 (j=1 to 3)〉n XT (=
6), no matching results can be obtained.

In this case, all the words in the word dictionary (7) are compared using the index function without referring to the word dictionary, as in the conventional setting. Sumire. At this time, the constant of 1 judgment condition is 0m=
3. n=2. O=10. Verification is performed with T=8. As a result, the same results as the conventional comparison were obtained, and @"P1=20 for Mizui υ.""Nagayama (P2 for I3
=11. @Mizukami”p3=12 for I.-Ice Island #(L
P4=12 for 9. Ps= for ``'Ice Field'' ae
It's 12a.

The minimum value among P, 1 (j = 1 to 5) is P2 = 1
1 [(nXT(=16)), and the correct answer is ``iceberg'' as a matching result.

In this way, by using the reading target word determining means (6), when matching recognition candidate characters with words in the word dictionary (7), it first selects a small number of words with a high possibility of matching. Verification is performed, and only when the verification is unsuccessful, all words in the word dictionary (7) are verified, so word verification can be performed at high speed without reducing word reading accuracy.

In addition, in the above example, 42. Whether the word to be read is extracted using the first character of the word as a key, or the character position and number of key characters can be set freely.

(For example, whether the first character or the last character of a word is 1, extracting the words contained in the u-heel candidate characters, etc.) Also, in the above example, the surname of the kanji character was mentioned. This invention can also be applied to words such as 1, names, company names, etc., such as katakana and English characters.

〔Effect of the invention〕

As described above, according to this invention, 111! By providing a means for determining target words, it becomes necessary to constantly check all the words in the seven-word dictionary for recognition candidate characters (this makes it possible to perform word checking at high speed without reducing word reading accuracy. .

[Brief explanation of the drawing]

Fig. 1 is a block diagram of a word reading device according to an embodiment of the present invention, Fig. 2 is a block diagram of a conventional word reading device, and Fig. 3 is an explanatory diagram showing an example of a form of the word reading device. and fifth
The figure shows an example BAV of input characters and recognition candidate characters in a word reading device. FIG. 6 is an explanatory diagram showing an example of a word string on the word length side in a word dictionary in a conventional word reading device, and FIG. This is an example of A-side control. In the figure, (3) is a character recognition means, and (5) is a word determination means. (6) is a reading target word determining means, and (7) is a word dictionary. In addition, in FIG. 1, the same reference numerals indicate the same or corresponding parts.

Claims (1)

[Claims] In a word reading device that recognizes and reads characters constituting a word, a character recognition means selects, for each input character, a plurality of recognition candidate characters ordered in descending order of similarity as input characters, and a word written on a form. A word dictionary that stores
a word determining means for determining a word by comparing the recognition candidate characters with the word dictionary; and a word in the word dictionary in which a character at a specific character position is included in the recognition candidate characters as a reading target word. a reading target word determining means that reads out a word from a word dictionary; first, the recognition candidate character is compared with the word read by the reading target word determining means, and only if a word cannot be determined at this stage; A word reading device characterized in that it is configured to determine a word by comparing recognition candidate characters with all words in a word dictionary.