JP3339879B2 - Character recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device for reading and recognizing characters of a sentence such as a document.

[0002]

2. Description of the Related Art In recent years, with the development of databases, demands for a high-speed and high-recognition character recognition apparatus have been increasing.

A conventional character recognition device is disclosed in, for example, IPSJ Transactions Vol.30 No.11 pp.1394-1401. FIG. 4 shows a conventional character recognition device.
The character correction unit of the character recognition device includes a word search unit 3, a word dictionary 4, a phrase search unit 6, a grammar dictionary 7, a phrase evaluation value calculation unit 9, and the like. It takes as input n candidate characters per character.
The word search unit 3 of the character correction unit obtains a set of candidate words from the set of candidate characters using the word dictionary 4. The phrase search unit 6 selects a candidate phrase from the set of candidate words using the grammar dictionary 7, and obtains a set of candidate phrases. The phrase evaluation value calculation unit 9 evaluates the evaluation value of the character recognition unit 1, the frequency of words, the length of characters, and the like for each candidate phrase, and derives a phrase evaluation value indicating the certainty of the phrase. The phrase selecting unit 10 selects a phrase that is considered to be the most correct based on the phrase evaluation value, and obtains a corrected character string 11.

As described above, by using a word dictionary or a grammar dictionary, a character that is difficult to determine only with the character recognition unit 1 can be corrected with knowledge of words and grammar.

[0005]

However, in the above-described character recognition device, the knowledge used for correcting the recognition character output from the character recognition unit 1 is knowledge of words and grammar of general sentences. is there. However, many documents have unique characteristics. For example, in the case of a patent document, many words unique to the patent are described. In this way, the style of sentences and the frequency of words used are usually different depending on the contents of the document, but such unique information possessed by the document was not used for correction of recognition characters .

Further, in the conventional method, a correct character is selected from the n candidate characters output from the character recognition unit 1 for correction. However, if the correct character is not included in the n candidates, the correction is impossible. there were.

[0007] Therefore, there is a problem that the character recognition rate is reduced because the unique information of the document is not used as described above.

SUMMARY OF THE INVENTION An object of the present invention is to provide a character recognition device which can increase the character recognition rate in consideration of the conventional problems as described above.

[0009]

SUMMARY OF THE INVENTION The present invention provides a candidate character recognizing means for reading a character string of a sentence to be recognized to obtain a candidate character group, and a candidate word based on a word dictionary and a candidate character string created from the candidate character group. A word phrase search means for obtaining a group of words and a candidate word group by the candidate word group and the grammar dictionary;
At least the candidate clause group, an evaluation value computing means for computing an evaluation value considering at least lexical and grammatical correctness for each clause, and selecting a clause from the candidate clause group according to the computation result; Phrase selecting means for outputting a selected character string created by the selected phrase, keyword extracting means for extracting a keyword from the output selected character string based on a predetermined criterion, the candidate character group and the keyword comprising a partial matching means for performing a partial match search between, and a candidate word adding means for adding to said candidate word group the keywords that the partial match as a candidate word and the
The selected character string output from the phrase selecting means is the candidate unit.
Only words included in the word group are composed of the predetermined
Is that the candidate word is included in the sentence to be recognized.
For all of the clauses, the evaluation value of the clause and one of the candidate words
Find the difference from the general frequency, the greater of the difference and 0
Is the sum of the values of
The character recognition device is characterized by recognizing a character string that has been extracted .

[0010]

[0011]

According to the present invention, the candidate character recognizing means reads the character string of the sentence to be recognized to obtain a candidate character group, and the word phrase search means obtains the candidate word group by the candidate character string created from the word dictionary and the candidate character group. And a candidate phrase group is obtained from the candidate word group and the grammar dictionary, and the evaluation value calculating means calculates an evaluation value for each phrase at least in consideration of at least lexical and grammatical correctness of the candidate phrase group. In accordance with the operation result, the phrase selecting means selects a phrase from the candidate phrase group, outputs a selected character string created by the selected phrase, and the keyword extracting means determines a predetermined character string from the output selected character string. The keyword is extracted based on the criteria of (1), the partial match search means performs a partial match search between the candidate character group and the keyword, and the candidate word adding means searches the partial match keyword for the candidate unit. As added to the candidate word group, to recognize the character string read by using the candidate word group. What
The selected character string output from the phrase selecting means is
It consists of only words included in the candidate word group,
The predetermined criterion includes the above-mentioned condition in the sentence to be recognized.
For all clauses containing complementary words, the evaluation value of the clause and the previous
The difference between the general frequency of the candidate word and the difference is calculated.
The sum of the larger value is used.

[0012]

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the drawings showing embodiments thereof.

FIG. 1 is a block diagram of a character recognition apparatus according to an embodiment of the present invention. That is, the character recognizing unit 1 performs character recognition from a character image, and sets a candidate character set 2 (n) having n character characters from a first candidate character to an nth candidate character per character in a character string of length m. (Candidate character group).

The word search unit 3 searches the word dictionary 4 to select a candidate word that matches a word existing in the word dictionary 4 from a combination (candidate character string) of the candidate character set 2 and selects a candidate word set. 5 (candidate word group) is created. The phrase search unit 6 refers to the grammar dictionary 7 to select a candidate phrase that is a combination of words that can be a phrase, and sets a candidate phrase set 8
(Candidate clause group). The phrase evaluation value calculation unit 9 calculates the lexical and grammatical correctness of the phrase retrieved by the phrase retrieval unit 6 based on the length and frequency of words in the phrase as an evaluation value (Equation 1).

[Number 1] _{E = aE a + bE b +} cE c + dE d However, E _a: the character recognition apparatus 1 Output E _b: the length of clause E _c: frequency of the independent words included in the phrase E _d: Word The lengths a, b, c, and d are constants. The phrase selection unit 10 selects a phrase having a large evaluation value from the phrase candidates and generates a corrected character string 11 (selected character string). The evaluation value of the phrase in the modified character string 11 is E _i (i = 1,
B). Here, B is the number of phrases in the entire document.

The keyword extracting unit 12 includes a phrase selecting unit 10
Then, the keywords of the document to be recognized are extracted from the corrected character string 11 output as described above, and a keyword set 13 is created. For this, the evaluation value of the phrase in the corrected character string and general frequency information of the word are used. For example, ease K _w becomes to keywords for word w can be determined by equation (2).
That is, the difference between the evaluation value and the general frequency of the word w is obtained for all the phrases including the word w in the document, and the sum of the difference and the larger value of 0 is obtained.

[0017]

(Equation 2)

Here, S _i represents a character string included in the phrase i, and F _w represents a general frequency of the word w. Keyword, (number 2) ease K _w made to the keywords that can be calculated
Are obtained by obtaining in order from the largest one. Table 1 shows the result of extracting keywords from documents related to the neural network according to the present embodiment.

[0019]

[Table 1]

[0020] Table 1 is obtained by arranging the words in descending order of ease K _w made to the keyword. As can be seen from this table, keywords contained in the document can be extracted. In addition, when you use only the frequency information of the word "thing", although the frequent words such as "thing" is extracted, since the extraction is carried out in search of ease of K _w made to the keyword, like this " It can be seen that frequent words such as "things" and "things" are not extracted as keywords, and only words representing the contents of the document are extracted as keywords, and the obtained keywords often express the features of the text.

The keyword partial match search unit 14 performs a partial match search between the obtained keyword and a set of candidate character strings. For example, if "recognition" is extracted as a keyword, "recognition *" and "* knowledge" in the corrected character string are extracted as partially matched characters.

The candidate word adding unit 15 adds the partially matched keyword to the candidate word set 5 as a candidate word of the candidate character string. In the example described above, partially recognized “*” and “* knowledge” are added to the candidate word set 5 as “recognition”. This allows characters not output from the character recognition unit 1 to be included in the candidate characters. This uses a keyword indicating the characteristics of a sentence, so that processing according to the characteristics of the sentence can be performed.

If there is a keyword in the phrase, the keyword weighting unit 16 increases the evaluation value of the candidate phrase according to (Equation 3). This enhances the evaluation of candidate phrases using keywords representing the features of the text. As a result, it is possible to make corrections using the features of the text.

[0024]

(Equation 3)

The character recognizing unit 1 constitutes a candidate character recognizing unit, the word retrieving unit 3 and the phrase retrieving unit 6 constitute a word phrase retrieving unit, and the phrase evaluation value calculating unit 9 constitutes an evaluation value calculating unit. , The phrase selection unit 10 constitutes a phrase selection unit, the keyword extraction unit 12 constitutes a keyword extraction unit, the keyword weight addition unit 16 constitutes a keyword weighting unit, and the keyword partial match search unit 14 comprises a partial match search unit. And the candidate word adding unit 15 forms a candidate word adding unit. Further, the portion excluding the above-described character recognition unit 1 constitutes a character correction unit.

Next, the operation of the above embodiment will be described.

First, the document to be recognized is processed by the character recognizing unit 1 to obtain, for example, a candidate character set 2 up to the tenth place. Next, the word search unit 3 selects a combination of candidate characters that match a word existing in the word dictionary 4, and
With reference to the grammar dictionary 7, a combination of words that can be a phrase is selected. Thereafter, the phrase evaluation value calculation unit 9 calculates the evaluation value of the phrase searched by the phrase search unit 6. The phrase selecting unit 10 selects a correct phrase from the phrase candidates based on the calculated evaluation value, and outputs the corrected character string 11 in the first pass.

[0028] In the next keyword extraction unit 12, calculates the ease K _w made to the keyword, to extract the keyword.

The keyword partial match search unit 14 performs a partial match search between the obtained keyword and the candidate character set 2. In order to reduce the amount of calculation, as the candidate character set 2, a character higher than the character recognition unit 1, for example, only the first place, may be set as the candidate character set. This is not a problem if the character recognition unit 1 has a high recognition rate.

Next, the candidate word adding unit 15 uses the keywords obtained by partial matching as candidate words as candidate words 5
To be added. Evaluation value E _a character recognition of the modified character
Here, it is appropriate to give the lowest evaluation value.

The keyword to be used is preferably about tenth. However, for long sentences, it is better to use more keywords.

Again, the phrase search section 6 and the phrase evaluation value calculation section 9
Then, a candidate phrase is selected using the candidate word set 5 to which the keyword is added as a candidate word, and a phrase evaluation value of the newly selected candidate phrase is obtained.

Next, the keyword weighting unit 16
The phrase evaluation value of (expression 3) is corrected according to the characteristics of the sentence.

Finally, the phrase selecting section 10 selects a phrase having a large evaluation value from among the phrase candidates, and outputs the final corrected character string 11 in the second pass.

FIGS. 2 and 3 show part of the result of character recognition according to this embodiment.

FIG. 2 shows the effect of the keyword weighting unit 16. In FIG. 2, seven phrases are included in the candidate for the character string “On-site learning function”. Although the phrase “learning” was not selected in the one-pass recognition result, the keyword “learning” was included in the keyword information. Could be corrected. As described above, in the one-pass recognition and the two-pass recognition, a phrase is selected from lexically and grammatically correct ones using the evaluation value, and in the two-pass recognition, a character string matching the contents of the document is obtained by using keyword information. Was able to choose.

FIG. 3 shows the result of applying the present embodiment to a sentence indicating the content of “post-processing effect of character recognition”. Using this result, the keyword partial match search unit 14 and the candidate word adding unit 15 Show the effect. In the output from the character recognition unit 1, the second "recognition" is not included in the candidate characters. Therefore, the word cannot be corrected by the conventional method. In this method, as shown in FIG. 3, a keyword based on the feature of a sentence can be extracted, so that “recognition” could be included as a candidate by using a keyword partial match search. Therefore, after the correction in the second pass, correct sentences could be obtained.

In the above-described embodiment, the keyword is extracted from the character string once processed by the character correction unit, and the two-pass recognition is performed by processing the character correction unit again using the extracted keyword information. A keyword may be extracted using the result of the recognition, and n-pass recognition may be performed by repeating the processing at least once in the character correction unit. in this case,
Further, correct keyword information is obtained, and the recognition rate is improved.

Also, in the above embodiment, for characters that can be extracted from outside the candidates (ie, characters obtained by adding a keyword as a candidate word to a candidate word set), the evaluation value E _a of the character recognition unit 1 is obtained. Gives the lowest value, but sends the extracted character to the character recognizing unit 1 and evaluates the character E _a
Recalculate the may be obtained evaluation value E _a. by this,
Further, accurate evaluation is possible.

In the above embodiment, the keyword is extracted by using the evaluation value of the phrase and the general frequency information of the word. However, the present invention is not limited to this. A keyword may be extracted using the number of appearances and a dictionary of the frequency of occurrence of the candidate word in a general sentence. Alternatively, other extraction methods may be used.

In the above embodiment, the character is recognized by using both the keyword partial matching search and the weighting. However, the present invention is not limited to this. Only one of the keyword partial matching searching and the weighting is used. A configuration may be used in which characters are recognized by using these.

[0042]

As apparent from the above description, the present invention has an advantage that the character recognition rate can be increased.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a character recognition device according to an embodiment of the present invention.

FIG. 2 is an output diagram of an experimental result showing an effect of keyword weighting of the embodiment.

FIG. 3 is an output diagram of an experimental result showing an effect of a keyword partial match search and a candidate word adding unit of the embodiment.

FIG. 4 is a configuration diagram of a conventional character recognition device.

[Explanation of symbols]

 Reference Signs List 1 character recognition unit 3 word search unit 4 word dictionary 6 phrase search unit 7 grammar dictionary 9 phrase evaluation value calculation unit 10 phrase selection unit 12 keyword extraction unit 14 keyword partial match search unit 15 candidate word addition unit 16 keyword weight addition unit

──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Eiji Maekawa 1006 Kadoma Kadoma, Kadoma City, Osaka Prefecture Inside Matsushita Electric Industrial Co., Ltd. (72) Yasuji Takaki 1006 Kadoma Kadoma, Kadoma City, Osaka Prefecture Matsushita Electric Industrial Co., Ltd. In-house (56) References “IEICE Technical Report PRL191-135” Vol. 91 No. 478 p. 71-78 (199.2.2.12) “Post-processing methods for character recognition and analysis of effects of post-processing” (58) Fields investigated (Int. Cl. ⁷ , DB name) G06K 9/72

Claims (1)

(57) [Claims] 1. A candidate character recognizing means for reading a character string of a sentence to be recognized to obtain a candidate character group, and obtaining a candidate word group from a word dictionary and a candidate character string created from the candidate character group. Word phrase search means for obtaining a candidate phrase group using a grammar dictionary, evaluation value calculation means for calculating, for each phrase, an evaluation value of at least the lexical and grammatical correctness of the candidate phrase group, and its calculation A phrase selecting means for selecting a phrase from the candidate phrase group according to the result and outputting a selected character string created by the selected phrase, and a keyword based on the output selected character string based on a predetermined criterion. Keyword extraction means for extracting a keyword; a partial match search means for performing a partial match search between the candidate character group and the keyword; Candidate word adding means for adding to the candidate word group, and the selected character string output from the phrase selecting means
And the predetermined criterion is the candidate word in the recognition target sentence.
For all the phrases that include, the evaluation value of the phrase and the candidate
Find the difference between the general frequency of the word and the difference
This is the sum of the larger values, and the read character string is recognized using the candidate word group.
Character recognition apparatus characterized by that.