JPH0696286A - Character recognizing device - Google Patents

Abstract

PURPOSE:To provide the character recognizing device which relieves a character in the parentheses recognized erroneously with respect to a result of character recognition including the parentheses, and can improve remarkably a recognition rate. CONSTITUTION:The character recognizing device is constitution by providing an image reading part 1, a character segmenting part 2 for extracting an image area of a character unit from data of the image reading part 1, a character recognizing part 3 for recognizing the data obtained from the character segmenting part 2 as a character and converting it to a character code, a language processing part 4 for executing phrase setting, a word collation, and a word connection test, a parentheses character array generating part 5 for storing a position of the character in which a parentheses character is a first candidate in an array, a pair deciding part 6 for deciding whether the parentheses character constitutes a pair or not, the other character searching part 7 for searching a character for forming a pair with the parentheses character, and a candidate character exchanging part 8 for positioning the character for forming a pair with the parentheses character in a first candidate character in a character position detected by the other character searching part 7.

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 converting printed characters, dot characters and handwritten character patterns of newspapers, magazines, novels etc. into code information such as JIS code.

[0002]

2. Description of the Related Art In recent years, with the development of office automation, a character recognition device is used to input a printed manuscript without using a keyboard.

The character recognition apparatus sets a provisional bunsetsu by character type transition of a candidate character, compares a word generated from a group of candidate characters in the provisional bunsetsu with a word dictionary to find a matching candidate word, and finds it by the above procedure. A word as a solution is obtained by testing whether or not the candidate words are connected.

A conventional character recognition device will be described below. A post-processing method of a conventional character recognition device will be described based on the recognition result as shown in (Table 1). A temporary clause is set according to the character type transition of the recognition result, and the words in the temporary clause are analyzed.

[0005]

[Table 1]

Tentative phrase 1 "" "Since" "" is a symbol, the phrase is established.

Tentative phrase 2 "ABCJ and" "ABCJ" are regarded as nouns because they are English character strings.

"To" is a postpositional particle and can be suffixed to a noun, so that a clause is established. Tentative phrase 3 ““ ”“ “” is a symbol, so the phrase is established.

Tentative phrase 4 "DEF" "DEF" is regarded as a noun because it is an English character string.

There is a provisional clause 5 "". Since "" "" is a symbol, the clause is established.

"Are." Is an affirmative auxiliary verb, and since it can be suffixed to a noun, a clause is established.

As described above, in the conventional character recognition device, the post-processing is performed at the time of the provisional phrase 2 without detecting "J" as a misrecognized character.

Next, the post-processing method of the conventional character recognition device will be described based on the recognition result as shown in (Table 2). A temporary clause is set according to the character type transition of the recognition result, and the words in the temporary clause are analyzed.

[0014]

[Table 2]

Tentative clause 1 "[" "[" is a symbol, so the clause is satisfied.

Tentative phrase 2 "ABCte da" "ABC" is considered as a noun because it is an English character string.

The first candidate, "te", is either a connective particle, a five-verb verb line hypothesis form, or a five-verb verb line imperative form, but all three cannot be suffixed to a noun.

The second candidate, "so", is in the form of a five-verb verb, which cannot be suffixed to a noun.

The third candidate, "ko", is a five-verb verb or incomplete form, and cannot be suffixed to a noun.

As described above, in the conventional character recognition device, 4
Although it can be pointed out that the column eye is an erroneously recognized character, post-processing was performed without relief because the correct character is not included in the candidate character group.

[0021]

However, in the above-mentioned conventional configuration, since the extraction of words according to the grammatical rules is mainly performed as described above, the characters that should be normally paired such as parentheses form a pair. If not, there is a problem that the abnormality cannot be detected. Further, even if it is possible to detect that the character position where the parenthesis should be recognized is erroneously recognized, there is a problem that the rescue is impossible unless the correct character is included in the candidate character group.

The present invention solves the above-mentioned conventional problems, and provides a character recognition device which relieves a erroneously recognized parenthesis character from a character recognition result including a parenthesis and significantly improves the recognition rate. The purpose is to

[0023]

In order to achieve this object, a character recognition device of the present invention is a recognition device that recognizes image data read from an image reading device and converts it into a character code. With respect to the result, among all the first candidate characters in the recognition result, a pair of symbols or characters such as parentheses is searched, and the character code of the symbol and the image data are stored in the array, and then stored. For each character, it is verified that the character paired with the character exists in the first candidate character of the recognition result, and if not, the image data of the character paired with the character stored in the array. And the character image data to be recognized is collated, and when the collation result is coincident and the character image data having a proper positional relationship with the character exists, the character image data is used as the first candidate character of the character image data. And a pair By adopting to character, it consists of correctly recognizing each means the sentence, including parentheses.

Specifically, an image reading section for photoelectrically converting an original image, a character cutting section for extracting an image area in character units from the image data of the image reading section, and image data obtained by the character cutting section. A character recognition unit that recognizes characters and converts them into a character code, a language processing unit that performs clause setting, word matching, and word connection verification, and the position of the character that makes the parenthesis character the first candidate from the verification result of the language processing unit. A parenthesis character array creation unit that stores the array, a pair determination unit that determines whether or not the parenthesis characters stored in the parenthesis character array creation unit form a pair, and the parentheses determined by the pair determination unit. A character having a partner character search unit that searches for a character paired with a character, and a candidate character exchange unit that positions a character paired with a parenthesis character to the first candidate character at the character position found by the partner character search unit. Recognition equipment It is.

[0025]

With this configuration, a character having a parenthesis in the first candidate of the recognition result is searched for, the image data, the position, and the character code of the character are stored, and if the stored parenthesis does not have the other parenthesis, the other parent The character image data at the position where the parenthesis is expected to exist is compared with the stored image data having the same character code as the other parenthesis. If they match, it is possible to correctly recognize the Japanese sentence including the parenthesis that replaces the first candidate character at the expected position with the partner parenthesis character.

[0026]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of a character recognition apparatus according to an embodiment of the present invention. Reference numeral 1 is an image reading unit for photoelectrically converting an original image, 2 is a character cutout unit for extracting an image area in character units from image data, and 3 is character recognition of the image data obtained from the character cutout unit 1 and converting it into a character code. A character recognition unit, 4 is a language processing unit that performs phrase setting, word matching, and word connection verification, 5 is a parenthesis character array creation unit that stores the position of a character whose parenthesis character is the first candidate in an array, and 6 is a parenthesis character A pair determination unit that determines whether or not a pair is formed,
7 is a partner character search unit that searches for a character paired with a bracket character, 8
Is a candidate character exchange unit that positions a character paired with a parenthesis character to the first candidate character at the character position found by the partner character search unit, and 9 determines whether or not all parenthesis characters have been processed. It is an end determination unit.

The operation of the character recognition apparatus of this embodiment having the above-described structure will be described below.

FIG. 2 shows image data to be recognized, and FIG.
3 is a flowchart showing a symbol recognition method of the character recognition device of this embodiment.

First, the image data of FIG. 2 photoelectrically converted (step 21) by the image reading unit 1 is converted into the character cutting unit 2.
Is used to cut out image data in character units (step 2
2), the character recognition unit 3 recognizes (step 23), the candidate character group is sent to the language processing unit 4, and the recognition result subjected to the relief processing (Step 24) by the language processing unit 4 is the parenthesis character array creation unit. 5 is sent. The recognition result sent out is shown in (Table 3).

[0031]

[Table 3]

In the table, the character positions are serial numbers assigned in the order in which they are cut out by the character cutting section, the first candidate shows the most probable recognition result, and the second candidate shows the second most probable recognition result. , The third candidate shows the third most probable recognition result.

The one-character word signature is determined by the number of characters of the word generated during language processing. When it is 0, it means that it is not a single letter word, and when it is 1, it is a single letter word.

In the recognition result after the language processing (step 24), parenthesis character processing (step 25) is performed by the parenthesis character array creating unit 5 to store the character position having the parenthesis character type as the first candidate in the array.

Next, a procedure for creating a parenthesis character array in the parenthesis character array creating unit 5 will be described. FIG. 4 is a flowchart showing a procedure for creating a parenthesis character array.

First, the beginning of the sentence is set as the character position to be processed (step 31, step 32).

The first candidate character code at the character position to be processed is obtained from the recognition result (step 33).

Whether the character code is a parenthesis character type is checked (step 34). If it is a parenthesis character type, the character position is stored in the parenthesis character array corresponding to the character code (step 35).
The above process is repeated until the character position reaches the last character of the sentence.

The results of the bracket character arrangement are shown in (Table 4).

[0040]

[Table 4]

In (Table 4), character codes are represented by characters for convenience. Here, the character code indicates a parenthesis character code. The partner character code indicates the character code of the parenthesis character paired with the parenthesis character indicated in the character code.

The character position is the position of the character whose parenthesis character shown in the character code is the first candidate character, and can be stored up to 5 places. -1 indicates that there is no memory character position.

In the text below, the character code is also expressed as a character. The pair determination unit 6 verifies the correspondence status of the parentheses. One parenthesis character position whose position has been confirmed is taken out from the parenthesis character array and stored in cNum (step 41).
0 is stored in cNum. The parenthesis character code at the cNum position
Store in code (step 42). "" is stored in code. The orientation of the parentheses is determined (step 43). If the parenthesis character array number is even, it is judged as left parenthesis, and if it is odd, it is judged as right parenthesis. Since the parenthesis character array number is 12, which is an even number, it is determined to be a left parenthesis.

Since it is a left parenthesis, the partner character searching unit 7 starts searching for the corresponding right parenthesis (step 4).
4).

FIG. 5 is a flow chart showing a method for searching for a right parenthesis. A method for searching for a right parenthesis will be described with reference to the flowchart of FIG. First, cNum is substituted for the search character position leftCNum (step 51). leftCNum becomes 0.

A parenthesis character code paired with code is taken out from the parenthesis character array and is substituted into opCode (step 5).
2). The opCode will be “”.

Move leftCNum forward one character (step 5
3). leftCNum = 1. It is confirmed whether leftCNum is less than or equal to the sentence end position (step 54). It is less than the last character of the sentence.

The first candidate character code at the leftCNum position is temp
Substitute for Code (step 57). The tempCode is "a".

Collate tempCode and opCode (step 5)
8). The collation results are inconsistent. The tempCode and code are collated (step 5a). The collation results are inconsistent.

Advance leftCNum to the position character ahead (step 53). leftCNum = 2. It is confirmed whether leftCNum is less than or equal to the sentence end position (step 54). It is less than the last character of the sentence.

The first candidate character code at the leftCNum position is temp
Substitute for Code (step 57). The tempCode will be “”.

Match tempCode and opCode (step 5)
8). The matching result is a match. Since the corresponding code is found, the analysis for the code ends.

It is determined whether analysis has been performed for all the parenthesized characters whose position has been confirmed (step 46). There are still parenthesis characters left.

One parenthesis character position whose position has been confirmed is extracted from the parenthesis character array and stored in cNum (step 4).
1). 5 is stored in cNum.

The parenthesis character code at the cNum position is stored in code (step 42). "" is stored in code.

The orientation of the parentheses is judged (step 43).
If the parenthesis character array number is even, it is judged as left parenthesis, and if it is odd, it is judged as right parenthesis. Since the parenthesis character array number is 12, which is an even number, it is determined to be a left parenthesis.

Since it is the left parenthesis, the search for the corresponding right parenthesis is started (step 44). At search character position leftCNum
Substitute cNum (step 51). leftCNum becomes 5.

A parenthesis character code paired with code is taken out from the parenthesis character array and is substituted into opCode (step 5).
2). The opCode will be “”.

Move leftCNum forward one character (step 5
3). leftCNum = 6. It is confirmed whether leftCNum is less than or equal to the sentence end position (step 54). It is less than the last character of the sentence.

Temp is the first candidate character code at the leftCNum position
Substitute for Code (step 57). tempCode will be "O".

Collate tempCode and opCode (step 5)
8). The collation results are inconsistent. The tempCode and code are collated (step 5a). The collation results are inconsistent.

Move leftCNum forward one character (step 5
3). leftCNum = 7. It is confirmed whether leftCNum is less than or equal to the sentence end position (step 54). It is less than the last character of the sentence.

The first candidate character code at the leftCNum position is temp
Substitute for Code (step 57). The tempCode will be "J".

Collate tempCode and opCode (step 5)
8). The collation results are inconsistent. The tempCode and code are collated (step 5a). The collation results are inconsistent.

Move leftCNum forward one character (step 5
3). leftCNum = 8. It is confirmed whether leftCNum is less than or equal to the sentence end position (step 54). It is less than the last character of the sentence.

The first candidate character code at the leftCNum position is temp
Substitute for Code (step 57). tempCode will be “of”.

Collate tempCode and opCode (step 5)
8). The collation results are inconsistent. The tempCode and code are collated (step 5a). The collation results are inconsistent.

Move leftCNum forward one character (step 5
3). leftCNum = 9. It is confirmed whether leftCNum is less than or equal to the sentence end position (step 54). It is less than the last character of the sentence.

The first candidate character code at the leftCNum position is temp
Substitute for Code (step 57). tempCode is “in”.

Match tempCode and opCode (step 5)
8). The collation results are inconsistent. The tempCode and code are collated (step 5a). The collation results are inconsistent.

Move leftCNum forward one character (step 5
3). leftCNum = 10. It is confirmed whether leftCNum is less than or equal to the sentence end position (step 54). It is less than the last character of the sentence.

The first candidate character code at the leftCNum position is temp
Substitute for Code (step 57). tempCode is ",".

Collate tempCode and opCode (step 5)
8). The collation results are inconsistent. The tempCode and code are collated (step 5a). The collation results are inconsistent.

Move leftCNum forward one character (step 5
3). leftCNum = 11. It is confirmed whether leftCNum is less than or equal to the sentence end position (step 54). It is less than the last character of the sentence.

The first candidate character code at the leftCNum position is temp
Substitute for Code (step 57). tempCode is "".

Collate tempCode and opCode (step 5)
8). The collation results are inconsistent. The tempCode and code are collated (step 5a). The matching result is a match. Since the left parenthesis is detected without the corresponding right parenthesis appearing, it can be determined that the right parenthesis is erroneously recognized.

Substitute "necessary" for the analysis signature. Substitute leftCNum-1 for the analysis end position endNum (step 5)
b).

It is determined whether the analysis signature is "necessary" or "unnecessary" (step 5c). Since it was "necessary", we proceed with the analysis.

FIG. 6 is a flow chart showing a method for processing parenthesis characters. First, cNum is substituted for leftCNum (step 61). leftCNum becomes 5.

Move leftCNum forward one character (step 6
2). leftCNum becomes 6. It is confirmed whether leftCNum is less than or equal to endNum (step 63). leftCNum is endN
It was below um.

Whether or not the leftCNum position is a one-letter word is confirmed (step 64). It is a one-letter word.

It is searched whether opCode is included in the candidate word group at the leftCNum position (step 65). Not included. Therefore, similar processing was performed for the character positions 7, 8, 9, and 10 in FIG. 2, but opCode was not included in the candidate word group. It is confirmed whether the opCode is found (step 67). Since it was not found, the corresponding parenthesis character is searched by collating the image data in the sentence with each other.

Substitute cNum for leftCNum (step 6
8). Advance leftCNum one character forward (step 62).
leftCNum becomes 6.

It is confirmed whether leftCNum is less than or equal to endNum (step 63). leftCNum was less than endNum.

It is confirmed whether the leftCNum position is a one-letter word (step 64). It is a one-letter word.

One character position having opCode as the first candidate is extracted from the parenthesis character array. 2 is obtained.

The character image data at the leftCNum position and the character image data at the two positions are collated. The verification results were in agreement.

Therefore, the candidate character exchange unit 8 positions opCode at the first candidate character at the leftCNum position (step 6).
d).

Next, the search for the left parenthesis will be described. The recognition result sent to the bracket character array creation unit 5 is shown in (Table 5).

[0090]

[Table 5]

The parenthesis character array created according to the procedure of the parenthesis character array creating unit 5 is shown in (Table 6).

[0092]

[Table 6]

The pair determination unit 6 verifies the correspondence status of parentheses. In FIG. 4, one parenthesis character position whose position has been confirmed is extracted from the parenthesis character array and stored in cNum (step 41). 2 is stored in cNum.

The parenthesis character code at the cNum position is stored in code (step 42). "" is stored in code.

The orientation of the parentheses is judged (step 43).
If the parenthesis character array number is even, it is judged as left parenthesis, and if it is odd, it is judged as right parenthesis. Since the parenthesis character array number is 13 and is an odd number, it is determined to be a right parenthesis.

FIG. 7 is a flow chart showing the method for searching the left parenthesis. A search method for the left parenthesis will be described with reference to the flowchart of FIG. First, cNum is substituted for the search character position rCNum (step 71). rCNum is 2.

A parenthesis character code corresponding to code is taken out from the parenthesis character array and assigned to opCode (step 7
2). The opCode will be "".

The rCNum is returned to the beginning of one character (step 73). rCNum = 1. It is confirmed whether rCNum is equal to or higher than the sentence head position (step 74). It is above the beginning of the sentence.

The first candidate character code at the rCNum position is tempCo
Substitute in de (step 77). The tempCode is "a".

Collate tempCode and opCode (step 7)
8). The collation results are inconsistent. The tempCode and code are collated (step 7a). The collation results are inconsistent.

RCNum is returned in the direction toward the beginning of one character (step 73). rCNum = 0. It is confirmed whether rCNum is equal to or higher than the sentence head position (step 74). It is above the beginning of the sentence.

The first candidate character code at the rCNum position is tempCo
Substitute in de (step 77). tempCode becomes “F”.

Collate tempCode and opCode (step 7)
8). The collation results are inconsistent. The tempCode and opCode are collated (step 7a). The collation results are inconsistent.

RCNum is returned to the beginning of one character (step 73). rCNum = -1. It is confirmed whether rCNum is equal to or higher than the sentence head position (step 74). It is below the beginning of the sentence.

Since the beginning of a sentence is detected without the corresponding left parenthesis appearing, it can be determined that the left parenthesis is erroneously recognized.

“Necessary” is substituted for the analysis signature. The sentence head position is substituted for the analysis end position endNum (step 7b).

It is determined whether the analysis sign is "necessary" or "unnecessary" (step 7c). Since it was "necessary", we will proceed with the analysis.

FIG. 8 is a flowchart showing a method of processing parenthesized characters. First, cNum is substituted for rCNum (step 81). rCNum is 2.

One character is returned from rCNum (step 82). rC
Num becomes 1. It is confirmed whether rCNum is equal to or larger than endNum (step 83). rCNum was greater than endNum.

It is confirmed whether the rCNum position is a one-letter word (step 84). It is a one-letter word.

It is searched whether opCode is included in the candidate word group at the position of rCNum (step 85). Not included.

Similar processing is performed for the 0 character position, but opCode is not included in the candidate word group.

It is confirmed whether or not the opCode is found (step 87). Since it was not found, the corresponding parenthesis character is searched by collating the image data in the sentence with each other.

Substitute cNum for rCNum (step 8)
8). Return rCNum by one character (step 89). rCNum is 1.

It is confirmed whether rCNum is equal to or larger than endNum (step 8a). rCNum was greater than endNum.

Whether or not the rCNum position is a one-character word is confirmed (step 8b). It is a one-letter word.

One character position having opCode as the first candidate is extracted from the parenthesis character array. 5 is obtained.

The character image data at the rCNum position is compared with the character image data at the 5th position. The matching results were inconsistent.

One character position having opCode as the first candidate is extracted from the parenthesis character array. 11 is obtained.

The character image data at the rCNum position is compared with the character image data at the 5th position. The matching results were inconsistent.

The rCNum is returned by one character (step 89). rC
Num becomes 0. It is confirmed whether rCNum is equal to or larger than endNum (step 8a). rCNum was greater than endNum.

It is confirmed whether or not the rCNum position is a one-character word (step 8b). It is a one-letter word.

One character position having opCode as the first candidate is extracted from the parenthesis character array. 5 is obtained.

The character image data at the rCNum position is compared with the character image data at the 5th position. The verification results were in agreement.

Therefore, the candidate character exchange unit 8 positions opCode at the first candidate character at the position rCNum (step 8d).

In the same manner, all parenthesized characters are analyzed in the same manner. The end determination unit 9 determines whether or not the above work is completed, and when the end is determined, the determination work is completed.

[0127]

As described above, according to the present invention, the character recognition result can be corrected so that the left and right parenthesis characters can be associated with each other, and the character recognition device having a significantly improved recognition rate can be realized.

[Brief description of drawings]

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

FIG. 2 is a diagram showing an example of image data of a recognition target used for explaining one embodiment of the present invention.

FIG. 3 is a flowchart showing a symbol recognition method of a character recognition device according to an embodiment of the present invention.

FIG. 4 is a flowchart showing a procedure for creating a parenthesis character array of the character recognition device in the embodiment of the present invention.

FIG. 5 is a flowchart showing a right parenthesis search method of the character recognition device in the embodiment of the present invention.

FIG. 6 is a flowchart showing a method for processing parenthesized characters of the character recognition device in the embodiment of the present invention.

FIG. 7 is a flowchart showing a method for searching for a left parenthesis of the character recognition device in the embodiment of the present invention.

FIG. 8 is a flowchart showing a method for processing a right parenthesis character of the character recognition device in the embodiment of the present invention.

[Explanation of symbols]

 1 image reading unit 2 character cutout unit 3 character recognition unit 4 language processing unit 5 parenthesis character array creation unit 6 pair determination unit 7 partner character search unit 8 candidate character exchange unit 9 end determination unit

Claims (1)

[Claims] 1. An image reading unit for photoelectrically converting an original image,
A character cutout unit for extracting an image area in character units from the image data of the image reading unit, a character recognition unit for character-recognizing the image data obtained from the character cutout unit and converting it into a character code, clause setting, word A language processing unit that performs matching and word connection verification, a parenthesis character array creation unit that stores the position of a character whose parenthesis character is the first candidate from the test result of the language processing unit in an array, and the parenthesis character array creation unit. A pair determination unit that determines whether or not the stored parenthesis characters form a pair, a partner character search unit that searches for a character that forms a pair with the parentheses character that is determined by the pair determination unit, and the partner character search unit A character recognition device, comprising: a candidate character exchange unit for locating a character paired with a parenthesis character, at the first candidate character at the character position found in 1.