JP3911942B2 - Character recognition device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a character recognition apparatus that performs character recognition of type characters, and more particularly to a technique for identifying similar characters that are difficult to identify.
[0002]
For example, characters with a similar pattern of uppercase and lowercase letters (hereinafter referred to as similar letters), such as the uppercase “tsu” and lowercase “tsu”, are normalized (compressed or expanded) to a certain size. Since the feature amount of the character pattern is extracted, the character recognition result may not be determined only with the pattern matching technique with the character dictionary that has extracted the feature amount of the character of the predetermined font. is there.
For this reason, there is a need for a technique for discriminating between uppercase and lowercase letters in addition to the character recognition by the pattern matching technique.
[0003]
[Prior art]
FIG. 5 is an explanatory diagram of a conventional method for identifying similar characters.
[0004]
The character recognition device determines an upper reference line, a lower reference line, and a standard width of the line from the input character pattern, and obtains an upper space ratio (upper space / standard width of the line) of each character, By comparing the value with a predetermined threshold value, it is distinguished whether it is uppercase or lowercase.
[0005]
As shown in FIG. 5, if the font and the number of points are the same, the upper space corresponding to uppercase letters and the upper space corresponding to lowercase letters are significantly different. Based on the percentage occupied by the corresponding upper space, it can be easily identified whether the character is uppercase or lowercase.
[0006]
[Problems to be solved by the invention]
However, in recent documents, various expression methods such as switching of a large number of fonts have been used.
[0007]
For example, there are many things whose height changes depending on the font.
[0008]
As shown in FIG. 6, for “sa”, for example, there is no difference in size between fonts, but “d” is lower in the Mincho style than the Gothic style.
[0009]
If the threshold value in FIG. 6 (the middle of the uppercase and lowercase letters) of the Mincho type “D” is identified, the lowercase letter of the Gothic type “D” may be erroneously identified as an uppercase letter.
[0010]
As described above, in the conventional similar character identification method, since the discrimination is performed based on the comparison result between the blank ratio of the character to be discriminated and the fixed threshold value, all the similar characters included in such a document are accurately detected. It is difficult to identify.
[0011]
The character recognition device of the present invention flexibly supports a document in which various fonts and emphasis expressions are mixed, and aims to improve the recognition rate of uppercase and lowercase letters of similar characters.
[0012]
[Means for Solving the Problems]
Character recognition means that extracts character pattern feature values from a document image and compares them with character dictionary feature values to select characters with similar feature values, and identifies and outputs uppercase and lowercase characters with similar shapes based on the recognition results A character recognition device having similar character identification means,
The similar character identifying means obtains a predetermined number of character patterns from among the character type marking means for classifying and marking each character type into similar character types and other character types according to the character recognition result, and the recognition result characters. A threshold setting means for setting a reference line in contact with the circumscribed rectangle, and setting a predetermined distance from the set reference line as a threshold, and a reference for a character pattern marked as a similar character type among a predetermined number of character patterns This is a configuration provided with similar character discriminating means for discriminating a character whose length from the line to the center of gravity is smaller than a threshold value as a lower case letter.
[0013]
With this configuration, paying attention to the fact that the distance from the reference line of the similar character to the center of gravity is different between uppercase and lowercase letters, and by setting a predetermined threshold, discrimination of the uppercase and lowercase letters of the similar characters can be realized easily and with high accuracy. .
[0014]
In addition, according to the character recognition result, the character type marking means for marking each character type into a similar character type and a similar character type uppercase type having a height equivalent to an uppercase letter and other character types, and marking among the characters of the recognition result Select the character pattern that is marked as uppercase from the predetermined number of character patterns obtained from, and set the shortest distance from the reference line that touches the circumscribed rectangle of the selected character pattern to the center of gravity of the uppercase character pattern It is a structure provided with the threshold value setting means which acquires and sets a threshold value based on the acquired shortest value.
[0015]
With this configuration, the threshold value is obtained from the shortest distance from the uppercase reference line having the height of the uppercase character of the similar character type to the center of gravity of the character. Can be easily distinguished from capital letters with high accuracy.
[0016]
Character type marking means for classifying and marking each character type into a similar character type and a similar character type uppercase type having a height equivalent to an uppercase letter and other character types according to the character recognition result,
When there is no character pattern marked as uppercase in the predetermined number of character patterns obtained from the recognition result characters, other characters from the reference line set in contact with the circumscribed rectangle of the predetermined number of character patterns The configuration includes a threshold setting unit that acquires the longest value of the distance to the center of gravity of the character pattern marked on the mold and sets a threshold based on the acquired longest value.
[0017]
With this configuration, if a capital letter type is not found in a certain number of character patterns, the capital letter and child characters of similar characters can be distinguished by selecting the one with the longest distance from the reference line to the center of gravity of the character such as Kanji. It is possible. This is because most of the characters such as kanji characters vary in the height of the character.
[0018]
DETAILED DESCRIPTION OF THE INVENTION
FIG. 1 shows a configuration diagram of a character recognition device 1 of the embodiment.
[0019]
The character recognition device 1 includes an image input unit 11, a character recognition unit 12, a similar character identification unit 2, and a character processing unit 13.
[0020]
FIG. 2 shows a flowchart of processing of the character recognition apparatus.
[0021]
The character recognition device 1 reads the document with the image input unit 11 (step S1).
Then, the character recognition unit 12 normalizes the character pattern read by the image input unit 11 to extract the feature amount, compares it with the character feature amount held in the character dictionary, and selects the character having the smallest difference. Then, the character recognition result and the character pattern are passed to the similar character identification unit 2 (step S2).
FIG. 3 shows a flowchart of the process of the similar character identification unit of the character recognition device (step S3).
[0022]
An example of “kana” will be described.
[0023]
As shown in FIG. 1, the similar character identification unit 2 includes a character information storage unit 3, a character type mark unit 4, a similar character identification control unit 5, and a discrimination result output unit 6.
[0024]
The character information storage unit 3 stores the character recognition result from the character recognition unit 12 and the input character pattern in units of lines (step S11).
The character type mark unit 4 determines the character type of all character information for “Kana” from the character recognition result stored in the character information storage unit 3, and stores the determination result as a mark in the character information storage unit 3. (Step S12).
An example of character type is shown below.
・ Similar character type-Kana Uetsuya Yuyawaii ぅ ぉ ゃ ゃ ウ エ 大 文字 大 文字 大 文字 大 文字 大 文字 大 文字 大 文字-Uppercase type- It is in contact with the reference line and is not a similar character.ク ドNext, the similar character identification control unit 5 first obtains a reference line of a line including a capital letter type.
[0025]
As the reference line, a straight line is determined by the least square approximation method from the set of midpoints of the lower sides of the circumscribed rectangular shape belonging to the uppercase type in one page for each page of the read document image (step S13).
Next, character information is identified line by line.
[0026]
It is determined whether the result of determining the character type is a line including a part of the similar character type, whether the character in the line does not include the similar character type, or whether the result is a line of only the similar type.
[0027]
Then, when the similar character type is not included in the characters in the line or the line is only the similar type, the recognition result is directly passed to the discrimination result output unit 6 (step S14).
If the result of determining the character type is a line that includes a part of the similar character type, it is checked whether or not there is an uppercase type in one line (step S15).
If there is a capital letter type in the line, the one with the shortest length from the reference line to the center of gravity of the rectangle is obtained. The threshold value is the length of the capital letter with the shortest length from the base line to the center of gravity multiplied by a predetermined constant (for example, 0.86). Then, it is passed to the discrimination result output unit 6 (step S16).
FIG. 4 is an explanatory diagram of the similar character identifying unit of the embodiment.
[0028]
FIG. 4 (a) shows threshold setting based on capital letters.
[0029]
The distance from the reference line of the capital letter “Nushi” to the center of gravity of the circumscribed rectangle is obtained. As a result, since the center of gravity of “shi” is lower than the center of gravity of “nu”, the center of gravity of “shi” becomes the shortest distance. Then, when the threshold value is obtained and the centroid distances of the similar characters “a” and “tsu” are compared, the centroid position of “a” is larger than the threshold value, and the centroid position of “tsu” is smaller than the threshold value. Is determined to be an uppercase letter, and “tsu” is considered to be a lowercase letter.
[0030]
If there is no capital letter type in a line, a reference line is set from the lower side of the circumscribed rectangle of the line.
[0031]
Then, the character having the longest distance from the determined reference line to the center of the circumscribed rectangle of the target-excluded character pattern in the row is obtained from kanji, numbers, and kana, and is regarded as an uppercase type. Similar processing is performed with the length multiplied by a predetermined constant (for example, 0.86) as a threshold (step S17).
FIG. 4 (b) shows threshold setting based on characters other than capital letters.
[0032]
Find the distance from the base line to the center of gravity of the Chinese character “Dai No.” and the excluding type Kana “Te”. As a result, the distance from the base line of the Chinese character “ban” to the center of gravity is the longest. Then, when the threshold value is obtained and the centroid distances of the similar characters “a” and “tsu” are compared, the centroid position of “a” is larger than the threshold value, and the centroid position of “tsu” is smaller than the threshold value. Is determined to be an uppercase letter, and “tsu” is considered to be a lowercase letter.
[0033]
Then, the discrimination result output unit 6 outputs the discrimination result of capital letters and child characters (step S18).
The character processing unit 13 corrects the result recognized by the character recognizing unit 12 based on the result determined by the similar character identifying unit 2, and outputs the character recognition result to a display or the like (step S4).
[0034]
【The invention's effect】
According to the present invention, the similar character determination of “Kana” in the similar character identification can be performed with high accuracy.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of a character recognition device of an embodiment. FIG. 2 is a flowchart of processing of a character recognition device of an embodiment. FIG. 3 is a flowchart of processing of a similar character identifying unit of the embodiment. Explanatory diagram of character identification unit [FIG. 5] Explanatory diagram of similar character identification method of conventional example [FIG. 6] Explanatory diagram of conventional problem [Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Character recognition apparatus 2 Similar character identification part 3 Character information storage part 4 Character type mark part 5 Similar character identification control part 6 Discrimination result output part 11 Image input part 12 Character recognition part 13 Character processing part

Claims (2)

Character recognition means that extracts character pattern feature values from a document image and compares them with character dictionary feature values to select characters with similar feature values, and identifies and outputs uppercase and lowercase characters with similar shapes based on the recognition results A character recognition device having similar character identification means,
Similar character identification means
Character type marking means for classifying and marking each character type into a similar character type and a similar character type uppercase type having a height equivalent to an uppercase letter and other character types according to the character recognition result,
Select a character pattern that is marked as uppercase from a predetermined number of character patterns obtained from the recognition result characters, from the reference line set in contact with the circumscribed rectangle of the selected character pattern to the center of gravity of the uppercase character pattern Threshold setting means for acquiring the shortest value of the distance, and setting a threshold obtained by multiplying the acquired shortest value by a predetermined constant;
Characters comprising: a similar character discriminating means for discriminating lowercase characters whose length from the reference line to the center of gravity is less than a threshold for character patterns marked as similar character types among a predetermined number of character patterns Recognition device. When there is no character pattern marked as uppercase in the specified number of character patterns obtained from the recognition result characters, other characters from the reference line set in contact with the circumscribed rectangle of the specified number of character patterns A threshold value setting means for acquiring a longest value of a distance to the center of gravity of a character pattern marked on a mold and setting a threshold value obtained by multiplying the acquired longest value by a predetermined constant. The character recognition device according to 1.

Images