JPH0496882A - Full size/half size discriminating method - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method for discriminating between full-width characters and half-width characters in a Japanese text in a character recognition device for Japanese texts.

[Prior Art] In recent years, character recognition devices have come to be used as means for inputting document data into a database. In such applications, faithful recognition of the manuscript is required.
In many cases, simply recognizing a character code accurately is not sufficient.

For example, in the case of Japanese texts, half-width characters are often used mixed in with full-width characters, but while recognizing half-width characters correctly as character codes, it is also necessary to recognize their size, that is, that they are half-width characters. be.

Conventionally, character recognition devices that consider full-width and half-width characters distinguish between full-width and half-width characters by focusing on the character width and space width detected from the vertical projection of character lines. There is a known character recognition device that recognizes characters using both a kanji dictionary and a non-kanji dictionary, and uses only the non-kanji dictionary for half-width characters (Japanese Patent Laid-Open No. 83887/1987). , Japanese Patent Publication No. 63-83
No. 888).

[Problem to be solved by the invention 1 Character width and character amount Blank width differs depending on the character and how the characters are arranged. For example, there is a big difference in character width between ■" and II MI+, and when it is followed by 'TIT', when it is followed by 'M I NI', it is 'Ni kl
The amount of characters and the blank width are different in the case where it continues with "M".

Therefore, the method of determining full-width and half-width characters by comparing the character width and character amount/blank width with a fixed character width threshold as in the prior art described above may have poor determination accuracy or Left and right separator characters (“〕
] It is also easy to misjudge the situation.

Further, in the above-mentioned conventional technology, when a left-right separated Kanji character is determined to be a half-width character, only a non-Kanji dictionary is used for that character 42, resulting in a fatal (unrecoverable) recognition error.

The purpose of the present invention is to solve the problem of accuracy of full-width/half-width judgment as described above and the problem of irrecoverable recognition errors due to judgment errors in a character H, 2 device that targets Japanese sentences. /An object of the present invention is to provide a half-width determination method.

[Means for Solving the Problems] According to the present invention, in a character recognition device for Japanese sentences, character recognition is performed on character images extracted from an input image without distinguishing between full-width and half-width characters. For the characters of a specific character type (for example, alphanumeric characters) obtained as a recognition result, the value related to the character size obtained when cutting out the character image and the full-width/half-width discrimination prepared in advance for each character of the specific character type. Use the value to determine whether it is full-width or half-width.

The full-width/half-width discrimination value is, for example, a threshold value for the ratio of the character width to the standard full-width character width or a threshold value for the ratio of the character height to the character width. The ratio of character width to width or the character height is determined by comparing the character width ratio with the corresponding half-width/full-width discrimination value.

Further, according to the present invention, after the above-mentioned full-width/half-width determination process for each character (this is the number of characters determined to be half-width characters and the number of characters determined to be full-width characters in a character string of a specific character type obtained as a recognition result), A final decision is made as to whether it is full-width or half-width based on the ratio, and the result is V!: (Thus, all characters in the character string are unified to full-width or half-width.

[Function] In general, the types of half-width characters that appear in Japanese texts are limited, and in most cases, kanji and hinagana are full-width characters, and half-width characters are limited to alphanumeric characters (letters and numbers).

The present invention performs character recognition without distinguishing between full-width and half-width characters,
When the result is a character type that may be half-width, such as an alphanumeric character (2■ is subject to full-width/half-width determination, it is possible to avoid the mistake of determining separated kanji, etc., as half-width characters. Furthermore, unlike the method of performing full-width/half-width determination before recognition as in the prior art described above, recognition failure (fatal chance, recognition error) can be avoided by erroneously determining a full-width character as a half-width character.

In addition, because the number of characters that can be used as half-width characters such as alphanumeric characters is limited, we also provide values for determining whether each character is full-width or half-width, such as the standard width of the full-width character. It is relatively easy to prepare in advance a threshold value for the ratio to the width and a threshold value for the aspect ratio (ratio of character height to character width), and since the settings are made for each character, they can be set in detail. Further, such a ratio is not directly influenced by the preceding and following characters.

Therefore, according to the method of the present invention that uses such character-specific discrimination thresholds, it is possible to accurately discriminate half-width characters such as alphanumeric characters that appear in Japanese sentences, and it is also possible to accurately discriminate half-width characters such as alphanumeric characters that appear in Japanese sentences. Unlike methods that use width, it is possible to avoid judgment errors due to the influence of character arrangement.

Also, alphanumeric characters I+ 1. For narrow characters such as 111+-, the difference between the full-width character width and the half-width character width is small, so a determination error may occur in the full-width/half-width determination process for each character.

According to the present invention, after full-width/half-width determination processing for each character, full-width/half-width determination is made for a character string based on the ratio of the number of characters determined to be half-width to the number of characters determined to be full-width in a character string of a specific character type. and, depending on the result, standardize all characters in the string to full-width or half-width characters. In general, in Japanese sentences, it is rare for a single alphanumeric character string (word) to contain both full-width and half-width characters, so by performing full-width/half-width determination processing for each character string, characters such as i I+ or '1 It is possible to correct misjudgment of narrow characters such as “.

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

In this character recognition device, an image input unit ]O reads an image of a document using a scanner or the like, inputs the binary image data, and stores it in an image memory ]1. The line/character cutting unit 12 cuts out a character image of a character line from the input image in the image memory 11, stores the cut out character image data in the character image memory 13, and also stores the character cutting position, character width, Cutting information such as character height and standard value for full-width characters for each line is stored in the cutting information memory 14.

The character recognition unit 15 reads the character image data from the character image memory 13, performs normalization processing, extracts the feature amount, and compares the extracted feature amount with the dictionary feature amount read from the character dictionary memory 16. , the recognition result candidates with the shortest distances between the feature amounts are obtained, and are stored in the recognition result memory 17 in order of decreasing distance. Note that at this stage, it does not distinguish whether the characters are half-width or full-width, and both a kanji dictionary and a non-kanji dictionary are used for character recognition, and a full-width character code is output as a recognition result.

The full-width/half-width determination unit 18 refers to the first candidate of the recognition result obtained in the recognition result memory 17 and the contents of the full-width/half-width discrimination table memory 19, and determines the first candidate of a specific character type (here, alphanumeric characters). Full-width/half-width determination is performed only for characters, and character codes of candidates determined to be half-width characters are rewritten to half-width character codes.

The full-width/half-width discrimination table memory 19 stores a full-width/half-width discrimination table in which full-width/half-width discrimination values for each character are registered for alphanumeric characters that are specific character types. In this example, full-width /! 4'', a threshold value of the ratio of the character width to the standard width of full-width characters, which is used as the value for corner discrimination.

The result output unit 20 outputs the recognition result data after the full-width/half-width determination process in the recognition result memory 21 to an output device such as a display or a printer.

FIG. 2 shows the overall processing flow of this character recognition device.

(2) is image input processing by the image input section 10, (2) is cutting processing by the line/character cutting section 12, and (2) is character recognition processing by the character recognition section 15 without distinguishing between full-width and half-width characters. ■ and ■ are processes performed by the full-width/half-width determination unit 18,
■ is a character-by-character determination process, and ■ is a character string amount determination process.

(2) is the output processing of the recognition result by the result output unit 20.

The contents of the character-by-character full-width/half-width determination process (2) are as shown in FIG. Also, full-width/side-width determination processing for each character string■
The processing contents are as shown in FIG.

Next, the horizontal text string “New Rjfax released” (
(Kanji characters are printed in full-width characters and alphabetic characters are printed in half-width characters.)
The full-width/half-width determination process will be explained in detail using as an example. However, to simplify the explanation, it is assumed that there are no other characters in the line where this string exists.

Assume that as a result of line cutting and single character cutting of the character string, cutting information having the contents shown in FIG. 5 is obtained in the cutting information memory 14.

Here, the character width is the horizontal size of the character circumscribing rectangle, and the character height is the vertical size of the character circumscribing rectangle. In addition, the full-width character standard width is a parameter given to the target line, and is calculated based on the maximum character width or height of characters within the line, the height of the line, etc. Ru.

Assume that as a result of character recognition for each character in the character string, a first candidate as shown in FIG. 6 is obtained and stored in the recognition result memory 19. Note that character recognition does not distinguish between full-width and half-width characters, so in this embodiment, all candidate character codes are full-width character codes, as described above.

Assuming that the full-width/half-width discrimination table memory 19 has the contents shown in FIG. 7, the character unit determination process (2) of [Full-width/half-width determination section]8 (FIG. 2) will be explained with reference to FIG.

First, the first recognition result candidate in the recognition result memory 19 is referred to to search for alphanumeric characters (uppercase letters, lowercase letters, and numbers) (steps 100, 101, and 102).

When an alphanumeric character is found, the character width and full-width standard width of that character are read from the cutout information memory 14, and the ratio A of the character width to the full-width character standard width is calculated (step 103). Next, the full-width/half-width discrimination table memory 19 is searched using the full-width character code of the target character, and the full-width/half-width discrimination threshold B for that character is read.

Then, A and B are compared, and if A<B, the target character is determined to be half-width, and if Ag3, the target character is determined to be full-width (
Step 105). In other words, half-width characters have a horizontally curved shape compared to full-width characters, so the ratio of half-width characters is A.
Since the value of ratio A is smaller than the value of ratio A for full-width characters, this difference is determined using threshold value B. and,
Since the appropriate value for threshold B naturally differs depending on the character, threshold B is prepared for each character. In other words, by character-specific, it becomes possible to finely set the threshold value B to the optimum value.

If it is determined to be half-width, rewrite the full-width character code of the first candidate of the target character to a half-width character code (step 1o6)
.

Similar processing is repeated until the final character is determined in step 1.07.

In the case of the above string, the third character "R" is A-0,5
Since 0,8=0.76, it is determined to be half-width. The alphanumeric characters from the 4th to the 7th are all A<B, so (
(see FIGS. 5 and 7), all are determined to be half-width.

Through such full-width/half-width determination processing for each character, the first candidates for most half-width characters are corrected to half-width character codes, so it is also possible to output the recognition result after this processing as the final result.

That is, it is also possible to omit the full-width/half-width determination process σ (FIG. 2) for each character string.

However, in this embodiment, in order to achieve more reliable full-width/half-width determination, full-width/half-width determination processing is further performed for each character string. The contents of this process will be explained with reference to FIG.

After the initial settings (steps 201 and 202), the first candidate for the target line in the recognition result memory 17 is examined in order from the first character, the alphanumeric string is searched, and the half-width and full-width characters in the found alphanumeric string are searched. (Step 20)
3-210). The alphanumeric flag is a flag for detecting the beginning and end of an alphanumeric character string. If the alphanumeric character string continues to the last character, proceed from step 209 to step 213, compare the number of full-width characters and the number of half-width characters in the alphanumeric character string, and if the number of full-width characters is equal to the number of half-width characters,
The character codes of the first candidates for all characters in the alphanumeric character string are rewritten to half-width character codes (step 214).In other words, in this embodiment, alphanumeric characters are Performs final determination of full-width or half-width for the entire column.

Further, if a character other than alphanumeric characters exists following the last character of the alphanumeric character string, the process proceeds to step 213 via steps 211 and 212, and similar processing is performed.

Then, if it is determined in step 215 that the character is the last character, the process ends.

As a result of the full-width/half-width determination process for each character, it is assumed that, for example, only 1'' in the character string 'Rjfax' is incorrectly determined to be full-width. Narrow characters such as 1" and '1" are
Because the difference in character width between full-width and half-width characters is small, incorrect determinations are likely to occur when character-by-character determination processing is performed.

However, since the number of full-width characters is l and the number of half-width characters is 4,
In the determination process for each character string, this "]" is corrected to half-width characters.

In general, it is rare for full-width and half-width characters to coexist in a single string of alphanumeric characters (word), so if you perform this kind of judgment processing on a string-by-string basis, in most cases You can get the correct answer.

According to another embodiment of the invention, the character height is a ratio of the character width (
Used to determine full-width and half-width (aspect ratio). This aspect ratio is larger for half-width characters than for full-width characters, so
Aspect ratio thresholds for determining this difference are prepared for each character and stored in the full-width/half-width discrimination table memory 19.

In the full-width/half-width determination process for each character, the aspect ratio of the character height obtained as cutting information is calculated from the character width, and by comparing that value with the corresponding threshold read from the full-width/half-width discrimination table memory 19, Full-width and half-width characters are determined.

[Effect of the invention] As explained above, according to the invention described in claim (1) or its dependent claims (3) or (4), half-width characters such as alphanumeric characters appearing in Japanese text can be translated with high precision. It is possible to avoid misjudgment of separated kanji, fatal recognition errors caused by misjudging full-width characters as half-width characters, and misjudgment due to the influence of character arrangement. According to the invention described in item (3) or (4), it is possible to correct the error in determining full-width or half-width characters of narrow characters such as alphanumeric characters "bi'".

[Brief explanation of drawings]

Figure 1 is a block diagram of a character recognition device according to an embodiment of the present invention, Figure 2 is a flowchart of the entire process, Figure 3 is a flowchart of full-width/half-width determination processing for each character, and Figure 4 is for each character string. Flowchart of full-width/half-width determination processing,
FIG. 5 is an explanatory diagram of the contents of the cutout information memory, FIG. 6 is an explanatory diagram of the contents of the recognition result memory, and FIG. 7 is an explanatory diagram of the contents of the full-width/half-width discrimination table memory. 1 Image input section, 11... Image memory, 12... Line/character cutting section, 13... Character image memory,
14... Cutting information memory, 15... Character knowledge department,
16. Character dictionary memory, 17. Recognition result memory,
18.Full-width/half-width determination unit, 19.Full-width/half-width discrimination table memory, 20. Result output unit. Figure 3

Claims (4)

[Claims] (1) In a character recognition device for Japanese sentences,
Character recognition is performed on the character image cut out from the input image without distinguishing between full-width and half-width characters, and for the characters of a specific character type obtained as a recognition result, the characters obtained when the character image is cut out. A full-width/half-width determination method characterized by determining whether the character is full-width or half-width using a value related to size and a value for determining full-width/half-width prepared in advance for each character of a specific character type. (2) In a character recognition device that targets Japanese texts, character recognition is performed on character images cut out from input images without distinguishing between full-width and half-width characters, and the specific character type obtained as a recognition result is After determining whether a character is full-width or half-width using the value related to the character size obtained when cutting out the character image and the full-width/half-width discrimination value prepared in advance for each character of a specific character type, recognition is performed. Based on the ratio of the number of characters determined to be full-width and the number of characters determined to be half-width in the resulting character string of the specific character type, final determination is made as to whether it is full-width or half-width, and all characters in the string are determined according to the result. A method for determining full-width or full-width angles, which is characterized by unifying them into full-width or half-width characters. (3) The value for determining full-width/half-width is the threshold of the ratio of the character width to the standard width of full-width characters, and the other determination of full-width or half-width for a specific character type is based on the ratio of the character width to the standard width of full-width characters, and The full-width/half-width determination method according to claim 1 or 2, characterized in that the determination is performed by comparing with a full-width/half-width determination value. (4) The value for full-width/half-width discrimination is the threshold value of character height and character width, and another determination of full-width or half-width for a specific character type is based on the ratio of the character height and width, and the corresponding full-width/half-width discrimination. Claims characterized in that the determination is made by comparison with the practical value (
The full-width/half-width determination method described in 1) or (2).