JP2963474B2 - Similar character identification method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION The present invention relates to a character recognition device (for example, Kanji OCR) for processing Japanese sentences, and particularly to a method for identifying similar characters which is difficult to identify by collating with a dictionary. About.

[Conventional technology]

In general, in a character recognition device, a character image cut out from a document image is compared with a dictionary after being compressed or decompressed (normalized) to a fixed size. There was a problem that it was not possible to distinguish “.”, Lowercase alphabetic characters “o”, uppercase alphabetic characters “O”, etc.

Regarding the identification of such similar characters, a method of focusing on the character on the left (Japanese Patent Publication No. 60-1676) and a method of using the position information of the frame for the characters entered in the form (Japanese Patent Publication No. 60-9314)
No.), a method using the absolute size of the character
No. 7584) has been proposed.

[Problems to be solved by the invention]

However, the method disclosed in Japanese Patent Publication No. 60-1676 uses local information, and cannot respond if, for example, two or more lowercase letters continue. The method of Japanese Patent Publication No. 60-9314 cannot handle general frameless sentences. The method disclosed in Japanese Patent Application Laid-Open No. 60-237584 cannot deal with a change in the character size for each line.

It is an object of the present invention to provide similar characters having different character sizes (for example, “tsu” and “tsu”, “yo” and “cho”) by simple processing without the restriction of the target sentence as described above, or The present invention provides a method for accurately identifying similar characters (for example, “゜” and “.”, “′” And “,”, etc.) having the same character shape and size but different positions in a line. .

[Means for solving the problem]

A similar character identification method according to the present invention is a character identification device for processing Japanese sentences, wherein in the case of a horizontal writing sentence, the ratio of the distance between the upper reference line of the line and the upper end of the character to the standard width of the line (upper blank) Ratio), and in the case of vertical writing, the ratio between the space between the left reference line of the line and the left edge of the character and the standard width of the line (left space ratio) is detected. Similar characters are identified by comparing the left blank ratio with a standard top blank ratio or left blank ratio prepared for a specific character in advance.

(Operation)

For example, if the upper space ratio or left space ratio is examined for characters that have almost the same shape, such as "tsu" and "tsu", "yo" and "cho", or "." There are considerable differences between letters.

For example, "゜" and ".", "'" And ",",
The same applies to similar characters such as "." And ".", Which have the same shape and size but differ only in position.

Therefore, a standard top space ratio or left space ratio is prepared for specific similar characters that are likely to be misidentified by matching with the dictionary, and for characters whose similar characters are candidates for comparison with the dictionary, the character Are compared with the standard top blank ratio or left blank ratio of each candidate, and the one with the smallest difference is taken as the first blank space ratio or left blank ratio.
By making the candidate a candidate, similar characters that are difficult to identify by collation with the dictionary can be identified and corrected with high accuracy to increase the recognition rate.

〔Example〕

FIG. 1 is a character recognition device (Kanji OCR) according to the present invention.
It is a block diagram of. In this character recognition apparatus, a document (horizontal writing) is input by an image input unit (scanner) 10.
Is read out, the image data is stored in the image memory 11, the line and character cutout unit 12 performs line cutout and character cutout from the document image, and stores the character image data in the character image memory 13 to be cut out. The character recognition unit 14 compares the normalized characteristics of the image data of each character with the data of the dictionary for each character registered in the character dictionary memory 15, and finds a candidate having a short distance (a high degree of coincidence). The tenth highest is selected and stored in the recognition result memory 16.
Up to this point, it is the same as the conventional character recognition device.

The difference between this character recognition device and the conventional character recognition device is that
The line / character extracting unit 12 determines the upper reference line, the lower reference line, and the standard width of the line at the time of the extraction processing, detects the upper blank ratio for each character, and stores these information in the extraction information memory 17. And having a similar character determination unit 18 and a similar character table memory 19, correcting the recognition result of the character recognition unit 14 by the similar character determination unit 18, and outputting the result from the output unit 20.

In FIG. 2, for example, the character string "Wait a moment." Is described, where Lu is the upper reference line of the line, Ld is the lower reference line of the line, and W is the standard line (height) of the line. Specifically, the upper side and the lower side of the circumscribed rectangle of the line character string are detected as the upper reference line Lu and the lower reference line Ld, and the interval between them is defined as the line reference width W. Alternatively, in order to cope with the skew, straight lines contacting the upper end and the lower end of each character of the line character string are defined as an upper reference line Lu and a lower reference line Ld, respectively.
The average interval may be set as the standard width W of the row.

From these values, for each character, Ask for. Here, as shown in FIG. 2, the upper blank is a space Du between the upper end of the character (the upper side of the character circumscribed rectangle) and the upper reference line Lu.
It is.

Now, it is assumed that the character recognition unit 14 obtains the recognition results shown in Table 1 for the character string "Wait a moment", and obtains the top blank ratio shown in Table 2 for each character. The processing of the similar character determination unit 18 in this case will be described below.

The similar character table memory 19 stores a table of similar characters that need to be processed by the similar character determination unit 18 and their standard upper blank ratios. An example of the similar character table is shown in Table 3.

The similar character determination unit 18 sequentially sorts the recognition result candidates (Table 1) obtained in the recognition result memory 16 into a similar character table (third table).
Table), and searches for a character in which a similar character registered in the similar character table appears as a candidate. In this example, the second character that has "yo" and "cho" as candidates is found first.

The similar character determination unit 18 reads the upper blank ratio detected for the second character from the cut-out information memory 17 and reads the standard upper blank ratios of the candidate “yo” and “cho” from the similar character table memory 19. Read and compare. The upper blank ratio of the second character is 0.4 according to Table 2, which coincides with the standard upper blank ratio of "yo" of 0.4, and is significantly different from the standard upper blank ratio of "yo" of 0.1. Therefore, the similar character determination unit 18 replaces the first candidate of the second character with “ょ” and the second candidate with “よ”.

The third character is found as the next character to be processed. The top blank ratio is 0.5, which is more than the standard top blank ratio of 0.2.
Since the blank ratio is close to 0.6 by default, the candidates are not replaced.

On the other hand, the sixth character is a similar character “tsu” and the first candidate “tsu” is the second candidate.
Swap the candidates.

By repeating the same processing, the recognition results shown in Table 1 are corrected as shown in Table 4.

Up to this point, a case has been described in which horizontally written sentences are to be processed. Next, a case in which a vertically written document is processed will be described.

In the case of processing vertical writing text, the line / character cutout unit 12
As illustrated in FIG. 3, the left reference line Ll of the row is used instead of the upper reference line of the row, and the right reference line Lr of the row is used instead of the lower reference line of the row.
Is determined from the interval between the reference lines, that is, the standard width W of the line, and the interval between the left end of the character (the left side of the circumscribed rectangle of the character) and the left reference line Ll of the line, that is, the left blank Dl. Ask for.

The similar character table memory 19 stores a table of standard left blank ratios of similar characters obtained in advance for a vertically written sentence.

Then, the similar character determination unit 18 refers to the similar character table for the vertically written text, and compares the left blank ratio with the standard left blank ratio for the similar characters in the recognition result, thereby making the correction as described above. I do.

〔The invention's effect〕

As described above, according to the present invention, similar characters that are difficult to identify in matching with a dictionary can be identified with relatively high accuracy by relatively simple processing, and their erroneous recognition can be corrected. The character recognition rate can be improved. Further, as is clear from the above description, the similar character identification method of the present invention can be applied to a general frameless text, a text in which the character size changes for each line, and a character string in which lowercase letters are continuous. Therefore, there is no limitation of the application target as in the related art.

[Brief description of the drawings]

FIG. 1 is a schematic block diagram of a character recognition device according to the present invention,
FIG. 2 is a diagram for explaining a top blank ratio of a horizontal writing sentence, and FIG.
The figure is a diagram for explaining the left blank ratio of a vertically written sentence. 10: Image input unit, 11: Image memory, 12: Line / character cutout unit, 13: Character image memory, 14: Character recognition unit, 15: Character dictionary memory, 16: Recognition result memory ,
17 ... cutout information memory, 18 ... similar character determination unit, 19
…… Similar character table memory, 20 …… Output part.

──────────────────────────────────────────────────続 き Continuation of front page (56) References JP-A-62-117090 (JP, A) JP-A-62-187988 (JP, A) JP-A-62-169285 (JP, A) 108691 (JP, A) JP-A-63-216189 (JP, A) JP-A-1-171080 (JP, A) JP-A-2-85985 (JP, A) JP-A-60-134394 (JP, A) JP-A-59-167783 (JP, A) JP-A-59-109979 (JP, A) JP-A-59-231681 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) G06K 9/62 G06K 9/46 G06K 9/68

Claims (1)

(57) [Claims] 1. A similar character identification method in a character recognition device for processing a Japanese sentence, wherein in the case of a horizontal writing sentence, a distance between an upper reference line of a line and an upper end of the character and a standard width of the line are determined. Ratio (hereinafter referred to as the top blank ratio), and in the case of vertical writing, the ratio between the space between the left reference line of the line and the left edge of the character and the standard width of the line (hereinafter the left blank ratio and A similar character is identified by comparing the detected top blank ratio or left blank ratio with a standard top blank ratio or left blank ratio prepared for a specific character in advance. Identification method.