JP2922949B2 - Post-processing method for character recognition - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a Japanese character recognition device, and more particularly to post-processing of a character recognition result by matching an input character with a dictionary.

[Conventional technology]

In general, when performing character recognition, a character image is normalized to a certain size and then matched with a dictionary.
Characters of the same shape but different sizes, for example, uppercase "A"
And lowercase “ぁ”, uppercase “ya” and lowercase “ya” cannot be distinguished.

To solve this problem, a method has been devised for identifying uppercase letters, lowercase letters, punctuation marks, and the like using the position and size of characters.

For example, as described in Japanese Patent Publication No. 60-1676, a method of determining a character based on the relative positional relationship with the character on the left side, as described in Japanese Patent Publication No. 60-9314,
A method of judging a character by a character position in a form entry frame, as described in JP-A-63-78287.
A method of determining a character based on a character arrangement, a position, and a size is known.

[Problems to be solved by the invention]

However, since the character size and the printing position are different depending on the font, it cannot be said that the problem has been solved when various documents are to be recognized.

SUMMARY OF THE INVENTION Accordingly, an object of the present invention is to provide a post-processing method that corrects an error in a character recognition result by matching with a character dictionary, particularly an error in uppercase and lowercase characters that is easily misunderstood by a simple process.

[Means for solving the problem]

The present invention provides a character recognition result by comparing a character arrangement rule describing a specific character and a character that can be prefixed or non-prefixable with a character recognition result by matching with a character dictionary. Is modified.

(Operation)

For example, kana lowercase letters “ゃ”, “ゅ”, “ょ”, etc. do not appear at the beginning of a sentence, but always appear before them with some capital letter. The capital letter does not mean anything; only limited characters are allowed. In other words, there is a rule as a character string.

Therefore, for specific kana lowercase letters, etc., in which the prefixed character is limited, the rules described above are prepared for the prefixable (or impossible) characters, and the rules are compared with the character recognition result, thereby violating the rule. Can be easily and reliably corrected.

〔Example〕

FIG. 1 is a schematic block diagram of a kanji OCR according to one embodiment of the present invention, and FIG. 2 is a schematic flowchart of a process.

In FIG. 1, image data of a document to be processed is input from an image input unit 10 and stored in an image memory 11 (processing step). Line cutout and character cutout are performed on the input image data by the line / character cutout unit 12,
The cut-out character image data is stored in the character image memory 13 (processing step). The character image data is subjected to preprocessing such as size normalization and feature extraction in a character recognition unit 14, and matching between the extracted feature vector and a character dictionary stored in a character dictionary memory 15 to obtain a distance. Character candidates from the smaller one to the N-th highest character are selected, and the result is stored in the recognition result memory 16 (processing step).

The post-processing 17 is a part for performing upper-case / lower-case correction processing on the character recognition result. The processing steps shown in FIG. 2 are the upper-case / lower-case correction processing.

The character arrangement rule used for the uppercase / lowercase correction processing is stored in the character arrangement rule memory 18.

This character arrangement rule describes characters that are permitted to come before a character (lower case of Hiragana and Katakana) whose preceding character is limited. An example is shown in Table 1.

Next, the uppercase / lowercase correction processing will be described by taking as an example a case where a character string “walking around” is recognized. In this case, it is assumed that the character recognition unit 14 has obtained the recognition results shown in Table 2. As a result of not considering the size of characters, "O"
And "ぉ", "yo" and "cho", "u" and "ぅ" are misrecognized.

First, the first candidates of the recognition result are sequentially compared with the character arrangement rule, and the process proceeds until a target character requiring correction processing appears. When the target character is found, the first candidate for the character immediately before that position is compared with the character arrangement rule to determine whether it is a prefixable character. As a result, if the preceding character does not exist (the target character is the head of the sentence) or if the preceding character is not a prefixable character, the second candidate at the target character position is replaced with the first candidate. If the second candidate is also in lower case,
Similarly, the previous character is compared with the rule for that character, and if it is a character that can be prefixed, it is replaced with the first candidate; otherwise, the same processing is performed for the next candidate.

In the case of this example, the first candidate “@” of the character number 1 is the target character. However, since this is the beginning of a sentence,
Is determined to be an abnormal character string, and is replaced with the second candidate “O”. Since this "@" is not a target character, the processing of character number 1 ends.

Since the first candidate "" of the next character number 2 is the target character, the immediately preceding first candidate character "" is compared with the contents of the character arrangement rule. However, since “O” is not in the rules, “O” is replaced with “Yo” as the second candidate. Since this “yo” is not a target character, the processing of character number 2 ends.

When the same process is repeated up to the character number 8, the recognition result corrects the erroneously recognized character as shown in Table 3.

Note that the character arrangement rule can be a description of a character that cannot be prefixed to the target character. However, in the case of Japanese kana, it is better to describe a character that can be prefixed as in the embodiment. Size can be reduced.

Misrecognitions other than capitalization can be corrected by creating similar rules.

In the post-processing unit 17, in addition to the upper / lower case correction process, a correction process based on validity as a word or a sentence may be performed.

The recognition result corrected by the post-processing unit 17 as described above is obtained in the recognition result memory 16, which is output from the output unit 19.
Is output by

〔The invention's effect〕

As described above, according to the present invention, it is possible to reliably correct an uppercase and a lowercase error in a character recognition result by matching with a character dictionary by a simple process.

[Brief description of the drawings]

FIG. 1 is a schematic block diagram of a kanji OCR according to an embodiment of the present invention, and FIG. 2 is a schematic flowchart of a process. 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 Post-processing unit 18 Character arrangement rule memory 19 Output unit

Claims (2)

(57) [Claims] 1. A post-processing method for character recognition for correcting a character recognition result obtained by matching an input character with a character dictionary, comprising the steps of: Prepare a character arrangement rule that describes a character that can be prefixed or non-prefixable to the character, compare the character string of the character recognition result with the character arrangement rule, and change the uppercase of the character to be corrected to lowercase or lowercase. Post-processing method for character recognition, wherein is corrected to uppercase. 2. A post-processing method for character recognition according to claim 1, wherein if no character exists before the character to be corrected and said character to be corrected cannot be the beginning of a sentence, the character is corrected. Post-processing method of character recognition characterized by the following.