JPH06208659A - Handwritten sentence recognizing device - Google Patents

Abstract

(57) [Summary] [Object] The present invention makes it possible to execute a series of processing including Japanese processing in a short time without burdening the writer. [Structure] The Japanese processing unit is composed of a character string composed of a combination of a 2-character long Kanji compound word dictionary 13 in which 2-character long Kanji compound words are registered and a recognition candidate character corresponding to each character obtained by the character recognition unit. Detecting a 2-character long Kanji word by referring to a Kanji compound word registered in the 2-character long Kanji compound dictionary 2
A noun is formed from the character length Kanji compound detection unit 12, the prefix / suffix detection unit 14 that detects the prefix and suffix for the detected 2-character Kanji word from the recognition candidate characters, and other recognition candidate characters. It is composed of a noun detection unit 16 that detects a character corresponding to a character string, and another part-of-speech detection unit 18 that detects a detected recognition candidate character and a character according to Japanese grammar from other recognition candidate characters.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a handwritten sentence recognition device for recognizing characters in a stroke sequence written and input via a coordinate input device integrated with a display.

[0002]

2. Description of the Related Art In recent years, an online handwritten text recognition device, which recognizes a stroke sequence written and input through a coordinate input device such as a tablet and is used for document creation, has been receiving attention. In the conventional online handwritten text recognition device, the following procedure is required to recognize the text input by the writer as a correct Japanese text. (1) A stroke string (coordinate data string) of a sentence written by the writer on the coordinate detection surface of the coordinate input device is input. (2) By the character cutout processing, the input stroke sequence is decomposed for each character. (3) By the character recognition processing, character recognition is performed on the decomposed stroke string for each character. (4) A correct character is selected from a plurality of candidate characters for each character obtained by the character recognition processing by the Japanese processing.

In such a procedure, if the correct answer character can always be output in the character recognition process of (3), the procedure of (4) is not necessary, but an erroneously recognized character is output in the character recognition process ( The first candidate character is incorrect).

Further, in the Japanese processing of (4), since it is necessary to collate with many independent word dictionaries (for example, about 100,000 words), much time is required. For this reason, if you execute Japanese processing as a series of processes, the reaction will be slow, so it is not suitable for online processing.Usually, when writing input is completed, the writer will start Japanese processing by button operation. Method is used. When Japanese processing is not performed, it is necessary to confirm whether each character recognition result is correct or incorrect, and correct any incorrect characters.

This is because, in the independent word dictionary of the conventional device, most of the kanji words are words of two-character length and combinations of two-character length words, but these words are registered in duplicate. (For example, "character" and "character recognition".
Both are registered in the independent word dictionary), so the number of words registered in the dictionary is very large. When the number of words registered in the dictionary is large, the processing load for searching for an independent word in Japanese processing is heavy, and the performance of the system is deteriorated.

[0006]

As described above, in the conventional online handwritten text recognition device, the writer must always look at the character recognition result when only the character recognition processing is performed without performing Japanese processing. Moreover, when the recognition result is incorrect, correction work is required, which hinders the writer's thinking in writing the sentence, and puts a heavy burden on the writer. Further, even when performing Japanese processing, the processing time becomes long due to the large number of registered words in the independent word dictionary, and if a series of processing up to Japanese processing is performed, online processing cannot be performed.

The present invention has been made in view of the above points, and provides an online handwritten text recognition device capable of executing a series of processes including Japanese processing in a short time without burdening the writer. The purpose is to provide.

[0008]

In order to solve the above-mentioned problems, the present invention provides a character cutting means for cutting out a stroke string for each character from a stroke string representing a character string written and input through a coordinate input device. A combination of character recognition means for performing character recognition on the stroke sequence output from the character cutout means and obtaining recognition candidate characters for each character, and recognition candidate characters corresponding to each character obtained by the character recognition means. And a Japanese processing means for detecting a 2-character long Kanji word and obtaining a correct character from a recognition candidate character corresponding to another character string portion based on the detected 2-character long Kanji word. To do.

According to the present invention, the Japanese language processing means is
A 2-character long Kanji compound word dictionary in which character length Kanji compound words are registered and a character string consisting of a combination of recognition candidate characters corresponding to each character obtained by the character recognition means are registered in the 2-character length Kanji compound word dictionary. A 2-character long Kanji compound word detecting means for detecting a 2-character long Kanji compound word by referring to the generated Kanji compound word; and a prefix and a suffix for the 2-character long Kanji word detected by the 2-character long Kanji compound word detecting means. Prefix / suffix detection means for detecting from the recognition candidate characters,
A character corresponding to a character string forming a noun is detected from the recognition candidate characters other than the character string portion in which the corresponding recognition candidate character is detected by the 2-character long Kanji compound word detecting unit and the prefix / suffix detecting unit. Noun detecting means for
Detected recognition candidate characters and characters according to Japanese grammar from the recognition candidate characters other than the character string part detected by the character length kanji compound detection unit, the prefix / suffix detection unit, and the noun detection unit. Another part-of-speech detecting means is provided.

[0010]

According to this structure, the collation is performed based on the 2-character long Kanji compound word dictionary consisting of only the 2-character length Kanji compound words that account for the majority of the Kanji words. Since only 2-character long Kanji compound words are registered, there is no content that is registered redundantly, so the amount of dictionary is reduced, and therefore the matching time is shortened. In addition, since the correct character is detected from the recognition candidate characters corresponding to other character string parts based on the 2-character long Kanji compound word detected by the collation based on the 2-character long Kanji compound dictionary, the dictionary amount It is possible to avoid an increase in false recognition due to the decrease in

[0011]

DETAILED 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 showing the configuration of an online handwritten text recognition device according to this embodiment. Figure 1
As shown in FIG. 1, the online handwritten text recognition device includes a display-integrated coordinate input device 1, a character input unit 2, a character cutout unit 3, a character recognition unit 4, a reference pattern storage unit 5, a Japanese language processing unit 6, and a display control. It is constituted by the section 7.

The display-integrated type coordinate input device 1 has both an input function 1a for writing a character string as a series of coordinate values and an output function 1b for displaying a writing / recognition result.
The display-integrated coordinate input device 1 has a coordinate input surface and a display surface that are superposed on each other and integrated so that display and coordinate input are possible on the same surface. The character input unit 2 controls writing input and inputs a character string as a series of coordinate values (stroke string). The character cutout unit 3 cuts out a stroke string for each character from the stroke string representing the character string input by the character input unit 2.

The character recognition unit 4 collates the stroke sequence for one character cut out by the character cutout unit 3 with the reference pattern stored in the reference pattern storage unit 5 to perform character recognition, and selects a plurality of candidate characters. Ask. The reference pattern storage unit 5 stores a reference pattern relating to a character to be recognized for character recognition. The Japanese processing unit 6 selects an appropriate character as a Japanese sentence from a plurality of candidate characters for each character obtained by the character recognition unit 4. The display control unit 7 displays the character selected by the Japanese language processing unit 6 in the output function 1b of the display-integrated coordinate input device 1.

FIG. 2 is a block diagram showing a detailed configuration of the Japanese language processing section 6. As shown in FIG. 2, the Japanese processing unit 6 includes a punctuation mark detection unit 11, a two-character long Kanji compound word detection unit 12,
2-character long Kanji compound dictionary 13, prefix / suffix detection unit 1
4, a suffix dictionary 15, a noun detection unit 16, a noun dictionary 17, and another part-of-speech detection unit 18. The punctuation mark detection unit 11 detects a punctuation mark indicating a delimiter of a character string as a processing unit, and starts Japanese processing.

The 2-character long Kanji compound word detection unit 12 stores two characters stored in the 2-character long Kanji compound word dictionary 13 for a character string composed of a combination of candidate characters of each character up to the punctuation mark detected by the punctuation mark detection unit 11. A long-character kanji compound word is detected by referring to the long-character kanji compound word. In the 2-character long Kanji idiom dictionary 13, about 10,000 words such as “character”, “recognition”, and “device” are registered.

The prefix / suffix detection unit 14 is stored in the suffix dictionary 15 from a plurality of candidate characters corresponding to the characters before and after the 2-character long Kanji phrase detected by the 2-character long Kanji phrase detection unit 12. Refer to the prefixes and suffixes
Detect the suffix.

The noun detecting section 16 is stored in the noun dictionary 17 from a candidate character group corresponding to characters other than the character string portion detected by the 2-character long Kanji compound word detecting section 12 and the prefix / suffix detecting section 14. A character string forming a noun having a length other than two characters is detected by referring to the noun.

The other part-of-speech detection unit 18 selects from the candidate character group corresponding to characters other than the character string portion detected by the 2-character long Kanji compound detection unit 12, the prefix / suffix detection unit 14, and the noun detection unit 16. , Detects characters in other grammatically correct parts of speech other than nouns. Next, the operation of this embodiment will be described using a specific example.

First, the output function 1b of the display-integrated coordinate input device 1 displays an underline 31 indicating the writing position so that the writer can easily write a character, as shown in FIG. 3, for example. Also, the character input unit 2 so that writing is always possible
Are operating in parallel with other parts.

During this standby, the display-integrated tablet 10
When a character is written with a pen on the input surface of, the stroke sequence (coordinate data sequence) forming the character is sequentially input. Here, it is assumed that the writer has handwritten the character string "Write a character string in the online sentence recognition device" as shown in FIG. 4, for example. The character string input by handwriting is sent to the character slicing unit 3 and the character recognizing unit 4.
A plurality of recognition candidates are obtained for each character.

As a method of separating each character from a character string written in one line in the character cutout unit 3, for example, "online character recognition of a frameless written character string by the candidate character lattice method" (IEICE paper) Magazine Vol.J68-D No.
4, p765 / 772, 1985) can be used.

As a method of character recognition processing in the character recognition unit 4, for example, the technique described in Japanese Patent Application No. 3-21157 "Online character recognition device" can be used.

The character recognition unit 4 outputs the recognition candidate characters corresponding to each character obtained by the recognition process for each character to the Japanese processing unit 6. FIG. 5 shows an example of the candidate characters output from the character recognition unit 4 when the handwriting is input as shown in FIG.

In FIG. 5, the left vertical writing character string is a character string based on the first recognition candidate for each character. The horizontal direction of each character in the vertically written character string indicates the recognition candidate characters whose recognition candidate ranks for each character are from 1st to 15th.

The punctuation mark detection unit 11 of the Japanese language processing unit 6 responds when the last "." Of the character string is written, and starts Japanese language processing. First, the 2-character long Kanji compound word detection unit 12
Performs matching between the 2-character long Kanji compound words stored in the 2-character long Kanji compound dictionary 13 and a character string formed by combining two consecutive Kanji characters in the recognition candidate characters shown in FIG.
Detects a character string that is semantically correct, which is the same as the character length Kanji compound word.

The character string (two-character long Kanji compound word) detected here has eight words as shown in FIG. For each word, the word recognition rate based on the character recognition rate in the character recognition of each character is obtained. For example, the character recognition rate for each character is converted so that the first-ranked candidate character is 100%, and the average of the character recognition rates after conversion of the respective characters forming the two-character long Kanji compound word is obtained. This allows
The word "sentence" consists of two letters, "sentence", which is ranked first in the candidate ranking, and "chapter", which is ranked second in the candidate ranking. The word recognition rate is 99%.
Has become.

The 2-character long Chinese character idiom detection unit 12 selects a preset threshold value α for the word recognition rate from the thus obtained 2-character long Chinese character idioms (for example, α =
Words that are less than 90%) have a low recognition rate with respect to the first word and are determined to be words due to a combination of erroneously recognized characters, and are deleted. As a result, the words "sentence" and "writing responsibility"
Only the five words "recognition", "device" and "character" remain. Further, since the words “sentence” and “sentence” are words formed by combining recognition candidate characters of the same character, only one word “sentence” having a higher word recognition rate is selected. Similarly, for the words "character" and "literature", only the word "character" is selected.

Next, the prefix / suffix detection unit 14 is activated. The prefix / suffix detection unit 14 refers to the suffix dictionary 15 to determine the prefix / suffix for the word obtained by the 2-character long Kanji idiom detection unit 12 from the recognition candidates of the characters before and after the word. To detect. By this matching, the word "character"
The "string" suffix for is detected.

Further, the noun detection unit 16 is activated. The noun detecting unit 16 compares the character strings other than the character string portions detected by the 2-character long Kanji compound word detecting unit 12 and the prefix / suffix detecting unit 14 with the noun dictionary 17 to have a length other than 2 character length. Detect the word. By this matching, the word "online" is detected.

In this way, the noun part is detected from the recognition candidate character string. Next, the other part-of-speech detection unit 18 is activated.
The other part-of-speech detection unit 18 detects the correct character of the other part-of-speech part excluding the noun from the recognition candidate character string. In the other part-of-speech detection unit 18, for example, “validation of vocabulary and syntax in recognition of handwritten manuscript” (Information Processing Society of Japan, Vol. 26 No.
5, p862 / 869, 1985) can be used.

The correct character string result ("Write a character string in the online sentence recognition device") thus obtained is sent to the display controller 7 and displayed by the output function 1b of the display-integrated coordinate input device 1. It

In this way, the Japanese language processing unit 6 uses the 2-character long Kanji dictionary 13 in which only the 2-character long Kanji idioms, which account for the majority of the Kanji words in the independent words, are registered. A 2-character long Kanji compound word is detected, and based on the detected 2-character Kanji word, collation is performed for a character string portion other than the 2-character Kanji word. That is,
The 2-character long Kanji dictionary 13 has a small number of registered words because only 2-character long Kanji idioms are registered, and therefore the processing time is shortened. For this reason, you can enter sentences online as a series of processes up to Japanese processing, so operations such as activating Japanese processing and confirming character recognition results and correcting erroneously recognized characters are not required. It is possible to create an efficient sentence without hindering the person's thinking of writing the sentence.

The present invention is not limited to the above embodiment. For example, other processing methods may be used for the character cutting unit 3, the character recognition unit 4, the other part-of-speech detection unit 18, and the like.

[0034]

As described above, according to the present invention, it is possible to execute online on a text recognition device by shortening the Japanese processing time, and the writer can see the result of character recognition while writing. It is possible to eliminate operations that have nothing to do with writing originally, such as being unable to start Japanese processing without continuing writing or performing extra button operations.

As a result, it becomes possible to efficiently input a sentence that is correct in Japanese, and to reduce the burden on the writer while inputting the sentence, which is a great practical effect.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an online handwritten text recognition device according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a detailed configuration of a Japanese language processing unit 6.

FIG. 3 is a diagram showing an example of a display screen for text input on the display-integrated coordinate input device 1.

FIG. 4 is a diagram showing an example of handwriting input on a display screen for text input.

FIG. 5 is a diagram showing an example of candidate characters output from the character recognition unit 4.

FIG. 6 is a diagram showing an example of a matching result by a 2-character long Kanji compound word detecting unit 12;

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Display-integrated coordinate input device, 2 ... Character input unit, 3 ... Character cutout unit, 4 ... Character recognition unit, 5 ... Character recognition reference pattern, 6 ... Japanese processing unit, 7 ... Display control unit, 11 ... Punctuation detection unit, 12 ... 2 Character length Kanji compound word detection unit, 13 ... 2
Character length Kanji compound dictionary, 14 ... Prefix / suffix detection unit, 1
5 ... letter dictionary, 16 ... noun detector, 17 ... noun dictionary, 1
8 ... Other part-of-speech detection unit, 30 ... Display input surface, 31 ... Underwriting character writing line, 51 ... Recognition candidate rank, 52 ... Recognition candidate character.

Claims (2)

[Claims] 1. A character slicing means for slicing a stroke string for each character from a stroke string representing a character string handwritten by a coordinate input device, and character recognition for a stroke string output from the character slicing means. , A 2-character long Kanji word consisting of a combination of character recognition means for obtaining a recognition candidate character for each character and a recognition candidate character corresponding to each character obtained by the character recognition means, and the detected 2-character long kanji character A handwritten sentence recognition device comprising: a Japanese processing means for obtaining a correct character from a recognition candidate character corresponding to another character string portion based on a word. 2. The Japanese language processing means comprises a combination of a 2-character long kanji compound word dictionary in which a 2-character long kanji compound word is registered and a recognition candidate character corresponding to each character obtained by the character recognition means. A two-character long Kanji compound word detection unit for detecting a two-character long Kanji compound word from a character string by referring to a Kanji compound word registered in the two-character long Kanji compound word detection unit, and detected by the two-character length Kanji compound word detection unit. It corresponds to the prefix / suffix detecting means for detecting the prefix and suffix for the 2-character long kanji word from the recognition candidate character, the 2-character long kanji compound word detecting means, and the prefix / suffix detecting means. A noun detecting means for detecting a character corresponding to a character string forming a noun from the recognition candidate characters other than the character string portion in which the recognition candidate character is detected; the two-character long kanji compound word detecting means; And a part-of-speech detecting means for detecting a detected recognition candidate character and a character according to Japanese grammar from the recognition candidate characters other than the character string portion detected by the acronym / suffix detecting means and the noun detecting means. The handwritten sentence recognition device according to claim 1, characterized in that.