JPH0247788B2 - - Google Patents

Description

[Detailed Description of the Invention] (1) Technical Field of the Invention The present invention relates to a character recognition device that recognizes handwritten Japanese text, and in particular to a character recognition device that improves the recognition accuracy of handwritten Japanese text by adding furigana only to kanji parts. This is related to the processing method.

(2) Background of the technology In general, character recognition devices first observe patterns. Next, characteristics such as whether the shape of the character is large or small, whether there are many or few lines, etc. are found, that is, the characteristics of the pattern are extracted. Then, the pattern characteristics are compared with the characteristics obtained from past data stored in the device, and the most similar pattern is selected. That is, features are processed and subsequent decisions are made to identify a given pattern from the results. Although character recognition is performed in this manner, recent character recognition devices have been developed not only for handwritten kana characters but also for handwritten kanji characters.

(3) Prior art and problems There is a problem in the recognition of kanji that the accuracy cannot be improved because it is more difficult to extract the features compared to the conventional recognition of katakana. Furthermore, Japanese sentences also have the following problems.

That is, there are many characters that are almost the same, especially between katakana and kanji, and it is very difficult to identify and recognize them. For example, ``tech'' in ``air'' and ``construction'' and ``power'' in ``sumo wrestler'' and ``casa'' are indistinguishable. Conventionally, as a method for this purpose, a method is known in which the recognition device judges the context of the sentence and distinguishes between non-kanji and kanji, but this method is complex and has the disadvantage that it cannot always be distinguished. ing.

Another method is to add identification marks for non-kanji and kanji to the handwritten Japanese text itself, but adding such marks within the character frame is troublesome and impractical. It has the disadvantage that it is not.

(4) Purpose of the Invention The purpose of the present invention is to provide a recognition processing method for handwritten Japanese text that can eliminate the above-mentioned conventional drawbacks, improve the recognition accuracy of kanji, and easily distinguish between non-kanji and kanji. It is said that

(5) Structure of the Invention According to the present invention, the purpose is to read the handwritten Japanese text and the handwritten furigana characters added only to the kanji corresponding to the handwritten Japanese text separately, and to recognize and output the handwritten Japanese text. The handwritten Japanese character recognition method is characterized in that a kanji is recognized by comparing the recognition output of the handwritten furigana character with a kana-kanji corresponding output, and a part where a corresponding furigana does not exist is recognized as a non-kanji character. This is achieved by providing a recognition processing method.

(6) Embodiment of the Invention An embodiment of the present invention will be described in detail below with reference to the drawings. FIG. 1 shows a block diagram for realizing a handwritten Japanese sentence recognition processing method according to the present invention.

In FIG. 1, reference numeral 1 denotes a medium to be read, such as a form, on which a handwritten Japanese text is written.As shown in FIG. There are 12 handwritten furigana characters added only to kanji. The characters on medium 1 to be read (furigana 12 and text 11) are
The image is scanned and read by a photoelectric conversion device 2 such as a CCD, and is input to the recognition preprocessing circuit 3 as image information.

The recognition preprocessing circuit 3 extracts the features of each character, selects and outputs candidates for each character. In this case, since most of the furigana answers are correct, it is sufficient to select only the first candidate.

However, for text containing kanji, candidates for each character are output as is. Buffers 4 and 5 store furigana and text, respectively, and the candidate outputs from the recognition preprocessing circuit 3 are sorted depending on whether they are furigana or text. Reference numeral 6 denotes a reading controller which accesses the reading dictionary memory 7 in response to the furigana output from the furigana buffer 4 and sequentially collates the read kana-kanji corresponding output with the text candidate group output from the text buffer 5. The candidates for kanji in the main text are selected and narrowed down to 2 to 3 candidates from the 20 candidate categories for each character. If the readout controller 6 recognizes that the corresponding furigana does not exist as a result of comparing the furigana for each character in the main text with the kana-kanji corresponding output, the readout controller 6 converts the character into a non-kanji character such as katakana, hiragana, alphabet, or Arabic. Judge it as a number or other symbol.

The character candidates selected by the readout controller 6 in this manner (two to three candidates for each character) are stored in the candidate buffer 8. The character candidate output from the candidate buffer 8 is input to the selection circuit 9,
Here, each character candidate is arranged horizontally to form a sentence, and processing is performed to select one final correct category. The character output from the selection circuit 9 is output to an output circuit 10 such as a floppy disk or a computer.

Figure 3 shows the readout controller 6 in Figure 1.
FIG. FIG. 3a shows an example of the main text 11 and its furigana 12, each of which is numbered consecutively in correspondence with the character frame. Figures 3b to 3e show how the kanji candidates of text 11 corresponding to the furigana part 11 in Figure 3a are recognized and determined. This indicates whether or not there is a match among the candidate categories from the kana-kanji corresponding output corresponding to the reading and the text buffer 6 (FIG. 1). In the case of the same figure, the kanji candidate corresponding to the furigana part ~ is determined to be "our company." Similarly, the kanji character corresponding to the furigana part ~ is determined to be "construction contractor", and therefore, the text part ~ "is air conditioner" is a non-kanji character, and in this case it is recognized to be katakana.

(7) Effects of the Invention As explained above, according to the present invention, furigana, which is easier to recognize than kanji, is used to distinguish between kanji and non-kanji with the same shape, thereby improving the recognition accuracy of kanji. This has the effect that non-kanji characters and kanji characters can be easily distinguished and recognized.

[Brief explanation of drawings]

FIG. 1 is a block diagram for realizing the recognition processing method of the present invention, FIG. 2 is an example of characters written on the read medium 1 in FIG. 1, and FIG. 3 is an example of the read controller 6 in FIG. FIG. 3 is a diagram for explaining the operation. In the diagram, 1 is Japanese sentence 11 and furigana 12.
2 is a photoelectric conversion device; 3 is a medium to be read on which is written;
4 is a reading buffer, 5 is a text buffer, 6 is a reading controller, and 7 is a reading dictionary memory.

Claims (1)

[Claims] 1. Distinguish and read the handwritten Japanese text and the handwritten furigana characters added to each word only for the kanji corresponding to the handwritten Japanese text, and shift the starting position of the kana obtained from the handwritten furigana characters for each character. By obtaining output corresponding to each kana-kanji and comparing the output corresponding to each kana-kanji with the recognition output of the above-mentioned handwritten Japanese sentence, the kanji is recognized, and parts where the corresponding furigana does not exist are recognized as non-kanji. A recognition processing method for handwritten Japanese sentences.