JPS61121159A - Character input device - Google Patents

Abstract

PURPOSE:To simplify the constitution of an input device for Japanese sentences by attaining the effective combination between the character pattern described on a form and the reading information on the character strings of the character pattern for processing and therefore ensuring the solid-written KANA (Japanese syllabary)/ KANJI (Chinese character) conversion with high performance. CONSTITUTION:An image input means 1 of a character input device reads optically the information given from an original and converts it to electric signals. A character pattern segmenting means 2 segments the characters of electric signals for each character. These segmented characters are compared with a character pattern of a standard character pattern memory means 4 by a character pattern comparing means 3. The coincidence signal given from the means 4 is applied to a control means 5 which performs the overall control. The means 5 detects the coincidence between said coincidence signal and the read code given from a read input means. The means 3 is controlled by the character code string given from the means 5. This character code string is used for control of a KANA character converting means 7 which supplies the character codes given from a Japanese dictionary memory means 8. Then the character code string given from the means 7 is applied to the means 5. Thus the solid-written KANA/KANJI conversion is possible with high performance.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention relates to a Japanese language input device used in many Japanese language information processing devices including Japanese word processors and personal computers.

Conventional technology The so-called OCR technology that recognizes kanji and hiragana that are printed or written on paper is known, for example, as described in ``Research on handwritten kanji recognition that has almost completed the stage of major classification'' (Nikkei Electronics, December 7, 1981). As described in 1996, achieving a high recognition rate either requires a long processing time or requires the use of a large amount of very expensive circuitry.

In particular, recognition rates for handwritten characters are still low and sufficient performance has not been achieved.

On the other hand, as a Japanese input method, a so-called kana-kanji conversion method is known, in which the pronunciation of a Japanese sentence is input from a keyboard or the like and the pronunciation information is converted into a sentence containing kanji and kana. However, this kana-kanji conversion generally requires an operator to specify the break between phrases, making the operation complicated. In order to solve this problem, research is being carried out in various fields on so-called kana-kanji conversion technology that does not require the input of phrase breaks. For example, Makino, Kizawa Nipeta written text kana-kanji conversion system and its homophone processing, Information Processing Academic Journal, Vol. 22. No, 1. pp, 59-67 (198
A typical method is described in 1). However, when trying to achieve a high conversion rate in the case of solid kana-kanji conversion,
After all, it requires a lot of memory to perform complex grammar processing, making it impractical in terms of processing time and 匍 1 case. As described above, with the conventional technology, it has not been possible to realize a Japanese input device that is easy to operate and has good performance at a low price.

The problem that the invention aims to solve The conventional technology has the following problems.

In the case of the above-mentioned solid-written kana-kanji conversion, in order to achieve a high conversion rate, it is necessary to use a large amount of memory and perform complex grammar processing, which is impractical in terms of processing time and cost. . Furthermore, if we were to realize solid kana-kanji conversion on a practical scale, there would be many errors when automatically dividing readings into phrases, and this, combined with errors in homophones, would result in extremely poor kana-kanji conversion performance. Become. On the other hand, when OCR technology is used to automatically read characters written on paper, achieving a high recognition rate requires a long processing time or requires the use of a large amount of extremely expensive processing circuitry. Ta. In particular, recognition rates for handwritten characters are still low and sufficient performance has not been achieved. As described above, the conventional technology has only been able to realize Japanese input with extremely limited performance.

It is an object of the present invention to eliminate the drawbacks of the first invention and to enable kana-kanji conversion to be performed with high probability without having to press keys to separate phrases.

Doubt.

In order to achieve the above-mentioned object, the present invention optically reads character patterns written on paper, collates them with information on a separately supplied character code string, and identifies discrepancies between the two characters. A method for checking whether there are any characters, a means for inputting the pronunciation of the character string you want to input, automatically determining the break between clauses from that pronunciation, and outputting a character code string of kanji or kana suitable for the obtained clause. , and a control means for exchanging information with each of the above two parts and finally outputting a Japanese character code such that the characters on the paper and the sentence given in pronunciation match without contradiction. It is something.

Effects of the present invention having the above configuration, the image input means optically reads the image from a sheet of paper on which characters and information are recorded, converts it into an electrical signal, and transmits the electrical signal to the character pattern cutting means. By cutting out each character recorded on the paper, you can calculate the area on the paper that each character occupies.

image information corresponding to each character is transmitted to the character pattern comparison means, information corresponding to the standard character shape is stored for each character in the standard character pattern storage means, and the image information corresponding to each character is transmitted to the character pattern comparison means. Detects the degree of character correspondence between the supplied character code string and the extracted character pattern,
A match signal indicating the degree of matching is output to the control means, the Japanese language dictionary storage means is connected to the kana-kanji conversion means, the character code is stored in association with the reading code representing the reading, and the reading supplied by the kana-kanji conversion means is stored. The code string is divided into a plurality of clauses, and the character code corresponding to the reading code obtained as a result of the division is retrieved from the Japanese language dictionary storage means and outputted to the control means. It is connected to an input means and a kana-kanji conversion means, and searches out and outputs a character code string that matches character pattern information and reading information without any contradiction.

Embodiment First, the operation of the embodiment of the present invention having the above configuration will be summarized in the first part.
This will be explained according to the diagram. FIG. 2 & is an example of the reading inputted from the reading input means. The interpretation of the Japanese text that corresponds to this reading is shown in Figure 2, and the two dishes shown in c are considered, but the interpretation is not determined. In this embodiment, the original shown in FIG. 1 or C is supplied as image information from the image input means. The kana-kanji conversion means first divides the reading in Figure 1 a into two phrases, 1 soikufuu, and outputs ``ingenuity.''

If Figure 1 is input as an image input, the above kanji string and the pattern of Figure 1 are combined using the character pattern ratio 1 method! 1.4j are combined to determine the correct kanji string. Furthermore, when the pattern shown in FIG. 1C is input as an image input, the character pattern comparison means combines the words "ingenuity" and "so go" (Qi), and returns the result that they do not match.

In this case, the next bunsetsu break is attempted and the kanji hiragana string ``Souyuki'' is obtained. This character string is shown in Figure 1C
The character string code is compared with the character pattern of , a match is confirmed, and a correct character string code is output from the control means. In this manner, the present invention has the effect of determining character strings that cannot be determined using the kana-kanji conversion method by referring to pattern information.

FIG. 1 is a diagram showing a specific embodiment of the present invention. The control means 5 controls the entire embodiment, and ultimately produces characters whose image information input from the image input means 1 and reading information input from the reading input means 6 match without contradiction. Output code string. The image input means 1 optically reads an image from a sheet of paper on which character information is recorded and converts it into an electrical signal, and the character information may be printed or handwritten. Also good. The character pattern cutting means 2 is connected to the image input means 1, and is supplied with image information converted into an electrical signal outputted from the image input means 1, and cuts out each character recorded on the paper one by one. By doing this, the area occupied by each character on the paper is determined, and image information corresponding to each character is output and supplied to the character pattern cutting means 2. The cutting may be performed using a writing frame provided on the paper in advance, or may be performed by referring to the graphical information of the characters. The standard character pattern storage means 4 stores standard graphic information of kanji and hiragana as average pattern information or as structural characteristic information of characters.
This is used as standard character pattern information when the character pattern comparison means 3 performs comparison processing. The character pattern comparison means 3 is connected to the character pattern extraction means 2, the standard character pattern storage means 4, and the control means 6, and determines to what extent the character code string supplied from the control means 5 matches the characters of the extracted character pattern. A matching signal indicating the degree of matching is outputted and supplied to the control means 5. This comparison operation of the character pattern comparison means 3 only needs to be performed on the characters indicated by the supplied character code string, so that conventional kanji recognition OCR can compare approximately 3000 characters (in the case of JIS 1st level kanji). The processing time is extremely fast compared to the previous one, and therefore it can be realized with a low-cost circuit. The reading input means 6 is for inputting the reading of characters corresponding to the character information recorded on the paper and transmitting it to the control means 6. The Japanese language dictionary storage means 8 stores character codes in association with reading codes representing reading methods, and may be an electronically constructed so-called Japanese language dictionary.

The 1-f1 word dictionary storage means 8 is referred to by the kana-kanji conversion means 7 when performing kana-kanji conversion. The kana-kanji conversion means 7 is connected to the Japanese language dictionary storage means 8 and the control means 5. The kana-kanji conversion means 7 divides the reading code string supplied from the control means 6 into a plurality of clauses, and searches the information in the Japanese language dictionary storage means 8 for character codes corresponding to the reading codes obtained as a result of the division. The character code string obtained as a result is supplied to the control means 5. The control means 5 is connected to the character pattern extraction means 2, the reading input means 6, and the kana-kanji conversion means 7. The control means 5 first transfers the reading code string supplied from the reading input means 6 to the kana-kanji converting means 7, and also sends a phrase separation command signal to the kana-kanji converting means 7, which instructs how to divide the reading code string into phrases. supply The bunsetsu break command signal supplied first may be one that commands a bunsetsu break that makes the length of 2-bunsetsu the longest, or it may command a bunsetsu break that makes the length of one bunsetsu the longest. It may be something that you do. Based on the command signal, the kana-kanji conversion means γ performs segmentation operations and searches for Japanese language dictionary information in the Japanese language dictionary storage means 8. The kana-kanji conversion means 7 supplies the resulting character code string to the control means 5. The control means S supplied with the character code string transfers the character code string to the character pattern comparison means 3. The character pattern comparison means 3 compares, based on the character code string, whether or not the character pattern information recorded on the sheet matches the character information represented by the character code string. The control means 5 transmits the coincidence signal obtained as a result.
supply to. Subsequently, the control means 5 receives the coincidence signal, and when the coincidence signal substantially indicates a coincidence, outputs the character code string outputted by the kana-kanji conversion means 7 as correct.

However, when the match signal does not substantially indicate a match,
The reading code string is again supplied to the kana-kanji converting means 7, and the bunsetsu break command signal is changed to a signal instructing a different bunsetsu break method from the previous time and transmitted. The clause break command signal in this case only needs to be different from the first clause break command signal, and may be one that commands a clause break that makes the length of the two clauses the longest, It may be possible to instruct the phrase break such that the length of one phrase is the longest. Alternatively, it may be something like simply instructing a shorter delimiter than the previous delimiter. As a result, the control means 5 receives from the kana-kanji conversion means 7 a clause break different from the previous one and a character code string corresponding thereto, and transfers this character code string to the character pattern comparison means 3 again. In this way, the control means 5 controls the kana-kanji conversion means 7 until the character pattern comparison means 3 outputs a match signal indicating a match with respect to the character code string outputted by the kana-kanji conversion means 7.
By repeating the commands to and the commands to the character pattern comparison means 3, a character code string is finally created in which the image of the character input from the image input means 1 and the reading input from the reading input means 6 match without contradiction. It operates to output .

Effects of the Invention As explained in detail above, the present invention solves the problems of conventional methods and effectively combines character pattern information written on paper with reading information of the character string. By processing them together, we are able to realize and provide a low-cost input device for Japanese text that converts solid kana-kanji with good performance.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a character input device according to an embodiment of the present invention, and FIG. 2 is a diagram showing an overview of the operation of the embodiment of the present invention. 1... Image input means, 2... Character pattern cutting means, 3... Character pattern comparison means, 4... Standard character pattern storage means, 5...
...Control means, 6...Reading input means, 7...
...Kana-kanji conversion means, 28...Japanese language dictionary storage means. Name of agent: Patent attorney Toshio Nakao (1st person)
figure

Claims (3)

[Claims] (1) A control means for controlling various signals, an image input means for optically reading an image from paper on which character information is recorded and converting it into an electrical signal, and an image input means connected to the image input means; The area occupied by each character on the paper is determined by cutting out each character recorded on the paper one by one by supplying the image information converted into an electrical signal output from the a character pattern cutting means for outputting corresponding image information; a standard character pattern storage means for storing information corresponding to a standard character shape for each character; the character pattern cutting means, the standard character pattern storage means, and the character pattern comparison means connected to the control means, detecting the degree of character correspondence between the character code string supplied from the control means and the cut-out character pattern, and supplying a coincidence signal indicating the degree of correspondence to the control means; , a reading input means for inputting the reading of the character corresponding to the character information recorded on the paper and transmitting it to the control means, and a Japanese language dictionary storing the character code in association with the reading code representing the reading. a storage means, connected to the Japanese language dictionary storage means and the control means, which divides the reading code string supplied from the control means into clauses and stores character codes corresponding to the reading codes obtained as a result of the separation in the Japanese language dictionary; Kana-Kanji conversion means searches from within the means and supplies the result to the control means, and the control means transfers the reading code string supplied from the reading input means to the Kana-Kanji conversion means, and A phrase separation command signal that instructs how to divide the code string into phrases is supplied to the kana-kanji conversion means, and the kana-kanji conversion means performs a phrase separation operation and a search in a Japanese dictionary based on the command signal. The obtained character code string is supplied to the control means, and then the control means supplies the character code string to the character pattern comparison means, and the character pattern comparison means compares the sheet with the character code string based on the character code string. The recorded character information and the character code string are compared to see if they match, and the resulting match signal is supplied to the control means, and then the control means receives the match signal and outputs the match signal. When the character code strings outputted by the kana-kanji conversion means are substantially matched, the character code string outputted by the kana-kanji conversion means is outputted as correct, and when the match signals do not substantially match, the reading code strings are converted into the kana characters again. By supplying the phrase to the kanji conversion means and changing the bunsetsu delimitation command signal to a signal instructing a bunsetsu delimitation method different from the previous one, the kana-kanji conversion means can output the bunsetsu delimitation different from the previous one and its corresponding character. A code string is received, the received signal is transferred to the character pattern comparison means again, and the received signal is transferred to the kana-kanji conversion means until the character pattern comparison means outputs a matching signal indicating a match for the character code string. By repeating the command and the command to the character pattern comparing means, the character code string in which the image of the character inputted from the image inputting means and the reading inputted from the reading inputting means match without contradiction is finally determined. A character input device characterized in that the character input device is configured such that a control means outputs the output. (2) The control means transmits a signal to the kana-kanji conversion means, as an initial bunsetsu break command signal, instructing a bunsetsu break method that makes the break of one bunsetsu the longest, and the character code string obtained as a result. If the character pattern comparison means indicates a mismatch, a phrase break that is shorter than the previous one is then issued as a bunsetsu break command signal, until the character pattern comparison means outputs a match signal indicating a match. 2. The character input device according to claim 1, wherein the character input device is configured to issue short phrase breaks one after another. (3) The control means transmits a signal to the kana-kanji conversion means as an initial bunsetsu break command signal instructing a bunsetsu break method that makes the break between two bunsetsu the longest, and the character code string obtained as a result. If the character pattern comparison means indicates a mismatch, then a signal is sent as a bunsetsu break command signal instructing a bunsetsu break method that makes the break of one bunsetsu the longest, and the resulting bunsetsu break command signal is transmitted. If the character pattern comparison means indicates a mismatch with respect to the character code string, a bunsetsu break that is shorter than the previous one is further commanded as a bunsetsu break command signal, and the character pattern comparison means outputs a match signal indicating a match. 2. The character input device according to claim 1, wherein the character input device is configured to issue commands for short phrase breaks one after another until the character input device reads the text.