JP3101153B2 - Japanese input device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a Japanese language input device, and more particularly, to a device capable of inputting Japanese language.
When inputting characters that you do not understand, use a dictionary that collects characters with similar shapes, first input characters that are similar in shape to that character, and then search for the target character using the dictionary. Japanese language input device. For example, a word processor,
The present invention is applied to all devices having a Japanese input method using a method of searching a dictionary, such as a personal computer, an office computer, a workstation, and an electronic organizer.

[0002]

2. Description of the Related Art Kana-Kanji conversion, which is currently mainstream,
You can get the desired character by inputting readings and converting the input kana, but there are methods of inputting characters that you do not understand, such as code input, radical input, stroke number input, etc. . As shown in FIG. 25, the code input is a code (JIS, JIS,
Shift JIS, Kuten codes, etc. are usually arranged in a list printed on paper (normally, the first level of kanji is arranged in the order of the Japanese syllabary of the reading of the phonetic alphabet, and the second level is arranged in the order of the number of strokes of the radical). ) To obtain the desired character by inputting the code and converting it. In some cases, the list is not printed on paper but is digitized.In this case, instead of entering a code, the target character is searched on the display in the same way as on paper, and a pointing device is used. The character may be specified using such as.

As shown in FIG. 26, a radical input is usually performed by picking up a radical to which a target character belongs from a list printed on paper and inputting a code written there. This is to get the desired character. In some cases, the list is not printed on paper but is digitized.In this case, instead of entering a code, the target character is searched on the display in the same way as on paper, and a pointing device is used. The character may be specified using such as.

[0004] The stroke number input collects characters having the same total stroke number, and visually searches for a target character from a table in which the kanji groups are arranged in the order of the stroke number. Find the code of the corresponding character from the list printed on paper and enter the code to obtain the desired character, or if the list is digitized and displayed on the display in the same way as on paper In some cases, a target character is searched for and the character is specified using a pointing device or the like.

As a known document describing a conventional Japanese input device, there is, for example, Japanese Patent Application Laid-Open No. Hei 2-18661. In this publication, candidate characters are shared by at least one of an on-sound input unit, a radical input unit, and a stroke number input unit, and a glyph component dictionary and a glyph component input unit. To represent,
That is, the number of candidate target kanji can be reduced. That is, information on readings, radicals, and the number of strokes are combined, a plurality of conditions are set, and target characters are narrowed down.

In Japanese Unexamined Patent Publication No. Hei 5-12249, when a kanji pattern displayed on a screen at the time of document creation is designated by, for example, a cursor, a radical analysis processing means performs a radical analysis based on a radical analysis table. The radical name of the designated kanji and the position of the radical are analyzed, and the retrieval output means searches the radical name of the designated kanji and the position of the radical based on the analysis result by the radical analysis processing means. All corresponding kanji codes are retrieved from the radical dictionary memory means. Thus, even if the radical name is not known, the target kanji can be obtained only by specifying the kanji having the same radical name. Further, even when only a part of the radical name constituting the target kanji is known, the target kanji can be obtained based on the radical name.

[0007]

In an information device such as a personal computer or an electronic organizer that can input Japanese, a method of inputting a reading in a kana and converting the input kana into a kanji or the like is usually used. However, that alone does not allow you to enter characters or symbols that you do not understand. In the case of inputting such characters, in the conventional method, a table in which kanji is arranged in the order of the total number of strokes is held in paper or electronic form, and the number of strokes is used as a clue to visually search the target character from the table. I was Or, a table in which kanji is arranged for each radical,
He held it on paper or in electronic form, and used the radicals as clues to search for the desired character visually. Normally, kanji groups having the same radical are arranged in the order of the number of strokes of the radical, and in the kanji group having the same radical, the characters are arranged in the order of the total number of strokes.

However, if the search method is based on the total number of strokes, the number of kanji having the same number of strokes increases to an enormous number, and it is very difficult to find one character from them, and the number of strokes cannot be clearly specified. Searching is even more difficult because of the presence of characters. Also, if you search from the radical,
The condition is that you need to know the radical of the target kanji. For some kanji, it is not clear what the radical will be (for example, the comet “com” has the radical “jun”), in which case it is very difficult to find the desired character. Also, for many radicals, the number of characters having the same radical becomes very large, and similarly, there is difficulty in searching.

In the case of code input, the list is usually
If it is the first level, it is arranged in the order of the Japanese syllabary of the reading of the phonetic alphabet, and since it is trying to input a character whose reading cannot be understood from the beginning, it can be understood that this is useless. In the case of the second level, the strokes are arranged in ascending order of the number of strokes of the radical, so that it takes time and effort to count the number of strokes of the target character, and it takes considerable time to search for a character whose radical is unknown. In the case of radical input, if the radical of the target character is not known, it takes a great deal of trouble to search for it, and it takes time to search when there are many characters having the same radical. In the case of inputting the total number of strokes, it is first troublesome to count the number of strokes of the target character. If the character is a multi-stroke character, the number of strokes will be erroneously counted, resulting in poor search efficiency. Also, since there are many characters having the same stroke number, it takes time and effort to search for the target character.

When the method disclosed in the above-mentioned Japanese Patent Application Laid-Open No. Hei 2-18661 is used, since the "reading" is not known, the target character is searched from the radical and the number of strokes. For those who do not know the head, the conditions are the same as entering the total number of strokes, and even if you know the radical, counting the number of strokes of the character is a very patient work, and it is a very efficient search method I can't say that.
The same is true when the list is digitized, and the number of characters that can be displayed on the display is often smaller than the number that can be displayed on paper, in which case it takes more time to find the desired character Will be.

In addition, when a similar-character dictionary for realizing the similar-character conversion is created manually, characters having similar shapes are collected by hand and their similarity is subjectively evaluated. Characters are missing from the dictionary even though there are still characters, the similarity evaluation is not unified in the similarity group, and there is a difference in the similarity evaluation between the user and the dictionary creator In some cases, the feeling of use may be significantly impaired. Needless to say, if this work is performed manually, the labor and time (cost) required for the work become enormous.

As described above, Japanese Patent Laid-Open No. 5-122
Japanese Patent Publication No. 49 only searches for a target character from kanji having the same radical. According to the present invention, in order to input a character which cannot be read, a similar character dictionary is searched using a key "character having a similar shape (= similar character)" as a key, and a target character is searched. The present invention also has means for searching for a target character using a common radical as a key, but this is only one means, and in the present invention, even if the radical differs, the shape is similar. Characters ("tile" and "mutual") are also searched.

In Japanese Patent Laid-Open Publication No. Hei 5-12249, the same character of a radical is simply picked up from a dictionary, but the present invention introduces the concept of "degree of similarity". Similar characters can be arranged before candidates. For example, in Japanese Unexamined Patent Publication No. Hei 5-12249, even if the user tries to input "En" and presses "Address" as a key, a large number of characters "Ukanmuri" usually appear, and "En" is efficiently input. I can't find it. In addition, if the position of the radical is specified, the candidates may be narrowed down a little more, but this work itself is inefficient. Therefore, if a similar character search is performed using a dictionary created according to the dictionary automatic generation method according to the present invention, it is possible to search for characters that cannot be read much more efficiently than in JP-A-5-12249.

The present invention has been made in view of such circumstances, and in a device capable of inputting Japanese, when a character which cannot be read is input, a dictionary which collects characters having similar shapes is used. It is an object of the present invention to provide a Japanese input device in which a character having a shape similar to that of a character is first input, and a target character is searched using the dictionary.

[0015]

SUMMARY OF THE INVENTION The present invention, in order to achieve the above object, when entering read Minowakara no characters,
Kanji input means for inputting characters with similar shapes and known readings as key characters, and for online handwritten character recognition
Using a dictionary and an optical character recognition dictionary, similar character dictionary and similar character dictionary automatically generated by collecting similar character degrees of each character of the character group, and input by the kanji input means has been to search for a character that is similar in form from the previous SL class character dictionary on the basis of the key characters, Japanese input that includes a Luigi search means for evaluating the degree of Luigi
Input device, wherein the kanji input means has a different viewpoint.
Input the number of key characters, and for each key character,
By searching for characters with similar shapes from
Narrow down character candidates and sort in order of key character and similarity
It is a Japanese input device that displays complementary characters.
I do.

[0016]

According to the present invention, there is provided a Japanese-language input device comprising:
It is possible to easily search for characters that cannot be read. That is, most kanji are not so many,
It is thought to consist of several parts. For example, "arm"
The character is composed of four parts: "moon", "shin", "evening", and "mi". Also, the word "en"

[0017]

[Table 1]

It consists of three parts, "evening" and "snake".
The parts of "evening" and "snake" are common in "arm" and "en",
Therefore, it can be said that these two characters are similar. It can be said that the greater the proportion of common parts, the higher the “similarity” of the two characters. A collection of similar characters of these types can be created using a recognition dictionary for online handwritten character recognition. In addition, even if the parts are not common, there are also characters having similar shapes, such as "mutual" and "tile". This can be created using a dictionary for optical character recognition.

As described above, a “similarity dictionary” having characters having similar shapes and an index (similarity degree) indicating how similar the characters are prepared is prepared. Search for unknown characters. For this purpose, first, a character similar to a character that the user wants to input but cannot understand is input using a normal kana-kanji conversion method or the like. Then, using that character as a key, a similar dictionary is drawn. Similarly, a similar character dictionary is drawn using another similar character as a key to narrow down a target character.
In this way, candidates are obtained from the (plural) key characters, and a target character is visually found therefrom. Compared to a conventional total stroke count table or radical table, the search time is overwhelmingly advantageous when searching by pulling a similar dictionary, since only a small number of candidate characters need to be searched.

[0020]

Embodiments will be described below with reference to the drawings. FIG. 1 is a configuration diagram for explaining one embodiment of a Japanese input device according to the present invention, wherein 1 is a kanji input means,
2 is a type search unit, 3 is a type dictionary, 4 is a buffer management unit,
Reference numeral 5 denotes a buffer unit, 6 denotes a candidate display / selection unit, and 7 denotes a kana-kanji conversion dictionary. The kanji input means 1 performs kanji input other than similar character input, and is realized by kana-kanji conversion means or the like. When a key character is given by the kanji input means 1, a similar character dictionary 3 is searched by a similar character search unit 2. The type dictionary 3 is a collection of characters having similar shapes and the degree of similarity in a device capable of inputting Japanese. When the candidate character is given by the type search unit 2, the buffer management unit 4 creates a new buffer from the character already stored in the buffer unit 5 and the new candidate character, and the candidate display / selection unit 6 Display and select. The kana-kanji conversion dictionary 7 is used as necessary when giving key characters by the kanji input means 1, and is not always necessary.

FIG. 2 is a diagram showing the structure of the type dictionary.
In order to perform type conversion, a type dictionary must first be prepared. Focus on a character and collect similar characters. For example, if you list the similar characters with the word "arm" as a key, you will find characters such as "Bai-fuen-en-en". These are similar in shape to each other, and a group of characters is called a “character group”. A set of type groups is a type dictionary. Then, points are assigned to the element characters of these subscript groups. The score is the degree of similarity with respect to a character serving as a key, and the higher the score, the more similar the shape. The type dictionary is a set of characters in which all the characters are used as keys and similar characters are listed, and a degree of similarity is added to each character.

In FIG. 2, first, a pointer to a similar character group for a key character is placed in the order of the JIS code. The head of the dictionary is the first of the JIS code, 2
For 121 (hexadecimal) characters (full-width space)
A pointer to the character group will be placed. At the end of the pointer, the characters of the double-byte space and the degree of the similar characters are arranged. Hereinafter, pointers to the character groups are listed in JIS code order.

For example, the type of “arm” is JIS of “arm”.
The codes are arranged from the location indicated by the pointer stored in the pointer storage location of the code 4F53 together with the degree of similarity. "Bowl", "Eupai", "To"
There are six characters, "En", "Snake", and "Grudge".
Assume 90, 80, 70, 60, 50, 40. Then, from the location indicated by the pointer to the character group of "arm", "bowl, 90, euphem, 80, addressed to, 70, garden, 60,
Snake, 50, grudge, 40, 0 "(the last 0 is an end code). Thus, all JI
For S characters, a dictionary in which the similar characters are arranged is used as a similar character dictionary.

Next, the operation of searching the type dictionary and outputting a character similar to the input character will be described. FIG. 3 is a flowchart showing the procedure of the type conversion. Hereinafter, description will be made in order according to each step (S). In order to input a character whose reading is unknown, in this method, a character whose shape is similar and whose reading is known is first input (S1). This is input using, for example, normal kana-kanji conversion.
Then, using that character as a key, a similarity dictionary is searched.

Next, based on the JIS code of the key character,
The address storing the pointer indicating the character group of the character is calculated. The pointer is read from the address to obtain the head address of the character group. The first character of the key character and its character degree are read from the address and stored in the buffer. The pointer is advanced to read the next type and its degree. This operation is continued until the end code is read. As a result, all the similar characters and similar character degrees of the key characters are stored in the buffer (S
2).

Similarly, when there is another key character (S
4) The same search is performed, and the type and degree of similarity are read out. Then, if any of the similar characters searched from the key character is the same as the character already in the buffer, the similarity degree is calculated as
In addition to the typology of the character in the buffer, make it the new typography of that character. In other words, when the same character is extracted from different key characters, the degree of similarity is added to increase the priority of candidate display. The newly read character class is
If not, add the character to the buffer. If the buffer overflows, the buffer with the lower similarity is deleted (S5).

In this manner, a plurality of key characters are designated,
By preferentially displaying a character that is likely to be searched for in the front, it is possible to easily search for a target character. The target character is selected from the candidate characters arranged in the order of the similarity degree, and this completes the similarity conversion (S6).

FIG. 4 is a flowchart showing an operation example of selecting a target character in accordance with the procedure of the type conversion shown in FIG. Hereinafter, description will be made in order according to each step (S).
Here, we want to input the word "grudge", but suppose that we do not know the reading. First, it is necessary to enter a character similar to this character as a key character. Here, the character "to" is first input as a key character. The input of the key character is performed by using a normal kana-kanji conversion or the like (S11). Then, the key characters are converted into similar characters. As a candidate character, "Bai Yi Yuen Jiangyuan ..." is found and displayed as a candidate character in order of similarity (S12).

Next, the key character "anger" is input (S1).
3). When this key character is converted into a typographical character, "Grave effort love forget ..."
And so on, but since candidates for which the character conversion of “address” has already been performed have already been made, new candidate characters integrated with the candidates are displayed (S14). here,
The “address” and the “anger” are converted into characters, and it is found that the target “grudge” is at the head of the candidate (S15).

Next, a method of automatically creating a type dictionary will be described. There are two methods for automatically creating this type dictionary: (1) a dictionary for optical character recognition, and (2) a dictionary for online handwritten character recognition. A character,
It is difficult to quantify how similar another character is. In addition, for each character, finding a character similar in shape to it, evaluating the similarity degree, and creating a similarity dictionary is performed manually, requiring a great deal of labor and time. . Here, a method of automatically generating a similar character dictionary by using an optical character recognition method or an online handwriting recognition method will be described.

First, a method of automatically generating a similar character dictionary using an optical character recognition dictionary will be described. This is usually
Holding dot information of each character to be recognized in vector form,
This is to match with a dot pattern input by an OCR (Optical Character Reader: scanner or the like). FIGS. 5A and 5B show the concept.
Here, it is assumed that a character of 8 dots × 8 dots is recognized.

First, a character pattern is recorded on a grid of 8 dots × 8 dots. It is assumed that the portion where the dot is ON is 1 and the portion where the dot is OFF is 0,
The values are arranged in order from the upper left, and expressed as a 64-dimensional vector. This vector exists for each character to be recognized. The set of vectors of each character is a character recognition dictionary.

Here, a character is input from the OCR. 1
Matching with dictionary data is performed using data normalized to be 8 dots × 8 dots of characters. As in the case of creating a dictionary, the input 8 dot × 8 dot data is
A four-dimensional vector is calculated, and the distance between the vector and each character vector recorded in the dictionary is calculated. The distance between these two vectors is the degree of coincidence. The basic method of optical character recognition is to calculate the degree of coincidence with all the characters in the dictionary and output the one with the highest degree of coincidence as a recognition result.

[0034]

(Equation 1)

Therefore, using this optical character recognition method,
Rather than matching dictionary data and input data from the OCR, matching is performed between each character in the dictionary. The matching method is exactly the same, and the distance between the 64-dimensional vectors of each character is determined, and is used as the similarity. For each character, a character having a certain degree of similarity and the degree of similarity are registered in the dictionary as a similarity group of the target character.

Hereinafter, a method of automatically generating a similar dictionary using an optical character recognition dictionary will be described in more detail. To explain the principle of optical character recognition, the degree of similarity between the matrix pattern of characters stored in the system and the input matrix pattern is determined by the vector obtained from the matrix. The evaluation is based on the distance of the data. For the sake of simplicity, the description will be made using data of 3 dots × 3 dots.

FIG. 6A shows a matrix pattern in a dictionary, and FIG. 6B shows an input pattern. FIG. 6 (a),
The matrix pattern of 3 × 3 dots in FIG.
If the part with dots is 1 and the part without dots is 0, 3 × 3
= 9 rows of 0s and 1s, that is, 9-dimensional vector data. The “distance” between these two pieces of 9-dimensional vector data is shown in Table 2 below.

[0038]

(Equation 2)

Therefore, if the pattern in the dictionary and the input pattern are exactly the same, the similarity degree becomes 1, and the closer the two patterns are shifted, the closer to 0. This value is used as the similarity degree.

Create a similar dictionary from matrix data
That method Figure 7 is a block diagram for creating Luigi dictionary from matrix data in Fig, 11 is an optical character recognition dictionary,
Reference numeral 12 denotes a type character evaluation unit, and reference numeral 13 denotes an optical character recognition version type dictionary. The optical character recognition dictionary 11 stores matrix pattern data of each character to be compared with the input matrix pattern, and is usually arranged in the order of character codes as shown in FIG. In order to automatically create a similar character dictionary from the matrix pattern data shown in FIG. 8, the similar character evaluation unit first calculates the first character (JIS code 2121) of the dictionary of FIG. ) And all the remaining characters in the dictionary. Then, for a character having a certain degree of similarity (for example, 0.7) or more, the character
The character code of the IS code 2121 is registered in the type dictionary.

Next, the degree of similarity between the character of the JIS code 2122 and all other characters included in the dictionary is calculated, and similarly, a character having a certain degree of similarity or more is converted to the JIS code 2122. Is registered in the type dictionary. Similarly, for all characters, a character dictionary whose similarity degree with the remaining characters is equal to or greater than a certain value is created as a character class of the character. In this way, the optical character recognition version type dictionary 13 is created.

The above is a description of a method of automatically creating a similar character dictionary using the optical character recognition dictionary. Next, a method of automatically creating a similar character dictionary using the online handwritten character recognition dictionary will be described. explain.

FIG. 9 is a diagram for explaining the method of online handwritten character recognition, and FIG. 10 is a conceptual diagram of a dictionary structure for online handwritten character recognition. Online handwritten character recognition is based on the recognition of characters as a combination of simple strokes (basic strokes), recognition of online handwritten input for each basic stroke, and a set of basic strokes as subpatterns. Is a character that recognizes characters from the bottom up like a character.

Therefore, the online handwritten character recognition dictionary has information indicating what kind of sub-patterns a character is composed of, and this information is used to generate a similar character dictionary. In other words, a sub-pattern is a component of a kanji that depends on the recognition algorithm, but is a part of the kanji that is different from the other kanji. It is composed of a combination of Therefore, the characters sharing these sub-patterns have similar shapes, and it can be said that the greater the sharing ratio, the higher the degree of similarity.

For example, the characters "arm" are "moon" and "shin".
It is composed of four sub-patterns, “evening” and “snake”.
The degree of similarity between the characters is set to be higher as the matching rate of the sub-pattern is higher. Further, a more accurate dictionary can be generated by increasing the weight to the similarity degree for a sub-pattern having a higher appearance frequency. In such a manner, for all characters, the degree of similarity with other characters is calculated, and characters having a certain degree of similarity or more and the degree of similarity are calculated.
Register in the dictionary as a similar character group of the target character.

Hereinafter, a method of automatically generating a similar character dictionary using the online handwritten character recognition dictionary will be described in more detail. Evaluation method of similarity degree The dictionary for online handwritten character recognition has a configuration as shown in FIG. In this example, "tie" is "yarn"
It is composed of three sub-patterns, “shi” and “mouth”,
“Zu” indicates that it is composed of three sub-patterns of “word”, “shi”, and “mouth”. More specifically, the dictionaries are configured as shown in FIGS.

The basic dictionary shown in FIG.
“Connection”) includes information indicating what kind of sub-pattern is formed. The sub-pattern dictionary shown in FIG. 12 includes information indicating what basic stroke each sub-pattern is composed of. The basic stroke dictionary shown in FIG. 13 includes information indicating what each basic stroke is.

Now, in the evaluation method of similarity, it can be said that the more common subpatterns are between two characters, the more similar the two characters are. For example, "En"
Consider the three letters "to" and "grass". It is assumed that each character is composed of a sub-pattern as shown in FIG. 14 (the selection of the sub-pattern differs depending on the recognition system).

The "evening" and the "address" are common between "en" and "address", and two of the three subpatterns are common. On the other hand, in "En" and "Kusa",

[0050]

[Table 2]

Are common, and one of the three sub-patterns is also common. Therefore, if we think very simply, the typographical degree of “En” and “To” is 2/3 = 0.
67. The degree of typology of “En” and “Kusa” is 1/3 = 0.33.

However, if only the ratio of common sub-patterns is the similarity of two characters, many characters having the same similarity will appear. In other words, there are many or few, which deviates from the original purpose of the present invention. Therefore, this problem can be solved by setting a weight for each sub-pattern. Now, suppose that weights are set as shown in FIG.

The principle of the weighting is that "things that appear in many characters in common are small, and those that rarely appear are increased in size". For example, there are so many characters that have "Kusakanri" or "Ukaneri", and characters that have these in common will not seem to be very similar. However, characters that have sub-patterns that do not appear in so many characters, such as “snake”, should feel more similar.

Therefore, by performing weighting in accordance with the appearance frequency of the sub-pattern, it is possible to increase the degree of similarity between characters that may be more similar. Regarding this weight, it is a rule that the appearance frequency of all character sub-patterns is examined and determined in comparison with the value. It does not always depend. Therefore, in order to create a dictionary that is easier to use, it is necessary to adjust the weight.

Now, according to the weights shown in FIG.
When the typologies of “En” and “To” and “En” and “Kusa” are calculated, the results are as shown in FIGS. FIG.
According to the weights shown in (a), since the common sub-patterns are "evening" and "snake", the typographic degree of "address" viewed from "en" is 0. 0 as shown in Fig. 16 (b). 86. Also,
According to the weights shown in FIG. 16C, the common sub-pattern is

[0056]

[Table 3]

Therefore, the degree of typology of "grass" viewed from "en" is 0.14 as shown in FIG. When this formula is generalized, the similarity of the character B as viewed from the character A is as shown in FIG. As can be seen from this equation, the degree of similarity takes a value from 0 to 1, and the closer to 1, the higher the degree of similarity.

From a dictionary for online handwriting recognition to a similar character dictionary
Method Figure 18 to create is a block diagram for creating Luigi dictionary online handwriting recognition dictionary, in the figure, 21 is line handwritten character recognition dictionary, 22 Luigi degree evaluation unit, 23-line handwritten Character recognition version type dictionary, 24 is a weight evaluation unit, and 25 is a weight table. FIGS. 11 to 1 described above.
The dictionary shown in FIG. 3 automatically generates a type dictionary using the type evaluation method shown in FIG.

First, the online handwritten character recognition dictionary 2
The sub pattern weight table 25 is created by the weight evaluator 24 using 1. All characters in the basic dictionary shown in FIG. 11 are checked, and the number is counted for each sub-pattern. The weight of the sub-pattern is normalized by setting the weight of the sub-pattern that has been counted the most as 1, and the weight table 2
Create 5. As described above, it is desirable to adjust the weight in consideration of the influence of the degree of similarity due to subjectivity. next,
According to the weight, the similarity evaluation unit 22 calculates the similarity of a certain character and the remaining characters. A character having a similarity degree of a certain degree or more (for example, 0.7 or more) is registered as a character having a similarity degree of a certain degree or more in a similarity dictionary, and a similarity dictionary 23 is created. .

The above is a description of a method for automatically generating a similar dictionary using an online handwritten character recognition dictionary. Next, integration of the optical character recognition version type dictionary and the online handwritten character recognition version type dictionary will be described. The similar character dictionary created from the optical character recognition dictionary generated as described above and the similar character dictionary created from the online handwritten character recognition dictionary are combined into one, and the similar character dictionary shown in FIG. A type dictionary.

More specifically, for one key character, the character classes are extracted from the optical character recognition version type dictionary and the online version type dictionary, respectively, and weighted appropriately, and the degree of similarity is added. The character type of “sound” is as shown in FIG. 19 (the character degree is normalized so that the perfect score is 100), the weight of the optical character recognition version is set to 0.4, and the weight of the online version is set. When the total typology of “sound” is calculated by setting to 0.6, the result is as shown in FIG. This value is registered in the type dictionary as the “character degree” of “sound”. Also in this case, by registering only numerical values having a similarity degree equal to or more than a certain value (for example, 30 or more) in the dictionary, the number of registrations is prevented from becoming too large. Also, the weights are tentatively set to 0.4 and 0.6, but this value should be adjusted to a value that seems to be the most appropriate after actually performing type conversion by a plurality of users.

Next, a method of searching / displaying the type dictionary will be described. The procedure of the type conversion is as shown in FIG. Here, a dictionary search and the like will be additionally described. The type dictionary has a structure as shown in FIG. Here, a numerical value obtained by normalizing the similarity degree from 0 to 100 is used. When a key character is given, the type search unit 2 searches the type dictionary 3 and stores the type of the key character and the degree of the type in the buffer unit 5. For example, if the key character is
If the type and degree of typology are as shown in FIG. 21, the buffer unit 5 will be as shown in FIG.

Next, it is assumed that another key character "anger" is given. The type and degree of similarity of “anger” are as shown in FIG. Add this type and degree to what is already in the buffer. If any of the new key characters has a common character with the character already in the buffer, the similarity is added to the similarity in the buffer, as shown in FIG. If the size of the buffer is limited, delete the buffer with the smallest similarity. Alternatively, the threshold for the degree of similarity may be adaptively changed so that only characters having a certain degree of similarity or more are left in the buffer. To adaptively change the threshold value means that the degree of similarity is 50 or more when one key character is specified, 70 or more when two key characters are specified, and 10 when three key characters are specified.
0 or more. In this way, by specifying a plurality of key characters, it is possible to bring the target character to the front of the candidate. By the above method, from optical character recognition dictionary and online handwritten character recognition dictionary,
A type dictionary can be automatically generated.

[0064]

As is apparent from the above description, the present invention has the following effects. (1) It is difficult to efficiently search for a target character by the currently used method of inputting a character that cannot be read. Therefore, if the “type conversion” function is used, these conventional methods can be used. The desired character can be found more efficiently than the search method used. Shortening of search time and simplification of operation are achieved. Also, even if you do not understand the reading,
Even if you do not know the radical, you can search for the desired character, and there is no need to count the number of strokes. (2) By automatically generating a type dictionary using optical character recognition and online handwritten character recognition, it is possible to prevent character omission in the type dictionary construction and objectively evaluate the degree of similarity.
It is possible to achieve unification of the degree of similarity among similarity groups, and to create a dictionary for similarity conversion with a comfortable feeling of use. In addition, an accurate dictionary can be created in a short time and with low labor (low cost). (3) Multiple key characters with different perspectives can be input from the kanji input means.
Enter a sentence that is similar in shape from the type dictionary for each key character
Search for characters, narrow down the target character candidates,
Since the candidate characters are displayed in order of character and similarity,
Can easily be found.

[Brief description of the drawings]

FIG. 1 is a configuration diagram for explaining one embodiment of a similar character search of a Japanese input device according to the present invention.

FIG. 2 is a diagram showing a structure of a type dictionary in the present invention.

FIG. 3 is a flowchart showing a procedure for converting characters according to the present invention.

FIG. 4 is a diagram illustrating an operation example of type conversion according to the present invention.

FIG. 5 is a diagram for explaining an optical character recognition method according to the present invention.

FIG. 6 is a diagram showing a dictionary pattern and an input pattern according to the present invention.

FIG. 7 is a block diagram for creating a similar dictionary from matrix data in the present invention.

FIG. 8 is a diagram showing similarity in the present invention.

FIG. 9 is a diagram for explaining a method of online handwritten character recognition according to the present invention.

FIG. 10 is a conceptual diagram of a dictionary structure for online handwritten character recognition according to the present invention.

FIG. 11 is a diagram showing a basic dictionary according to the present invention.

FIG. 12 is a diagram showing a sub-pattern dictionary according to the present invention.

FIG. 13 is a diagram showing a basic stroke dictionary according to the present invention.

FIG. 14 is a diagram (part 1) for describing a method of evaluating the degree of similarity in the present invention.

FIG. 15 is a diagram (part 2) for explaining a method of evaluating the degree of similarity in the present invention.

FIG. 16 is a diagram (No. 3) for explaining the method of evaluating the degree of similarity in the present invention.

FIG. 17 is a diagram showing calculation of similarity in the present invention.

FIG. 18 is a block diagram for creating a similar character dictionary from the online handwriting recognition dictionary according to the present invention.

FIG. 19 is a diagram showing an optical character recognition version type dictionary and an online handwritten character recognition version type dictionary according to the present invention.

FIG. 20 is a diagram illustrating an example of calculating a total similarity degree for a key character according to the present invention.

FIG. 21 is a diagram showing a similar character and a similarity degree in the present invention.

FIG. 22 is a diagram illustrating a state of a buffer of a similar character and a similarity degree in the present invention.

FIG. 23 is a diagram showing a similar character and a similarity degree in the present invention.

FIG. 24 is a diagram illustrating a state of a buffer of a similar character and a similarity degree in the present invention.

FIG. 25 is a diagram showing a conventional list for code input.

FIG. 26 is a diagram showing a conventional list for radical input.

[Explanation of symbols]

1 ... Kanji input means, 2 ... Type search section, 3 ... Type dictionary, 4
... Buffer management unit, 5 ... Buffer unit, 6 ... Candidate display / selection unit, 7 ... Kana-Kanji conversion dictionary, etc. 11 ... Dictionary for optical character recognition, 12 ... Type evaluation unit, 13 ... Type dictionary for optical character recognition , 21 ... Dictionary for online handwritten character recognition, 22 ...
Type character evaluation unit, 23: online handwritten character recognition version type dictionary, 24: weight evaluation unit, 25: weight table.

Claims (1)

(57) [Claims] When inputting a character which cannot be read,
And the Chinese character input means for character key character type the you know the reading of the characters are similar in form, for online handwriting recognition
Using a dictionary and an optical character recognition dictionary, similar character groups and similar character dictionaries that are automatically generated by collecting similar character degrees of each character in the character group, and input by the kanji input means has been to search for a character that is similar in form from the previous SL class character dictionary on the basis of the key characters, Japanese input that includes a Luigi search means for evaluating the degree of Luigi
Input device, wherein the kanji input means has a different viewpoint.
Input the number of key characters, and for each key character,
By searching for characters with similar shapes from
Narrow down character candidates and sort in order of key character and similarity
A Japanese input device characterized by displaying complementary characters .