JPS5884372A - Recognizing method for on-line character written by specific writer - Google Patents

Abstract

PURPOSE:To increase recognition factor regardless of the writing habit of the writer, by recognizing the partial gathering of fractionated characters to discriminate the KANJI (Chinese character) in the form of a group of the partial gatherings, and therefore reducing the storage capacity necessary for a dictionary. CONSTITUTION:A pseudo radical dictionary 3 containing the features of pseudo radicals and pseudo radical codes is used when the stroke information fed through a tablet 1 is recognized at a recognizing part 2. The output of recognition is recognized and stored in an input register 4 in the form of a pseudo radical code. A selecting circuit 5 selects the characters from the contents of a character dictionary 6 and the pseudo code of the output of the register 4. In this case, the contents of the dictionary 3 are corrected in terms of only the pseudo radicals registered to the dictionary 8 when the registering is finished to the dictionary 8 by a writer. The information of a standard dictionary 7 is previously transferred to the dictionary 3 for other pseudo radicals. Therefore the input characters written by the writer are all recognized.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a recognition method related to an online handwritten character recognition device used as an input device for information processing equipment.

Conventional handwritten character input word processors use online handwritten character recognition technology to recognize handwritten characters, but for example, when recognizing numbers, alphabets, hiragana, and kanji, it is necessary to recognize 10 numbers and 26 alphabets. There are about 200 characters including 46 hiragana characters and symbols such as voiced and handakuten, and kanji are written in JIS C622.
6 There are 2,965 characters in the first level kanji set. Even if we narrow down the kanji to commonly used kanji, there are 1945 characters, a total of 2000.
It goes beyond words.

Now, a handwritten character input Japanese word processor that uses kanji as regular kanji and recognizes a total of over 2000 characters has been announced, but the recognition process uses, for example, the K-L expansion method. The drawback was that it required too much hardware, such as not being able to directly recognize kanji. Furthermore, in addition to the above example, when handwritten kanji are directly recognized using the same algorithm as numbers, alphabets, hiragana, etc., the amount of calculation required for recognition is enormous and processing time is required. There was a serious drawback from the point of view of practical application, as the reduction in time required an increase in the amount of hardware.

Furthermore, in the past, when recognizing handwritten kanji, even if the stroke order could be recognized even if the stroke order was slightly incorrect, it was necessary to force the scribe to write it strictly in the handwriting style, making it easier to use. I couldn't say yes. This was a measure to eliminate as much as possible the handwriting habits of individual scribes, and was supposed to be used for online handwritten character recognition, but it was found to be painful for the scribes, and it was recognized that the scribe's habits would come out. °There was a serious drawback that the recognition rate decreased.

As explained above, the conventional online handwritten character recognition method has the serious drawback that it is expensive when attempting to recognize Chinese characters, and furthermore, it is almost impossible to recognize irregular characters.

In order to eliminate the above drawbacks, the present invention subdivides kanji, recognizes subsets of the subdivided characters, and identifies kanji as a collection of the recognized subsets, comprising:
This reduces the number of kanji characters to be recognized under equal constraints and also reduces the storage capacity required for the dictionary.Since the character dictionary to be recognized can be registered in the form of a series of subsets of the aforementioned characters, the number of characters to be recognized increases. In addition to having the feature of minimizing the increase in memory capacity, the present invention is a dictionary that recognizes a subset of characters by subdividing kanji, and extracts features from the standard dictionary and the handwriting results of each scribe. Because it has a personalized dictionary, it also has the feature of being able to identify characters, including kanji, with a high recognition rate, regardless of the presence or absence of handwriting quirks by the scribe.

FIG. 1 is a block diagram showing an embodiment of the present invention,
1 is a tablet, 2 is a recognition unit, and 3 is a subset of characters obtained by subdividing kanji and characters other than kanji (hereinafter referred to as pseudo radicals)
4 is an input register that stores the pseudo radical code recognized by the recognition unit 2, and 5 is one or more of the characters stored in the input register 4. A selection circuit that selects characters from pseudo-radical codes;
6 is a character dictionary storing pseudo radical codes and character codes;
7 is a standard pseudo-radical dictionary that stores the standard features of pseudo-radicals and pseudo-radical codes, and 8 is a list of individual features of pseudo-radicals extracted from the handwriting results of each reporter, and pseudo-radicals. It is structured like a personal pseudo-radical dictionary that stores neck codes.

Stroke information input from the tablet 1 is sent to the recognition section 2. The recognition unit 2 works to recognize pseudo radicals by a well-known stroke analysis method suitable for online handwriting. When the recognition unit 2 performs recognition, a pseudo radical dictionary (hereinafter referred to as a dictionary) '3 that stores features of pseudo radicals and IN radical codes is used. The output of the recognition unit 2 is outputted and stored in the input register 4 in the form of a pseudo-radical code each time it is recognized. Thereafter, the stored pseudo radical codes are sequentially output from the input register 4 and input to the selection circuit 5. The selection circuit 5 selects a character based on the contents of the character dictionary 6 and the pseudo radical code output from the input register 4, and outputs the resulting character code.

FIG. 2 shows an example of the dictionary 3. Numbers in hexadecimal digits following 000 indicate pseudo-radical codes, and each pseudo-radical is written on the right side of the pseudo-radical code. In an actual dictionary, feature data is written at the position of each pseudo-radical according to the recognition algorithm of the recognition unit 2, but each pseudo-radical itself is shown here for convenience of explanation. In addition, in Dictionary 3, hiragana, numbers, alphabets, etc. other than kanji are not subdivided.
It is stored directly in its original form.

Here, the standard pseudo-radical dictionary (hereinafter referred to as standard dictionary) 7 and the individual pseudo-radical dictionary (hereinafter referred to as personal dictionary) 8 in FIG. 1 will be explained.

In general, character recognition, whether online or not, has been configured to recognize the handwriting of an unspecified scribe. However, considering the case where online handwritten character recognition is applied to a Japanese word processor to configure an online handwritten character input word processor, the word processor is used only by a limited number of people, and not necessarily by an unspecified number of people. There is no need to think of it as an object.

Based on this idea, one feature of the present invention is to propose the idea of registering the handwriting of the individual who will use it in advance.

Now, in the standard dictionary 7, hiragana,
Numbers, alphabets, etc. are stored directly as they are without being subdivided (approximately 200 characters), and kanji are stored in 296 kanji.
Five characters are subdivided into over 400 pseudo radicals and stored. The characteristics of each pseudo-radical stored in the standard dictionary 7 are determined by using feature data for each pseudo-radical collected from several hundred people, and the recognition algorithm of the recognition unit 2 uses the contents of the standard dictionary 7 as a dictionary. 3, common sense characters can almost be recognized.

Next, the personal dictionary 8 is configured to be able to transfer the standard dictionary 7 to the dictionary 3 and register the handwriting of the scribe if it cannot be recognized due to the scribe's habits or the like when the scribe writes in one state. There is. Therefore, in addition to the recognition function, the recognition unit 2 also needs an analysis function for input characters, but even if it is expressed in the word analysis, the content is the calculation of feature quantities, and functionally it is not a recognition function. There is no change. The only difference is that in the case of recognition, the features of the input character are compared with the features stored in the dictionary, but in the case of analysis, the features of the input character are stored in the personal dictionary 8 without comparison. In this way, only characters that give a recognition result contrary to the scribe's will even if written by the scribe are registered in the personal dictionary 8.

As mentioned above, when the scribe completes the registration in the personal dictionary 8, if the contents of the dictionary 3 are corrected only for the pseudo radicals registered in the personal dictionary 8, other pseudo radicals can be set as standard in advance. Since the information in the dictionary 7 is transferred to the dictionary 3, all characters input by the scribe can be recognized.

Each dictionary functions as explained in detail above, so Dictionary 3
may be a random access memory, and the standard dictionary 7 and personal dictionary 8 may be constructed from a file device such as a floppy disk.

The relationship between the standard dictionary 7, the personal dictionary 8, and the dictionary 3 has been explained in detail above, so the following explanation will explain how the contents of the standard dictionary 7 are transferred to the dictionary 3 before use, and are further registered in the personal dictionary 8. The explanation will proceed assuming that the contents of the δ-reading dictionary 3 have been modified for the pseudo radical.

FIG. 3 shows details of the input register 4. Reference numeral 9 indicates an output from the recognition unit 2, 10 to 17 indicate an Io register to an I register in the input register 4, 18 indicates a switching path, and 19 indicates an output from the input register 4.

FIG. 4 shows a portion of the character dictionary 6. The third line of Figure 4 is
The pseudo radical code 176 indicates the character ``tate'' and the pseudo radical ``日'', which is expressed by the pseudo partial ko and -do OFB, indicates that it is the character ``on.''
The character code for the character "One" is JIS C6226
0-I+-5, 'Proverb 3B' code. Note that the characters in 0 in FIG. 4 are shown for convenience of explanation, and the actual dictionary is composed of pseudo radical codes and character codes.

FIG. 5 shows the processing according to the present invention when the character "Akira" is input, and the input register 4 is filled with ■. register 10
~■7 This shows the pseudo radical code input to the register 17.

Hereinafter, the online handwritten character recognition method according to the present invention will be described in detail with reference to FIG. 5, taking the character "Akira" as an example. First, timing T.

Then, when the operator inputs "'" from Tablet 1, "
゛'' is output to the recognition unit 2, and the recognition unit 2 uses a pseudo radical dictionary (hereinafter referred to as dictionary) 3 to check whether it is registered as a pseudo radical. Because it is not defined, the code is undefined. Register IO
KRegister. Next, when "-" is input from the tablet 1 at timing ';f T2, the recognition unit 2 checks whether the pseudo radical "-" exists in the dictionary 3 along with the undefined "゛" at timing T1. When tested, as shown in Figure i, the pseudo radical "1" is registered as pseudo radical code 050, so ■. 05 in register 10
A code of 0 is set. Since the pseudo radical of "\" input at timing T3 is not registered in the dictionary 3, ■. An undefined code is registered in register 11, leaving register 10 as it is. The [/J input at timing T is not registered in the dictionary 3, but at timing T, the pseudo radical ``\l'' is added to the dictionary 3, along with the undefined ``\''. Since it is registered with the head code, the undefined code box in register 11 is deleted and a new code of 014 is set. Note that whether or not the pseudo radical codes 050 and 014 are new pseudo radicals is tested using Dictionary 3 using the character ``tate'', but the character ``tate'' is tested as a new pseudo radical by using Dictionary 3. Therefore, the contents of the I0 register 10 and 11 register 110 are held unchanged. When "-" is input at timing T, the character "-" is found to be a pseudo radical code 004 from the dictionary 3, so 004 is set in the register 12, and then "
It is verified whether pseudo-radical codes are registered in the dictionary 3 for Mame 1 and "tachi".

In other words, a test is performed to express all strokes within the character using the minimum number of pseudo-radical codes. In this case, "mame" is not registered as a pseudo-radical code, and "tate" is registered as pseudo-radical code 176 in the dictionary. Therefore, ■.

Register 10, ■, Register 11. Reset I2 register 12, ■. Register 176 in register 10. Thus, the character ``tate'' indicates that it is a single pseudo-radical as indicated by the pseudo-radical code 176 shown in FIG.

Similarly, as shown in FIG. 5, the pseudo radical code for the character "Akira" is 176, OFB.

It is confirmed that the character is represented by 045.065. Furthermore, at timing T, the input character consisting of three pseudo radical codes is re-verified and one pseudo radical code 17 is obtained.
6, timing T.

Then, the character consisting of the pseudo radical code 021 and the undefined "
-" and "-" to identify the character "day" with the pseudo radical code OFB, and also the timing T,..., T, 3°T, 4
Also, one pseudo-radical code is identified from input characters that have been identified as 2° pseudo-radical codes.

In this way, pseudo-radical codes are tested using a type of longest match method. Generally, in the case of the longest match method, a match is generally checked after all of the inputs have been input. In other words, the number of pseudo radicals that are judged for all input strokes and registered as pseudo radicals in Recognition 3 is over 600, even including numbers, alphabets, hiragana, symbols, etc., which is a large number. In consideration of the fact that the speed of inputting characters is slow, the verification is performed in the order of input.

As explained above, by inputting "Akira" from Taplet 1, I0 registers 10 to 10 in input register 4 are
Pseudo radical code 176°OFB in I3 register 13
, 045, and 065 are stored. I in these input registers 4. The contents of the registers 10 to 17 are sequentially output to the output 19 by the switching circuit 18 and input to the selection circuit 5. The selection circuit 5 uses the input pseudo radical code to select JI using the character dictionary 6 shown in FIG.
Select the character code according to SC6226.

That is, I in input register 4. Since the pseudo-radical code stored in registers 10 to 7 registers 17 is 176°OFB, 045,065, when the character dictionary 6 is checked by the selection circuit 5, the pseudo-radical code is found as shown in FIG. It turns out that the characters whose codes are 176, OFB, and 045,065 are the kanji character "Ang", which is indicated by the character code 3E34. In the above manner, J becomes 3E34.
By outputting the IS C6226 character code from the selection circuit 5, it is recognized that the handwritten character input from the tablet 1 is the Chinese character "Ang".

As explained in detail above, in the above embodiment, a kanji is subdivided, a subset of the subdivided characters named pseudo-radicals is recognized, and a kanji is identified as a collection of the recognized pseudo-radicals. We will show you how to do it.

In order to recognize pseudo-radicals here, it is sufficient to use a recognition unit 2 with a simple algorithm and a pseudo-radical dictionary 3 consisting of a very small number of pseudo-radicals compared to the number of Chinese characters. For example, numbers, alphabets, hiragana, symbols, and JIS
The number of pseudo radicals for the 2965 characters in the C6226 first level kanji set is over 600, of which over 400 are 29 kanji.
It is a pseudo-radical for 65 characters.

In this way, the present invention has the effect of reducing the number of 2965 Chinese characters to the same constraint only during recognition. An expert in the field can easily understand that the function of the recognition unit 20 for recognizing the contents of the pseudo-radical dictionary 3 shown in Figure 2 can be implemented using a simple algorithm such as the well-known stroke analysis method. Think of it as something. Furthermore, the character dictionary 6 for identifying characters including kanji as a collection of these recognized pseudo-radicals can be composed of only pseudo-radical codes and kanji codes according to JIS C6226, as shown in FIG. Another advantage of the present invention is that even when the number of characters including kanji increases, only a very small amount of memory is required. Unlike conventional dictionaries that record features, which have a large capacity, the amount is extremely small. Therefore, online handwritten character recognition has been developed, with the target being 2965 kanji characters.
Even if it is extended to the JIS C6226 first level kanji set consisting of characters, it can be provided at a low cost.

Furthermore, as a feature of the present invention, the contents of the pseudo-radical dictionary 3 include a standard pseudo-radical dictionary 7 consisting of features related to pseudo-radicals collected from the handwriting of several hundred people, and a standard pseudo-radical dictionary 7 consisting of features related to pseudo-radicals collected from the handwriting of several hundred people. It is possible to modify the information using the stored personal pseudo-radical dictionary 8, and to create a configuration that is easy for each scribe to recognize.

In general, an unspecified number of characters are recognized. In this case, the entire device becomes large due to the wide variety of quirky characters, and it is not possible to provide something that satisfies the scribe, such as regulating stroke order, strokes, etc., in order to make it easier to recognize. Ta. Under these circumstances, in applications such as online handwritten character input word processors where the number of scribes is not necessarily an unspecified number, the recognition rate itself can be dramatically improved, and the recognition method itself is also special, taking into account characters such as C characters. Since there is no need to do this, the recognition algorithm becomes simple, and an input device suitable for information processing devices in general can be provided at low cost.

Although the basic elements have been explained in the above embodiment, by implementing various improvements as shown below, a better online handwritten character recognition method can be provided, which will be explained below.

First, in the above embodiment, the content of the character dictionary 6 is a combination of only pseudo-radical codes and character codes, but as is clear from the example in FIG. Things that make up one character by themselves: ``concubine'', ``on'',
"iL" is located at the top of the character such as "Chapter", "I", and "Dou", and "iL" is located at the top left of the character such as "Akira" and "Rime".
There are some where "five" is located. These ``tachi'' are treated exactly the same in the above example, and there is no problem if the number of characters is about the same as in the example, but if you want to further increase the number of characters, the position of these ``tachi'' characters may be changed. If the information is included in the character dictionary 6, the recognition rate can be further improved. As the position information, it is sufficient to use the glyph structure in item 4 of the JIS C6226 glyph index.

, Second, as explained using the fifth example in the above example, every time a new stroke is input from the tablet L, all the steps up to the first stroke and the new stroke written are one pseudo stroke. We have explained how to test input strokes to determine whether they constitute a radical. For example, the pseudo radicals "-", "\l", and "-"
21- is the pseudo radical code 050,014,004
However, ``tate'', which is a collection of these three pseudo radicals, is also a pseudo radical, and its code is 176, which was determined by referring to the pseudo radical dictionary 3 using the stroke of ``tate''.

However, in addition to the pseudo-radical dictionary 3 as a dictionary referred to by the recognition unit 2, (pseudo-radical 050) + (pseudo-radical 014) + (pseudo-radical 004) = (pseudo-N
Providing a pseudo-radical correlation dictionary that shows the correlation between pseudo-radicals such as the head 176) has the effect of extremely shortening the recognition processing time.

As explained in detail above, the present invention subdivides kanji that are difficult to recognize, and recognizes the subdivided pseudo radicals to identify kanji, and also has a personal dictionary for quirky characters. As a result, the recognition rate can be improved, so the cost can be lowered while keeping the recognition rate high, thereby making it possible to provide an inexpensive and high-performance online handwritten character recognition device.

[Brief explanation of drawings]

Fig. 1 is a block diagram of an embodiment of the present invention, Fig. 2 is a diagram showing an example of a pseudo radical dictionary, Fig. 3 is a detailed diagram of an input register, Fig. 4 is a partial diagram of a character dictionary, and Fig. 5 is a diagram showing an example of a pseudo radical dictionary. The figure is a diagram showing a recognition method according to the present invention. 1... Tablet, 2... Recognition section, 3... Pseudo radical dictionary, 4... Input register, 5... Selection circuit. , 6...
・Character dictionary, 7...Standard pseudo-radical dictionary, 8...Personal pseudo-radical dictionary, 9...Output from recognition unit 2, 18...1 ...Switching circuit, 19...Output of input register 4 Patent applicant
Oki Electric Industry Co., Ltd.

Claims (1)

[Claims] In an online character identification method that includes a tablet that extracts stroke information of handwritten characters and a dictionary that stores character features, the information from the tablet is compared with the features of the dictionary to identify handwritten characters. It has a first dictionary that stores characteristics of subdivided subset patterns and the entire pattern of characters other than kanji, and a second dictionary that stores character codes for collections of subsets of kanji and characters other than kanji. and the first dictionary comprises a memory that integrates a standard dictionary that stores patterns in the standard handwriting of characters and a personal dictionary that stores patterns in the handwriting of at least one specific scribe. , furthermore, a recognition unit that compares the information from the tablet with the output of the first dictionary, an input register that stores the recognition result 4, and a character identification unit that compares the contents of the input register with the second dictionary. A selection circuit, a specific scribe online character recognition method characterized in that after the contents of the first dictionary are modified by a specific individual dictionary, characters input by the specific individual are compared and identified for each subset. .