JPH0565912B2 - - Google Patents

Description

[Detailed description of the invention]

TECHNICAL FIELD The present invention relates to character recognition, and more particularly to an online character recognition device including Chinese characters. (Technical Background) In the conventional field, there is a method of character recognition as a combination of subset patterns obtained by subdividing kanji, as shown in Japanese Patent Laid-Open No. 58-82373. This method, as shown in Japanese Patent Application Laid-Open No. 58-82373, subdivides kanji and recognizes subsets of the subdivided characters, and identifies kanji as a collection of the recognized subsets, and recognizes them equivalently. This method reduces the number of target characters, reduces the storage capacity required for the dictionary, and as a result, reduces the processing time required for recognition. However, in the above method, the number of kanji to be recognized is
Up to 2000 characters or less of commonly used kanji, including kanji used in place names, surnames, company names, etc. that are not included in the JIS 1st level kanji set, 2965 characters, or the JIS 1st level kanji set, etc. A problem occurs when you want to use it as a recognition target. For example, the character "Akira" has a partial pattern of 176 (standing),
It is composed of 0FB (Japanese), 045 (10), and 065 (彡), and according to the explanation in Japanese Patent Application Laid-Open No. 58-82373, it is identified as the character ``Ang''. However, if the target characters are up to the JIS 1st level kanji set, for example, the characters ``xiao'' and ``koro'' differ only in ``ko'' and ``匕'', and ``xiao'' is also written as ``〓'', so ``koro'' and ``koro'' are written as ``〓''. ”, making identification difficult. In addition, the character ``Kan'' has the same stroke as the character ``桔'', which is determined from the stroke combination, and if the stroke order is free, the partial patterns of ``田'', ``han'', ``ki'', and ``kichi'' will be formed. Characters become difficult to distinguish. There are approximately 1,500 types of confusion shown in the above example in the JIS Level 1 Kanji set alone. The types mentioned here include not only the identification of two characters, but also the words ``上'', ``earth'', and ``工''.
Most of the combinations are more than two characters, such as ``shi'', ``ki'', ``han'', ``cho'', ``satoru'', and ``tung'', and although it is possible in principle to identify these characters, it is impractical in practice. It can be said that this process is close to possible. In order to solve this problem, there are proposals such as fixing the stroke order to the contents written in the stroke order instruction manual published by the Ministry of Education, but this is also not practical. In other words, if the characters to be recognized are expanded to include the JIS Level 1 Kanji set, it can be said that it is practically impossible to recognize the characters using conventional recognition logic. Also, some characters have the same or similar shape but have different meanings (duplicate characters), such as E (e).
There were many characters such as ko (ro), kuchi (ro), ha (ha), and hachi (hachi), and it was difficult to recognize them because all the character types were mixed together. (Object of the Invention) The object of the present invention is to solve the above problems and to identify target characters using a simple method even when the number of target characters is 3000 or more. (Summary of the Invention) The main point of the present invention is to provide a detailed partial pattern dictionary for verifying whether or not a subset pattern (hereinafter referred to as a partial pattern) itself is the partial pattern at the partial pattern stage. After checking the necessary and sufficient conditions for a partial pattern, it is recognized as a partial pattern and then character recognition is started. Also, a detailed character dictionary is provided for multiple characters composed of exactly the same partial pattern. It is characterized by being designed to be identified. Furthermore, the present invention is characterized in that a plurality of characters having the same or similar shapes are provided with a duplicate dictionary and are identified through some processing. (Embodiment) Fig. 1 is a block diagram showing an embodiment of the present invention, in which 1 is a tablet that shows the coordinates of a writing instrument online when writing characters, and 2 is a tablet that limits the strokes written from the coordinates of the writing instrument. A stroke recognition unit, 3 is a stroke dictionary in which stroke information is stored in advance, 4 is a stroke string register that accumulates the output of the stroke recognition unit 2, 5 is a partial pattern recognition unit that identifies partial patterns, and 6 is a combination of strokes. a partial pattern dictionary storing partial patterns; 7 a partial pattern detail check dictionary storing conditions for being a partial pattern; 8 a redundant dictionary storing a plurality of characters having the same or similar shapes; 9 is a partial pattern string register; 10 is a character recognition unit that identifies characters from the identified partial patterns; 11 is a character dictionary storing characters formed by combinations of partial patterns; and 12 is a positional relationship of the partial patterns. This is a character detail check dictionary. Next, the operation will be explained. The strokes written on the tablet 1 are output to the stroke recognition section 2, compared with reference data in the stroke dictionary 3, recognized as a single stroke, and input into the stroke string register 4. The stroke here refers to the period from when the pen touches the tablet until it leaves the pen. In the embodiment of the present invention, strokes are classified into a total of 76 strokes using angle, direction, rotation, and loop. Therefore, the stroke dictionary 3 describes the characteristics of these strokes. FIG. 2 is an explanatory diagram showing strokes to be recognized. The stroke recognition method shown in FIG. 2 is not the purpose of the present invention, so its explanation will be omitted, but it can be easily recognized by the well-known stroke analysis method, K-L expansion method, or the like. In the figure, SXX indicates the stroke code, and the symbol to the right of the stroke talk code indicates its shape.
Furthermore, since each stroke has a width in terms of direction, length, etc., both "→" and "〓" are defined as S04. The following explanation will be given regarding the case of writing the character "member". FIG. 3 is an image diagram of the input. The numbers circled in FIG. 3 indicate the stroke input order, and SXX indicates the input stroke code. Here, before explaining the recognition method in the partial pattern recognition unit 5, the description method of each dictionary will be explained to facilitate understanding. Figure 4 a and b are examples of partial pattern dictionary 6.
SC indicates the stroke code, which is shown in Figure 2. The first line is the stroke code S04, which means that one stroke constitutes a partial pattern 001-0. Similarly, the second line is a combination of stroke codes S01 and S03! ,×,〓,〓,……
This means that partial patterns such as
On the right side of the stroke code and partial pattern code, typical shapes represented by each code are shown. Also, if a partial pattern code is not displayed and only a figure is displayed, this indicates that it is not a partial pattern but an independent character. As is clear from FIGS. 4a and 4b, one partial pattern is composed of a maximum of seven strokes. In addition, in order to avoid redundancy in Figures 4a and 4b, only the parts necessary to explain the present invention are extracted, and the actual number of items is approximately 650 items and the number of partial patterns is approximately 370. and corresponds to the independent characters that display only the figures described above.
Although a JIS6226 code is attached, it has been omitted to simplify the explanation. Next, as described above, the strokes written on the tablet 1 are recognized by the stroke recognition section 2 and input into the stroke string register 4, but the stroke string register 4 stores the strokes in the registers assigned the same number as the input order. The stroke code recognized by the recognition unit 2 is stored. Using the stroke code stored in the stroke string register 4, the coordinate data of the start point and end point of this stroke, and information in the partial pattern dictionary 6, partial pattern detail check dictionary 7, and duplication dictionary 8, the partial pattern recognition unit 5 generates a partial pattern. or 7
Characters that do not form a partial pattern that is smaller than a stroke are recognized, and the manner in which this is recognized will be explained using FIGS. 5a, b, and c. FIGS. 5a, 5b, and 5c are partial pattern recognition state diagrams showing how the process progresses each time a stroke code is input. In the figure, "input order" is the same as the input order shown in FIG. 3, "input stroke" indicates its shape, and "stroke code" indicates the stroke code recognized by the stroke recognition unit 2. . As mentioned above, the stroke string register 4 has a group of registers corresponding to the "input order" in FIG. 5, and each register stores a stroke code. If S02 “｜” is input first in the input order, the fourth
As is clear from the figure, S02 alone does not constitute a partial pattern, so there are no partial pattern candidates regarding the input order. If S06 “” is input in the input order, the stroke code will be 011-1 from the partial pattern dictionary 6 (Fig. 4) by combining the strokes with the input order.
011-2, 015-1, 015-2, 015-3, 015-4,
The partial pattern and the characters "7" and "u" are candidates. Here, it is possible that S02 in the input order and S06 in the input order are each independent partial patterns, but since S02 in the input order is not registered as an independent partial pattern, there is no need to consider partial patterns with the input order as the first. This is because if there are less than 7 characters from the first stroke (input order) to the last stroke, all of them will form a single character or a combination of partial patterns will form a character. Partial Pattern Dictionary 6, Character Dictionary This is because 11 is configured. Therefore, the input up to the input order is a stroke.
Limited to partial patterns consisting of S02 and S06. Note that the partial pattern code in Figure 5 has a branch number of 3 digits or 3 digits followed by a branch number, but this means that when there is only one partial pattern expressed by a certain stroke group combination, If there are multiple branches, branch numbers are added and displayed. To further explain Fig. 5, No. 1,
5, 7, 16, 23, 27, and 40 cannot be partial patterns or characters because their stroke combinations do not exist in the partial pattern dictionary 6. Further, in No. 8, the stroke S02 in the input order cannot be one partial pattern in one stroke, so there is no need to consider the partial pattern starting from the input order. Next, if the strokes are from the input order or later, the combination of those strokes cannot form a character. This is because a single partial pattern that becomes one character always includes the first stroke. Therefore, in the case of a combination of strokes after the input order, even if there is a candidate for one partial pattern and one character from the partial pattern dictionary 6, it may be ignored. Therefore, No. 2 and No. 6 are a combination of the same strokes, but the two candidates for the characters “7” and “u” in No. 2 are
It does not exist in No. 6. To explain further, the stroke S04 is shown as "-", but it also includes not only "→" but also "〓". So S04 like No.30
With the combination of and S04, "n" and "ni" are output as partial pattern candidates. As mentioned above, stroke S04 is the most extreme example, but all strokes are defined within a certain range. Furthermore, since the partial pattern is defined with 7 strokes or less, there is no need to consider partial patterns with more than 7 strokes. In addition, the stroke order within a partial pattern and the stroke order of partial patterns within a character are free, but even if the stroke order within a single partial pattern is free, it must be written consecutively, so the input order number can be skipped to create multiple strokes. The strokes of ke do not form one partial pattern. Therefore, the lines indicating the partial patterns in FIG. 5 are continuous. Next, the partial pattern detail check dictionary 7 will be explained. FIG. 6 is an explanatory diagram showing the contents of the partial pattern detail check dictionary 7. The partial pattern detail check dictionary 7 is used to check which partial pattern is a combination of written strokes. For convenience of explanation, only the items to be checked have been picked up and rewritten to correspond to the numbers in Figure 5, and the No. column in Figure 6 is based on the numbers in Figure 5.
The numbers after the hyphen in the No. column in FIG. 6 indicate the check order, the stroke A and B columns describe the strokes to be checked, and the check contents column describes what to check. In this way, the coordinate data detected by the tablet 1 can be used to check the start point, end point, length, etc. of each stroke. In the B code column, if the candidates are narrowed down by the check, the corresponding partial pattern code will be entered. In the JIS code column, if the candidates are narrowed down, the partial pattern code will be entered for those that can become characters.
JIS6226 code is listed. The candidate column shows the images of the candidates remaining at the time of the check. As a person skilled in the art would easily understand, the dictionary used for this process actually has a check item column corresponding to the check contents in FIG. 6 for each partial pattern column, and a check item column corresponding to each check item.
A column for reference data of the check result is provided. When there is an input stroke, the above-mentioned check item column is checked to see if the result matches the above-mentioned reference data column. A procedure is used in which a candidate is determined based on a match, but since it would be complicated to explain it according to the regular procedure, the method of description as shown in FIG. 6 is used. The manner in which partial pattern candidates are narrowed down using the detailed partial pattern check dictionary 7 shown in FIG. 6 described above will be explained with reference to FIGS. 3 and 5 again. Now, the case where the strokes are written in the input order in FIG. 3 will be explained. In this case, as explained above, the partial patterns are 011-1, 011-2,
015-1, 015-2015-3, 0154, and the characters "7" and "u" are candidates, and from the stroke shape
It can be considered as 011-1. That is, Fig. 6 Nos. 2-1 and 2-2 are partial patterns.
It shows the process in which 011-1 is selected. No.2-
From 1, "The starting point of S02 and the starting point of S06 are close"
There is a condition that when checked under this condition, the partial patterns selected are 011-1 (), 011-2 (〓),
, and the character is "7". Further, from No. 2-2, 011-1() remains and all other candidates are eliminated based on the condition that "the Y-direction lengths of S02 and S06 are equal". In this way, as shown in No. 2-2, the partial pattern 011-1 has no remaining characters. Further, as explained above, only those patterns that satisfy the necessary and sufficient conditions to be recognized as partial patterns are left as partial patterns. As described above, the candidates appearing from the partial pattern dictionary 6 shown in FIG. 4 are checked by the partial pattern detailed check dictionary 7 shown in FIG. 6, and the remaining candidates are shown in FIG. Figure 7 is a diagram showing partial pattern candidates, and the "input order", "input stroke", and "stroke code" in Figure 7
is the same as that in FIG. 5, and below the stroke code there is provided an image column showing the images from the first stroke to the current stroke. Also, for No. 9 and below, the shape of a partial pattern starting from a stroke in the middle of the input order is shown in brackets [ ] to the left of each line. As described above, the candidate partial patterns shown in FIG. 7 are selected. This information is sent to the partial pattern string register 9 and further processed by the character recognition section 10. It may be considered that the partial pattern sequence register 9 stores the state shown in FIG. The character recognition unit 10 processes the following steps. That is, the partial pattern string register 9 has 10 strokes, which exceeds the number of strokes 7 needed for one partial pattern to become one character, so it cannot become a character by itself. Also, since it becomes one character, it is expressed by a combination of some partial patterns, so the lines showing the partial patterns in FIG. 7 must be connected from the input order to form one line. That is, in principle, even a combination of and and and and in the input order can be characters. In other words, this combination becomes a candidate character. The combinations of candidate partial patterns as described above are as follows.

[Table] In this case, the 10 combinations are rearranged in numerical order and checked to see if they are registered in the character dictionary 11. An example of the character dictionary 11 is shown in FIG. FIG. 8 shows what characters are formed by the combination of partial pattern numbers and their shapes. The partial pattern column shows a combination of a 3-digit partial pattern and a 1-digit branch number, which is a total of 4 digits. The shape of the partial pattern is shown on the right side of the partial pattern number for reference. The character column shows the JIS code of the character formed as a result of combining the partial patterns in the partial pattern column and the shape of the character, and the shape is shown for reference. The detailed character dictionary number indicates the item number of the detailed character check dictionary 12, and indicates that the detailed character check dictionary 12 should be referred to. Further, in the character dictionary 11, character candidates are identified using three-digit partial patterns excluding branch numbers. Now, in the case of the example shown in FIG. 7 which is now being explained, FIG. 8 shows that the partial patterns 046, 197, and 018 are the character "member" or the character "uta". In this way, when a plurality of character candidates appear, the character recognition section 10 narrows down the candidates using the character detail check dictionary 12. FIG. 9 shows an example of the character detail check dictionary 12, and "No." indicates a registration number, which is the number indicated by the character dictionary 11 (FIG. 8). The number of candidates is the number of character candidates, and 2 is shown at No. m. This means that there are two candidates. "Partial patterns A, B" indicates which of the plurality of partial patterns is to be checked, and here 046 (mouth)
and 197 (eyes) to indicate checking. The branch number column will be explained later. The positional relationship indicates the positional relationship between partial patterns, and in No. m, A is above B, so based on the data based on the written characters, partial pattern 046 (mouth) is above partial pattern 197 (eyes). If it is, JIS in the character field
This is the character ``member'' represented by 3077 shown in the code column. Similarly, when partial pattern 046 (mouth) is to the left of partial pattern 197 (eyes), it is the character "uta". Partial pattern 046 as shown in Figure 3
(mouth) is above partial pattern 197 (eye), so the written character is "member". In order to check the mutual positional relationship between the partial patterns in this way, the coordinate data detected by the tablet 1 can be used. Next, the branch number column in FIG. 9 will be explained. As explained in FIG. 8, the character dictionary 11 identifies character candidates using three-digit partial pattern numbers without using branch numbers. This is because if ``mouth'' is written as ``〓'', it becomes ``〓'' and becomes unrecognizable, so in order to prevent it from becoming unrecognizable, two similar shapes are grouped together. For example, the partial pattern 005 is 005-1 (T), 005-2 (T), 005-3 (10), 005 as shown in No. 10 of Figure 5.
−4(). Therefore, for example, as is clear from FIG. 8, there are three characters consisting of partial pattern 005 and partial pattern 046: ``fortune'', ``kichi'', and ``kano''. In the case of these kinds of characters, especially when writing ``fortune'' and ``kichi'', people generally write them carefully while being aware of the differences between these characters.005
-2 (G) and 005-3 (10) are rarely mistaken. Therefore, there is a branch number section to identify characters by examining this feature. In No. n of FIG. 9, the branch number of partial pattern A is 2, which means that if the described positional relationship conditions are satisfied, it is the character ``Zan''. In this way, the characters "fortune", "kichi", and "kan" are identified from the partial pattern 005 and the partial pattern 046 using the three digits starting from No. n in FIG. In addition, in Figure 9 -
The parts marked with are not involved in identification. As is clear from the above description, the character recognition unit 10 recognizes and identifies the written character code and outputs it. Next, the duplicate dictionary (DJ) 8, which is a feature of the present invention regarding recognition of partial patterns, will be explained. Some characters have the same or similar shapes and cannot be distinguished by humans alone. These are called duplicate characters, and an example of a duplicate dictionary is shown in FIG. The numbers, alphabets, katakana, hiragana, kanji, and symbols listed in the columns of Figure 10 are examples of duplicate characters, and the range can be expanded depending on the degree of variation in writing method, but here The examples given are all written in exactly the same way, or with only slight differences, making it difficult to distinguish them from the letters alone. These letters are basically the 10th
Selected in the order shown in the diagram, that is, numbers, alphabets, katakana, hiragana, kanji, and symbols, but the immediately preceding character is an alphabet,
In the case of katakana, hiragana, kanji, symbols, or numbers, the partial pattern recognition unit 5 functions to select alphabetic characters, katakana, hiragana, kanji, and symbols as the first candidate. Of course, if the first candidate is not the character intended by the scribe, the scribe will use a switch (not shown) to select alphabetic, katakana, hiragana, kanji, or symbols, and print the desired output. Obtainable. In the above explanation, a dictionary format was used to store various types of information, but instead of using a dictionary format, it is also possible to use a processing format in which each piece of generated information is sequentially processed. That is, since the stroke dictionary 3, duplicate dictionary 8, character detail check dictionary 12, etc. have small contents, they do not necessarily need to be in the form of dictionaries. Further, in the present invention, the partial pattern recognition section 5 and the character recognition section 10 have been described as separate units, but it is also possible to process both processing units as a single unit without having to separate them. This can be easily realized by those skilled in the art, including the fact that it can be realized without using the dictionary format described above, so a detailed explanation will be omitted. As explained in detail above, this recognition method
By adopting the configuration shown in the figure, it is possible to easily recognize a large number of characters (more than 3000 characters), and all character types are mixed.
It is possible to provide an online character recognition device that requires less processing time for recognition processing. Furthermore, as those skilled in the art will easily understand, the present invention is not only realized by a hardware configuration, but can also be easily realized by an input section, a microcomputer, and a memory. It can greatly contribute to the company as a powerful input device in the man-machine interface. (Effects of the Invention) As explained above, the present invention provides a partial pattern detailed dictionary at the partial pattern stage in a method in which characters are hierarchically configured as strokes, partial patterns, and characters. Since character recognition is performed after verifying the necessary and sufficient conditions of That is, when the method of the present invention was not used, it was necessary to verify combinations for all candidate partial patterns in FIG. 5, but the number of combinations that need to be verified using the detailed partial pattern dictionary is now 10. The processing time required for recognition has been dramatically reduced. Next, the character dictionary is verified using the 3-digit representative code of the partial pattern, so even if you write a little carelessly, it will be correctly identified. Furthermore, if there are different characters in the same partial pattern representative codes, they can be easily distinguished by providing a detailed character dictionary. Furthermore, a redundant dictionary is set up that collects the same or similar characters, and the characters are identified based on the context of the characters, and if the identification result is not the character that the scribe intended, the scribe's intention is to identify the character that the scribe intended. By making it possible to obtain the desired output by selecting the character type, it is now possible to recognize a mixture of all characters, and it also eliminates the troublesome process of selecting the character type in advance and then writing the character. Therefore, it is possible to provide a device that can greatly escape from the constraints on the man-machine interface.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing an embodiment of the present invention, Fig. 2 is an explanatory diagram of strokes to be recognized, Fig. 3 is an input image diagram when writing the character "member", Fig. 4
Figures a and b are explanatory diagrams of a partial pattern dictionary, Figures 5 a, b, and c are partial pattern recognition state diagrams, Figure 6 is an explanatory diagram of a detailed partial pattern dictionary, and Figure 7 is an explanatory diagram showing partial pattern candidates. , Figure 8 is an explanatory diagram of the character dictionary, Figure 9 is an explanatory diagram of the character detail check dictionary,
FIG. 10 is an explanatory diagram of a duplicate dictionary. 1... Tablet, 2... Stroke recognition unit,
3... Stroke dictionary, 4... Stroke string register, 5... Partial pattern recognition unit, 6... Partial pattern dictionary, 7... Partial pattern detail check dictionary, 8... Duplicate dictionary, 9... Partial pattern string register. , 10...Character recognition unit, 11...Character dictionary,
12...Character details check dictionary.

Claims (1)

[Scope of Claims] 1. A tablet that extracts the strokes of handwritten characters and sequentially outputs them as stroke information, a dictionary that stores characteristics of characters, and recognition that performs recognition by comparing the stroke information and the contents of the dictionary. an online character recognition device for identifying handwritten characters, comprising: a stroke dictionary storing stroke information; and a stroke recognition device for identifying stroke information of handwritten characters output from the tablet using the stroke dictionary. A partial pattern dictionary that stores partial patterns consisting of combinations of strokes whose number is a maximum of a predetermined natural number n; and a partial pattern detailed dictionary that stores positional relationships between strokes so that a combination of strokes can become a partial pattern. and a step of comparing the combination of output information of the stroke recognition unit that is continuously outputted with the partial pattern dictionary, and extracting partial pattern candidates within a range not exceeding the natural number n strokes each time the stroke information is input. 1 and a second process of comparing the positional relationship of the output information of the stroke recognition unit with the detailed partial pattern dictionary, excluding those that do not meet the conditions from among the partial pattern candidates, and selecting a partial pattern. a pattern recognition unit; a character dictionary storing characters formed by combinations of partial patterns arranged in a predetermined code order; and a character dictionary that rearranges the combinations of output information from the partial pattern recognition unit in a predetermined code order and compares the character dictionary. An online character recognition device comprising: a character recognition unit for identifying characters, and identifying characters written on the tablet. 2. It has a tablet that extracts strokes of handwritten characters and sequentially outputs them as stroke information, a dictionary that stores characteristics of characters, and a recognition unit that performs recognition by comparing the stroke information and the contents of the dictionary, An online character recognition device for identifying handwritten characters, comprising: a stroke dictionary that stores stroke information; a stroke recognition unit that uses the stroke dictionary to identify stroke information of handwritten characters output from the tablet; and a predetermined natural number. Continuously outputs a partial pattern dictionary that stores partial patterns consisting of combinations of strokes with a maximum number of n, and a detailed partial pattern dictionary that stores positional relationships between strokes so that a combination of strokes can become a partial pattern. a first process of comparing the combination of output information of the stroke recognition unit and the partial pattern dictionary and extracting partial pattern candidates within a range not exceeding the natural number n strokes each time the stroke information is input; a partial pattern recognition unit having a second process of comparing the positional relationship of the output information of the recognition unit with the detailed partial pattern dictionary and selecting a partial pattern by excluding those that do not meet the conditions from among the partial pattern candidates; A character dictionary that stores characters made up of combinations of patterns arranged in a predetermined code order, and identification conditions for a plurality of characters made of combinations of the same partial patterns are stored with the number of character candidates added for each concatenated code with the character dictionary. and a character recognition unit that rearranges the combinations of output information from the partial pattern recognition unit into a predetermined code order and compares the character dictionary and the detailed character dictionary to identify characters; An online character recognition device characterized by identifying written characters. 3. It has a tablet that extracts strokes of handwritten characters and sequentially outputs them as stroke information, a dictionary that stores characteristics of characters, and a recognition unit that performs recognition by comparing the stroke information and the contents of the dictionary, An online character recognition device for identifying handwritten characters, comprising: a stroke dictionary that stores stroke information; a stroke recognition unit that uses the stroke dictionary to identify stroke information of handwritten characters output from the tablet; and a predetermined natural number. A partial pattern dictionary that stores partial patterns consisting of combinations of strokes with a maximum number of n, and registers similar partial patterns as groups with branch numbers, and positions between strokes so that a combination of strokes can become a partial pattern. A partial pattern detailed dictionary storing relationships, a combination of continuously outputted output information of the stroke recognition unit, and the partial pattern dictionary are compared, and each time the stroke information is input, a range not exceeding the natural number n strokes is determined. a first process of extracting partial pattern candidates; comparing the positional relationship of the output information of the stroke recognition unit with the partial pattern detailed dictionary; and selecting a partial pattern by excluding those that do not meet the conditions from among the partial pattern candidates; a character dictionary storing characters formed by combinations of partial patterns arranged in a predetermined code order; and a character dictionary storing identification conditions for a plurality of characters formed by combinations of the same partial patterns. A detailed character dictionary in which the number of character candidates is added and stored for each concatenated code, and the combination of output information from the partial pattern recognition unit is rearranged in a predetermined code order, and the combination of codes and the character dictionary are compared. , a character recognition unit that identifies a character by comparing the branch number information included in the output information of the partial pattern recognition unit and the position information for each partial pattern with the detailed character dictionary, An online character recognition device characterized by identifying characters that are