JPS63213094A - On-line character confirming device - Google Patents

Abstract

PURPOSE:To correctly select a character by testing the size relation and the positional relation of the candidate character of a character string when discriminate the character string inputted by hand writing into respective characters. CONSTITUTION:Every punctuation of the characters presumed among the hand written character string whose punctuations inputted from a tablet 1 are obscure characters are recognized based on the shape characteristic of one or plural strokes and the positional characteristic among the plural strokes included every period unit by a character recognition part 3, and the candidate of the character is obtained. The characters are recognized based on the discrimination of factors such as the number strokes detection 3a, stroke characteristic extracting 3b, positional relation decision among the strokes 3c, collation 3d with a dictionary 2 and character size detection 3e.

Description

[Detailed Description of the Invention] [Object of the Invention] (Industrial Application Field) The present invention is capable of effectively recognizing a plurality of characters continuously input by hand via a coordinate input device such as a tablet. Relating to an online character recognition device.

(Prior Art) Recently, online character recognition devices that recognize characters input by hand via a coordinate input device such as a tablet and are used for document creation and the like have been attracting attention.

This type of online character recognition device basically focuses on the stroke characteristics and combinations of strokes in a character pattern, and uses information on stroke characteristics and their combinations stored in a dictionary for each character and the tablet, etc. Characters are recognized by comparing the information on the writing strokes of the characters input through the computer.

However, various online character recognition devices that have been developed generally have a system that indicates character separation by inputting the next character after a predetermined period of time each time a character is input by hand, or by inputting a character in a character entry frame. A restriction is set such that one character can be written in each character entry frame. However, such restrictions impose a considerable burden on scribes, and are hindering the widespread use of online character recognition devices.

Therefore, recently, for example, online recognition of frameless written character strings using the r candidate character lattice method" (Electronic Communication Ugan Papers Vol. 1. J68-D No. 4. P2O3, 19
As introduced in 85), character boundaries are estimated from multiple freely written characters, that is, character strings with unknown character boundaries, and all combinations of strokes for one character are estimated. One idea is to recognize the characters from the computer, find candidate characters, and select characters from these candidate characters to generate an optimal character string.

However, in this method, an optimal character string is generated after selecting a character from candidate characters, so the calculation process becomes enormous. For this reason, a major problem remains in real-time processing.

Therefore, the present inventor previously proposed a character recognition device capable of solving the above-mentioned problems in Japanese Patent Application No. 81-301495. This character recognition device estimates character divisions from information on each stroke of a plurality of character patterns continuously inputted via a coordinate input device, and identifies one or more characters included in this estimated division unit. Character candidates for each delimiter are determined from the characteristics of the strokes and the positional relationships between the strokes.

After that, characters are selected from the obtained connection relationship of character candidates in each delimiter unit, and a character string is generated as a recognition result, for example, a combination of candidate characters in consecutive delimiter units with the smallest number of characters. This is how it was done.

According to such a method, a character string with unknown spaces between characters can be easily and effectively recognized, and a great practical effect can be achieved.

However, the size of the characters written by different scribes varies, and when trying to accommodate this, for example, the ``dakuten'' and ``ha'' characters are different.
This causes problems such as ``Yo'' and ``Yo'' being misrecognized, and ``Yo'' and ``Yo'' becoming indistinguishable. Furthermore, if the writing intervals between characters are narrow, "tono" may be mistakenly recognized as "tono".

(Problems to be Solved by the Invention) When recognizing each character from a string of characters that are continuously input by hand with unknown delimiters, there is a problem of misrecognition caused by the size of the characters and the spacing between the characters. However, improvements are desired in order to put a simple online character recognition device into practical use.

The present invention was made in consideration of these circumstances, and its purpose is to quickly and accurately recognize characters from freely written character strings with unknown character delimiters. An object of the present invention is to provide an online character recognition device that is suitable for practical use and can also correctly recognize voiced marks, syllables, etc.

[Structure of the Invention] (Means for Solving the Problems) The present invention estimates the delimitation of characters from information on each stroke of a plurality of character patterns that are continuously input by hand via a coordinate input device, and uses this estimation. 1 included in the delimited unit
After determining character candidates in each delimiter unit from the characteristics of one or more strokes and the positional relationship between the strokes, characters are selected from the connection relationship of the character candidates in each delimiter unit. When generating a character string as a recognition result that is a combination of candidate characters in a delimiter unit with the smallest number of characters, the size and positional relationships of the candidate characters in the character string are examined in particular, and the character is selected. It is characterized by being able to perform correctly.

−6= (Function) According to the present invention, for each character delimiter estimated from a character string with unknown delimiters, the shape of one or more strokes included in that delimiter unit, and the Character recognition is performed based on positional features, and candidate characters are determined. From the combinations of these candidate characters, when determining the character string that is the combination of consecutive delimited candidate characters and has the smallest number of characters as a recognition result, the size relationship of each candidate character in the character string and the The positional relationships are each verified.

As a result, candidate characters with many strokes are preferentially selected from among the candidate characters correctly determined for each character delimiter, and a character string is generated. It becomes possible to recognize character strings at high speed.

Furthermore, by examining the size relationship of candidate characters, it is possible to correctly recognize sulky sounds, etc., and it is possible to prevent erroneous recognition of, for example, "yo" and "yo".

Furthermore, by testing the size and positional relationships of candidate characters, it is possible to correctly recognize voiced marks, etc., making it possible to prevent erroneous recognition of ``tono'' as ``tono'' or ``tohano,'' for example. Become. As a result, it becomes possible to accurately recognize a character string freely input by hand.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings.

FIG. 1 is a schematic configuration diagram of an embodiment device. This device is
A coordinate input device (tablet) for inputting characters by hand as a series of coordinate values (strokes), a character recognition unit 3 that recognizes characters by comparing the stroke information with a dictionary 2, and a character string based on the character recognition results. and a character string generation section 4 that generates.

The dictionary 2 stores stroke information of characters to be recognized, for each character, as shown in FIG. 2, for example. For example, if the character type to be recognized is katakana,
The number of strokes is 2, and its shape is "F".
It is configured by registering information on the positional relationship of these strokes, as well as information such as `` and ノ''. Information on the positional relationship of these strokes is, for example, ``The starting point of ``is above the starting point of ``.'' r The starting point of ``fu'' is to the left of the starting point of ``.''. ” and other information.

For example, as shown in FIG. 3, the character recognition section 3 includes a stroke detection section 3a, a stroke feature extraction section 3b, a positional relationship determination section 3c between strokes, a dictionary dictionary section 3d, a character size detection section 3e, and a buffer 3f. It is composed of the following. The stroke detecting section 3a detects character strokes input by hand via the tablet 1 as a series of coordinate values, and detects the starting point and ending point, respectively. At the same time, character recognition processing is started assuming that the end point of the stroke is the end of the character.

If the character type to be recognized is, for example, katakana, and the recognition process is performed with each voiced and handakuten as one character, the maximum number of strokes is 4, so the end of the character assumed (estimated) as described above. (Character separation)
1 stroke from.

The steps are traced back to 2 strokes, 3 strokes, and 4 strokes, and each set of 1 to 4 strokes is compared with the stroke information registered in the dictionary 2.

For example, if "A" is input by hand, first the first
When the stroke "fu" is detected, it is presumed to be a character break and compared with the dictionary 2. As a result, "F" is found as a candidate character because it matches the stroke characteristics of the character "F".

After that, when the second stroke "ノ" is detected,
Recognition processing is performed on one stroke "NO" using the end point of the stroke as a character delimiter, and since it matches the stroke characteristics of the character "TONO", "NO" is determined as a candidate character. Furthermore, in this case, recognition processing is performed for the two strokes ``fu'' and ``no'' tracing back to the previous stroke using the end point of the second stroke as a delimiter.

In this case, from only the shape of the strokes, it corresponds to the characteristics of "A" r ka" registered in the dictionary 2, but only "A" is found as the candidate character from the positional relationship between the strokes. That is, if we pay attention to the positional relationship between the first stroke "F" and the second stroke "NO", (2) the starting point of "F" is above the starting point of "NO".

■The starting point of "fu" is to the left of the starting point of "ノ".

■The bending point of "fu" is to the right of the starting point of "ノ".

■The ending point of "fu" is near the starting point of "ノ".

The following information is obtained, and only the recognition target character "a" having features corresponding to these features is obtained as a candidate character. Regarding the character to be recognized, ``Chikara'',
Since the positional characteristics between the strokes are different, they are excluded from the candidate characters.

In this way, candidate characters up to the second stroke are determined.

In addition, if the katakana characters "F" and "ノ" are written in succession, there is a possibility that "A" will be required as a candidate character based only on the shape characteristics of these two strokes, but as mentioned above, Since the positional relationship between the strokes is also verified, the present device only requires ``fu'' and ``ノ'' as candidate characters.

After that, when the third stroke is detected, one stroke, two
Stroke and three-stroke character recognition is performed, and candidate characters are determined for each set of strokes.

When the fourth stroke is detected in the same way, the end point of the fourth stroke is used as the delimiter for characters, and then one stroke, two
Stroke, 3-stroke, and 4-stroke character recognition is performed. Then, when a recognition target character that matches the characteristics of each set of strokes (shape characteristics and positional relationship characteristics) is obtained from the dictionary 2, that character
Alternatively, it is obtained as a candidate character for a set of strokes. Note that if no candidate character is found, the process proceeds to the next stroke detection.

Thereafter, such processing is sequentially performed for characters with a maximum of four strokes. As a result, only one stroke or a candidate character of a combination of multiple strokes that matches the stroke characteristics of the recognition target character registered in the dictionary 2 from continuously input strokes is the delimiter of the estimated characters. Each can be reliably found as a candidate character in a unit. At this time, the size of each candidate character is determined by the character size detection section 3e.

For the candidate characters thus obtained by the character recognition section 3, the character string generation section 4 obtains a candidate character string that is a combination of candidate characters in which character delimiters are consecutive and has the minimum number of characters, The size and positional relationships of each candidate character in the candidate character string are examined to determine the recognition result of the character string continuously input by hand via the tablet 1.

In other words, the characters are separated so that all of the detected strokes are included in one of the characters in the character string, and the number of characters is the minimum, and the size relationship and positional relationship of each candidate character are determined. Character strings that pass the test are considered to be highly reliable recognition results.

Figure 4 schematically shows such a recognition process, and shows the recognition process when the katakana character string ``Kohokukukukatsutachō'' is continuously written as an example of an address name. .

With respect to a katakana character string that is continuously input by hand as shown in FIG. 4(a), its strokes are detected as shown in FIG. 4(b). Candidate characters as shown in FIG. 4(c) are obtained by character recognition processing for a set of one to four strokes, assuming that the end point of each stroke is a character delimiter.

From these candidate characters, the character string generation unit 4 calculates a character string in which character breaks are continuous, that is, all strokes are included in one of the candidate characters, and the number of characters is the minimum, as follows. .

In this example, the following two strings are found: ■ Kouhokukuri Tsutachicho ■ Kouhokukuganotacho.

This process is achieved, for example, by sequentially detecting candidate characters having the maximum number of strokes in successive character division units.

Thereafter, the size relationship and positional relationship of the candidate characters are tested for the candidate character string obtained in this manner. This verification process is performed by sequentially comparing the character sizes determined as described above and by examining the spacing between candidate characters from the position coordinates of the writing strokes.

Specifically, as shown in Figure 5, for the candidate character string ``Kouhokukurikitachou'', since there is a possibility that the candidate character ``tsu'' can be decomposed into ``i'' and ``rノ'', first "
From the size and position of "1X", confirm that it is not a misrecognition of "r".

Thereafter, it is checked whether "i" is close to the previous candidate character "chikara" or the decomposed character "r no" based on their positional relationship, and it is confirmed that it is not "ga".

As a result, it is verified that the candidate character "tsu" is not a character other than "rtsu".

On the other hand, for the candidate character "yo", the values before and after it = 15
- Compare the size of the candidate character to verify that it is "Yo". This test, for example, calculates the size and height a) of the nine candidate characters before and after it, and the size and height a) of the corresponding character, and also calculates the space dimension C) of the margin of the corresponding character with respect to the characters before and after it. . For example, if b/a < 315, it is determined that the corresponding character is a small character such as a consonant or consonant, and if c/a < 215, it is determined that the character is written lower than the preceding and following characters. , it may be determined that the corresponding character is a consonant or a consonant. In this way, regardless of the size of the handwritten characters themselves, it becomes possible to accurately recognize consonants and consonants based on the size ratio of the characters.

In addition, when testing the above-mentioned "tsu", the character height d
and the character width e, and determine whether the candidate characters that can be decomposed into ")1" and "ノ" satisfy the condition e/d < 815.
What is necessary is to test whether or not they form two characters, that is, whether or not "\\" and "rノ", which can be decomposed as described above, are sufficiently close in position.

As described above, this device estimates character divisions from the end points of strokes, and sequentially determines candidate characters based on the shape characteristics of the strokes in each division unit, and for combinations of multiple strokes, based on their positional relationships. , from the connection of these candidate characters, find a character string with consecutive character breaks and the minimum number of characters, and then test the candidate characters forming this character string according to the size relationship and positional relationship of the characters. We are looking for the final recognition result. Therefore, compared to the conventional process of finding the optimal character individually from among the candidate characters found by simply recognizing the shape characteristics of the stroke, the candidate character itself can be determined with high reliability, so it is possible to find the optimal character. Character strings can be obtained through very simple processing and in a short time. In addition, it is possible to accurately recognize consonants, consonants, etc. without erroneously recognizing voiced marks.

Therefore, it is possible to efficiently and accurately recognize character strings that have been continuously input by hand in real-time processing, which is extremely practical.

Note that the present invention is not limited to the embodiments described above. Although the character types to be recognized are described here as katakana, it is also possible to recognize character types other than katakana, and it is sufficient to introduce a similar method. Further, the above-mentioned test conditions may be determined empirically, for example, and the method thereof is not particularly limited. In addition, the algorithm for selecting a character string with the minimum number of characters can be modified in various ways, and in short, the present invention can be implemented in various modifications without departing from the gist thereof.

[Effects of the Invention] As detailed above, according to the present invention, it is possible to quickly and accurately recognize and process character strings freely written in any size and character spacing, with unknown character divisions. In addition to greatly improving recognition performance, the present invention has great practical effects, such as greatly reducing the burden on the scribe and improving its handling.

[Brief explanation of the drawing]

The figures show one embodiment of the present invention, in which Fig. 1 is a schematic configuration diagram of the embodiment device, Fig. 2 is a diagram showing an example of the configuration of a dictionary, and Fig. 3
4 is a diagram showing an example of the configuration of a character recognition section, FIG. 4 is a diagram schematically showing a processing procedure of a character recognition procedure, and FIG. 5 is a diagram for explaining a verification process for candidate character strings. l... Coordinate input device (tablet), 2... Dictionary,
Death...Character recognition unit, 3a...Stroke detection unit, 3b
...Stroke feature extraction unit, 8c...Positional relationship determination unit between strokes, 3d...Dictionary matching unit, 3e...
-Character size detection section, 3r...Buffer, 4...Character string generation section.

Claims (3)

[Claims] (1) A means for estimating character divisions from information on each stroke of a plurality of character patterns continuously inputted via a coordinate input device, and characteristics of strokes included in the estimated division units and stroke intervals. a character recognition unit that determines character candidates in each delimiter unit according to the positional relationship between the
and a character string generation unit that selects characters based on the size relationship of character candidates and generates a character string. (2) The character recognition unit recognizes the stroke characteristics and positional relationship characteristics between multiple strokes of the recognition target character registered in the dictionary, and the stroke characteristics included in the delimiter unit and the positional relationship characteristics between multiple strokes. Compare with
2. The online character recognition device according to claim 1, which obtains candidate characters in each delimiter. (3) The character string generation unit tests the connection of candidate characters in consecutive delimiting units based on the size relationship and positional relationship of each candidate character, and generates a character string that is a combination of the candidate characters and has the smallest number of characters. 2. The online character recognition device according to claim 1, wherein the on-line character recognition device is performed by selecting.