JPS6242290A - Recognizer for handwritten character - Google Patents

Abstract

PURPOSE:To attain a learning function based on variance data intrinsic to a writer by obtaining this data for each partial character pattern serving as a component element of a character recognized mistakenly in case the result of recognition is wrong. CONSTITUTION:A writer supplies a character to a tablet handwriting input part 1 in a correct order of strokes. Thus a stroke recognizing part 2 obtains normalized stroke data and this data is recognized by a stroke dictionary part 6. Thus a basic stroke code is obtained. These stroke data and code are separated from each other by a partial character pattern recognizing part 3. Then the partial character pattern code is obtained by a partial character pattern dictionary part 7. These candidate groups are compared with each character of a character pattern dictionary part 8 by a character pattern recognizing part 4 to be specified and delivered to a display part 5. When the character pattern is recognized mistakenly, a partial character pattern variance detecting part 10 compares the normal pattern with the candidate groups of partial character patterns recognized mistakenly for detection of the variance data intrinsic to the writer. This variance data is stored in a partial character pattern variance dictionary 9. Thus the correct recognition is possible after the variance data on the writer is once stored even if a character is supplied in a wrong order of writing strokes.

Description

DETAILED DESCRIPTION OF THE INVENTION (A) Field of Industrial Application The present invention relates to a handwritten character recognition device that can recognize handwritten characters.

(B) Prior Art In recent years, various literary magazine A1 devices have been developed in which characters, figures, etc. are inputted onto a tablet using a pen-type jig, and the input information is recognized and used as input information. In this type of character recognition device, erroneously recognized characters or unrecognized characters occur due to variations in the stroke shape, number of strokes, and stroke order of input characters.

As a method to reduce the number of misrecognized and unrecognized characters, ■ Preparing the contents of the magazine dictionary to accommodate changes in stroke shape, number of strokes, and stroke order = data that may occur for each character; ■ Recognition. It is conceivable to use an algorithm that takes fluctuations into account. However, in method (■), the recognition dictionary becomes enormous, and it is realistically impossible to describe all possible variations. In method (2), more recognition processing than necessary occurs because all variations are taken into account regardless of the variation characteristics of each character. To solve this problem, ■ have a Japanese dictionary and perform word processing from the characters before and after the character to be recognized; ■ store the misrecognized characters and other regular input character data in correspondence; It is possible to learn the variations unique to each scribe. However, method (■) requires a Japanese dictionary and also has problems with processing speed, while method (■) can be expected to have a learning function for characters that are misrecognized.

It has no effect on other characters and does not improve the overall recognition rate.

(c) Problems that the invention attempts to solve The present invention is based on the above considerations.

The object of the present invention is to provide a character recognition device with a simple configuration, which enhances the learning function in the case of misrecognition, can recognize variable characters unique to the author.

B) Means for Solving the Problems When the recognition result is an erroneous recognition, the handwritten character recognition device of the present invention obtains fluctuation data unique to the writer for each character partial pattern that is a constituent element of the erroneously recognized character. , which has a learning function based on this.

(E) Function According to the handwritten character recognition device of the present invention, due to its learning function, characters once recognized will not be misrecognized thereafter;
It is even possible to prevent the recognition of other characters that share the unrecognized character part of the incorrect W character.

(F) Embodiment FIG. 1 shows the configuration of a handwritten character recognition device of the present invention. In the figure, (1) is the tablet handwriting input section, (2)
(3) is a character part pattern recognition unit, (4) is a character pattern recognition unit, and (5) is a display unit.

(6) is a stroke dictionary section, (7) is a character partial pattern dictionary section, and (8) is a character pattern dictionary section, and the basic operation of handwritten character recognition is performed by one or more of these components.

That is, as shown in Fig. 2, when a writer inputs the character "right" in the correct stroke order into the tablet handwriting input section (1), the input section (1) outputs the trajectory of the pen as a coordinate signal and calculates the stroke. The recognition unit (2) receives this. At this time,
The stroke recognition unit (2) separates the coordinate signal sequence for each stroke from the time series of coordinate signals to obtain normalized stroke data for each stroke, and converts this into stroke data for which a large number of stroke data are stored in advance. Dictionary Department (6
) to obtain each basic stroke code. In this case, the first to fifth strokes [)], (-), (1], (
Basic stroke codes corresponding to (-) and (-) can be obtained.

Each of the basic stroke codes obtained in this way is input to a character part pattern recognition unit (3), which combines and separates these code strings, so that a large number of character part patterns can be recognized in advance as basic strokes. A character part pattern code is obtained by selecting a candidate group of character part patterns that may be established as a character based on each character part pattern in the character part pattern dictionary section (7) that is expressed and stored in the form of a code string. . In this case, as shown in FIG. 2 (1)), nine character partial pattern codes are obtained corresponding to the character partial pattern 人 [su] and the character partial pattern B [mouth]. Note that this character pattern part (7) has a frequently occurring stroke shape.

In order to cope with variations in stroke order, number of strokes, etc., character part patterns corresponding to variations are stored in the character part pattern dictionary section (7) in advance, and stroke order-free or stroke-number-free processing is performed. However, there is no such measure against fluctuations for low t.

The character part pattern code string for each of the candidate groups thus obtained is input to the single character pattern recognition unit (4),
The recognition unit (4) compares each character pattern based on each character pattern in the character pattern dictionary unit (8) in which a large number of character patterns are expressed and stored in the form of character part pattern board rows in advance, and selects one of the candidate groups. A character pattern that is likely to be established as a character is identified from the above, and a character pattern code is output to the display section (5).

In this case, as shown in FIG. 2(C), a character pattern code corresponding to the character "right" is output, and "right" is displayed in the i-receiving part (5).

In the above configuration, when the recognized handwritten characters are misrecognized, it is necessary to correct this, and in history, the configuration is designed to be able to absorb changes in the stroke shape, stroke order, number of strokes, etc. that are unique to the author. Therefore, in addition to the above-mentioned configuration, the present invention has a character part pattern variation dictionary part (9), a character part pattern variation detection part C1Cl, and a regular character input part a.
υ is provided.

The regular character input section αυ is used when there is a misrecognition.

■ Request rewriting on tablet (1).

(2) Relax the recognition degree in each recognition unit f2+, (31, (41), display a plurality of candidate characters on the display unit (5), and allow the writer to select a regular input character from among them.

Then, the erroneously recognized characters displayed in the 0 display section (5) are changed to regular characters.

The character partial pattern variation detection unit (II is a combination of the regular character partial pattern of the regular character obtained from the regular character input unit αυ and the recognition result of the erroneously recognized handwritten character by the character partial pattern recognition unit (3)). A candidate group of character part patterns stored in the character part pattern dictionary (7) that has become,
along with unrecognizable character part pattern codes that led to misrecognition.

Detect the author's own fluctuation data. For example '1JIJ3
As shown in the example above, when trying to input the single character "right", the stroke order of the first and second strokes is switched,
1st stroke (-), 2nd stroke ()], this character part pattern A [su] will be detected as an unrecognizable character part pattern, and along with this pattern code, the Stroke order variation data can be obtained.

The character part pattern variation detection unit (9) is the character part pattern variation detection unit! The author's own variation data in this character part butter/ is stored in association with the unrecognizable character part pattern code obtained from 1ω. For example, in the example of the character "right" mentioned above, the stroke order free variation data for the character part butter / "su" is stored in this dictionary section (9), and the character part pattern [ This variation data becomes the link to be combined with the code.

Once the author's work data is stored in this way,
As shown in Figure 3, even if the character "stone" is entered in the wrong stroke order into the tablet's handwriting input section (1), the 9-character partial pattern recognition section (3) will add it to the character partial pattern dictionary section (7). This will also refer to the variation data in the character part pattern variation dictionary section (9), so enter the letters that are "right" in the normal stroke order as shown in Figure 1gIIz, and enter the one character part pattern [su] as at 9 o'clock. Recognition is performed, and as a result, the correct character [right] is displayed on the display section (5). In this case, one character part pattern A [
When inputting the character 〔〕 that has 〕, ′! PJ4 diagram (
Even if the stroke order of this character partial pattern A [su] is incorrect as shown in IL), the recognition result of the character [a] can be obtained as shown in Figure 4 (C).

(G) Effects of the Invention As is clear from the above description, the handwritten character recognition device of the present invention converts the writer's own variation data into character part patterns when there is an erroneous recognition by comparing the normal character and the erroneously recognized character. Since it is calculated in units and stored, the recognition rate can be improved without causing erroneous recognition for the same fluctuation in subsequent handwriting input. Moreover, since the author's unique learning function for variable data is performed not for each character but for each character partial pattern, the memory capacity for storing this variable data can be small, reducing the cost of increasing the size of the device. It doesn't invite up.

[Brief explanation of drawings]

Figure 1 shows the components of the handwritten character recognition device of the present invention, Figure 82 (a-) (1)) (Q), Figure 3, and Figure 4 (tL).
(b) (0) No husband? FIG. 1 is a schematic diagram showing the process of handwritten character recognition according to the present invention. (11...Tablet handwriting input section, (2)...Stroke recognition section, (3)...Character part pattern recognition section, (4)...Character pattern recognition section, (5)...Display Part, (7)... Character part pattern dictionary part, (9)...
-Character part pattern variation dictionary part. α1...Character partial pattern variation detection section, aυ...Regular character input section.

Claims (1)

[Claims] 1) In a handwritten character recognition device that recognizes handwritten character information in character partial pattern units and combines them to recognize characters, a regular character input section that inputs regular characters when an erroneous recognition occurs; A character part pattern variation detection unit that detects variation data unique to the author by comparing the character part pattern of the regular character from the input unit and the character part pattern of the misrecognized handwritten character, and the variation data obtained from the detection unit. and a character part pattern variation dictionary section that stores character part pattern variations, and performs handwritten character recognition that allows for variations unique to the author by referring to the variation data of the character part pattern variation dictionary part during recognition processing in character part pattern units. A handwritten character recognition device featuring: