JPH0619449A - Handwritten character output device - Google Patents

Abstract

PURPOSE:To automatically generate handwriting-like characters with small memory capacity. CONSTITUTION:When >=1 handwritten characters are inputted from a handwritten character input part 1, a feature extracting means 6 extracts features from the handwritten characters and stores them in a feature storage means 7, the handwriting-like characters are automatically generated from a standard font data part 4 for character information to be outputted according to the feature information on the characters stored in the feature storage means 7, and the handwriting-like characters on which the features of the handwritten characters are reflected are outputted from an output means.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a handwritten character output device capable of recognizing handwritten characters and reflecting the handwriting of a handwritten character input person.

[0002]

2. Description of the Related Art In recent years, computers have come into widespread use as indispensable devices for business such as document creation devices.

FIG. 8 is a block diagram showing an example of this type of document creating apparatus.

In the figure, 31 is a control unit for controlling the entire apparatus, which controls each unit based on a control program stored in a storage medium such as a ROM (not shown). A font memory (Font memory) 32 stores the data of the characters to be output, for example, in a dot format or an outline format corresponding to each character code. When a plurality of typefaces (Mincho typeface, Gothic typeface, etc.) can be used, the above data are stored for each typeface. A keyboard 33 is used to input desired character information. A display unit 34 displays character information and control information input from the keyboard 33. A text memory 35 stores the input character information. A printer 36 prints out character information and the like expanded in the text memory 35.

In the document processing, the text entered in the keyboard 33 or the digitizer by the control unit 31 is used as the text memory 35.
It is configured to be stored on the screen and displayed on the display unit 34 for editing using the font (character style) information of the font memory 32, or to be printed by the printer 36. Also, illustrations and drawings can be edited in the same manner by a digitizer or a pointing device such as a mouse (not shown).

When characters are output from a device such as a computer in this way, the fonts such as Mincho, Gothic, cursive, and round fonts registered in the font memory 32 are used. As shown in Japanese Patent Publication No. 219945 or the like, the handwritten character of the writer himself is registered as a font in the memory and output is performed, or the handwritten character itself is treated as image information.

[0007]

However, according to the conventional character processing, if fonts such as Mincho typeface, Gothic typeface, cursive typeface, and round typeface type are used, the same typeface is always output regardless of the writer of the sentence. Therefore, it is suitable for business-like document creation because it does not reflect the personality of the writer's character, but there is a problem that it is not suitable for personal letters etc. to transmit even the writer's personality. .

In addition, when the letter character of the writer himself is used as a font, although the character of the writer character can be reflected,
There was a problem in registering the characters to be used for output, even if the kanji were decomposed into crepe and crease to reduce the registration amount.

Further, when handwritten characters are treated as image information, there are many problems that a large memory is required to record a long sentence even if the image compression technique is used, and it cannot be easily edited. there were.

The present invention has been made to solve the above-mentioned problems, and analyzes a small amount of handwritten character input information to generate a handwritten character while transforming existing character information, thereby reducing the memory capacity. It is an object of the present invention to obtain a handwritten character output device capable of automatically generating handwritten characters with.

[0011]

A handwritten character output device according to the present invention extracts an input means for inputting a handwritten character and a feature of a handwritten character of a writer from one or more handwritten characters input from the input means. Extracting means, storage means for storing the characteristic information of the characters extracted from the extracting means, and generation for automatically generating handwritten characters from predetermined font data based on the characteristic information of the characters stored in the storing means. And means for outputting the handwritten character generated by the generator.

Further, the extraction means, the storage means and the generation means are constructed by a neural network.

[0013]

In the present invention, 1 input from the input means
When more handwritten characters are input, the extraction means extracts the characteristics of the handwritten characters and stores them in the storage means, and outputs the handwritten character based on the characteristic information of the characters stored in the storage means with respect to the output character information. By automatically generating from the predetermined font data, it becomes possible to output from the output means handwritten characters reflecting the characteristics of the characters of the writer.

Further, by constructing the extraction means, the storage means, and the generation means by a neural network, it becomes possible to output a handwritten character reflecting the characteristics of the character of the writer from any input print character.

[0015]

1 is a block diagram showing the configuration of a handwritten character output device according to an embodiment of the present invention, in which the same parts as those in FIG. 8 are designated by the same reference numerals.

In the figure, reference numeral 1 is a handwritten character input unit, which is composed of, for example, a digitizer. Reference numeral 5 is a handwriting-like character control unit, which extracts and stores the characteristics of the font using one or more of the handwritten characters input from the handwriting character input unit 1.
Generates handwritten characters that reflect the characteristics of the font when outputting characters.

The handwritten character processing operation in the handwritten character output apparatus according to the present invention will be described below with reference to the detailed block diagram shown in FIG.

2 and 3 are detailed block diagrams for explaining the essential structure of the handwritten character output device shown in FIG.

As shown in FIG. 2, handwritten characters are input in the handwritten character input unit 1 such as a digitizer. The character recognition unit 2 for the input handwritten character
The character corresponding to the handwritten input character is converted from the standard font data section 4 by performing character recognition on the keyboard or by designating the handwritten input character from the keyboard 3. call.

The feature extraction means 6 of the handwritten character control unit 5 uses the character information 9 from the handwritten character input unit 1 and the character information 10 from the standard font data unit 4 to identify the handwriting character specific to the writer. Extract features. The extracted feature amount is stored in the feature storage means 7 of the handwritten character control unit 5. The stored characteristic amount information can be stored in another storage medium or stored in the characteristic storage means 7 of the handwritten character control unit 5 of another device of the present invention for use. To output a handwritten character using the stored handwritten character feature, a character input from the keyboard 3
Alternatively, in accordance with the characters sequentially read from the output text 11, the character information 12 output from the standard font data unit 4 is called and input to the handwriting-style character generation means 8 of the handwriting-style character control unit 5 to store the stored handwritten characters. The handwritten character is generated using the feature amount, and the handwritten character information 13 is output.

FIG. 3 shows an input data conversion section 14, a teacher data conversion section 15, and output data as means for realizing the feature extraction means 6, the feature storage means 7, and the handwritten style character generation means 8 of the handwriting style character control section 5. This corresponds to an example using the learning type neural network 17 having the conversion unit 16.

In the handwritten character output device configured as described above, when one or more handwritten characters input from the input means (handwritten character input section 1) are input, the extraction means (feature extraction means 6) handwrites. Character features are extracted and stored in storage means (feature storage means 7), and handwritten characters are stored in predetermined font data (standard font data section) based on the character information stored in the storage means for the output character information. By automatically generating from 4), it is possible to output, for example, a handwritten character reflecting the characteristics of the writer's character from the printer.

Further, the extraction means, the storage means, and the generation means are constructed by a neural network (learning type neural network 17), and handwritten characters reflecting the characteristics of the characters of the writer are output from any input print characters. Is possible.

The learning type neural network 17 automatically acquires the relationship between the input and the output in the learning phase as a time series, stores the relationship as the structure of the neural network itself (coupling amount between layers), and finishes the learning. By giving the subsequent input data, the output according to the relationship between the input and the output obtained even for the unlearned data is generated, so that the characteristic extraction means 6, the characteristic storage means 7, and the handwritten character generation means 8 can be covered. The input data conversion unit 14 uses the standard font data unit 4
The character information 10 and 12 are converted into data suitable for input to the learning neural network 17. In FIG. 3, the character information 10 and 12 from the standard font data section 4 are drawn independently, but they may be realized by a common signal line. The teacher data conversion unit 15 converts the character information 9 from the handwritten character input unit 1 into data suitable for teacher data in the learning phase of the learning neural network 17. The output data conversion unit 16 converts the output data generated by the learning neural network 17 into the handwritten character information 13.

On the other hand, in FIG. 2, a handwritten character control unit 5
The feature extracting means 6 is realized by using the learning phase of the learning type neural network 17. The handwritten character information 9 input from the handwritten character input unit 1 is converted into teacher data by the teacher data conversion unit 15, and the character information 10 from the standard font data unit 4 of the character corresponding to the input handwritten character is input. The data conversion unit 14 converts the input data for learning and inputs the learning input data to the learning-type neural network 17 to perform learning (see FIG. 4), thereby acquiring the characteristics unique to the writer of handwritten characters.

FIG. 4 shows the input data converter 1 shown in FIG.
4 is a schematic diagram showing a learning process of a character converted into bitmap data by a teacher data conversion unit 4 and a teacher data conversion unit 15.

As shown in this figure, the learning input data 18 corresponds to the output from the input data conversion unit 14, and the teacher data 19 corresponds to the output from the teacher data conversion unit 15.

Since the feature amount storage means 7 of the handwritten character control unit 5 shown in FIG. 3 stores the input / output relationship acquired by the learning type neural network 17 as the connection amount between layers as described above. It is realized as the storage element group. It is also possible to install only this memory element group outside the learning type neural network 17. In addition, it is possible to read the coupling amount between layers and save it in another storage medium, or copy and use the information in another device.

Further, the handwritten character control unit 5 shown in FIG.
The handwritten character generation means 8 is realized as a learning type neural network 17 which has acquired the characteristics peculiar to the writer of handwritten characters through the learning phase. The learning type neural network 17 is provided by converting the character information 12 from the standard font data unit 4 according to the character designated by the keyboard 3 or the output text 11 by the input data conversion unit 14.
Output data obtained by passing through the learning-type neural network 17 and converted by the writer's unique characteristics, and converted by the output data conversion unit 16 Character information 13 is obtained. As described above, the learning-completed neural network outputs output data converted by the input / output relationship acquired by learning when given arbitrary input data for output, and therefore any neural network that reflects the peculiar characteristics of the handwritten character You can get the output of handwritten characters.

FIG. 5 shows the input data converter 1 shown in FIG.
4 is a schematic diagram showing an example of handwritten characters generated as bitmap data by a learning-type neural network 17 using the character information converted into bitmap data in 4. FIG.

As shown in this figure, the output input data 20 corresponds to the output from the input data conversion unit 14, and the output data 21 corresponds to the output from the learning type neural network 17.

In the above embodiment, the input data conversion unit 1
4. The output of the teacher data conversion unit 15 is not particularly limited, but the spectrum of the two-dimensional Fourier transform is output as the output of the input data conversion unit 14 and the teacher data conversion unit 15, and the two-dimensional spectrum is output as the conversion output of the output data conversion unit 16. The character pattern may be output by inverse two-dimensional Fourier transform of the Fourier spectrum.

The feature extraction processing operation of the font in the handwritten character output device according to the present invention will be described below with reference to FIGS. 6 and 7.

6 and 7 are schematic diagrams for explaining the operation of the feature extracting means 6 and the feature storing means 7 shown in FIG. In particular, FIG. 6 corresponds to the case where the learning type neural network 17 is used to acquire the character characteristics of the writer, and FIG. 7 is used to deform the character using the learning type neural network 17 and output a handwritten character. Corresponds to the case.

The learning type neural network 17 prepares data representing the relationship between the input and the output even if the relationship between the input and the output is unclear, and performs learning by using the output data corresponding to the input data as the teacher signal. , The relationship between input and output is automatically acquired. In the present embodiment, for example, one or more handwritten characters input by a device for inputting handwritten characters of a writer, such as a digitizer, is used as teacher data for learning a neural network, and the input data is a standard corresponding to handwritten characters. By learning using the printed character 51 (see FIG. 6), the characteristics of the character of the writer can be acquired. The learning-type neural net 17 stores the acquired input / output relationship as a net connection amount, and can read out and save the same connection amount or copy the information to another device of the same type.

On the other hand, the learning type neural network 17 shown in FIG. 7 which has been learned, when given arbitrary data to the input,
It is possible to obtain an output converted according to the acquired input / output relationship. Therefore, by inputting an arbitrary print character 53 to the neural network, it is possible to obtain the output of the handwritten character 54 that reflects the characteristics of the handwritten character of the writer.

In the above embodiment, the case where the handwritten character is output by acquiring the feature of how the character collapses based on the printed character as the feature of the individual character is shown. The following (1), (2 The present invention can also be applied to an apparatus that acquires a feature on the basis of a printed character by any of the above.

(1) An apparatus that uses the thickness and direction of a line of a printed character as reference information as the characteristic of an individual character and uses the characteristic of the deviation from the reference information.

(2) A device that uses a method of change in a two-dimensional spectral pattern with respect to a printed character as a feature of an individual character.

[0040]

As described above, according to the present invention, when one or more handwritten characters input from the input means are input, the extraction means extracts the characteristics of the handwritten characters, stores them in the storage means, and outputs them. With respect to the character information, a handwritten character is automatically generated from predetermined font data based on the character characteristic information stored in the storage means, and the handwritten character reflecting the character characteristics of the writer is output from the output means. Since it is configured, it is possible to easily generate a handwritten character reflecting the characteristics of the writer from a small amount of input handwritten character information. Further, since the extraction means, the storage means, and the generation means are configured by the neural network, it is possible to output the handwritten character reflecting the characteristics of the character of the writer from any input print character.

Therefore, by transforming the existing print character based on the feature information of the character extracted from the small amount of input handwritten character information, it is possible to easily obtain the character reflecting the handwritten character of the writer. it can. In addition, since it is not necessary to perform the process for all the characters when acquiring the characteristics of the characters, it is possible to significantly reduce the character labor load of the writer. Further, since it is not necessary to store all handwritten characters used by the writer, the memory ratio in the handwritten character feature information can be reduced, the device cost can be reduced, and the memory can be used for other purposes. Produce the effect of.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of a handwritten character output device according to an embodiment of the present invention.

FIG. 2 is a detailed block diagram illustrating a main part configuration of the handwritten character output device shown in FIG.

FIG. 3 is a detailed block diagram illustrating a main part configuration of the handwritten character output device shown in FIG.

FIG. 4 is a schematic diagram showing a learning process of a character converted into bitmap data by the input data conversion unit and the teacher data conversion unit shown in FIG.

5 is a schematic diagram showing an example of handwritten characters generated as bitmap data by a learning type neural network using the character information converted into bitmap data by the input data converter shown in FIG.

FIG. 6 is a schematic diagram for explaining the operation of the feature extraction means and the feature storage means shown in FIG.

FIG. 7 is a schematic diagram for explaining the operation of the feature extraction means and the feature storage means shown in FIG.

FIG. 8 is a block diagram showing an example of this type of document creation apparatus.

[Explanation of symbols]

 1 Handwritten character input unit 2 Character recognition unit 3 Keyboard 4 Standard font data unit 5 Handwritten character control unit 6 Feature extraction unit 7 Feature storage unit 8 Handwritten character generation unit

Claims (2)

[Claims] 1. An input unit for inputting handwritten characters, an extracting unit for extracting characteristics of a handwritten character of a writer from one or more handwritten characters input from the input unit, and a character extracted by the extracting unit. A storage unit for storing the characteristic information, a generation unit for automatically generating a handwritten character from predetermined font data based on the character characteristic information stored in the storage unit, and a handwritten character generated by the generation unit. A handwritten character output device, comprising: 2. The handwritten character output device according to claim 1, wherein the extraction means, the storage means and the generation means are constituted by a neural network.