JPH11102254A - Pen type input device and pattern recognizing method therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To directly discriminate an input pattern from the inclination of a pen. SOLUTION: Two reflection type inclination sensors 11a and 11b respectively irradiate a writing plane, make reflected light incident from the writing plane and output signals showing the intensity of that reflected light. A dynamic planning(DP) matching part 23 performs DP matching to data showing a change in the intensity of reflected light detected by two reflection type inclination sensors 11a and 11b and data prepared in a pattern data storage part 22 and recognizes the moving pattern of a pen tip part.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pen-type input device for inputting figures, characters, and the like, and a pattern recognition method for the pen-type input device.

[0002]

2. Description of the Related Art Keyboards, mice, digitizers, light pens, tablets and the like are used as input devices such as computer devices. With the miniaturization of computer devices, the need for portable terminal devices has increased and the number of users has been increasing year by year. Therefore, a small input device has been required.

[0003] There is a limit in miniaturizing a keyboard in terms of a human interface, and it is not practical as an input device of a portable terminal device. Although a mouse can be downsized as a pointing device, it is not suitable for inputting figures, characters, and the like.

For this reason, a pen-type input device using a tablet and a pen is often used as an input device for a portable terminal device. In order to further reduce the size of the pen-type input device using the tablet, the size of the tablet becomes a problem. Therefore, for example, in a pen-type computer input device disclosed in Japanese Patent Laid-Open No. 6-67799, the direction and amount of movement of the input device are checked based on signals from two acceleration sensors, and the signals of the piezoelectric vibrating gyroscope are determined. The influence of the rotation of the input device on the moving direction and the moving amount detected by the acceleration sensor is detected and corrected.

[0005]

However, two acceleration sensors cannot be attached to the tip of the pen,
The mounting position of the acceleration sensor does not match the position of the pen tip. For this reason, in the pen-type computer input device disclosed in JP-A-6-67799, an error may occur in the detection result, and the locus of the pen tip on the paper surface may not be accurately detected. is there.

[0006] Further, by accumulating the above error,
It becomes more difficult to accurately detect the trajectory of the pen tip.

SUMMARY OF THE INVENTION The present invention has been made to solve such a disadvantage, and by detecting an input pattern directly from the inclination of a pen, it is possible to prevent errors from accumulating and making it difficult to determine the input pattern. It is intended to perform accurate recognition of characters and the like.

[0008]

A pen-type input device according to the present invention has two reflection-type tilt sensors, a data preprocessing unit, and a matching unit, and each of the two reflection-type tilt sensors has a writing surface. Irradiation, the reflected light from the writing surface is incident, a signal indicating the intensity of the reflected light is output, and a signal indicating the tilt of the pen by the intensity of the reflected light is output. A signal indicating the intensity of the reflected light from each of the reflection type tilt sensors is digitally converted, data indicating the intensity of the reflected light corresponding to the input pattern is cut out, and the matching unit calculates the intensity of the reflected light cut out by the data preprocessing unit. DP matching is performed on the data prepared in advance, and the movement pattern of the pen tip is recognized.

Another pen-type input device has two reflection-type inclination sensors, a data preprocessing unit, a neural network operation unit, and a post-processing unit, and each of the two reflection-type inclination sensors has a writing surface. Irradiation, the reflected light from the writing surface is incident, and a signal indicating the intensity of the reflected light is output. The data preprocessing unit digitally converts the signal indicating the intensity of the reflected light from the two reflection-type tilt sensors. Convert and cut out the data indicating the intensity of the reflected light corresponding to the input pattern, the neural network operation unit extracts the features of the input pattern based on the change in the data indicating the intensity of the reflected light cut out by the data preprocessing unit, The intermediate recognition result of the input pattern is calculated based on the features of the extracted input pattern and the result learned in advance, and the post-processing unit calculates the intermediate recognition result of the input pattern by the neural network operation unit. And determining an input pattern, conveniently using neural network calculation, and performs accurate pattern recognition.

Another pen-type input device has two reflection-type tilt sensors, a data preprocessing unit, a wavelet conversion unit, a neural network operation unit, and a post-processing unit.
Each of the reflection-type tilt sensors irradiates the writing surface, receives the reflected light from the writing surface, and outputs a signal indicating the intensity of the reflected light. The data preprocessing unit includes two reflection-type tilt sensors. Digitally converts the signal indicating the intensity of the reflected light from
The data indicating the intensity of the reflected light corresponding to the input pattern is cut out, and the wavelet transform unit performs a wavelet transform on the data indicating the intensity of the reflected light cut out by the data preprocessing unit to obtain a coefficient representing the input pattern, thereby obtaining a neural network. The arithmetic unit extracts the features of the input pattern based on the coefficients obtained by the wavelet transform unit, calculates the intermediate recognition result of the input pattern based on the extracted features of the input pattern and the result learned in advance, and the post-processing unit The input pattern is determined based on the intermediate recognition result of the input pattern by the neural network operation unit, and the pattern recognition rate is further improved.

The pen-type input device has a pattern recognition method that irradiates a writing surface from two points of a pen tip, makes reflected light from the writing surface incident, and obtains an intensity of reflected light corresponding to the input pattern. Is detected, and the data of the change in the intensity of the reflected light detected from two places is DP-matched with the data prepared in advance to recognize the movement pattern of the pen tip.

In another pattern recognition method for a pen-type input device, a writing surface is radiated from two places of a pen tip, reflected light from the writing surface is incident, and reflected light corresponding to the input pattern is reflected. Of the input pattern is extracted based on the data of the change in the intensity of the reflected light detected from two places,
An intermediate recognition result of the input pattern is calculated based on the extracted characteristics of the input pattern and a result learned in advance, and the input pattern is determined based on the calculated intermediate recognition result of the input pattern.

Another pen-type input device pattern recognition method is to irradiate a writing surface from two places of a pen tip portion, to input the reflected light from the writing surface, and to reflect the reflected light corresponding to the input pattern. , The data representing the change in the intensity of the reflected light detected from two locations is subjected to wavelet transform to obtain a coefficient representing the input pattern, and a neural network operation is performed based on the obtained coefficient to determine the characteristics of the input pattern. Extract,
An intermediate recognition result of the input pattern is calculated based on the extracted characteristics of the input pattern and a result learned in advance, and the input pattern is determined based on the calculated intermediate recognition result of the input pattern.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS A pen-type input device according to the present invention is for inputting characters, symbols, figures, and the like to a computer or the like. In particular, the pen inclination is obtained, and the input is made directly based on the obtained pen inclination. The detection error is eliminated by detecting the pattern.

The pen-type input device has, for example, a pen-type input unit and a pattern recognition device. The pen-type input unit includes two reflection-type tilt sensors provided at the tip of the pen such that the detection directions are substantially orthogonal to each other. The reflection-type tilt sensors respectively irradiate the writing surface, enter the reflected light from the writing surface, and output a signal indicating the intensity of the reflected light,
The tilt of the pen is detected by using the intensity of the reflected light. This utilizes the fact that the illuminance for irradiating the writing surface varies depending on the tilt of the pen, and detects the tilt of the pen from the intensity of the reflected light.

Further, the reflection type tilt sensor includes, for example, a light emitting element, two optical fibers, and a light receiving element, and applies light generated by the light emitting element to a writing surface via one optical fiber. Is incident on the light receiving surface via another optical fiber. This eliminates the need to provide a light emitting element and a light receiving element at the pen tip.

The pattern recognition section includes, for example, a data preprocessing section, a pattern data storage section, and a matching section. The data preprocessing unit converts the signal indicating the intensity of the reflected light from the pen-type input unit into a digital signal, and cuts out the data indicating the intensity of the reflected light corresponding to the input pattern. Here, the data indicating the intensity of the reflected light is cut out, for example, by making the pen core 19 slightly move up and down, contacting the core with the writing surface and moving the core upward, and detecting this with a microswitch or the like. To determine whether or not the document is being written.

The pattern data storage section previously stores a change pattern of the reflected light intensity when a standard pattern is drawn for a plurality of standard patterns. The matching unit performs pattern recognition by matching the pattern of the change in the data of the reflected light intensity extracted by the data preprocessing unit and the pattern stored in the pattern data storage unit by DP matching.
As a result, pattern recognition can be performed with high accuracy, and the number of prepared patterns can be reduced. Here, DP matching refers to dynamic programming (Dynami
This is a matching that performs normalization using c Programming), in which the distance between the data of the standard pattern and the detected data is calculated, and the pattern of the shortest data is used as the pattern of the recognition result.

Further, instead of having the pattern data storage section and the matching section, a wavelet transform section and a neural network operation section may be provided.

The wavelet transform section performs a wavelet transform on a signal indicating the intensity of the reflected light from the pen-shaped input section cut out by the data preprocessing section to obtain a plurality of coefficients representing the input pattern. Here, the wavelet transform is defined by convolving a function called a basic wavelet with a translation and an expansion / contraction operation,
It has the feature of localizing the square integrable function and its Fourier transform at the same time, makes it possible to create a real-time algorithm for calculating the coefficient sequence of the wavelet series, and can easily handle information on the occurrence time of events. Also,
In the wavelet transform, the number of terms required to represent a function value at one point is not significantly affected by the degree of discontinuity at another point. This is because the basis functions are localized. In addition, since local correspondence can be performed in an unsteady part, the boundary of the pattern can be detected.

The neural network operation unit extracts the features of the pattern to be recognized based on the coefficients obtained by the wavelet transform unit, and calculates an intermediate recognition result of the input pattern based on the features of the extracted input pattern and the result learned in advance. I do.
Here, the neural network operation unit is configured by a neural network that analyzes an input signal into some features based on cooperation, competition, and self-organizing ability, and integrates them. For example, the neural network operation unit actually inputs information to be recognized in advance, samples the data, inputs the data to the input layer, prepares neurons for the number of patterns to be classified into the output layer, and prepares each pattern and neurons. To correspond. Learning is performed by, for example, a back propagation method (back propagation method) so that when a certain information is input, the output of the corresponding neuron becomes “1” and the outputs of the other neurons become “0”. The neural network operation unit performs the neural network operation using the same coupling coefficient as the coupling coefficient thus obtained. The post-processing unit selects the one with the largest neuron output based on the intermediate recognition result of the input pattern of the neural network operation unit, determines the input pattern, increases the pattern recognition rate, and shortens the pattern recognition time.

[0022]

FIG. 1 is a block diagram of a pen-type input device according to an embodiment of the present invention. As shown in the figure, the pen-type input device has a pen-type input unit 1 and a pattern recognition device 2. As shown in FIG. 2, the pen-type input unit 1 has an Xs-axis reflection type inclination sensor 11a and a Ys-axis reflection type inclination sensor 11b having a light emitting element 12, optical fibers 13, 14 and a light receiving element 15, respectively. As shown in FIG. 3, the Xs-axis reflection type inclination sensor 11a and the Ys-axis reflection type inclination sensor 11b each have an Xs of a pen coordinate system (Xs, Ys, Zs) with the pen longitudinal direction as a Zs axis.
Light emitting elements 12a, 12a
b is applied to the writing surface via the optical fibers 13a and 13b, and the reflected light is incident on the light receiving elements 15a and 15b via the optical fibers 14a and 14b.
a and 15b output a signal indicating the intensity of the reflected light. By providing the Xs axis reflection type inclination sensor 11a and the Ys axis reflection type inclination sensor 11b in this way, it is possible to detect the inclination of the pen in the Xs axis direction and the Ys axis direction.
In addition, since light is transmitted using the optical fibers 13 and 14, the light emitting element 12 and the light receiving element 15 do not need to be provided at the pen tip, and an error due to the mounting position as in the case of using an acceleration sensor is prevented. Can be prevented.

The pattern recognition device 2 receives a signal indicating the intensity of reflected light from the pen-type input unit 1 through, for example, a cable and performs pattern recognition. The data preprocessing unit 21 and the pattern data storage unit 22 and a DP matching unit 23. The data pre-processing unit 21 is, for example,
As shown in FIG. 7, amplifiers 211a and 211b, A / D converters 212a and 212b, and noise filters 213a and 213a.
13b, and a data extracting unit 214a, 214b, etc., which digitally converts a signal indicating the intensity of the reflected light from the pen-type input unit 1 and cuts out data indicating the intensity of the reflected light corresponding to the input pattern. Here, the data extracting units 214a and 21
The cutout of the data indicating the intensity of the reflected light by 4b is performed, for example, by making the pen core 19 slightly move up and down, contacting the core 19 with the writing surface and moving the core upward, and detecting this with a microswitch or the like. This is performed by determining whether or not the user is currently writing.

The pattern data storage unit 22 previously stores a change pattern of the reflected light intensity when a standard pattern is drawn for a plurality of standard patterns. Here, the standard pattern includes a kanji pattern of the first level of Japanese Industrial Standard.

The DP matching unit 23 is a data preprocessing unit 2
The pattern of the change in the data of the reflected light intensity cut out by No. 1 and the pattern stored in the pattern data storage unit 22 are represented by DP.
Pattern matching is performed by matching. Since the pen tilt angle changes during writing, the input pattern is recognized based on the change pattern. As a result, pattern recognition can be performed with high accuracy, and the number of prepared patterns can be reduced. Here, the DP matching is a matching that performs normalization using dynamic programming (Dynamic Programming), and calculates the distance between the data of the standard pattern and the detected data to calculate the pattern of the shortest distance data. Is a pattern of the recognition result.

The pattern recognition device 2 may include a data pre-processing unit 21, a neural network operation unit 24, and a post-processing unit 25, as shown in FIG.

The neural network operation unit 24 includes a recurrent neural network 241 as shown in FIG. 6, and extracts and extracts features of an input pattern based on a change in reflected light intensity cut out by the data preprocessing unit 21. An intermediate recognition result of the input pattern is calculated based on the characteristics of the input pattern and a result learned in advance. Here, the neural network operation unit 23 is configured by a neural network that analyzes an input signal into some features based on cooperation, competitive action, and self-organizing ability, and integrates them. For example, the neural network operation unit 24 actually inputs information to be recognized in advance,
The data is sampled, these data are input to the input layer, neurons are prepared for the number of patterns to be classified into the output layer, and each pattern is associated with the neuron.
Learning is performed by, for example, a back propagation method (back propagation method) so that when a certain information is input, the output of the corresponding neuron becomes “1” and the outputs of the other neurons become “0”. The neural network operation unit 24 performs the neural network operation using the same coupling coefficient as the coupling coefficient thus obtained. The post-processing unit 25 selects the one with the largest neuron output based on the intermediate recognition result of the input pattern of the neural network operation unit 24 and determines the input pattern. By performing the pattern recognition by the neural network operation in this way, accurate pattern recognition can be performed with a simple configuration.

As shown in FIG. 7, the pattern recognition unit 2 includes a data preprocessing unit 21, a wavelet conversion unit 26,
A neural network operation unit 24 and a post-processing unit 25 may be provided.

The wavelet transform unit 26 performs a wavelet transform on the data indicating the intensity of the reflected light extracted by the data preprocessing unit 21 to obtain a coefficient representing the input pattern.

The neural network operation unit 24 extracts the features of the input pattern based on the coefficients obtained by the wavelet transform unit 26, and based on the extracted features of the input pattern and the result learned in advance, the intermediate recognition result of the input pattern. Is calculated. The post-processing unit 25 includes the neural network operation unit 2
The input pattern having the largest neuron output is selected based on the intermediate recognition result of the input pattern No. 4 to determine the input pattern.

The operation when the input pattern recognition is performed by the pen-type input device having the above configuration will be described with reference to the flowchart of FIG.

The pen-type input unit 1 outputs a signal indicating the intensity of light reflected from the writing surface to the data preprocessing unit 2 of the pattern recognition device 2.
Send to 1. When the user starts writing (step S
1) The data preprocessing unit 21 receives a signal indicating the reflected light intensity from the pen-type input unit 1, converts the input signal into a digital signal, cuts out the signal being written from the signal, and sends the signal to the wavelet conversion unit 26. Send (steps S2, S3).

The wavelet transform unit 26 normalizes the intensity of the reflected light cut out by the data pre-processing unit 21 using, for example, a time axis and a frequency (step S4). For example, the intensity of the reflected light is normalized on the time axis of 1 second, and the frequency is 1 Hz, 2 Hz,
When calculating the coefficients at 4 Hz and 8 Hz, the coefficients are each 1
, 2, 4, and 8 are obtained, for a total of 15 pieces. The intensity of the reflected light includes the intensity of the reflected light in the Xs axis direction and Y
Since there are two types of intensity of the reflected light in the s-axis direction, a total of 30 coefficients can be obtained. As described above, since the acceleration data is subjected to the wavelet transform and the simultaneous analysis of the time and the frequency is performed, the information of the occurrence time of the event can be easily handled. In addition, since local correspondence can be performed in an unsteady part, the boundary of the pattern can be detected.

The neural network operation unit 24 extracts the features of the input pattern based on the coefficients from the wavelet transform unit 23, and calculates the intermediate recognition result of the input pattern based on the extracted features of the input pattern (step S).
5). The post-processing unit 25 receives the input pattern intermediate recognition result of the neural network operation unit 24 and determines the input pattern (Step S6). This allows characters,
You can enter figures and their components,
The input pattern can be recognized accurately and quickly.

The pen-type input device operates as described above (step S2).
Steps S6 to S6) are repeated while the data preprocessing unit 21 cuts out data, and characters and the like are input (Step S7). As described above, by performing the recognition of the input pattern by directly using the signals from the reflection type tilt sensors 11a and 11b, it is possible to prevent the occurrence of erroneous recognition due to the accumulated error.

Here, the data preprocessing unit 21 may be provided with a memory for temporarily storing the data cut out by the data cutout unit 214. Thereby, the neural network operation unit 24 can be configured by a hierarchical neural network. In addition, by temporarily storing the acceleration data extracted by the data extraction unit 214, it becomes possible to recognize an input pattern requiring a large amount of time-series data.

In the above embodiment, the pattern recognition device 2
Provided one wavelet conversion unit 26 and one neural network operation unit 24, but two wavelet conversion units 26 and two neural network operation units 24 may be provided according to the number of inputs.

[0038]

As described above, according to the present invention, the writing surface is irradiated from two places of the pen tip, the reflected light from the writing surface is incident, and the intensity of the reflected light corresponding to the input pattern is reduced. Since the detection is performed, the inclination of the pen can be detected from two different directions.

Since the inclination of the pen is detected by using light, the writing surface may be separated from the sensor by using an optical fiber or the like, and the occurrence of accumulated errors due to the mounting position of the sensor can be prevented.

The data of the change in the intensity of the reflected light detected from two places is DP-matched with the data prepared in advance, and
Since the movement pattern of the pen tip is recognized, pattern recognition can be performed with a small number of patterns and with high accuracy.

The feature of the input pattern is extracted based on the data of the change in the intensity of the reflected light detected from two places, and the intermediate recognition result of the input pattern is obtained based on the feature of the extracted input pattern and the result learned in advance. Is calculated, and the input pattern is determined based on the intermediate recognition result of the calculated input pattern.
Pattern recognition can be performed with a simple configuration.

Further, since the data representing the change in the intensity of the reflected light detected from two places is subjected to wavelet conversion to obtain a coefficient representing the input pattern, information on the occurrence time of the event can be easily handled. In addition, since the wavelet transform enables local correspondence at an unsteady portion, it is possible to detect a pattern boundary.

Further, a neural network operation is performed based on the obtained coefficients to extract the characteristics of the input pattern, and an intermediate recognition result of the input pattern is calculated based on the extracted characteristics of the input pattern and a result learned in advance. Since the input pattern is determined based on the intermediate recognition result of the input pattern, the input pattern can be directly recognized from the data indicating the intensity of the reflected light.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a pen-type input device.

FIG. 2 is a configuration diagram of a reflection type tilt sensor.

FIG. 3 is a layout diagram of a reflection type tilt sensor.

FIG. 4 is a configuration diagram of a data preprocessing unit.

FIG. 5 is a configuration diagram of a pattern recognition device including a neural network operation unit.

FIG. 6 is a configuration diagram of a neural network.

FIG. 7 is a configuration diagram of a pattern recognition device including a wavelet operation unit.

FIG. 8 is a flowchart showing the operation of the pen-type input device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Pen-type input part 11 Reflection type inclination sensor 2 Pattern recognition part 21 Data pre-processing part 214 Data extraction part 22 Pattern data storage part 23 DP matching part 24 Neural network operation part 241 Neural network 25 Post-processing part 26 Wavelet conversion part

Claims (6)

[Claims] The present invention has two reflection-type tilt sensors, a data preprocessing unit, and a matching unit. Each of the two reflection-type tilt sensors irradiates a writing surface, and receives the reflected light from the writing surface. And outputs a signal indicating the intensity of the reflected light, and the data preprocessor converts the signal indicating the intensity of the reflected light from the two reflection-type tilt sensors into a digital signal, and indicates the intensity of the reflected light corresponding to the input pattern. The data is cut out, and the matching unit performs DP matching on data prepared in advance with data indicating the intensity of the reflected light cut out by the data preprocessing unit,
A pen-type input device for recognizing a movement pattern of a pen tip portion. 2. It has two reflection-type tilt sensors, a data pre-processing unit, a neural network operation unit, and a post-processing unit.
Each of the two reflection type tilt sensors irradiates the writing surface, receives the reflected light from the writing surface, and outputs a signal indicating the intensity of the reflected light. The signal indicating the intensity of the reflected light from the sensor is converted into a digital signal, and the data indicating the intensity of the reflected light corresponding to the input pattern is cut out.The neural network operation unit outputs the data indicating the intensity of the reflected light cut out by the data preprocessing unit. The features of the input pattern are extracted based on the change, and the intermediate recognition result of the input pattern is calculated based on the features of the extracted input pattern and the result learned in advance. The post-processing unit performs the intermediate recognition of the input pattern by the neural network operation unit. A pen-type input device for determining an input pattern based on a result. 3. A reflection type inclination sensor, a data pre-processing unit, a wavelet conversion unit, a neural network operation unit, and a post-processing unit, each of which illuminates a writing surface, The reflected light from the writing surface is incident, and a signal indicating the intensity of the reflected light is output. The data preprocessing unit converts the signal indicating the intensity of the reflected light from the two reflection type tilt sensors into a digital signal, The data indicating the intensity of the reflected light corresponding to the input pattern is cut out, and the wavelet transform unit performs a wavelet transform on the data indicating the intensity of the reflected light cut out by the data preprocessing unit to obtain a coefficient representing the input pattern, thereby obtaining a neural network. The arithmetic unit extracts the features of the input pattern based on the coefficients obtained by the wavelet transform unit, and based on the features of the extracted input pattern and the results learned in advance, the input pattern A pen-type input device, wherein an intermediate recognition result is calculated, and the post-processing unit determines the input pattern based on the intermediate recognition result of the input pattern by the neural network operation unit. 4. A writing surface is irradiated from two places of a pen tip portion,
The reflected light from the writing surface is incident, the intensity of the reflected light corresponding to the input pattern is detected, and the data of the change in the intensity of the reflected light detected from two places and the data prepared in advance are converted to DP.
A pattern recognition method for a pen-type input device, comprising matching and recognizing a movement pattern of a pen tip. 5. A writing surface is irradiated from two places of a pen tip portion,
The reflected light from the writing surface is incident, the intensity of the reflected light corresponding to the input pattern is detected, and the characteristic of the input pattern is extracted based on the data of the change in the intensity of the reflected light detected from two places. A pen-type input device characterized in that an intermediate recognition result of an input pattern is calculated based on characteristics of an extracted input pattern and a result learned in advance, and an input pattern is determined based on the intermediate recognition result of the calculated input pattern. Pattern recognition method. 6. A writing surface is irradiated from two places of a pen tip,
The reflected light from the writing surface is incident, the intensity of the reflected light corresponding to the input pattern is detected, and the data of the change in the intensity of the reflected light detected from two places is subjected to wavelet transform to obtain a coefficient representing the input pattern. Is calculated, and a neural network operation is performed based on the obtained coefficients to extract the features of the input pattern.The intermediate pattern recognition result of the input pattern is calculated based on the features of the extracted input pattern and the result learned in advance. A pattern recognition method for a pen-type input device, wherein an input pattern is determined based on an intermediate recognition result of (1).