JPH1021340A - Handwritten character recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device recognizing a character peculiar to an individual, the wrong order of strokes and continuous character by storing input patterns in a memory and updating a standard dictionary from stored data. SOLUTION: A temporary registering dictionary 70 for storing the input patterns is provided within a handwritten character recognizing device 10. Then whenever a character pattern is inputted, similarity with each standard pattern in a standard dictionary 50 is calculated to register the number, a distance value, a character code and a registered date of a standard pattern judged to be at a shortest distance in the dictionary 70. In addition, the number within the dictionary 50, coordinate information, the character code, the number of erroneously recognizing times and the registered date (updated date) of the standard pattern can be registered. Then this device generates and updates the standard pattern within the dictionary 50 from stored data every prescribed time. And the device counts the number of the recognizing times with each standard pattern and when the number is large, erases the standard pattern.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to handwriting recognition technology, and more particularly to learning handwritten characters for learning input characters unique to each individual, updating standard patterns in a standard dictionary, and improving the recognition of input characters. It relates to a recognition device.

[0002]

2. Description of the Related Art In a conventional on-line handwritten character recognition device, an input pattern includes a unique character, a continuous character, a different stroke order, and the like. The recognition rate did not improve.

For this reason, a method for improving a standard pattern in a standard dictionary has been proposed as follows.

[0004] 1) Various writing methods are registered in advance as standard patterns from the collected input patterns or expected patterns.

[0005] 2) As described in JP-A-59-165190, a correlation coefficient between a dictionary (a standard pattern in a standard dictionary in the text) and character data (input pattern) is calculated, and the value is set to an allowable value. If not, the dictionary is automatically updated with a weighted average. If the value is equal to or larger than the allowable value, the operator interactively operates the displayed character data to register, update, and discard in the dictionary.

3) As described in JP-B-7-111732, a method has been proposed in which a predetermined recognition rate is determined for each category and the number of categories (standard patterns) having a low recognition rate is increased. .

[0007]

The above method 1) requires an excessive memory for the standard dictionary and is not practical.

[0008] The method 2) is not optimal because registration, updating, and discarding in the dictionary are performed by the operator interactively, so that the operator has to perform a lot of manual work and is subjective.

The method of 3) increasing the number of categories (standard patterns) up to a predetermined recognition rate also has a disadvantage that an excessive memory is required for the standard dictionary.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above-described problems. An input pattern is stored in a small memory, and a standard dictionary is updated from the stored data at predetermined time intervals. An object of the present invention is to provide a handwritten character recognition device capable of recognizing input characters.

[0011]

The above object can be achieved by the following method.

(1) Each standard pattern in the standard dictionary is provided with a predetermined distance value. In FIG. 3, the standard pattern "U"
If the distance value between the standard pattern "U" and the nearest standard pattern "ra" is 2R, the predetermined distance value of the standard pattern "U" is R.
Now, when a plurality of character patterns are input and recognized as "U" but the distance value is larger than R, a new "U" of A1 different from the original "U" of A0 is newly generated. As to which input pattern is used as the standard pattern, a pattern having the largest distance value, a pattern temporarily registered first, a pattern having the average value, and the like can be selected.

(2) Next, as shown in FIG. 4, when a plurality of input patterns are within the range of "U" of the original A0, the coordinates of the standard pattern are smoothed as shown by the following equation. Moving.

[0014]

[Equation 1] New _xy = (1−α) Old _xy + α Input _xy (Equation 1) Here, New _xy is an arbitrary coordinate of the smoothed standard pattern Old _xy is an arbitrary coordinate of the standard pattern before smoothing Input _xy is an arbitrary coordinate α of the input pattern is a constant.

(3) After calculating the similarity between the input pattern and the standard pattern, if the output character code is rewritten due to erroneous recognition, if the number of times is large, the standard pattern is deleted.

In order to realize the means (1) to (3), the following configuration is adopted.

A temporary registration dictionary for storing input patterns is provided in the handwritten character recognition device. Each time a character pattern is input, a similarity calculation is performed with each standard pattern in the standard dictionary. The number of the standard pattern determined,
The distance value, character code, and registration date are registered in the temporary registration dictionary.

The standard pattern is configured so that the number, coordinate information, character code, number of incorrect recognitions, and registration date (update date) in the standard dictionary can be registered.

In the above configuration, the following processing is performed after a predetermined time has elapsed.

(1) When the distance value of the accumulated input pattern is
If the distance is larger than a predetermined distance value of the standard pattern, the input pattern and the character code are registered as a new standard pattern.

(2) After the processing of (1), even when the output character code is rewritten due to erroneous recognition, the input pattern and the rewritten character code are registered as a standard pattern. In this case, the number of times of erroneous recognition of the erroneously recognized standard pattern is updated.

(3) In (1), if the distance value is smaller than or equal to a predetermined distance value of the recognized standard pattern,
The coordinates obtained by smoothing the coordinates of the standard pattern and the coordinates of the input pattern based on the above equation (1) are defined as the coordinates of the standard pattern.

(4) The number of times of erroneous recognition of the standard pattern is checked. If the number is large, the standard pattern is erased.

As described above, the temporary registration dictionary for storing the input patterns in the handwritten character recognition device is provided, and the stored character patterns are collated with the standard patterns to generate and update the standard patterns. (1) by comparing character patterns entered online with standard patterns without storing them in
The above object can be achieved even by performing the processes (3) to (3).

[0025]

BEST MODE FOR CARRYING OUT THE INVENTION

(Embodiment 1) An embodiment of the present invention will be described below with reference to FIGS. In FIG. 1, an input device 20 such as a tablet
Outputs the coordinate point sequence data of the input pattern to the preprocessing unit 30 in the handwritten character recognition device 10. The pre-processing unit 30
The coordinate point sequence data is approximated to N points as a representative point of each stroke, and is output to the character recognition unit 40. The character recognition unit 40 calculates the similarity between the input pattern approximated to the N points and the standard pattern registered in the standard dictionary 50.

This similarity calculation is performed, for example, in Japanese Patent Laid-Open No. 58-12 / 1983.
Since it is disclosed in Japanese Patent Publication No. 3177, it is omitted here.

The standard pattern may be a number in the standard dictionary 50 (the order of registration, the number of strokes, the order of character codes, or a combination thereof), coordinate information approximated by N points, character codes, It consists of elements for writing the number of times an input pattern was erroneously recognized (supported by software), the date registered in the standard dictionary, and the date on which the processing described later is updated.

After calculating the similarity, the character code of the standard code having the smallest distance value is stored in the character display unit 60.

Reference numeral 70 denotes a CRT or LCD display device. In recent years, an input integrated display device in which the input device 20 and the display device 70 are layered has been commercialized.

Reference numeral 80 denotes a temporary registration dictionary for realizing the function of the present invention. Each time character data is input from the input device 10, an input sequence number, an input pattern approximated to N points,
Describe the character code, standard pattern number, and registration date after similarity calculation.

FIG. 2 shows a processing flow for newly registering, updating, and deleting a standard pattern in the standard dictionary from the input data stored in the temporary registration dictionary. This process is started at an arbitrary time (every week, every day, or every hour).

In step 100, a predetermined number of deletions is set for use in determining whether to delete the standard pattern. In step 110, information on an input pattern temporarily registered is input first.

Steps 120 to 200 are loop processing in which information on all of the temporarily registered input patterns and all standard patterns are referred to. First, in step 120, information on the first standard pattern is input to a work area (not shown).

Steps 130 to 190 also refer to the information of the tentatively registered arbitrary input pattern and all the standard patterns by loop processing. In step 130, it is determined whether or not the temporarily registered input pattern is the same as the standard pattern to be processed by the stored standard pattern number.
If “Yes”, step 140 is performed, and if “No”, jump to step 170.

At step 140, it is determined whether or not the distance value of the temporarily registered input pattern is larger than a predetermined distance value written in each standard pattern. Is registered in the standard pattern, and if "No", a process of updating the coordinates of the standard pattern in step 160 described later is performed. After the processing of steps 150 and 160, step 170 is performed.

In step 170, the number of erroneous recognitions written in the standard pattern is compared with the number of deletions set in step 100.
That is, if “Yes”, the standard pattern is deleted from the standard dictionary in step 180. If “No”, it is determined whether reference to all the standard patterns in step 190 has been completed, and if not, the standard patterns are updated (step 200) and the process returns to step 130.

When the loop processing of steps 130 to 190 is completed, the loop processing of steps 120 to 210 is executed to terminate the reference of all the temporarily registered input patterns and standard patterns.

Next, the number of erroneous recognitions of all the standard patterns is cleared, the registration date is rewritten (step 220), the mutual distance value of all the standard patterns is calculated, and the predetermined distance value of each standard pattern is updated. (Step 230).

Further, all the temporary registration patterns are cleared, and the process is terminated.

In the process of registering the temporarily registered pattern in step 150 in the standard dictionary, if there are a plurality of input patterns, the one with the largest distance value, the one that is temporarily registered first, the most average one, etc. , Can be arbitrarily selected.

In the process of updating the coordinates of the standard pattern in step 150, the process of smoothing the standard pattern and the input pattern may be performed as described above. The pattern may be exchanged. The process of updating the predetermined distance value for each of all the standard patterns in step 230 is の of the distance value between adjacent standard patterns from the calculation of the similarity between all the standard patterns.

Second Embodiment In the first embodiment, a temporary registration dictionary is provided, input character data is accumulated for a predetermined time, and a standard pattern in the standard dictionary is newly generated, updated, and deleted.
Rather than temporarily registering the character data input online, immediately refer to all the standard patterns, find the distance value, compare the distance value with the predetermined distance value of the standard pattern, and newly generate and update the coordinates of the standard pattern. It is possible. If you make a mistake,
Rewriting is performed within the standard pattern, and the number of rewriting is determined within a predetermined time.

FIG. 5 shows an example of updating the coordinates of the standard pattern in step 160 in FIG. 1. Initially, the coordinates of the N-point standard pattern (a) registered in the standard dictionary are used to generate the input pattern. It is updated according to the number of times.

[0044]

According to the present invention, a temporary registration dictionary for storing character data to be input into the handwriting input device is provided, and the standard dictionary is learned and updated from the character data, or the standard dictionary is obtained by online processing. To update the dictionary,
A handwritten character recognition device capable of recognizing input characters unique to an individual can be provided.

[Brief description of the drawings]

FIG. 1 is a flowchart for updating a standard dictionary according to an embodiment of the present invention.

FIG. 2 is a flowchart of the handwriting recognition device.

FIG. 3 is an explanatory diagram of generation of a new standard pattern.

FIG. 4 is an explanatory diagram of coordinate movement of a standard pattern.

FIG. 5 is an explanatory diagram of an embodiment of coordinate movement of a standard pattern.

[Explanation of symbols]

10: Handwriting recognition device, 20: Input device, 30: Preprocessing unit, 40: Character recognition unit, 50: Standard dictionary, 60: Character display unit, 70: Temporary registration dictionary

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Akihiro Katsura 7-1-1, Omika-cho, Hitachi City, Ibaraki Prefecture Within Hitachi Research Laboratory, Hitachi, Ltd.

Claims (5)

[Claims] 1. A similarity calculation is performed between an input means capable of performing handwriting input in a time series and a standard pattern in a standard dictionary in which a handwritten pattern input by the input means is set in advance, and a character pattern which is most similar is obtained. In the handwritten character recognition device that outputs the code of the above, having a means for storing the input pattern, from a plurality of stored input pattern information,
Alternatively, each time an input pattern is input, a standard pattern in a standard dictionary is generated, changed, or deleted. 2. A handwriting to be registered as a standard pattern, wherein a predetermined distance value is provided for each of the standard patterns, and the distance value of the input pattern after the similarity calculation is larger than the predetermined distance value. Character recognition device. 3. The method according to claim 1, wherein after calculating the similarity of the input pattern, if the output character code is rewritten due to erroneous recognition, the input pattern and the rewritten character code are registered as a standard pattern. Handwritten character recognition device. 4. When the input pattern after the similarity calculation according to claim 2 is smaller than or equal to a predetermined distance value of each standard pattern, the coordinates of the standard pattern and the coordinates of the input pattern are determined. A handwritten character recognition device that uses the smoothed coordinates as the coordinates of the standard pattern. 5. The standard pattern according to claim 1, further comprising a function for counting the number of times the input pattern has been correctly recognized and the number of times the input pattern has been incorrectly recognized, and an element for retaining the number. A handwritten character recognition device that erases the standard pattern if the number of erroneous recognitions is large.