JPH0950491A - Method for recognizing handwritten character on-line and device therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method for recognizing handwritten character on-line guaranteeing the sure improvement of recognition accuracy without increasing the capacity of a character recognition dictionary and to provide a device therefor. SOLUTION: A standard sub-temperature for alternation selection part 11 selects a template whose importance is the lowest as a standard sub-template for alternation, based on the use frequency and the final use time in a standard sub-template, as for the character category in which the number of sheets of the standard sub-template becomes equal to the upper limit number of K pieces. As for a case where a writing person selects a right answer from the candidate character categories which are set at the second and succeeding number-th of places for the recognition result of an input pattern, a standard sub-template storage part 6 additionally registers the input pattern as the standard sub-template in which the use frequency is once and the final use time is the present time in the character recognition dictionary of a right answer character category. At the same time, when the total number of sheets of the standard sub-template is equal to K pieces before the additional registration, the part 6 deletes the standard sub-template for alternation from the character recognition dictionary.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention stores a standard template in advance as a character recognition dictionary for each character category to be recognized, and performs template matching with an input pattern of written handwritten characters to perform pattern matching. When the inter-pattern distance or the similarity is calculated and a plurality of candidate character categories are output as recognition results in the ascending order of the inter-pattern distance or in the descending order of the similarity, and the writer misreads the first candidate character category. To the second
The present invention relates to an on-line handwritten character recognition method and apparatus that enables selection of a correct answer from a candidate character category of ranks below.

[0002]

2. Description of the Related Art Conventionally, a method for learning a character recognition dictionary is
The standard template stored in advance for each character category is corrected as appropriate using the pattern correctly read and the pattern misread in the character category, and the misread character pattern is additionally registered as a standard sub-template. There are two main methods. In addition, conventionally, when the application of the above learning method is open to writers, it is necessary to specify whether to use a specific input pattern to modify the standard template or to specify whether to additionally register the standard subtemplate. An operation mode for dictionary correction or dictionary registration was prepared for the person himself.

[0003]

However, in the method of correcting the standard template of each character category with the input pattern group of the writer of the character category, the standard template is large because the variation of the handwritten character shape is large even within the individual. Will be modified to represent those average blurred shapes, and instead of being able to recognize roughly written character patterns to some extent, carefully written character patterns will also have the same level of recognition accuracy. It will deteriorate. For this reason, there is a problem that the reliability of the recognition device using the original standard template before correction is impaired, that is, if the user wants to input the data surely, it needs to be written carefully.

On the other hand, in order to solve the above-mentioned problems, the standard sub-template additional registration method holds the original standard template as it is and registers the misread input pattern as the standard sub-template of the character category. But,
In this case, at present, the decision as to which misreading pattern should be registered is left to the discretionary judgment of the writer himself, so that the improvement of the recognition accuracy has not been guaranteed. Further, when setting an upper limit for the number of registered standard sub-templates, no rational processing method has been found when standard sub-templates exceeding the upper limit are generated. Further, the writer is required to make an extra operation for correcting a dictionary or registering a dictionary, which is far from the original purpose of inputting characters. This does not mean that the learning of the character recognition dictionary is automated, and there is a problem in that the writer is heavily burdened.

An object of the present invention is to provide an on-line handwritten character recognition method and apparatus which guarantees a reliable improvement of recognition accuracy without enlarging the capacity of the character recognition dictionary.

[0006]

In the online handwritten character recognition method of the present invention, in addition to the standard template stored in advance for each character category to be recognized, the writer recognizes the input pattern as a second result. When the correct character category is selected from the candidate character categories below the rank, the input pattern is automatically additionally registered as a standard subtemplate of the correct character category, and at that time, the upper limit of the total number of standard subtemplates of each character category is registered. If the number of standard sub-templates in the correct answer category has reached the upper limit before the additional registration, the one with the lowest importance is replaced based on the frequency of use and the last use time of each standard sub-template. Register a newly misread input pattern after deleting it as a standard sub template And wherein the door.

Further, the on-line handwritten character recognition device of the present invention sets the usage frequency of the standard subtemplate to 1 when the template of the first candidate character category of the input pattern is the standard subtemplate additionally registered by then.
For the standard sub-template usage frequency and last-use time storage means that both add times and store the current time as the last-use time, and for the character category in which the number of standard sub-templates is equal to the upper limit of K And a replacement standard sub-template selecting means for selecting the least important one as a replacement standard sub-template based on the frequency of use and the time of last use, and the recognition result of the input pattern by the writer.
When a correct answer is selected from the candidate character categories below the rank, the input pattern is additionally registered in the character recognition dictionary of the correct answer category as a standard subtemplate whose usage frequency is once and the last use time is the current time When the total number of standard sub-templates is equal to K before registration, standard sub-template additional registration means for deleting the substitute standard sub-template from the character recognition dictionary is provided.

As the substitute standard sub-template, it is possible to select the standard sub-template having the oldest last use time among those having the lowest use frequency.

[0009]

BEST MODE FOR CARRYING OUT THE INVENTION The present invention uses a natural operation in which a writer selects a correct answer from a plurality of candidate character categories output as a recognition result when erroneous reading occurs. We will automatically register the pattern as a standard sub template of the correct answer category, at that time,
In order to select and retain only a certain number of optimal standard sub-templates for each character category, manage the frequency and time of last use of each standard sub-template, and replace them with those with the lowest importance. By registering the input pattern that is misread, the guarantee of the recognition accuracy is guaranteed without enlarging the capacity of the character recognition dictionary.

[0010]

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram of an online handwritten character recognition apparatus showing an embodiment of the present invention.

The online handwritten character recognition apparatus of this embodiment includes a character information input section 1, a preprocessing section 2, a feature point extraction section 3, a standard template storage section 4, an input pattern storage section 5, and a standard subtemplate storage section 6. The character recognition unit 7, the first-ranked candidate category display unit 8, the correct answer character category selection unit 9, the standard sub-template use frequency / last use time storage unit 10, and the replacement standard sub-template selection unit 11 are included. The operation of each processing unit will be specifically described below.

The character information input unit 1 is composed of an existing data tablet, and the time series data of the position coordinates of the writing point is input for each stroke with respect to the handwriting of one character pattern drawn by a dedicated pen on the data tablet. , Preprocessing unit 2
To send to.

The preprocessing unit 2 performs noise removal, position / size normalization, etc. on the brush point position coordinate sequence of one character pattern sent from the character information input unit 1 based on a known technique. After applying, it is sent to the feature point extraction unit 3.

The feature point extraction unit 3 has a constant value that increases with the stroke length including the start point and the end point from the writing point sequence forming each stroke with respect to the writing point position coordinate sequence of one character pattern sent from the preprocessing unit 2. The writing points closest to the positions obtained by equally dividing the stroke by the number of points are sequentially extracted as characteristic points, and they are stored as a characteristic point sequence of one character pattern. The feature point sequence for each stroke of one character obtained here is sent to the standard template storage unit 4 when the standard template is created. On the other hand, when the recognition processing is executed, the feature point sequence is sent to the input pattern storage unit 5.

The standard template storage unit 4 is a feature point extraction unit 3 for a large number of character patterns that are collected for each character category to be recognized and are written with a certain number of strokes and stroke order.
A standard template of the character category is obtained by calculating an average position coordinate of each feature point of each stroke that constitutes the character category using the feature point sequence sent from Store as. Here, the total number of character categories to be recognized is N, and the standard template of the i-th (1 ≦ i ≦ N) character category is described as R _io . A set of standard templates {R _io } (1≤i
≦ N) is sent to the character recognition unit 7 for template matching with the input pattern.

The input pattern storage unit 5 sets, as an input pattern of the character pattern, a series of position coordinate sequences of the feature points forming each stroke sent from the feature point extraction unit 3 for the character pattern to be recognized, in accordance with the handwriting. Store. The input pattern is sent to the character recognition unit 7 for the recognition processing. Further, when the writer selects the correct answer character category from the candidate character categories of the second rank and below in the correct answer character category selecting unit 9, the input pattern is newly added to the standard subtemplate storage unit 6 as a standard subtemplate to be additionally registered. Sent out.

The standard sub-template storage unit 6 is intended for automatic additional registration of the writing habits of the writer in the character recognition dictionary, and has an upper limit of two or more arbitrarily specified K for each character category to be recognized. Store the standard sub-template in. Let the total number of standard sub-templates of the i-th (1 ≦ i ≦ N) character category be k _i (k _i ≦ K), and let the standard sub-templates of the i-th character category be {R
_ij } (1 ≦ j ≦ k _i ). All standard sub-template set _{{R ij} (1 ≦ i} ≦ N, 1 ≦ j ≦ k i)
Is sent to the character recognition unit 7 for template matching with the input pattern. Here, the additional registration processing to the standard sub template is performed when the writer recognizes the candidate character category for the input pattern and selects the correct character category from the second or lower candidate character categories in the correct character category selecting unit 9. Only, the input pattern is stored as a standard subtemplate of the correct character category. However, when the total number of standard sub-templates registered in the correct answer character category before the additional registration reaches K, the substitute standard sub-template selecting unit 1 for the correct answer character category.
The replacement standard sub-template selected in 1 shall be deleted at the same time. Hereinafter, the standard template and the standard sub-template, which are templates of the i-th character category, are described as R _im , so that i is referred to as a category number and m is referred to as a template number. However, m = 0 indicates a standard template, and m ≧ 1 indicates a standard sub-template. Here, the template number m of the standard sub-template is managed by the standard sub-template usage count and the last use time storage unit 10, and the template number when newly registering the standard sub-template is the standard sub-template usage count and the last use time. The use time storage unit 10 specifies.

The character recognizing unit 7 receives a standard template set of all character categories sent from the standard template storage unit 4 and all characters sent from the standard sub-template storage unit 6 for the input pattern sent from the input pattern storage unit 5. A known technique, template matching, is performed with each template of the standard sub-template set of the category to calculate the distance between patterns or the degree of similarity. The candidate character categories from the first rank to the Lth rank are output as recognition results in the ascending order of the calculated inter-pattern distance values or in the descending order of the similarity values. The recognition result is composed of a category number, a template number, a character code, a pattern distance or a similarity value, and the current time for each candidate character category in each rank. Of the recognition results of these input patterns, the category number of the first-ranked candidate character category, the template number, the character code, the distance between patterns, or the value of similarity,
And the current time are sent to the first-ranked candidate character category display unit 8, and the category number, template number, character code, inter-pattern distance or similarity value of the candidate character categories from the second rank to the Lth rank, and The current time is sent to the correct answer character category selecting unit 9.

The first-ranked candidate character category display section 8 has a category number, a template number, and a first-ranked candidate character category of the input pattern sent from the character recognition section 7.
The character code, the distance between patterns or the value of similarity, and the current time are temporarily stored, and the type font corresponding to the character code of the first candidate character category is displayed so that the writer can confirm whether the recognition result is correct. Display on top. At the same time, the category number of the first-ranked candidate character category, the template number, and the current time are sent to the standard sub-template use frequency and last use time storage unit 10.

The correct answer character category selecting unit 9 selects the correct answer from the second and lower candidate character categories without the writer accepting the first candidate character category displayed in the first candidate character category displaying unit 8. In doing so, a typeface font corresponding to the character code of the candidate character category from the second rank to the Lth rank is displayed on the display to allow the writer to select the correct character category. If the correct answer is not included, the writer is made to input the correct character code by a known technique, and the category number corresponding to the character code is determined. The category number of the correct character category and the current time are sent to the standard sub-template use frequency and last use time storage unit 10.

The standard subtemplate use frequency and last use time storage unit 10 stores and manages the number of standard subtemplates, the use frequency and last use time of each standard subtemplate for each character category. First,
Regarding the category number, the template number, and the current time of the first-rank candidate character category of the input pattern sent from the first-rank candidate character category display unit 8, the template number is 1 or more, that is, a standard sub-template is indicated. Only in this case, the use frequency of the standard sub-template designated by the category number and the template number is added once, and the current time is stored as the last use time of the standard sub-template. At that time, when the total number of standard sub-templates of the character category is equal to the upper limit of K, the frequency of use and the last use time of each of the standard sub-templates are sent to the substitute standard sub-template selecting unit 11. On the other hand, regarding the category number of the correct character category of the input pattern additionally registered as the standard sub template sent from the correct character category selecting unit 9 and the current time, the standard sub template of the correct character category before the additional registration If the number k is less than the upper limit number K, the additional registration template number of the input pattern is set to k + 1, the frequency of use is set once, and the last use time is set to the current time, and the total number of standard sub-templates is set to the upper limit K. If they are equal, assuming that the template number of the substitute standard sub-template designated by the substitute standard sub-template selection unit 11 is p, the additional registration template number of the input pattern is p and the frequency of use is 1
After setting the time and the last use time as the current time, the use frequency and the last use time for each of the standard subtemplates of the correct answer character category after additional registration are sent to the substitute standard subtemplate selecting unit 11.

Replacement standard sub-template selection unit 11
Is the most important based on the frequency of use and the time of last use of each standard subtemplate for a character category for which the total number of standard subtemplates sent from the standard subtemplate use frequency and last use time storage unit 10 is equal to the upper limit of K. Select the lower standard sub-template as the alternate standard sub-template for the character category. Here, as the selection criterion of the standard sub-template having the lowest importance, for example, it is conceivable to select the standard sub-template having the oldest last use time among the ones having the lowest use frequency. When the template number of the replacement standard sub-template of the selected character category is written as p, the value of the template number p is sent to the standard sub-template usage frequency and last use time storage unit 10,
It is used when additional registration of the standard sub-template to the character category occurs.

FIG. 2 shows an example of the character category "knowledge", where the maximum number of standard subtemplates is K = 3, the standard template storage unit 4, the standard subtemplate storage unit 6, and the standard subtemplate usage frequency. A specific example of the contents stored in each of the last use time storage unit 10 will be shown.

The operation of the specific example of FIG. 2 will be described.
First, the writer writes an input pattern of the character category "knowledge" and recognizes it with this device. As a result, the operation when it is correctly recognized is as follows. If the template of the character category "knowledge" that has output the minimum inter-pattern distance or the maximum similarity is, for example, the standard sub-template of template number 1, the usage frequency of the standard sub-template is added once and 19 The current time, for example, “1995 08 07 9:30” is stored as the last use time. However, if the template of the character category "knowledge" that outputs the minimum inter-pattern distance or the maximum similarity is the standard template, nothing is done. On the other hand, the operation when it is not correctly recognized and misread is as follows. Since the number of standard sub-templates of the character category “knowledge” has already reached the upper limit number K = 3, the replacement template is deleted. In FIG. 2, the standard sub-template of template number 2 is selected as the replacement standard sub-template having the lowest degree of importance based on the frequency of use and the time of last use, and is deleted. Next, the input pattern that is misread is newly used as a standard subtemplate of the template number 2 of the character category “know”, the usage frequency is once, and the last usage time is the current time, for example, “1995 08
07 9:30 ”and store.

As described above, in this embodiment, a standard template prepared from a large number of character patterns is stored in advance for each of the character categories to be recognized, and the character recognition dictionary is prepared according to the writing habits of the writer. The optimum standard sub-template registration method has been implemented, and when the writer confirms the recognition result of the input pattern, the correct character category is selected from the second and lower candidate character categories. , The input pattern is additionally registered as a standard subtemplate of the correct character category, and at the same time, the upper limit K of the number of standard subtemplates for each character category is set to manage the use frequency and the last use time of each standard subtemplate. If the number of standard sub-templates in the correct answer category reaches K, Before additional registration, delete the least important one of the standard sub-templates based on the frequency of use and time of last use, and then register a new standard sub-template to increase the capacity of the character recognition dictionary. The recognition accuracy can be surely improved by automatically registering the writing habit of the misread character pattern into the standard sub-template optimally without considering the writing frequency and writing time. In particular, when the writer is specified, the writing habit of the writer is automatically registered in the character recognition dictionary within the range of the maximum number of standard sub-templates very efficiently. In addition, since the automatic learning of the character recognition dictionary is realized by using the natural operation of selecting the correct character category for the misreading pattern by the writer, the writer has no extra burden for correcting or registering the dictionary. It is clear that it has eliminated the instability when leaving the dictionary correction or dictionary registration to the writer.

[0027]

As described above, according to the present invention,
Accurate reading of the writing habits of writers It is possible to automatically learn the character recognition dictionary by additionally registering patterns, and it is guaranteed that the recognition accuracy will surely improve as the usage time increases. The advantage is extremely large when the online handwritten character recognition device to which the present invention is applied is incorporated in such a portable personal information communication device or a window work terminal.

[Brief description of drawings]

FIG. 1 is a block configuration diagram of an online handwritten character recognition device showing an embodiment of the present invention.

FIG. 2 is a diagram showing a specific example for explaining operations of a standard template storage unit 4, a standard sub-template storage unit 6, a standard sub-template usage frequency and a last use time storage unit 10 in FIG.

[Explanation of symbols]

1 Character information input unit 2 Pre-processing unit 3 Feature point extraction unit 4 Standard template storage unit 5 Input pattern storage unit 6 Standard sub-template storage unit 7 Character recognition unit 8 1st candidate category display unit 9 Correct answer character category selection unit 10 Standard Sub-template usage frequency and last-use time storage unit 11 Replacement standard sub-template selection unit

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Kazumi Otaka 1-1-6 Uchisaiwaicho, Chiyoda-ku, Tokyo Nihon Telegraph and Telephone Corporation

Claims (4)

[Claims] 1. A standard template is stored in advance as a character recognition dictionary for each character category to be recognized, and template matching is performed with an input pattern of written handwritten characters to determine the distance between patterns or the degree of similarity. A plurality of candidate character categories are calculated and output as recognition results in the ascending order of the distance between patterns or in the order of increasing similarity, and when the writer misreads the first candidate character category, the candidate character category In the online handwritten character recognition method that enables the selection of the correct answer from the candidate character categories, in addition to the standard template that is stored in advance for each character category that is the recognition target, Select the correct character category from the second or lower candidate character categories In this case, the input pattern is automatically additionally registered as a standard subtemplate of the correct character category, and at that time, the upper limit of the total number of standard subtemplates of each character category is set and the correct character category of the correct character category is set before the additional registration. If the number of standard sub-templates has reached the upper limit, the one with the lowest importance is deleted as a substitute standard sub-template based on the frequency of use and the time of last use of each standard sub-template, and then newly misread An online handwritten character recognition method characterized by registering a pattern. 2. The online handwritten character recognition method according to claim 1, wherein a standard subtemplate having the earliest last use time is selected as the substitute standard subtemplate among the least frequently used ones. 3. A standard template is stored in advance as a character recognition dictionary for each character category to be recognized, and template matching is performed with an input pattern of written handwritten characters to determine the distance between patterns or the degree of similarity. A plurality of candidate character categories are calculated and output as recognition results in the ascending order of the distance between patterns or in the order of increasing similarity, and when the writer misreads the first candidate character category, the candidate character category In an online handwritten character recognition device capable of selecting a correct answer from a candidate character category, use of the standard subtemplate when the template of the first candidate character category of the input pattern is the standard subtemplate additionally registered by then The frequency is added once and the current time is used as the last use time. Memorizing standard sub-template usage frequency and last use time storage means, and character categories for which the number of standard sub-templates is equal to the upper limit of K, importance based on the usage frequency and last use time in the standard sub-template And a case in which the writer selects a correct answer from the second or lower candidate character category for the recognition result of the input pattern, The input pattern is additionally registered in the character recognition dictionary of the correct answer character category as a standard subtemplate having a use frequency of once and the last use time is the current time, and at the same time, the total number of standard subtemplates is equal to K before the additional registration. Character recognition of the replacement standard sub-template Online handwriting recognition apparatus characterized by comprising a standard sub-template additional registering means for deleting from the book. 4. The on-line handwritten character recognition device according to claim 3, wherein a standard subtemplate having the earliest last use time is selected as the substitute standard subtemplate among the least frequently used ones.