JPH1091727A - Method and device for recognizing handwritten character and its recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To shorten time required for recognition without dropping a recognition ratio in on-line handwritten character recognition. SOLUTION: A specific stroke pattern is extracted from ink data for one inputted character (step 102), what number is the extracted stroke pattern appearing in the total number of input pictures (the total number of strokes), i.e., the appearance position of the stroke pattern, is discriminated and a retrieving range in a recognition dictionary is converged in accordance with the appearance position (step 103). Then the recognition dictionary is retrieved within the converged range based on the featured value of ink data and a candidate character for the recognition of a handwritten character is outputted (step 104).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to handwritten character recognition, and more particularly, to a handwritten character recognition method and apparatus for improving recognition speed by narrowing a search range in a recognition dictionary, and a recording medium therefor.

[0002]

2. Description of the Related Art In handwritten character recognition, a recognition dictionary is prepared in advance, a candidate character is selected by searching a recognition dictionary based on one character of handwritten character data input by a user, and the candidate character is recognized. Result. In the process of retrieving the recognition dictionary, generally, a feature amount which is a vector amount representing a feature of a handwritten character pattern is used. That is, for each character, a characteristic amount representative of the character is stored in the recognition dictionary, a characteristic amount is calculated from the input handwritten character data, and the characteristic amount closest to the characteristic amount of the input character (for example, Euclid (The one with the smallest norm) is searched from the feature amounts stored in the recognition dictionary, and the character corresponding to the searched feature amount is set as a candidate character. As the feature amount, for example, there is a value obtained by performing orthogonal function expansion such as a discrete cosine transform or a discrete Fourier transform to obtain each coefficient, and arranging the calculated coefficients in order to form a vector.

In the following description, "ink data" refers to locus data input by a user using an input device such as a pen, and "stroke" refers to one stroke (one stroke).
Locus data.

[0004]

However, in the case of the above-described conventional handwritten character recognition method, after calculating the characteristic amount from the handwritten character input by the user, it is necessary to find the one closest to the characteristic amount of the input character. In addition, the feature amount in the recognition dictionary is searched almost in brute force, and when the number of characters to be recognized is large or when a large number of feature amounts are stored in the recognition dictionary with the aim of improving the recognition rate, the feature There is a problem that the recognition time becomes longer when the number of dimensions of the feature vector, which is a quantity, is large.

An object of the present invention is to provide a handwritten character recognition method, a handwritten character recognition device, and a recording medium that can shorten the recognition time without lowering the recognition rate.

[0006]

SUMMARY OF THE INVENTION A handwritten character recognition method according to the present invention is a handwritten character recognition method for recognizing handwritten characters for one character input by a user using a recognition dictionary. A specific stroke pattern is extracted, the search range in the recognition dictionary is narrowed according to the appearance position of the stroke in which the extracted stroke pattern appears in the total number of strokes, and handwritten character recognition is executed to obtain candidate characters.

As a result of narrowing the search range in the recognition dictionary, an appropriate candidate character may not be found. In such a case, it is assumed that the appearance position of a specific stroke pattern has changed. The search range may be narrowed down and handwritten character recognition may be executed again. If the range of the narrowing-down is widened by changing the appearance position, the recognition rate is correspondingly improved, but the time required for recognition becomes long. Therefore, in order to allow the user to control such a trade-off relationship, a level value indicating the depth of search is set in advance by an input from the user, and an appropriate candidate character is not searched. In addition, it is preferable that the narrowing of the search range and the handwritten character recognition are repeated while changing the appearance position within the range of the number of repetitions indicated by an appropriate candidate character or the level value. However, even if the appearance position is changed excessively, the recognition rate will reach a plateau and the required time for recognition will only increase, so the repetition of narrowing down by changing the appearance position and handwritten character recognition will be discontinued at a certain point in time, and then use If the user re-enters the ink data or trains the recognition dictionary so that it is recognized as a candidate character, the overall efficiency is improved.

A handwritten character recognition device according to the present invention is a handwritten character recognition device for recognizing handwritten characters with respect to one character of ink data input by a user. A recognition dictionary used to perform character recognition using feature amounts,
Extract a specific stroke pattern from the ink data,
Performs handwritten character recognition by narrowing the search range in the recognition dictionary according to the appearance position of the extracted stroke pattern in which stroke of the total input strokes, and referring to the recognition dictionary within the narrowed search range Recognizing means.

In the handwritten character recognition device of the present invention,
If an appropriate candidate character is not found, it is effective to change the appearance position and narrow the search range again, and within the range of the number of repetitions represented by the level value input by the user, It is effective to sequentially change the appearance position and repeatedly execute the narrowing of the search range and the handwritten character recognition.

A recording medium according to the present invention is a computer-readable recording medium having an input unit for inputting ink data, and performs character recognition on one character of ink data input by a user. A specific stroke pattern from the recognition dictionary and the ink data, and narrow down a search range in the recognition dictionary according to the appearance position of the extracted stroke pattern in which stroke of the total input strokes. And a program for causing a computer to execute a step of executing handwritten character recognition and obtaining candidate characters with reference to the above.

(Operation) The present invention seeks to shorten the search time by narrowing the search range in the recognition dictionary before executing character recognition using the recognition dictionary.
At this time, the amount of calculation for narrowing down the search range must be small, not to mention lowering the recognition rate. Therefore, the present inventor has conducted intensive studies, and as a result, extracted a specific stroke pattern from the ink data, and in the total number of strokes (the total number of strokes), the extracted stroke pattern appeared in what number of strokes. That is, it is necessary to identify the appearance position of the stroke pattern, narrow the search range in the recognition dictionary according to the appearance position, and refer to the recognition dictionary within the narrowed search range. Is effective for online handwriting recognition.

In the present invention, specific stroke patterns include the following. That is, a linear (non-curved) stroke extending in one of the vertical direction and the horizontal direction and having the largest stroke length, the second stroke, and the third stroke.
The second one,..., Is a key-shaped stroke. Further, when considering kanji, the ink data of the kanji is composed of strokes for the number of strokes, but strokes are generally arranged from the upper left to the lower right in the stroke order. However, in a portion corresponding to a break between radicals, the stroke between two successive strokes changes greatly from bottom to top or from right to left. Therefore, even when the moving direction between strokes is from bottom to top or from right to left and the moving distance exceeds a predetermined threshold, a specific stroke pattern can be obtained.

After extracting such a specific stroke pattern from the ink data, a pointer from the appearance position to the search range is obtained by using a table including the appearance position of the stroke pattern and a pointer to the search range in the recognition dictionary. The recognition dictionary may be searched only within the range of the pointer. The number of appearance positions for narrowing the search range is not limited to one, and the search range in the recognition dictionary may be narrowed based on a combination of the appearance positions. For example, if a key-shaped stroke is a specific stroke pattern, the ink data may include a plurality of stroke patterns and may have a plurality of appearance positions (for example, a character of "art" may have a key-shaped pattern). There are three strokes), but it is also possible to narrow down the search range based on a combination of these plural appearance positions. In addition, a search range in the recognition dictionary may be narrowed by a combination of an appearance position of the longest one and an appearance position of a second one in a linear stroke in either the vertical direction or the horizontal direction. It may be.

[0014]

Next, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a handwritten character recognition device according to one embodiment of the present invention.

This handwritten character recognition device executes online handwritten character recognition, and includes an input / display panel 11 for inputting ink data by a user and displaying a recognition result and the like, and an input / display panel. 11
An input processing unit 13 for receiving ink data input to the printer, an ink data storage unit 14 for storing ink data of handwritten characters, and a recognition dictionary for storing representative characteristic amounts for each character and used for character recognition 21 and a recognition dictionary 21
The narrowing-down table storage unit 22 storing a narrowing-down table used for narrowing down the search target range in the search, and the ink dictionary in the ink data storage unit 14 by referring to the recognition dictionary 21 while narrowing down the search target range by the narrowing-down table. Character recognition engine unit 15 that performs character recognition by using
And a recognition result storage unit 16 for storing a character recognition result by the character recognition engine unit 15 and an output processing unit 17 for displaying on the input / display panel 11 according to the character recognition result.
And is constituted by.

The input / display panel 11 is a so-called display-integrated input panel in which a display section and an input section for ink data 20 are integrated. For example, the input / display panel 11 has a configuration in which a transparent tablet is laminated on a liquid crystal flat display. Can be used. The user operates the input / display panel 11 with the pen 12.
The desired ink data 20 can be input by tracing the input / display screen.

The character recognition engine unit 15 searches the recognition dictionary 21 based on the feature amount of the ink data to find candidate characters. In particular, the character recognition engine unit 15 executes the following processing. That is, the character recognition engine unit 15
Extract a specific stroke pattern from the input ink data and find the appearance position of the stroke pattern,
A search range in the recognition dictionary 21 is obtained by referring to the narrowing-down table from these appearance positions (or combinations of appearance positions). Then, the character recognition engine unit 15 searches the recognition dictionary 21 in the obtained search range based on the feature amount of the ink data to find a candidate character. FIG. 2 is a diagram illustrating extraction of a specific stroke pattern from ink data (“plant”) 20 for one character. Here, three straight strokes extending in the vertical direction and extending from the longest stroke to the third stroke are extracted. In the figure, the numbers indicate the longest stroke (second stroke), the second length stroke (seventh stroke), and the third length stroke (sixth stroke).

Hereinafter, handwritten character recognition in the handwritten character recognition apparatus of the present embodiment will be described with reference to the flowchart of FIG. Here, from the ink data for one character, as shown in FIG. 2 described above, the longest linear stroke extending in the vertical direction and the next linear stroke are extracted as specific stroke patterns. The search range in the recognition dictionary is narrowed by the combination of the appearance position of the longest object and the appearance position of the second object.

First, when a user inputs ink data 20 to the input / display panel 11 using the pen 12, ink data for one character is obtained (step 101). Then, the character recognition engine unit 15 normalizes the size of the ink data, and further extracts a stroke pattern as shown in FIG. 2 (step 10).
2). The character recognition engine unit 15 narrows down the dictionary search range by referring to the narrowing-down table in the narrowing-down table storage unit 22 in accordance with the appearance position (or combination of appearance positions) of the extracted specific stroke pattern (step 103). ).

FIG. 4 shows a narrowing-down table 30 used when the longest and second linear strokes extending in the vertical direction have a specific stroke pattern.
An example is shown. In this narrowing-down table 30,
Among the linear strokes extending in the vertical direction, the stroke position is classified according to the stroke position (stroke order) of the longest stroke and the stroke position (stroke order) of the second longest stroke. Stroke is 2 in input order
If the stroke of the second stroke is the seventh stroke, it is described that the dictionary search range can be narrowed down to “plant”, “forest”, “respect”, “grip”, etc. . For the sake of explanation, the dictionary search ranges are "plant", "forest",
Although "Kei" and "Kikari" are described, these "plants", "forests", "Kei", and "Kikari" are actually stored in the recognition dictionary 21.
Is stored.

After narrowing down the dictionary search range in this way, the character recognition engine unit 15 searches the recognition dictionary 21 within the narrowed down range based on the characteristic amount of the ink data and outputs candidate characters. (Step 104). These candidate characters are displayed on the input / display panel 11, and it is determined whether or not the user intends to input, that is, whether or not a hit has been made (step 105). Whether a hit is determined by the user confirming the candidate character or instructing re-recognition. In the case of a hit, the character is decided as it is and the process is terminated. It is determined whether the narrowing down and the character recognition have been repeated a limited number of times or more (step 10).
6). If the number of repetitions is equal to or greater than the limit number, the process is terminated; otherwise, the dictionary search range is narrowed down by the narrowing-down table 31 again assuming that the appearance position of the specific stroke pattern has changed (step 107). ), And return to step 104.

Here, the process of changing the appearance position will be described. For example, in the case of the example shown in FIG. 2, the linear strokes extending in the vertical direction are the second stroke,
The seventh and sixth strokes. Due to writing habits, etc., there may be a stroke in which the relationship of length is reversed from the original. For example, it is assumed that the second longest and the third longest are reversed, Based on the combination of the second and sixth strokes, the narrowing of the dictionary search range and the handwritten character recognition are executed again. If it doesn't hit, replace the longest one with the second one
Combination of the stroke and the second stroke (Since the stroke length is considered here, the order of appearance in the combination is also significant)
To narrow down the search range and execute handwritten character recognition. In this way, the narrowing of the dictionary recognition range by changing the appearance position and the handwritten character recognition are repeated at step 106 until the number of times reaches the limit. Of course, if a hit is made even within the limited number of times, the repetition ends there.

Next, a method for setting the limit number of times will be described. Increasing the number of restrictions increases the recognition rate, but increases the recognition time. Even if the number of restrictions is increased unnecessarily, the recognition rate reaches a plateau. Further, if the number of candidate characters selected in one search to the recognition dictionary 21 is changed and the number of candidate characters is increased, the possibility that the intended character is contained therein increases, but the determination operation is troublesome. Thus, if the number of candidate characters is reduced, the intended character is often not output as a candidate character. In the handwritten character recognition device of the present embodiment, in order to allow the user to set desired parameters in the relationship of the song trade-off, as shown in FIG.
A setting window 40 is displayed on the display panel 11, and the user can set the limit number of times and the number of candidate characters by inputting the setting window 40 with the pen 12. In FIG. 5, “recognition depth” is used to set a limit number. The limit number can be set small when “shallow”, and large when “deep”. Specifically, the user can change the parameter value to a desired value by dragging the tip of the bar graph display as shown in FIG.

Since the user's writing habits and the like are reproduced almost every time for each user, for example, the character recognition engine unit 15 is provided with a learning function to frequently change the appearance position. In this case, the narrowing-down may be performed by a combination of the stroke order in which the writing habits of the user and the like are considered with priority.

Although the embodiment of the present invention has been described above, the specific stroke pattern extracted from the ink data is not limited to the above-described one. FIG. 6A shows an example in which a stroke pattern is extracted according to a general change in the stroke direction between strokes. In the case of kanji, the stroke direction between strokes is generally from top to bottom and left to right. However, the stroke direction between radicals is different. In the example shown in the figure, the movement is from top to bottom as indicated by the arrow inside the radical (unbalanced), but it moves greatly from the lower left to the upper right as indicated by the arrow between the radicals. ing. Here, a stroke pattern is detected by detecting a movement like an arrow.

In the case shown in FIG. 6B, a specific stroke pattern is detected by detecting a key-shaped change (key portion 35) among the strokes. In the case of the character "", as shown, there are two key portions 35, and these two strokes are respectively detected as stroke patterns.

As is apparent from the above description, the handwritten character recognition device of the present embodiment is configured as a portable information communication device, and is connected to a portable information communication device as a computer and to a recording medium. This can be realized by reading stored software programs and recognition dictionaries. FIG. 7 is a block diagram showing a configuration of such a portable information communication device.

The CPU 1 is connected to the input / display panel 11, and the memory 2 and the disk drive 3 are connected to the CPU 1. Then, a recording medium 4 containing the data of the recognition dictionary 21 and storing a software program for executing the above-described processing is prepared. The recording medium 4 is mounted on the disk drive 3 and the data and the program are stored in the memory 2. , The portable information processing device can be used as the handwritten character recognition device of the present embodiment.

[0029]

As described above, the present invention extracts a specific stroke pattern from ink data, identifies the appearance position of the extracted stroke pattern in the total number of input strokes, and responds to the appearance position. By narrowing down the search range in the recognition dictionary and referring to the recognition dictionary within the narrowed search range, there is an effect that the time required for recognition can be significantly reduced without lowering the recognition rate.

[Brief description of the drawings]

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

FIG. 2 is a diagram illustrating extraction of a stroke pattern from ink data.

FIG. 3 is a flowchart illustrating a procedure of handwritten character recognition.

FIG. 4 is a diagram illustrating an example of a narrowing-down table.

FIG. 5 is a diagram illustrating a method of setting a recognition depth.

FIGS. 6A and 6B are diagrams each showing another example of a stroke pattern.

7 is a block diagram showing a configuration of a portable information communication device in which the handwritten character recognition device of FIG. 1 is configured.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 CPU 2 Memory 3 Disk drive 4 Recording medium 11 Input / display panel 12 Pen 13 Input processing unit 14 Ink data storage unit 15 Character recognition engine unit 16 Recognition result storage unit 17 Output processing unit 20 Ink data 21 Recognition dictionary 22 Refinement table storage Unit 30 narrowing-down table 35 key unit 40 setting window 101 to 107 steps

Claims (7)

[Claims] 1. A handwritten character recognition method for performing handwritten character recognition on a single character of ink data input by a user using a recognition dictionary, comprising: extracting a specific stroke pattern from the ink data; A handwritten character recognition method characterized by narrowing a search range in the recognition dictionary according to an appearance position of a stroke in a total number of strokes of a stroke pattern and executing handwritten character recognition to obtain candidate characters. 2. The handwritten character recognition method according to claim 1, wherein when an appropriate candidate character cannot be obtained, the search position is narrowed down by changing the appearance position, and handwritten character recognition is performed again. 3. A level value indicating the depth of search is set in advance by an input from a user, and if an appropriate candidate character is not searched, an appropriate candidate character is searched or the level value is displayed. 2. The handwritten character recognition method according to claim 1, wherein the appearance position is sequentially changed within the range of the number of repetitions to narrow the search range, and the handwritten character recognition in the narrowed search range is repeated. 4. A handwritten character recognition apparatus for performing handwritten character recognition on one character of ink data input by a user, comprising: input means for inputting ink data; and a feature value extracted from the ink data. A recognition dictionary used for performing character recognition, and a specific stroke pattern is extracted from the ink data, and the extracted stroke pattern is determined according to an appearance position of which stroke in the total number of input strokes. A handwriting character recognition device, comprising: a recognition unit that narrows a search range in a recognition dictionary and executes handwritten character recognition by referring to the recognition dictionary within the narrowed search range. 5. The handwritten character recognition device according to claim 4, wherein when an appropriate candidate character is not obtained, the appearance position is changed to narrow the search range, and handwritten character recognition is executed again. 6. A level value indicating a search depth is set in advance by an input from a user, and if an appropriate candidate character is not searched, an appropriate candidate character is searched or said level value is set. 5. The handwritten character recognition device according to claim 4, wherein the appearance position is sequentially changed within the range of the number of repetitions represented by the expression, and the narrowing of the search range and the handwritten character recognition are repeatedly executed. 7. A computer-readable recording medium having an input section for inputting ink data, comprising: a recognition dictionary for performing character recognition on one character of ink data input by a user; Extracting a specific stroke pattern from the ink data, narrowing a search range in the recognition dictionary in accordance with an appearance position of the stroke in which the extracted stroke pattern appears in the total number of strokes, and A program for causing a computer to execute a step of executing handwritten character recognition with reference to obtain candidate characters, and a recording medium storing the program.