JPH0863553A - Character string recognizing method - Google Patents

Abstract

PURPOSE: To recognize plural characters even when dot string data forming a character string is not largely divided timewisely and spatially. CONSTITUTION: Input dot string data is converted into stroke data and stored based on pen down information (steps 11 and 12). A base line expressing the inclination of an inputted character string is obtained to rotate the inputted character strings so as to be parallel to an x-axis, to form a lump being the lump of a smallest dot string (steps 13 and 14). The lump is successively combined from a left end to obtain a character code and an evaluation value and when the aspect ratio of the lump or the combined lump is not less than or a next lump does not exist, the combination of a highest evaluation value among the combinations of the lump and the combined lump is decided as one character. Time operation is repeated (steps 15 to 22). Finally, a recognized character string is displayed (step 23).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an on-line handwritten character recognition method, and in particular, recognizes a character string input by the pen based on information from an input device capable of periodically obtaining position information of the pen. Character string recognition method and device.

[0002]

2. Description of the Related Art In online character recognition, when inputting and recognizing a plurality of characters at a time, how to separate characters becomes a problem.

Conventionally, as a method of marking characters, 1) a method in which a frame for each character is provided and the frame serves as a character break,
2) A method of setting a character break when the distance of stroke data obtained from an input device is more than a certain standard. 3) Adding time information to the input stroke information, and separating the character by the time length between strokes. There was a way to decide.

[0004]

In all the above-mentioned methods, the input point sequence data must be divided temporally and spatially so that the character delimiter information is clear. However, the actual handwritten character string point sequence data is not data that can be distinguished temporally and spatially.

An object of the present invention is to provide a character string recognizing method and apparatus capable of recognizing a plurality of characters even if the point sequence data forming the character string is not largely separated temporally and spatially. .

[0006]

A character string recognition method according to the present invention comprises a first procedure for converting input point sequence data into stroke data based on pen-down information, and an input character based on the stroke data. A second procedure for obtaining a base line representing the inclination of a row, a third procedure for forming a ramp that is a minimum point row mass from the stroke data based on the base line, and a first ramp of the formed ramp group. Lamp and 2nd lamp, 1st lamp and 2nd lamp and 3rd lamp, ..., 1st lamp and 2nd lamp and 3
The second lamp and the last lamp are combined and combined, and the letters are recognized for all combinations of the first lamp and the above lamp, and of these, the evaluation value showing the closeness to the standard pattern is the highest. Fourth, a thing is selected as one character, and the same process is performed for the next lamp of the lamp or the combination of lamps selected as one character.
And the procedure.

In the third procedure, the stroke data is rotated in the x-axis direction by an angle formed by the base line and the x-axis, and the stroke data having an overlap in the x-axis direction is formed as a ramp.

The fourth procedure is to combine lamps that overlap in the x-axis direction, or combine lamps until the aspect ratio reaches a predetermined value.

[0009]

According to the present invention, character recognition is performed by forming a lamp, creating a combination of the lamps, and selecting the most probable lamp or the combination thereof.
Multiple characters can be recognized even if the point sequence data forming the character string is not significantly separated in terms of time and space.

[0010]

Embodiments of the present invention will now be described with reference to the drawings.

FIG. 1 is a block diagram of a character string recognition apparatus according to an embodiment of the present invention, FIG. 2 is a flow chart showing its operation,
FIG. 3 is an explanatory diagram of a base line, FIG. 4 is an explanatory diagram of rotation of stroke data, and FIG. 5 is an explanatory diagram of a ramp and an aspect ratio.

The on-line handwritten character string recognition apparatus of this embodiment comprises an input / storage unit 1, a baseline calculation unit 2 and a lamp formation unit 3.
And a recognition unit 4, an automatic selection unit 5, and a presentation unit 6.

The input / storage unit 1 inputs the coordinate data of the pen (point sequence data) together with the information on whether or not the pen is on the table (pen down information), and the point sequence data is converted into a single image of a character by the pen down information. The stroke data is converted into constituent stroke data and stored, and when a certain time has elapsed, the stroke data is output to the baseline calculation unit 2.

The base line calculating unit 2 performs the following calculation based on the stroke data to obtain a base line (a value representing the inclination of the character string). Figure 3 shows the hiragana character string "aiueo".
Is input, and the real number in the figure is the base line in this case.

The point coordinate data in the stroke S _i

[0016]

[Equation 1] And the number of point data included in the stroke is Ns _i . The number of strokes input and stored within a certain time is N. When the base line is aX + bY + c = 0, the coefficient a,
b and c are obtained by solving the following equations.

[0017]

[Equation 2] Since the above equation is a simultaneous linear equation, the coefficients a, b, and c are obtained by using the inverse matrix of the coefficient matrix on the left side. Here, ε is any sufficiently small number that is determined when obtaining a straight line. The obtained coefficients a, b, c are sent to the ramp forming unit 3.

The ramp forming unit 3 uses the coefficients a, b and c to determine the rotation angle θ of the entire character string, that is, the angle θ formed by the base line with the x axis.
(See FIG. 4) is obtained by the equation (4).

Θ = arctan (−a / b) (4) The ramp forming unit 3 uses the obtained rotation angle θ to rotate the coordinate data of the stroke data so that the character string is parallel to the x axis. (See Figure 4). Next, the lamp forming part 3
Is one lump (ramp) if the projections about the x-axis of strokes rotated so as to be parallel to the x-axis partially overlap each other. For example, for "a", x of all strokes that compose it
Since the projections on the axes overlap each other, they are used as one ramp.
As for “i”, the projection of the stroke on the left side on the x-axis and the projection of the stroke on the right side on the x-axis do not overlap with each other, so that each is regarded as one ramp.
Since the projections of all strokes constituting “u” and “e” on the x-axis overlap each other, they are regarded as one ramp. As for “o”, the projections of the right image on the x-axis do not overlap with the projections of the left image on each other, so that they are each regarded as one lamp. Eventually, the handwritten character "aiueo" in FIG. 3 forms seven lamps surrounded by a rectangular frame as shown in FIG. If there is overlap for the x-coordinates between the lamps, combine them into one lamp.

The automatic selection unit 5 sends point sequence data to the recognition unit 4 with the leftmost lamp as the lamp of interest to obtain the evaluation value and character code of the lamp, and the aspect ratio of the lamp of interest or the combined lamp is α. Until the above or until there is no next lamp, the lamps at the head (the left end) of the lamp group formed by the lamp forming unit 3 are connected to the lamps on the right side in order. Here, the aspect ratio of the lamp is the horizontal length / vertical length of a rectangular frame that surrounds the lamp, and is R ₁ / L for the lamp on the left side of “o” in FIG. 5, for example. When the left lamp and right lamp of "o" are combined, the aspect ratio is R
It becomes ₂ / L. The evaluation value indicates how close the stroke data is to the standard pattern, and is 1 when the stroke data completely matches the standard pattern and 0 when the pattern does not match the standard pattern at all. In this way, the automatic selection unit 5 selects the one having the highest evaluation value from the combination of one lamp and the lamp including the lamp.
Determined as one character. The automatic selection unit 5 repeats the same processing from the lamp to the right of the lamp or the combination of lamps determined as one character, and when the processing is completed up to the rightmost lamp, the presentation unit 6 presents the character code of the determined character.
Output to.

The presentation unit 6 displays a character string corresponding to the character code output from the automatic selection unit 5 on a screen such as a CRT.

Next, the operation of this embodiment will be described with reference to FIGS.

The coordinate data of the point sequence is input from the pen to the input / storage unit 1, and the pen-down information is converted into stroke data and stored (steps 11 and 12). The stroke data input within a certain period is the baseline calculation unit 2
Is input to the equation, and the equations (1) to (3) are solved to obtain the coefficients a, b, and c.
Is required (step 13). The coefficients a, b, c and stroke data are input to the ramp forming unit 3, the rotation angle θ of the character string is obtained by using equation (4), the stroke data is rotated so as to be parallel to the x axis, and From the maximum and minimum x-coordinates, the overlap in the x-coordinates for each stroke is calculated. Strokes with overlap are set as one ramp, and the overlap with respect to the x coordinate is similarly obtained between the ramps. The lamp where there is no overlap is finally the lamp. In this way, the lamp group is generated (step 14). Next, the lamp group is input to the automatic selection unit 5, the lamps are combined, and the character division is determined. First, the lamp at the left end of the lamp group is set as the lamp of interest (step 15), and the point sequence data of the lamp is sent to the recognition unit 4 to obtain the evaluation value and the character code (step 1).
6). Next, the aspect ratio is determined (step 17),
If it is equal to or greater than the constant value α, the lamp is regarded as one character (step 20). If the aspect ratio of the rumb is less than α,
It is determined whether or not there is a next (right adjacent) lamp (step 18), and if it exists, the leftmost lamp and the next lamp are combined (step 19). The process of steps 16 to 19 is repeated for this combined lamp combination, and if the aspect ratio becomes a certain value α or more or there is no next lamp, the evaluation value is the highest among the leftmost lamp or the combined lamp combination. The high ramp (leftmost ramp) or combination of ramps is determined as one letter (step 20). next,
Determining whether there is a lamp next to the lamp or combination of lamps determined as one character (step 21),
If there is, the next first lamp is set as the lamp of interest (step 22) and the above steps 16 to 20 are repeated. If it does not exist, it means that all the character delimiters have been added, and the automatic selection unit 5 outputs the character codes of all the determined characters to the presentation unit 6, and the presentation unit 6 displays the recognized character string. (Step 23).

In the example of FIG. 5, the second lamp and the third lamp are combined and determined as one character "i",
6th lamp and 7th (last) lamp are combined,
One character is determined as "O", and other lamps are determined as one character, and the input character string is recognized as "AIUEO".

[0025]

As described above, according to the present invention, the ramp is formed for the input stroke data, the combination of the ramps is created, and the ramp or the combination of the ramps having the highest evaluation value is selected. Since character string recognition is performed, there is an effect that a plurality of characters can be recognized even if the point sequence data forming the character string is not largely separated in terms of time and space.

[Brief description of drawings]

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

FIG. 2 is a flowchart showing an operation of the character recognition device in FIG.

FIG. 3 is an explanatory diagram of a baseline calculation method.

FIG. 4 is an explanatory diagram of stroke data rotation.

FIG. 5 is an explanatory diagram of an aspect ratio.

[Explanation of symbols]

 1 Input / storage unit 2 Baseline calculation unit 3 Lamp formation unit 4 Recognition unit 5 Automatic selection unit 11 to 23 steps

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

Claims (4)

[Claims] 1. A character string recognition method for recognizing a character string handwritten by a pen from input point sequence data representing position information of the pen, the input point sequence data being converted into stroke data based on pen down information. And a second procedure for obtaining a base line representing the inclination of an input character string based on the stroke data, and a minimum point sequence block from the stroke data based on the base line. The third procedure for forming a certain lamp, the first lamp and the second lamp of the formed lamp group, the first lamp, the second lamp and the third lamp,
..., the first lamp, the second lamp, the third lamp, ... and the last lamp are combined and combined, and the letters are recognized for all combinations of the first lamp and the lamp. The fourth procedure for selecting the one having the highest evaluation value indicating the closeness to the standard pattern as one character and performing the same processing for the next lamp of the lamp or the combination of lamps selected as one character, A character string recognition method having. 2. The third procedure comprises rotating the stroke data in the x-axis direction by an angle formed by the base line with the x-axis, and forming stroke data having an overlap in the x-axis direction as one ramp. The method for recognizing a character string according to claim 1. 3. The fourth step comprises combining lamps that overlap in the x-axis direction.
Handwritten character string recognition method described. 4. The character string recognition method according to claim 1, wherein the fourth step includes combining lamps until the aspect ratio reaches a predetermined value.