JPS58165181A - Method for character recognition - Google Patents

Abstract

PURPOSE:To eliminate noise component easily, by setting a nearby region for a coordinate point and using the number of coordinate point existing in this region to decide the validity of said coordinate point and adopting the point of the center of gravity of coordinate in the region as a smoothed point. CONSTITUTION:A character pattern consists of coordinate points P0-Pn, and X coordinate value and Y coordinate value of these coordinate points are stored in a coordinate memory. A smoothed point G0 is calculated for a set of coordinate points P0-P3. That is, when the nearby region is expressed with the inside of a circle having a radius (r), numbers N0-N3 in nearby coordinate existing in respective nearby regions of coordinate points P0-P3 are obtained, and coordinates g0-g3 of points of centers of gravity of the region are obtained. In this case, the largest number N of nearby coordinates is N and N2 (=2), and the point of the center of gravity is additionally obtained in respect to coordinates g1 and g2 of points of centers of gravity corresponding to N1 and N2, and the obtained point is defined as a smoothed point. A coordinate of the smoothed point G0 is expressed with G0=(g1+g2)/2. Next, a smoothed point G1 is obtained for a set of coordinate points P1-P4. Hereafter, smoothed points G2, G3... are obtained similarly and are stored in the coordinate memory.

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field of the Invention The present invention relates to a character recognition method, and more particularly to a handwritten character recognition method suitable for recognizing handwritten characters drawn on a tablet by real-time processing.

Background technology and its problems> In order to recognize handwritten characters drawn on a tablet, the characters are read in as multiple coordinate point data, feature points are extracted from this coordinate point data, and then feature points are extracted from the coordinate point data. A common method is to measure the degree of similarity between a set of patterns and a standard/mother turn stored in advance in a recognition device.

However, this type of conventional recognition method has the disadvantage that character recognition cannot be performed correctly even if noise and distortion are caused by the structure of the tablet or extra extraction is performed. If an attempt was made to eliminate the drawbacks, the device would become expensive.

<Object of the Invention> An object of the present invention is to provide a character recognition method in which noise components are removed by improving the processing method of coordinate point data without changing the structure of the tablet itself.

<Summary of the Invention> In this invention, a region within a certain distance from a coordinate point constituting a character is defined as its neighborhood area, a plurality of sets of consecutive coordinate points are created, and for each set, For the coordinate points belonging to this set, calculate the number of coordinate points and barycentric coordinates in the neighboring area, select the coordinate point with the largest number of coordinate points and its corresponding important coordinates, and select a smooth point that represents the set. By determining and performing this operation on all sets, a smooth point set is formed, and subsequent processing such as sampling and feature extraction is performed on this smooth point set. A recognition method was realized. Embodiments of the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a block diagram showing an example of a character recognition device used in carrying out the present invention.

1 is a coordinate memo that stores character coordinate point data 1 C2 is a coordinate memo that stores the coordinate point data of a character Sequentially reads the coordinate point data in the coordinate memory, adds the number of coordinate points in the neighboring area and the coordinate point data, and stores it again A neighborhood coordinate number calculator includes a difference calculator 212 and a determiner 22. It has a counter 23, an addition value register 24, and read counters 25 and 26. A comparator 3 compares the contents of the counter 23 and the contents of the register 4.

5 is a neighboring coordinate number memory that stores the number of neighboring coordinates for each coordinate point, and 6 is a counter 2 that stores the contents of the addition value register 24.
A barycenter point calculator that calculates the barycenter coordinate value by dividing by the contents of 3, and 7 is an address counter 'i''l.

The barycenter point coordinate memory stores the value calculated by the barycenter point calculator 6 at the address indicated by 8, and 9 is the smooth point needle:':
'□) In the calculator, comparator 919 determiner 92. It has a counter 93 and an addition value register 94. 1o is a barycenter point calculator that calculates the barycenter coordinate value by dividing the contents of the addition value register 94 by the contents of the counter 93; 12 is a write counter 1;
This is a coordinate memory that stores the value calculated by the gravity center point calculator 10 at the address indicated by 1.

Assuming that the number of coordinate points belonging to the set is 4, and the neighboring area is within a distance of 0.8 from coordinate point 0, that is, the resolution of the character is H, and this is 10 bits, then the X coordinate value between the two coordinate points, Y The operation will be described below assuming that the differences in coordinate values are both within ±8 bits.

Figure 2 (,) shows an example of a character pattern, which is composed of coordinate points p, p,...P, P, and is 0 1 n-I n
X coordinate value of this coordinate point
Xn-1+Xn) and Y coordinate value Y (yo+)
'i l +++ yn-1+ yn) is shown in Figure 2(b)
They are stored in the coordinate memory 1 from address 0 to address n+1 as shown in FIG. Further, a character end flag LE indicating the last coordinate point of the character/mother turn and a pen down flag PD indicating the continuous state of strokes are added to each coordinate point and stored together.

FIG. 3 is a partially enlarged view of FIG. 2(a).

First, for the coordinate point set consisting of Po-P3, smooth point G
. We will explain the procedure for calculating .

Assuming that the aforementioned nearby area is represented inside a circle of radius r shown in FIG. 3, the coordinate point P. ,Pl.

Number of neighboring coordinates existing in each neighboring region of P2 s P3 ) N1t N2 + N3 is N, respectively. =1
(Po itself only) e N1 = 2 (P, and P2)
, N2=2 (Pl and P2).

N3-1 (P3 itself only) and its center of gravity coordinates go
* gl r g2 p g5 are gO” (x
(3t yo) g3 = (N31 y3).

Here, the largest number of neighboring coordinates N is N, , N2 (=2
), the corresponding coordinates of the center of gravity g11 g2
If we further find the center of gravity for , it becomes the smooth point.

That is, smooth point G. teeth, becomes.

In this way, the first coordinate point set P is created. - Smooth point G for P3. After finding the second coordinate point set P
If we perform the same procedure for −P to find the smooth 4-point G1, we get: In the same manner, all smooth points are found.

The smooth point coordinate string obtained in this way includes Po, P
The coordinates of isolated points such as 5 are not included.

Next, the operation of the block diagram shown in FIG. 1 will be explained with reference to the flowchart shown in FIG. Note that the symbols used in the flowchart of FIG. 4 are defined as follows.

i... Content of read counter 25 j... Content of read counter 26 1... 6 Content of counter 23 2... Content of counter 93 t... Content amount of address counter 8... Write Contents of counter 1 S,...Contents of added value register 24 S2...Contents of added value register 94, R...Contents of register 4 A (i)...Contents of coordinate memory accessed by i ',': II.

A (j)...Contents of coordinate memory 2 accessed by j h)...Contents of coordinate memory 2 accessed by m C (4...Memory 5 of neighboring coordinates accessed by t) Contents D (/=)...Contents of the center of gravity coordinate memory 7 accessed at t First, read counters 25, 26, counter 23,
93, addition value register 24, 94, address counter 8. The contents of the register 4 and write counter 11 are cleared and initialized (step 1).

Next, the difference calculator 21 in the neighborhood coordinate number calculator 2 reads the coordinate memory 1 indicated by the read counters 25 and 26.
, and calculate the difference between their X and Y coordinate values. If the differences are both 8 bits or less, the determiner 22 determines that they are in the adjacent region, increments the contents of the counter 23 by 1, and further adds the contents of the coordinate memory 1 indicated by the read counter 26. The X coordinate value and Y coordinate value are respectively added to the X coordinate value and Y coordinate value of the addition value register 24, and the values are stored again. If it is determined that it is not within the nearby area, do nothing (step 2
).

After finishing, step 3) increments the contents of the read counter 26 by 1, and repeats the above process. This process is repeated four times, and for the first coordinate among the four coordinates, the number of neighboring coordinates in the set and the added value of their X coordinate value and Y coordinate value are stored in the counter 23 and the added value register. 24.

Further, the contents of the counter 23 are stored in the neighboring coordinate number memory 5 at the address indicated by the address counter 8 (step 4).

The comparator 3 compares the contents of the counter 23 and the contents of the register 4, and if the former is larger, it is stored in the register 4, and if the latter is larger, nothing is done (step 5).

The centroid point calculator 6 reads out the contents of the addition value register 24 and inputs the X coordinate value and Y coordinate value to the counter 23.
Calculate the center of gravity fish coordinates by dividing by the contents of . The result is stored at the address in the center of gravity point coordinate memory 7 indicated by the address counter 8 (step 6). Next, increase the contents of the read counter 25 and address counter 8 by 1,
The contents of the read counter 26 are decremented by 3 in order to be set to the storage address of the first coordinate of the coordinate set, and the contents of the counter 23 and the addition value register 24 are cleared (step 7).

Subsequently, in the same manner as described above, for the second coordinate point, the number of neighboring coordinate points and the coordinates of the center of gravity are calculated and stored, and the number of neighboring coordinate points is compared in magnitude, and stored in each memory and register. By repeating this process four times, the number of neighboring coordinates of each of the four coordinate points is stored in the neighboring coordinate number memory 5, the maximum value is stored in the register 4, and the center of gravity of the set of neighboring coordinates of each coordinate point is stored. The point coordinates are stored in the center of gravity point coordinate memory 7.

Next, clear the contents of address counter 8 in step 8.
), in the smooth point calculation, the comparator 91 reads the contents of the register 4 and the contents of the neighboring coordinate number memory 5 indicated by the address counter 8, and compares the values. Judgment device 92
If the two values are equal, the contents of the counter 93 are incremented by 1, the contents of the Atmory 7 are read out, and the X and Y values are added to the X and Y coordinates of the addition value register 94, respectively. Add and store. If the two values are not equal, do nothing (step 9). After the completion, the contents of the address counter 8 are incremented by 1 (step 10), and similar processing is performed. This process is repeated four times and the counter 93 calculates the number of coordinate points having the number of neighboring coordinates equal to the maximum number of neighboring coordinates indicated in the register 4, and the sum of the X and Y coordinate values of the corresponding barycentric point coordinates. It is stored in the addition value register 94. When this process is completed, the center of gravity calculator 10 reads the contents of the addition value register 94, divides the X coordinate value and the Y coordinate value by the contents of the counter 93, calculates the coordinate value of the center of gravity, and calculates the coordinate value of the center of gravity. Determine it as a smooth point of the set of four □ coordinate points, and write
ll, i, 7 are stored at the ::[ address of coordinate memory 2 indicated by data 11 (step 11). To subsequently determine the next set of coordinate points and determine the smooth point, first decrement the contents of the read counters 25 and 26 by 2, and make the second coordinate point in the previous set of coordinate points the first coordinate of the current set of coordinate points. The contents of the write counter 11 are incremented by 1, and the counters 23, 93, the addition value register 24, and the address counter 8. The contents of register 4 are cleared and initialized, and the process of calculating the number of neighboring coordinates is executed again (step 12). The smoothed points obtained in this way are stored in the coordinate memory 12, and when the processing for one character's data is completed, the coordinate memory 12
The content is read out, sampled, features are extracted, and character recognition is performed.

<Effects of the Invention> As explained above, in this invention, a neighborhood area is set for a coordinate point, and the number of coordinate points existing within that area is used to determine the validity of the coordinate point, and the coordinates within the area are Since the center of gravity of is used as the smoothing point, noise components can be easily removed and a high-quality smoothing pattern can be obtained.

In addition, this invention has the advantage that character recognition can be performed by smoothing character coordinate e-turns containing noise and distortion into high-quality mother turns, so a high recognition rate can be achieved even when using a low-cost tablet. Therefore, it can be used in an online handwritten character recognition device for various characters, symbols, etc.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an example of a character recognition device used in carrying out the present invention, FIG. FIG. 4, which is a partially enlarged view of the input cover shown in the figure, is a flowchart of an embodiment of the present invention. DESCRIPTION OF SYMBOLS 1... Coordinate memory, 2... Neighborhood coordinate number calculator, 2... Difference calculator, 22... Determiner, 23... Counter, 24... Addition value register, 25.26 ...Read counter, 3...Comparator, 4...Register,
5... Neighborhood coordinate number memory, 6... Centroid point calculator, 7
... Center of gravity coordinate memory, 8... Address counter,
9...Smooth point calculator, 91...Comparator, 92...
Determiner, 93... Counter, 94... Addition value register, 1o... Centroid point calculator, 11... Write counter, 12... Coordinate memory. Figure 2 (C1) (b) Figure 3 " ~ P5 Commissioner of the Japan Patent Office 21 Invention name character recognition method 3, relationship with the case of the person making the amendment Patent application office (105) Port of Tokyo Toranomon 1-7-12-4
Agent address (105) 1-7-1 Toranomon, Minato-ku, Tokyo
No. 2 1) "Coordinates of emphasis" on page 3, lines 13 to 14 of the specification is corrected to "coordinates of center of gravity." 2) In the 6th line of page 5 of the ministry, the phrase "distance within 0.8 vm" will be corrected to "both the X difference and Y difference are within ±0.8 m." 3) In the 6th and 7th lines of page 5 of the same ministry, the statement ``O, 2■, which is 10 bits'' should be corrected to ``rO, 1wm, which is 1 bit''. ,,1

Claims (1)

[Claims] In a character recognition method that performs character recognition by reading coordinate data of coordinate points that constitute a character, a region within a predetermined distance from the coordinate point is defined as a nearby area, a plurality of consecutive coordinate points are grouped together, and this set is 1. Calculating the number of coordinate points belonging to the neighboring area and the barycenter point coordinates as an average value of coordinate data indicating the coordinate points for each of the coordinate points in the area.
a second step of determining the coordinates of the center of gravity from the coordinates of the center of gravity corresponding to the coordinate point having the maximum number of coordinate points belonging to the neighboring area to obtain smooth point coordinates; A set of smooth points corresponding to the set of consecutive coordinate points is obtained by reconstructing the set including the next consecutive coordinate points, and repeating the first step and the second step sequentially for this reconstructed set. A character recognition method characterized in that character recognition is performed using the set of smooth points as input data to be recognized.