JPS60201485A - Handwritten character recognizing device - Google Patents

Abstract

PURPOSE:To attain character recognition of a form with high precision by extracting a slope of a character stroke from a character sample of a writer at each form and dividing the inside of the character frame from the center of barycentric coordinate of the character pattern and the stored slope into a partial area so as to calculate the feature element. CONSTITUTION:A sub-pattern extracting section 6 applies vertical scanning to a pattern register 2 so as to extract a vertical sub-pattern VSP from the relation between the consecutive length of black bit and the line width calculated at a line width calculation section 3. Similarly, a horizontal sub-pattern HSP, a right slope sub-pattern RSP and a left slpe sub-pattern LSP are extracted. A character frame detection section 4 detects a character frame circumscribed to the character pattern in the pattern register 2, gives the result to a character frame division deciding section 5, a feature matrix extraction section 7 divides the character frame area on the sub-patterns of VSP, HSP, RSP, LSP into JX1 partial area by using the division point coordinate decided by the character frame division decision section 5, the barycentric coordinate of the character pattern and the slope stored in the slope register. Then each data is normalized and finally a feature matrix of JX1X4 dimensions is generated.

Description

DETAILED DESCRIPTION OF THE INVENTION (Technical Field) The present invention relates to a high-speed and highly accurate handwritten character recognition device.

(Background Art) In conventional handwritten character recognition devices, in order to absorb variations in features due to variations in the slope of character lines due to differences in scribes, the variations in features were absorbed by using a plurality of dictionary masks. . However, with this device, the time required to compare extracted features with a dictionary during identification increases in proportion to the number of dictionary masks, resulting in a reduction in the processing speed of the device.

In order to eliminate this drawback, the slope of the scribe's character line is extracted from the horizontal sub-main turn and vertical sub-turn, which have been extracted from the scribe's character sample, and is registered at a predetermined address in the memory. During normal character reading, the handwritten character recognition device calls up the registered amount of inclination and determines the dividing area of the character frame according to this inclination, thereby absorbing variations in characteristics between scribes. Proposed.

However, this method required operations for registering and recalling the slope of the character line, which not only made the character reading procedure troublesome, but also had the disadvantage that different scribes' forms could not be read consecutively. .

(Objective and Summary of the Invention) The present invention aims to eliminate such conventional drawbacks and provide high-speed, high-precision character recognition. The slope of the character line is extracted and saved for each form from the scribe's character sample, and during normal character reading, the division point coordinates determined by the character frame division determination unit and the center of gravity of the character/mother turn are used. The purpose of this method is to recognize characters on the form by dividing the character frame into partial regions based on the coordinates and the stored inclination amount, and calculating characteristic elements.

(Embodiment of the Invention) FIG. 1 is a block diagram showing an embodiment of the handwritten character recognition device of the present invention. In the figure, 1 is a photoelectric conversion section, 2 is a turn register, and 3 is a line width calculation section. , 4 is a character frame detection unit,
5 is a character frame division determination unit, 6 is a sub pattern extraction unit, 7 is a feature matrix extraction unit, 8 is an identification unit, 9 is a control unit, 1o
1 is a memory section, 11 is a character output, 12 is a stroke extraction section, and 13 is a slope extraction section.

When a form as shown in the example in Figure 4 is loaded into the reading mechanism, it reads the scribe's character sample first, then the normal characters.Following this order, (a) the scribe's character sample The operation will be explained in the following order: extraction of the amount of inclination of a character line, and then ←) recognition operation of a normal character.

Extracting the amount of inclination using the scribe's character sample in (b) is performed as follows.

The scribe's character sample written in a predetermined entry frame on the form is converted into a binary quantized dinotal signal in the photoelectric conversion unit 1, and is stored in the quantan register 2. At the same time, the line width of the input pattern is calculated in the line width calculation section 3. In this example, the character samples are the kanji ``木'' and shita. The sub-guturn extraction unit 6 is...
Vertical sub-guturn (vsp
). Similarly, horizontal sub-scanning by horizontal scanning
Make an e-turn () (SP) and perform a right diagonal sub-ya turn (H8F) by scanning diagonally 45 degrees to the right and make a diagonal left 45 degrees.
A left diagonal sub pattern (LSP) is extracted by scanning. Figure 2 shows an example of the original pattern and each sub pattern (
a) is the original pattern, (b) is the vertical zabuno and turn (vs.
p), (c) are horizontal sub-mother turns (H4F), (
d) is a right diagonal sub-pattern (RAP), and (e) is a left diagonal sub-pattern (LSP).

The stroke extracting unit 12 performs a horizontal or vertical scan on each sub-gutern reno star, detects the change points from the white (5) point to the black point (with the character winning part as the black point), and from the black point to the white point. Extract the stop edge from the relationship between the number of changing points and changing point coordinates in the previous scan of the column (or row) and the number of changing points and changing point coordinates of the current column (or row), and i at both end points of the stroke
The coordinates in the two-dimensional coordinate system defined by the 9-turn register (the origin is at the lower left of the pattern register) are sent to the slope extraction unit 13.

The slope extracting unit 13 calculates the slope by referring to the coordinates of both end points of the stroke in the sub-pattern register extracted by the stroke extracting unit 12.

That is, the coordinates of both end points of the stroke extracted from the horizontal sub-turn H8P are (HXS, HY8) (HXEn
, HYEn) where n-1, . . . , N, N is the number of strokes, and the slope θ□ is calculated using equation (1). (However, HXE > ΣHLG n (6) HLGn in equation (1) represents the length of the stroke, which is determined by approximate equation (2).

HI, Gn2 MAX (l HXEn-T(XSnl
, l HYEn−HYSnl )+[MIN(IHX
En-HXSnl, IHYEn-)ffSn11]
/2(2) Equation (2) is an approximation equation in which the distance between two points is the sum of the smaller one of the differences in horizontal and vertical coordinates between the two points and the other one. Similarly, 07 is calculated using equation (3). However, vYEm>vYSm.

Note that M in the above is the number of strokes extracted from vertical dips and turns. When the number of groove strokes is 0, the slope is also 0. Further, the stroke length VLGm is calculated using an approximate equation similar to equation (2).

In the slope extraction unit 13, the slope θ of the sub-e turn H8P and VSP calculated by the above equations (1) to (3)
, 1° θ7 is controlled by the feature matrix extraction unit 7 via the control unit 9.
sent to and saved.

Next, the normal character recognition operation (b) is performed in the following steps.

Ordinary reading characters are converted into a binary quantized digital electrical signal in the photoelectric conversion section l and stored in the turn register 2. At the same time, the line width of the input/turn is calculated in the line width calculation section 3.

Zabu i? The turn extracting section 6 performs a vertical scan on the turn register 2, and extracts a vertical sub-gturn (vsp) from the relationship between the continuous length of black bits and the line width calculated by the line width calculation section 3. Extract. Similarly,
Horizontal sub-pattern (H4F) is created by horizontal scanning.
Right diagonal 45° scan allows right diagonal sub-turn (H
8F) to the left diagonally by scanning 45 degrees diagonally to the left.
and turns (LSP).

The character frame detection unit 4 detects a character frame circumscribing the character/mother turn in the turn register 2, and sends the result to the character frame division determination unit 5. The character frame division determination unit 5 divides the detected character frame into a JXI area (J and I are integers, in this embodiment, J=
The coordinates of the dividing point on the X-axis and Y-axis for dividing into 1=5) are determined. Here, the X axis indicates the horizontal direction of the character frame, and the Y axis indicates the vertical direction.

The division points are determined as follows. First, project the characters and patterns onto the X and Y axes to obtain the distribution of the number of black bits, respectively, and divide the first moment value of these distributions by the total number of bits to find the centroid coordinates x and y of the distribution. Ru. Furthermore, the distribution is divided into two by the obtained center of gravity, and the coordinates of the center of gravity of each divided region are calculated. Thereafter, the distribution is divided into a plurality of regions using the centroid determined so far, and the process of determining the centroid of each region is repeated to obtain a plurality of centroid coordinates. The division coordinates necessary for the desired number of divisions are selected from this barycenter coordinate series, and this and the barycenter coordinates Xc and Yo of the first character pattern are output to the feature matrix extraction section 7.

The feature matrix extraction unit 7 extracts the dividing point coordinates X, , X2 . ...XJ-4
(9) and Yl, Y2. . . . Yl-1, the barycentric coordinates Xc, Yc of the character pattern, and the tilt amount θ□ stored in the tilt register.

By θ7, VSP, H4F, H8F, L
JX character frame area on each supno and turn register of SP
Divide into partial regions of I.

The method for determining divided regions will be explained below.

1. Division point coordinates X1. X2...X, -1 and Yl,
Y2゜...Yl, , 1, Xc, Yc, inclination θ□, θ7, on a two-dimensional plane with the origin at the lower left of the character frame, (J+l-2) is expressed by equation (4) Define the straight lines of the book. (However, Y,〈Y2...〈Yl-1,X,
<X2...<X, , ) Any point A(x
o, ye) satisfies the inequality of equation (5), point A is included in the partial region (j, j).

θH(Xe-Xc)+Y, -4〈ye≦θn (x e
X c ) 10Y tk θv (ye-Yc) 10X-
1〈Xe≦θv(y8-Yo) 10
4) Divide the character frame into several (10) partial areas using the straight line in the equation as a boundary. FIG. 3 shows an example of division in the case of l=J=5.

Feature Ma) The IJ mark extraction unit 7 counts the number of black dots Bij in each area of each sub- and P-turn register in which the character frame is divided into partial areas using the method described above, and calculates the line width calculated by the line width calculation unit 3. Using W, calculate the features that determine the character line length using (6) to create a J X I X four-dimensional feature matrix.

Li j = Bij/W (6) After sonography, the VSP feature matrix is the length ΔY of the character frame in the Y-axis direction, the H8P feature matrix is the length ΔX in the X-axis direction, and the R8P and LSP feature matrix are ) IJ food is (Δχ
Normalize each with 1ΔY)/2 and finally JXI
Create an X4-dimensional feature matrix.

The identification unit 8 applies a distance (D)ffi defined by equation (7) between the dictionary mask (f, ) read out from the dictionary memory unit 10 and the extracted feature matrix (f),
The category name of the dictionary mask for which D has the minimum value is sent to the control unit 9. The control unit 9 is configured to output the received category name to the character output 11.

D=li〒〒〒100 cores (7) (Effect of the invention) As explained above, this invention calculates the slope of the scribe's character line from the scribe's character sample written within a predetermined frame of the form. By extracting and storing the character lines for each form, it is possible to efficiently absorb variations in characteristic elements caused by the slope of character lines caused by scribes, without having to extract the slope of character lines for each character. I can do it. Furthermore, the present invention
The method automatically extracts and saves the slope of the extracted character line for each form without registering or recalling it, making it ideal even when processing forms with different scribes in succession. It is possible to determine divided areas within a character frame in a short time, and it can be used in a high-speed, highly accurate handwritten character recognition device.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the handwritten character recognition device of the present invention, FIGS. 2(a) to (e) are diagrams showing examples of the original pattern and each sub-pattern, and FIG. 3 is a character frame. FIG. 4 is a diagram showing an example of a form used in the present invention. ■...Photoelectric conversion unit, 2... and turn register, 3
. . . line width calculation unit, 4 . . . character frame detection unit, 5 . . . character frame division determination unit, 6 .
...Feature matrix extraction section, 8...Identification section, 9...
Control unit, 10... dictionary memory, 12... stroke extraction unit, 13... slope extraction unit. Patent Applicant: Oki Electric Industry Co., Ltd. Patent Attorney: Megumi Yamamoto - (13) Amendment to Special V-Motion Procedures (Voluntary) August 7, 1980 Manabu Shiga, Commissioner of the Patent Office, Case 1 Indication of 1981 Patent Application No. 57/l 19 No. 2, Name of the invention Handwritten character recognition device 3, Relationship with the case of the person making the amendment Patent applicant name (029) Patent claim in the specification of Oki Electric Industry Co., Ltd. Scope column, Detailed explanation of the invention column, Brief explanation of drawings column, Drawing 6, Contents of amendment (11*The scope of claims is amended as shown in the attached sheet. (2) Page 4 of the specification “9 is...11” in line 9 is changed to “9
is the dictionary memory section, and is corrected as 10. (3) "12" on the 10th line of page 4 is corrected to "11", and "13" is corrected to "12". (4) On page 5, line 19, "12 is for each" and 111 is horizontal. and vertical”. (5) "Each" on page 6, line 5 of the same page: 1 horizontal and 1 vertical line.
and correct it. (6) Correct “13” on page 6, line 9 of the same page to “]2”. (Correct "13" in line 10 of page 6 of the same document to "12", and "12" to "11". (8) "13" in line 19 of page 7 of the same document to "12". (9) Delete “via the control unit 9” in the first line of page 8. (10) Change “Sent and saved” in the second line of page 8 to “
sent and stored in the slope register. ” he corrected. Old) "10" on page 11, line 15 is corrected to "19." (12) Correct "send to control unit 9" in line 19 of page 11 to "output to character output 10." (13) Delete "The control unit..." from the 19th line of the 11th page to the 1st line of the 12th page. (14) 19 from page 13, line 6 to line 7 of the same page...
・Control unit... ] 3... Incline" 19... Dictionary memory, 10... Character output, 11... Stroke extraction unit,
12...Inclination" is corrected. (Figure 1 of Drawing 151 is corrected as shown in the attached sheet.) (2. Photoelectrically converting the handwritten character pattern entered in the claims form and quantizing the digital signal obtained as the original pattern in the pattern register. The subpatterns in each direction extracted from the original pattern are stored in the character frame (for each partial area obtained by dividing, the number of black bits in that partial area is stored in the character frame corresponding to the character line width and stroke direction). A handwritten character recognition device that creates a feature matrix using the size of the character as a feature element, and recognizes the handwritten character pattern by referring to a dictionary in which standard character masks are described in the same format as the feature matrix,
The form 9 has a sample character entry frame at a predetermined position, and a digital signal obtained by photoelectrically converting and quantizing the scribe's character sample written in the entry frame is stored in a pattern register as an original pattern, and from the pattern. A subpattern that produces strokes in a desired direction is extracted, and the slope of the stroke obtained from the horizontal and vertical subpatterns is weighted and averaged using the length of the stroke as a weight. A handwritten character recognition device comprising means for extracting and storing a slope for each form, and determining division of the partial area according to the slope. (2)

Claims (1)

[Claims] The digital signal obtained by photoelectrically converting the f-turn of handwritten characters written on the form and quantizing it is stored in the mother turn register as the original and turf, and the sub and digits in each direction extracted from the original and mother turn are stored. For each partial area obtained by dividing the turn in the character frame, create a feature matrix using the number of black bits in the partial area as the character line width and the size of the character frame corresponding to the stroke direction as feature elements, In the handwritten character recognition device that recognizes the handwritten character pattern by referring to a dictionary in which a standard character mask is described in the same format as the feature matrix, the form has a sample character entry frame at a predetermined position,
A digital signal obtained by photoelectrically converting and quantizing the scribe's character sample written in the writing frame is stored in the turn recorder as an original pattern, and a sub-pattern that produces a stroke in a desired direction is extracted from the pattern. , a means for extracting and storing a value obtained by weighted averaging of stroke inclinations obtained from horizontal and vertical sub-graphs using the stroke length as a weight, as a character line inclination unique to the scribe for each form. A handwritten character recognition device 0 characterized in that the handwritten character recognition device 0 has: