JPH03670B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Technical Field) The present invention relates to a stable and accurate character recognition method.

(Background Art) Conventional character recognition devices absorb variations in extracted features caused by the inclination of character lines due to differences in handwritten characters by different writers, as shown in the example in Figure 1, and the inclination of printed characters. Therefore, variations in the characteristics have been absorbed by creating a plurality of dictionary masks.
However, with this method, the time required to compare the extracted features with the 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 point, a method has been proposed in which the slope is extracted from the horizontal sub-pattern and the vertical sub-pattern and divided regions are determined according to the slope.
In addition, the dividing point coordinates are generally determined using the distribution of the number of black dots obtained by projecting the character pattern onto the vertical or horizontal axis, so in the example shown in Figure 5, the dividing point coordinates are
It is located in the center of the two strokes as shown in Y ₁ and Y ₂ . Here, line a indicates a conventional division boundary.
As described above, in the conventional technique, the dividing boundaries were determined according to the above-mentioned inclination based on the left side and the bottom side of the character pattern, which had the disadvantage that the stroke was cut off as shown in FIG. 5, and its characteristics became unstable.

(Problem to be solved by the invention) The present invention extracts the slope of each sub-pattern in order to eliminate defects as described above, and furthermore, in the step of creating a feature matrix, the slope is extracted based on the dividing coordinates on the axis passing through the center of gravity of the character pattern. By defining division boundaries according to the following, instability of features due to stroke cutting is removed, and the purpose is to provide a stable and highly accurate character recognition method.

(Structure and operation of the invention) FIG. 2 shows a block diagram of an embodiment of the character recognition system of the invention. In the figure, an optical signal of a character is converted into a binary quantized digital electrical signal from an optical signal input 1 by a photoelectric converter 2, and is stored in a pattern register 3. At the same time, the line width W of the input pattern is calculated in the line width calculating section 4. The sub-pattern extraction unit 5 performs a vertical scan over the entire surface of the pattern register 3, and extracts vertical sub-patterns based on the relationship between the continuous length of black dots (character line portions are black dots) and the line width calculated by the line width calculation unit 4. Extract the pattern (VSP) and store it in the subpattern register. Similarly, the horizontal subpattern (HSP) is extracted by horizontal scan, the right diagonal subpattern (RSP) is extracted by right diagonal 45° scan, and the left diagonal subpattern (LSP) is extracted by left diagonal 45° scan. Store in. Figure 3 shows an example of the original pattern and each sub-pattern, where a is the original pattern, b is the vertical sub-pattern (VSP), c is the horizontal sub-pattern (HSP), d is the right diagonal sub-pattern (RSP), and e is the left diagonal. It is a subpattern (LSP).

The stroke extraction unit 6 performs a horizontal or vertical scan over the entire surface of each sub-pattern register, detects points of change from a white point to a black point, and from a black point to a white point,
Strokes are extracted from the relationship between the number of change points and change point coordinates in the scan of the previous column (or row) and the number of change points and change point coordinates in the current column (or row), and the strokes in each extracted subpattern register are The coordinates of both end points in the two-dimensional coordinate system defined by the pattern register 3 (with the lower left of the pattern register as the origin) are sent to the slope extractor 7. The slope extraction unit 7 refers to the coordinates of both end points of the stroke in each subpattern register extracted by the stroke extraction unit 6, and calculates the average slope of the horizontal subpattern HSP and the vertical subpattern VSP. In other words, the coordinates of both end points of the stroke extracted from the horizontal subpattern HSP are (HXSn, HYSn), (HXEn, HYEn),
However, n=1,..., P, P is the number of strokes (1)
Calculate the slope Q _H by the formula. (However, HXEp>
HXSp) HLGp in equation (1) represents the length of the stroke, and is determined by the approximate equation of equation (2).

HLGp＝MAX｜HXEp−HXSp｜、｜HYEp
−HYSp｜｝+
MIN{|HXEp−HXSp|, |HYEp−HYSp|}/2 (2) Equation (2) calculates the distance between two points as 1/2 of the smaller of the horizontal and vertical coordinate differences between the two points. This is an approximate expression that is the sum of the other one.

Similarly, θ _V is calculated using equation (3). However, VZEq>
Let it be VZSq.

Note that Q in the above formula is the number of strokes extracted from the vertical sub-pattern. Furthermore, when the number of strokes is 0, the slope is also 0. Further, the stroke length VLGq is calculated using a formula similar to formula (2).

The slope extraction unit 7 sends the slope of each sub-pattern calculated from the above equations (1) to (3) to the feature matrix extraction unit 10. The character frame detection section 8 is the pattern register 3
A character frame circumscribing the character pattern within is detected, and the result is sent to the character frame division determining section 9.

The character frame division determining unit 9 divides the inside of the detected character frame into M
×N area (M, N are integer constants, in this example, M=
Coordinates of dividing points on the X-axis and Y-axis for dividing into N=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, the character pattern is projected onto the X-axis and Y-axis to obtain the distribution of the number of black bits, respectively, and the center of gravity of the distribution is determined by dividing the first moment value of the distribution by the total number of black bits. Furthermore, by repeating the procedure of dividing the distribution into two at the obtained center of gravity and calculating the center of gravity of each distribution, a plurality of coordinates of the center of gravity are obtained. 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 YC of the first obtained character pattern are output to the feature matrix extraction section 10.

The feature matrix extraction section 10 extracts the division point coordinates X ₁ , X _{2 .} . . , X _M-1 determined by the character frame division determination section 9.
and each sub-pattern of VSP, HSP, RSP, and LSP using Y ₁ , Y ₂ , ..., Y _N-1 , the barycentric coordinates XC, YC of the character pattern, and the tilt amounts θ _H , θ _V obtained from the tilt extraction unit 7. Divide the character frame area on the register into M×N partial areas. The method for determining divided regions will be explained below.

first. Dividing point coordinates X ₁ , X ₂ ...X _M-1 and Y ₁ ...Y _N-1
Then, (M+N-2) straight lines shown by equation (4) are defined on a two-dimensional plane with the origin at the lower left of the character frame by XC, YC and the inclinations θ _H and θ _V. (However, Y ₁ <
_{Y 2} ... _< Y _{N - 1 , X 1 <} (y-YC) + _{X 1} 〓 x=θ _V (y-YC) + Included in partial area (m, n).

{(xe−XC)・θ _H +Yn ^- ₁ }＜ye≦{(xe−XC)・
θ _H +Yn} and {(ye−YC)・θ _V +Xm ^- ₁ }＜xe≦{(ye−YC)・
θ _V +Xm} (5) However, Y ₀ =X ₀ =0, Y _N =X _M =∞ In other words, the inside of the character frame is divided into a plurality of partial areas using the straight line in equation (4) as a boundary. FIG. 4 shows an example of division in the case of M=N=5. Feature matrix extraction unit 10
is calculated by counting the number of black dots Bij in each area of each sub-pattern register in which the character frame is divided into partial areas using the above method, and using the line width W calculated by the line width calculation unit 4 to form an equation.
Calculate the feature representing the character line length using (6) and calculate M×N
×Create a four-dimensional feature matrix.

Lij=Bij/W (6) Furthermore, the length of the VSP feature matrix in the Y-axis direction of the character frame is ΔY, the HSP feature matrix is the length in the X-axis direction is ΔX, and the RSP and LSP feature matrices are (ΔX+ΔY)/2. Each is normalized to finally create an M×N×4-dimensional feature matrix. The identification unit 12 applies the distance (D) defined by equation (7) between the feature matrix (F) extracted by the feature matrix extraction unit 11 and the dictionary mask (f) prepared in advance, and calculates the distance (D) defined by equation (7), and determines whether D is the minimum. The category name of the dictionary mask serving as the value is output to the character name output 12.

D=√(-) ² (7) (Effects of the Invention) As explained above, in this embodiment, the slope of the stroke of the horizontal sub-pattern and the vertical sub-pattern is extracted, and the dividing coordinates are further set on the axis passing through the center of gravity of the character. Since the dividing area is determined according to the inclination of the stroke using the coordinates as a reference, there is an advantage that stable features can be obtained without cutting the stroke.

In other words, since the character frame is divided along the inclination by focusing on the center of gravity of the character, there is an advantage that features that follow the deformation of handwritten characters can be extracted.

The present invention extracts the inclination of strokes in each direction within a character pattern, and furthermore, in the step of creating a feature matrix, dividing regions within a character frame pass through division coordinates determined on an axis passing through the center of gravity of the character pattern. Since the dividing boundary is determined based on the inclination of the stroke, character lines are less likely to be cut and stable features can be obtained, which can be used in a stable character recognition device with high recognition accuracy. .

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an example of handwritten characters, FIG. 2 is a structural diagram showing an embodiment of the character recognition device of the present invention,
FIG. 3 is a diagram showing an example of the original pattern and each sub-pattern, FIG. 4 is a diagram showing an example of area division, and FIG. 5 is a diagram explaining the drawbacks of the conventional technique. 1: Optical signal input, 2: Photoelectric conversion section, 3: Pattern register, 4: Line width calculation section, 5: Sub pattern extraction section, 6: Stroke extraction section, 7: Slope extraction section, 8: Character frame detection section, 9: Character frame division determination section,
10: Feature matrix extraction section, 11: Identification section, 1
2: Character name output.

Claims (1)

[Claims] 1. A sub-pattern representing a stroke in a desired direction is extracted from a character/figure pattern, and the slope of the sub-pattern is obtained by weighting the slope of the stroke obtained from each sub-pattern using the length as a weight. For each partial area obtained by extracting and dividing each sub-pattern within the character frame according to the slope, the number of black bits in the partial area is divided into a character frame corresponding to the character line width and stroke direction. In a character recognition method, a feature matrix is created using the quantities obtained by normalizing with the size of the feature element, and character patterns are recognized by referring to a dictionary in which standard character masks are described in the same format as the feature matrix. , When creating a feature matrix, the division coordinates are determined on the axis passing through the center of gravity of the character pattern, and by the division boundary that passes through this and follows the slope of the sub-pattern,
A character recognition method characterized by dividing into partial areas.