JPS603070A - Character recognition system - Google Patents

Abstract

PURPOSE:To improve the precision of a character recognition system by removing the unstableness of a feature due to the break of a stroke. CONSTITUTION:A subpattern extraction part 5 extracts a vertical subpattern (VSP) on the basis of the relation between the length of the continuation of a black dot obtained by making a vertical scan over the entire surface by a pattern register 3 and the line width calculated by a line width calculation part 4. Similarly, a horizontal subpattern (HSP) is extracted by a horizontal scan, a right slanting subpattern (RSP) is extracted from a right slanting 45 deg. scan, and a left slanting subpattern (LSP) is extracted by a left slanting 45 deg. scan. A stroke extraction part 6 makes a horizontal or vertical scan over the entire surface by respective subpattern registers to detect the change point from a white dot to a black dot and a black dot to a white dot.

Description

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

(Background Art) In conventional character recognition devices, the inclination of the character line due to the difference in size of handwritten characters, as shown in the example in Fig. In order to absorb all of the resulting variations in the extracted features, the variations in the features have been absorbed by means of creating a plurality of masks. However, in 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 that extracts the slope from the horizontal dip pattern and the vertical leap pattern and defines divisions and regions according to the slope. The division coordinates are generally determined using the distribution of the number of sunspots obtained by projecting the character pattern on the vertical or horizontal axis (2 lines, as shown in Y, , Y2 in the example in Figure 5). Here, the line ta+ indicates the conventional dividing boundary. However, in the conventional technique, the left side and bottom edge of the character pattern were defined as the base (the dividing boundary was defined according to the above-mentioned slope, so as shown in Fig. 5) The stroke was cut off,
The drawback is that its characteristics are unstable.

(Problem to be solved by the invention) In order to eliminate such drawbacks, the present invention extracts the slope of each sub pattern, and furthermore, in the stage of creating a feature matrix, Mi+ is written based on the dividing coordinates on the axis passing through the center of gravity of the character pattern. By defining division boundaries whose slopes follow F, the instability of features due to cut strokes is removed, and the purpose of Section 7 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 device of the invention. In the figure, the original signal for the characters is optical signal human power 1
In Gendenhen Kyoto 2, 2 Hani's sleeves are transformed into Teijikuru'11 Genki Signal, and ζ Turn Register:
Stored in 3. At the same time, the line width (w) of the line 1i1,5; the line width (w) of the metering section 4 is applied by force 4j≦9−. The sub pattern extraction unit 5 performs a vertical scan over the entire surface of the nocturne resistor 3, and calculates the length of the continuous black dots (letter line portions are black dots) and
1' The relationship between the calculated width and the vertical line width is 7.
Extract SI). Similarly, from the horizontal scan point, make a horizontal sub-turn (IISP) diagonally 45 degrees to the right.
Right diagonal sub < 7 (R8P) due to scan
, left; f, mi [me45'' Sugiyan extracts the left rfFme subtsuki turn (LS P) 'e.The third worst is Haranoku turn and examples of each Zabu pattern, (a) is Haranoku turn, (blV vertical sub pattern (VSP), (C)
is a horizontal sub' turn (IISL)), ((1) is a right diagonal double pattern (1LsP), (ei is a left diagonal double pattern (LS, 1)).

The stroke extraction capital 6 is a horizontal or "Ii) direct scan in each turn register for a total of 1r1j rows (4
, detect the change points from a white point to a black point, and from a black point to a white point, calculate the number of change points and the coordinates of the change points in the previous scan of one column (or row), and the number and change of change points in the current column (or row). Strokes are extracted from the relationship of point coordinates, and pattern register 3 of both end points of the stroke in each extracted sub-pattern register is created.
The coordinates in the two-dimensional coordinate system defined by (with the origin at the lower left of the pattern register) are sent to the slope extraction unit 7.

The slope extraction section 7 refers to the subpattern register (ISI) extracted by the stroke extraction section 6 and the coordinates of both end points of the stroke of the VSP and converts the coordinates of the stroke of the VSP and the equal slope 7 of the VSP to the original A]. ν1j, the stroke number both end points j414 extracted from the horizontal sub-pattern 118P (I-1, XS, , H, YS, , ).

(HXE, , )IYEn) , however, n21, -,
P, ]-) is the number of strokes, and the expression 'tQ+( is incremented by 1 using the formula (]).

(However, IIXEp) llX5 p) Σ IILGp In equation (1), T-I L G p represents the length of the stroke, which is determined by the approximate equation of equation (2).

1.1LG1) MA Kariin XI Yumiichi 14Xsp l
, l IIY, Ep-11YSp l )M, lN(1
jlX, No. 1,, 1lXsp l, IIIYE, -
ILYS, I) (2) Equation (2) is the distance between two points 1ζ1[, the horizontal and vertical pressure between the two points, the sum of the smaller knee and one of the pieces at the top of t. This is a similar expression to Tozuru.

11.'' θV is calculated by Ni'l using the equation (:3).

However, let V'I' be disease>VYS.

Note that Q in the notation is the number of strokes extracted from the vertical sub pattern. When the number of strokes cut is O, the inclination is also O. The length of the stroke, VLG, is calculated using a formula similar to formula (2).

The slope extracting unit 7 sends the slope of the valley pattern calculated using equations (1) to (3) to the feature matrix extraction group 10. Character frame detection 1? b8 is a pattern register; detects all the character frames circumscribing the character pattern in the box, and sends the results to the all character frame division determining section 9;

The character frame division determining unit 9 determines dividing points on the X It'll and Y axes for dividing the detected character frame into MxN areas CM, an N-digit integer (M=N-5 on the false male side). Determine the coordinates. Here, the X axis indicates the water/1' direction of the character frame, and the Y+axis does not indicate the vertical direction.

The division points are determined as follows. The distribution of the number of n black pins is calculated by projecting the sweet character/cutan onto the Xl'1III and Y axes and dividing the first moment area of the distribution by the total number of bits. Find the center of gravity of. Furthermore, by repeating the procedure of dividing the entire distribution of the selected car into two and calculating the center of gravity of the distribution of 10 points, the center coordinates of the multiple points are obtained. From this barycenter coordinate series, all the division coordinates necessary for the desired number of divisions are selected, and the barycenter +=marks XC, YC of the first character turn made by Koga are output to the feature matrix extraction unit IO.

The feature matrix extraction capital 10 is based on the division point coordinates X+, X2., XM-1, and Y determined by the character frame division determination unit 9.
+, Y2, ..., YN-1 and the center of gravity of the character pattern IGI≦marks
P, ft5I', , r, SP The character frame area on each sub iZ turn register is iVl,
1'f'J for a certain minute. Below (111, l・15Cχ
Explain how to determine the area.

1, dividing point j・14. Drift X, , X2・XM-1 and Y, -YN-1 and XC, YC and flJ-IIN:
1 Set the origin at the bottom left of the character frame by using Europe On + θV 2
Indicate on the dimension Δ1/- plane using equation (4) (M + N-2)
The straight line of a book is silent. (However, yt<Y2-<"l'
N + + X+ <
When the i-phrase is added, point A ends up in the partial song range (m r n ).

(xe-XC) (x'e-XC) 'lI 10Y11-] < ye≦olI10Yn and (
ye-YC) (yc-YC) θv +X,, -1 (xe≦θV +, Xm (5) However, YO=XO=O, YH=X+v+-", i.e. (4)
Assume that the entire boundary of the equation is a straight line - the letter C is divided into multiple subregions. The 41st number j is an example of division when M=N=5. Features 1. l) The line width extraction unit 10 calculates all the black dots in each sub-pattern region of each sub-pattern register divided into 1-11S regions using the method described above, and the line width calculation unit 4 calculates the total number of black dots. Using all the line widths \■, calculate the character line length using 2 (6), and create a four-dimensional feature matrix.

Furthermore, the length of the VSP feature matrix in the Y-axis direction of the character frame is ΔY, the length of the 11SP four-character matrix is ΔX in the x #III+ direction, and the R3P and LSP feature matrices are (
Perform general normalization using Δχ+ΔY)/2 and finally i
vl X N X Create a trix with four-dimensional features.

The identification unit 12 calculates a distance (I) defined by the formula (7λ) between the feature matrix extracted by the feature matrix extraction unit 11 (1) and the dictionary mask (f)) prepared in advance.
? r T, which outputs the category name of the dictionary mask for which D is the minimum value for JM, character name output 12-\)
−jΣage1−fj)” (71 (Effect of the invention) As explained above and in 7., in the Kokuho example, the slope of the stroke of the horizontal sub-pattern and the vertical sub-pattern is extracted, and furthermore, it is Set the division coordinates on 1iQh,
Since the divided regions are determined according to the slope of the stroke based on the ratio of each stroke, there are 7111 points in that stable features can be obtained without cutting the stroke.

In other words, since the character frame is divided diagonally into i~ by focusing on the character's one-ton flow, 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 the character Bakun, and furthermore, in the stage of creating a feature matrix, the divided regions within the character frame are passed through the divided coordinates determined on the axis tracing the center of gravity of the character pattern. Since the division boundary is determined based on the inclination of the stroke, the character line is less likely to be cut and has a stable feature (4), which makes it suitable for character recognition devices with stable and high recognition accuracy. ” can be used.

[Brief explanation of the drawing]

Fig. 1 is a diagram showing an example of handwritten characters, Fig. 2 is a block diagram showing an embodiment of the character recognition device of the present invention, and Figs. I-J are diagrams showing examples of the original pattern and each sub-pattern. , Fig. 4 is a diagram showing an example of area division, Fig. 51
Y1 is a diagram that explains all the shortcomings of the conventional technology 1: Light 13
Goujin Tsuba 2: Photoelectric conversion section, 3: Pattern resistor, 4
:Line IM pattern part, 5: Sub pattern extraction 4 prefectures 6 2 stroke extraction part, 7: Slope extraction part, 8: Character frame detection t1
13.9: Character frame division determination unit, 10: Feature matrix extraction unit, 11: Identification unit, 12: Character name output. Patent applicant Oki Electric Industry Co., Ltd. Patent attorney Megumi Yamamoto −

Claims (1)

[Claims] A sub-pattern that produces a stroke in a desired direction is extracted from the character figure pattern, and the slope of the stroke obtained from the valley pattern is weighted and averaged using the length as a weight, and the slope of the sub-pattern is taken as the slope of the sub-pattern. 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 determined by the character line width and the size of the character frame corresponding to the stroke direction. In an electronic character recognition method, a feature matrix is created using the quantities obtained by normalization as feature elements, 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. A character recognition method characterized in that, when creating a matrix, division coordinates are determined on an axis passing through the center of gravity of a character pattern, and division into partial areas is carried out using division boundaries that pass through the coordinates and follow the inclinations of the sub-patterns described in 111S.