JPS603077A - Inclination extraction system - Google Patents

Abstract

PURPOSE:To extract an inclination stably at a high speed by calculating the weighted mean value of inclinations of stroke components extracted from a character pattern on the basis of the length of each stroke component as weight. CONSTITUTION:A stroke extraction part 6 performs a horizontal or vertical scan on the entire surface of each sub pattern register to detect change points from a white dot to a black dot and from a black dot to a white dot. Then, a stroke is extracted on the basis of relations between the number of change points in scanning on the last column (or row) and coordinates of the change points, and the number of change points on the current column (or row) and coordinates of the change points. The coordinates (based upon the left lower point of a pattern register as an origin) of both terminal points of an extracted stroke in each sub pattern register on a two-dimensional coordinate system defined by a pattern register 3 are sent out to an inclination extraction part 7. The inclination extraction part 7 refers to the coordinates of both terminal points of the stroke in each sub pattern register which is extracted by the stroke extraction part 6, and calculates the mean inclination of each sub pattern.

Description

DETAILED DESCRIPTION OF THE INVENTION (Technical Field) The present invention relates to a fast and stable slope j'' extraction method.

(Background Art) Conventionally, in a character recognition device, the inclination of the character line due to the difference in the number of characters of various handwritten characters as shown in the example of Fig. 1, and the inclination of the character line due to the inclination of the printed character are In order to absorb the resulting variations in the extracted features, the variations in the features have been absorbed by means of multiple layer masks. However, with this method, the time required to compare the extracted η phrases and signs with the boundary layer during identification increases in proportion to the number of layer masks, resulting in a reduction in the processing speed of the device. In order to eliminate this drawback, there is a method to improve the processing speed by extracting the character line slope in each direction and selecting a dictionary mask to use.However, in the conventional slope extraction method C1, for each direction, Since it is calculated by a simple average of the slope of each stroke, the #i slope to be extracted is unstable due to the influence of short strokes with unstable slope trends fJ:I)
This in turn leads to a decline in recognition performance.

(Objective of the Invention) In order to eliminate all such defects, the present invention calculates a weighted average of the slopes of stroke components extracted from character patterns, using the length of each stroke component as a weight. It is characterized by calculating the slope of character lines, and its purpose is to provide a fast and stable slope extraction 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, an optical signal of a character is converted into a binary quantized digital electrical signal by an optical signal input 1 and a photoelectric conversion section 2, and is stored in a pattern register 3. At the same time, the line width (W) of the human pattern is calculated in the line width calculating section 4. The sub-pattern extractor 5 vertically scans the entire surface of the putter 7 register 3 and calculates the relationship between the continuous length of black dots (all black dots in the character line area) and the line width calculated by the line width calculator 4. C,
Extract vertical subpattern ~7 (VSP)-i.

Similarly, by horizontal scanning, horizontal sub pattern 7 (1-1
S P ) 2, right diagonal 45° scan right diagonal sub putter 7 (RS P )-q, left diagonal 45° ski/
4 left diagonal sub patterns (LSP) are extracted. Third
The figure shows examples of the original pattern and each sub pattern, where (a) is the original pattern, (1)) is the vertical sub pattern (vsp), and (C
) is a horizontal sub-pattern (IISP), (d) fi right diagonal sub pattern (1 is also SP), and (c) qx left diagonal sub pattern 7 (LSP).

The stroke extraction unit 6 performs horizontal or vertical scanning over the entire surface of each sub-pattern register, and extracts from the white point to the black point,
Detect the change point from a black point to a white point, and calculate the stroke based on the relationship between the number of change points and the coordinates of the change point in the previous column (or row), and the number of change points and the coordinates of the change point in the current column (or row). Extract the coordinates (with the lower left of the pattern register as the origin) in the two-dimensional coordinate system defined in the pattern register 3 of both end points of the stroke in each of the extracted Zabu putter 7 registers] 1 to the old B part 7 Send. The slope extraction section 7 refers to the coordinates of both end points of the stroke in each sub-pattern 7 register extracted in the stroke extraction capital 6, and calculates the average slope for each subpattern. That is, the coordinates f (HXS
n, , HYS rl), (1-IXEn.

HYEn), where n-11--+ P, Pta, and the number of strokes, calculate the reverse slope QH using equation (1) (
However, 1-IXEp') f(XS p ).

f-IL G p in equation (1) represents the length of the stroke, which is determined by the approximate equation of equation (2).

1-(LGp = MAX (1l-IXEp -HX5
p 1.11-1. YEp-1-IYS p I)
Equation (2) is an approximate expression in which the distance between two points is the sum of 1/2 of the smaller one of the horizontal and vertical coordinate differences between the two points and the other one. Similarly, vVlθR, 0t, ff1(3)
~ (5) Equation 2 calculation table. However, VYEq) V
YSq.

RXE t) RXS t, LXE k) L
Let it be X8k.

ΣVLGq ■ Note that in the above formula, Q, L, and 1st are the number of strokes extracted from the vertical sub pattern, right diagonal sub pattern, and left diagonal sub/turn, respectively. Furthermore, when the number of strokes is O, the slope is also assumed to be Ω. Ma7j Strokes Long VLGq
, i are also calculated using the same calculation as LGt, LLGktrL (2). .

The slope extraction section 7 sends the slope of each sub putter 7 calculated according to the above equations (1) to (5) to the dictionary selection section 8. , the character frame detection unit 9 detects a character frame circumscribing the character pattern in the pattern register 3, and sends the result to the character frame division determination unit 1.
0-\Send.

The character frame division determining unit 10 converts the inside of the detected character frame into MX.
N(7) area (M, NtrJ, integer, in this example ■\
4-N-5) Xjl! ll+, 'Y
Determine the coordinates of the division points on the bridge. Here, the X axis indicates the horizontal direction of the character frame, and the Y axis indicates the vertical direction.

The feature matrix extraction unit 11 extracts VSP and flsP based on the division point coordinates determined by the character frame division determination unit 10.

Divide the character frame area on each sub-pattern register (J, SP, LSI) into 1\+I x N areas, count the number of black dots in each area (131), and use the line width calculation unit 4 to calculate the number of black points (131).
Express the character line length using the following equation (6) using the line width W obtained by +-. 1! Divide all i characteristics to create a fvlxNx4-dimensional feature matrix.

Li j = Bi j /W (6) Then, V S
The P feature matrix is the length ΔY of the character frame in the Y-axis direction,
H8P feature matrix BX axis direction length ΔX, ILS
The P and LSP feature matrices are each normalized by (Δχ+ΔY)/2 to finally create an M×N× four-dimensional feature matrix.

The slope 1' selection unit 8 is output from the slope extraction unit 7 and the slope θ is output from the slope extraction unit 7.
With reference to H, θ■, θR+θany k, a selection signal for selecting the /ζ dictionary suitable for the input character pattern is sent to the dictionary memory section 13. Dictionary 1: In the memory unit 13, several types of dictionary masks are prepared for each rudder machine, which are classified and created in advance for each /[M' characteristic. In this example, g is υ
In the memory section of the two words, there are VSP η temples, Aokikusu, 1-I.
SP chronological matrix, 1 is also SP % characteristic matrix, I
, 1 masks corresponding to three types of slopes were prepared for each of the SP feature matrices. I and I of each q blue mold matrix are respectively) θn<-0,25, -0,25<θH<0.25, 0
．． 25 <3 types of θH θv< 0.25. -0.25
Three types of θR: <θv<0.25, 0.25 <θ■
< 0.7, 0.7 <θR・te14 ・】4 kuθI? 3 ff1f class θL<-1,,4,-1,4
, <θL<-0,,7, -0,7≦θI7. The dictionary selection unit 8 outputs a signal for selecting a dictionary for each feature matrix corresponding to the slopes θH4θ■ and θR7θL obtained from the slope extraction unit. The dictionary memory unit 13 causes the identification unit 2 to refer to the dictionary for each feature specified by the selection signal. The distance (1) defined by equation (7) is applied between the feature matrix (fl) and the category name of the word 1 mask such that D is the minimum value is output to the full character name output 14. It is something.

D issue Since it is extracted by weighted averaging of all the strokes, the advantage is that it can remove the influence of the instability of the slope of strokes where the slope angle of the character line in each direction in the character pattern 7 is short and extract it stably. There is.

In other words, the process is simple as a method for extracting all the slope trends of character lines due to differences in scribes, and if Zabbata-7 used in the major classification method is used, it is consistent with the major classification method. There is no problem with this point either.

(Effect of the invention) ``Dufu'' extracts the slope I of the character line in each direction in the character pattern / and calculates the slope I of each stroke in this subpattern.
1 slope is extracted by weighted averaging using the length of the stroke cork as a weight, so the slope remainder 1 of a short stroke is
This method has the advantage that slope extraction is stable without being affected by instability, and can be used in character recognition devices with high speed and high recognition accuracy.

[Brief explanation of drawings]

Figure 1 is an example of handwritten characters, Figure 2 is an example of character recognition system M of the present invention.
, a configuration diagram showing an example of the raw butter/
and FIG. 7 is a diagram showing an example of each sub putter 7. ■, Kojogo Jinriki, 2. Photoelectric conversion, 3 Batarji 7'l
, 4: Line width word calculation section, 5. Zabu putter 7 extraction capital, 6: stroke extraction section, 7 slope extraction section, 8: dictionary selection section, 9:
Character frame detection unit, 100 character frame division determination unit, 11: Feature matrix extraction unit, 12: Identification unit, 13: Dictionary memory, J
4: Character name output.

Claims (1)

[Claims] In order to select a dictionary mask for a character recognition device, character/figure patterns are scanned and detected in multiple predetermined directions.
This is a slope extraction method that extracts stroke components from each of the 7 sub-butters created by extracting all cross-sections whose cross-sectional length is sufficiently longer than the line width in the character figure pattern from among the character line breaks...j. Q, and the weighted average of the stroke component of each extracted sub-pattern with the length of each stroke component, 7, (R'H1), is taken as the slope of the 9th turn of the sub-pattern. The trend 1 extraction method is characterized by extraction.