JPH04591A - Method for feature extraction - Google Patents

Abstract

PURPOSE:To stabilize extracted features even in the case of an inclined character pattern and to attain rapid character recognition to be executed without using a dictionary by extracting the inclination of an auxiliary subpattern based upon the inclination of a line element included in the subpattern, scanning a character pattern based upon the inclination and extracting the feature of the pattern. CONSTITUTION:The 1st subpattern extracting part 6 scans a character pattern stored in a pattern register 4 in vertical and horizontal directions. A subpattern expressing a line element component in the scanning direction is extracted based upon the relation between continuous black bits on the scanning line and the line width of the pattern. A character inclination extracting part 7 finds out the inclination of the subpattern based upon the inclination of the line element included in the subpattern. The 2nd subpattern extracting part 10 scans the character pattern in respective directions based upon the precedently obtained inclination, extracts the subpattern and then extracts a feature. Thereby, the extracts feature can be stabilized even in an inclined character pattern, the preparation of a dictionary corresponding to the deformation of character inclination in an identification part 12 is made unnecessary, a collating time can be shortened and a processing speed can be increased.

Description

Detailed Description of the Invention (Industrial Application Field) The present invention provides a feature extraction method for scanning a character pattern obtained by photoelectrically converting characters on a medium in multiple directions and extracting features from the character pattern. , especially regarding a fast and stable feature extraction method.

(Prior art) Conventionally, this type of feature extraction method was disclosed in Japanese Patent Application Laid-open No. 57-2.
Publication No. 3185 (Reference 1) and Special Publication No. 38-5555
There was one described in Publication No. 1 (Reference 2).

In the technique of Document 1, subpatterns representing line element components (stroke components) in each direction are extracted from a character pattern obtained by photoelectrically converting characters on a medium. Then, the sub-pattern is divided into multiple regions, and for each divided region,
A feature matrix is generated by extracting features representing the amount of line elements in the sub-pattern. Then, the feature matrix and
The characters were recognized by comparing them with a dictionary prepared in advance.

Further, in the technique of Document 2, the number of intersections between a scanning line and a stroke is extracted as a feature amount by scanning a character pattern.

(Problems to be Solved by the Invention) However, the above feature extraction method has the following problems.

(i) FIG. 2 is a diagram showing an example of printed characters in italics. The third diagram is a schematic diagram of the numbers ゛○゛ in this figure.
Shown in Figures (a) to (d). That is, FIG. 3(a) is an example of the non-italic character pattern '○゛, and FIG. 3(b) is an example of the italic character pattern "0°".
An example of the same figure (C) is the same figure (a
), and (d) in the same figure is a vertical sub-pattern extracted from the character pattern in (b) using the method of Document 1.

When the vertical sub-patterns of the character pattern shown in FIG. 3 (a>) are extracted using the feature extraction method of Reference 1, the vertical components can be faithfully extracted as shown in FIG. 3 (C).

However, when a vertical sub-pattern is extracted for the italic letters in FIG. 3(b), the line element components are tilted as shown in FIG. 3(d). Moreover, since the line element component shown by the broken line in FIG. 3(d) is to be extracted, the line element is scanned perpendicularly to the inclined line element, so both ends of the line element are missing. Because of this, there was a problem that the extracted features were different (in other words, the extracted features were unstable) because the same glyph shape was "0°".In addition, as a deformation of handwritten characters, Since most of the features are upward-sloping to the right and the horizontal line elements are tilted, there is a problem that the extracted features are unstable, similar to the above.

Therefore, when character recognition is performed by comparing characters with dictionaries after feature extraction, it is necessary to prepare a large number of dictionaries corresponding to the deformation of the character slope. As a result, the dictionary capacity increases, the processing speed decreases due to longer verification times, and the hardware scale (device scale) increases, which are disadvantageous.

(ii) In the technique of Document 2, the number of intersections between a scanning line and a line element is used as a feature quantity, so if a line element has an inclination as shown in Fig. 3(b), it is difficult to extract when the scanning direction is fixed. The number of intersections (features) that are generated varies greatly, and the same problem as in (i) above occurs.

The present invention provides a feature extraction method that solves the problems of the prior art, such as instability of extracted features, reduction in processing speed, and increase in device size.

(Means for Solving the Problem) In order to solve the problem, a first invention scans a character pattern obtained by photoelectrically converting characters on a medium in a plurality of directions and extracts features from the character pattern. In this feature extraction method, the following measures are taken.

That is, the character pattern is scanned in the vertical direction and/or the horizontal direction, and line elements in the scanning direction are determined based on the relationship between the number of consecutive black pixels in the scanning direction and the line width of the character pattern. After extracting an auxiliary sub-pattern representing the component and extracting the slope of the auxiliary sub-pattern from the slope of line elements included in the auxiliary sub-pattern, the character pattern is scanned in a plurality of directions based on the slope of the auxiliary sub-pattern. Features are extracted from the character pattern.

In the second invention, a character pattern obtained by photoelectrically converting characters on a medium is scanned in multiple directions, and line elements in the scanning direction are determined based on the relationship between the number of consecutive black pixels in the scanning direction and the line width. In a feature extraction method of extracting a subpattern representing a component, dividing the memorized subpattern into a plurality of regions, and extracting a feature representing the amount of line elements of the subpattern for each divided region, the following means is used. This is a lecture on

That is, the character button is scanned in both the vertical direction and the horizontal direction, and the line elements in the scanning direction are determined based on the relationship between the number of consecutive black pixels in the scanning direction and the line width of the character pattern. After extracting an auxiliary sub-pattern representing a component and extracting the slope of the auxiliary sub-pattern from the slope of a line element included in the auxiliary sub-pattern, the sub-pattern is extracted based on the slope of the auxiliary sub-pattern.

(Operation) According to the first and second inventions, the feature extraction method is configured as described above, so that when a character pattern is input, the character pattern is moved vertically and/or horizontally. After scanning the direction and extracting an auxiliary sub-pattern representing a line element component in the scanning direction, the slope of the auxiliary sub-pattern is extracted.

In the first invention, the character pattern is scanned according to the extracted slope, and feature quantities such as the number of intersections between scanning lines and strokes are extracted.

Further, in the second invention, a sub-pattern is extracted according to the extracted slope, and then the sub-pattern is divided into a plurality of regions, and the characteristics of the sub-pattern are extracted for each of the divided regions.

This makes it possible to stabilize the extracted features even for character patterns with slants in handwritten characters or printed characters, and to solve the above-mentioned problem.

(Embodiment) FIG. 1 is a functional block diagram of a character recognition device using a feature extraction method showing an embodiment of the present invention.

This character recognition device has a photoelectric conversion unit 1 that converts an optical signal IN of a character image on a medium such as a form into a quantized electric signal (digital signal), and a line buffer 2 is provided on the output side of the photoelectric conversion unit 1. It is connected. Row buffer 2 is, for example, Fuku 20
It has a size of 48 bits x height of 128 bits, and is configured to store the digital signal of one line of character image, and the output side is connected to a button register 4 via a character cutting section 3.
is connected. The character cutting section 3 has a memory, and has a function of storing a digital signal for one character (hereinafter referred to as a "one character pattern") from the output of the line buffer 2 into the button register 4.

The button register 4 has a storage capacity of, for example, 64x64 bits, and a line width measuring section and a first sub-button extracting section 6 are connected to its output side. A character slope extraction section 7 is connected to the output side of the first sub-battle extraction section 6 . Further, to the output side of the button register 4, a dividing point determining section 9 is connected via a character frame detecting section 8, and a second sub-butter extracting section 10 is also connected. On the output side of the dividing point determining unit 9 and the second sub-battle extracting unit 10,
A feature matrix extraction unit 11 is connected, and on its output side,
An identification unit 12 is connected.

The line width measurement section 5 has a function of measuring the line width of the output of the button register 4 and providing the measurement results to the first and second sub-button extraction sections 6.10. The first sub-pattern extraction unit 6 scans the button register 4 in the vertical and horizontal directions to extract a first vertical sub-pattern (auxiliary vertical sub-pattern) and a first horizontal sub-pattern (auxiliary horizontal sub-pattern).
It is composed of a vertical sub-pattern extraction section 6a and a horizontal sub-pattern extraction section 6b. Each extraction part 6
+a and 6b each have a memory for storing a button.

The character slant extraction unit 7 extracts the slants of the first vertical sub-battens and first horizontal sub-battens extracted by the first sub-pattern extraction unit 6, and transmits the extraction results to the second sub-pattern extraction unit 10 and the feature matrix extraction. It has a function of outputting to the section 11.

The character frame detection unit 8 has a function of scanning the character pattern in the button register 4 to detect a circumscribed frame, that is, a character frame, and providing the detection result to the division point detection unit 9. The division point determination unit 9 has a function of determining division point coordinates for dividing the inside of the circumscribed frame into a plurality of partial regions, and outputs the determination result to the feature matrix extraction unit 11.

The second sub-pattern extraction unit 10 scans the button register 4 in a plurality of directions based on the slope extracted by the character slope extraction unit 7 to extract a second vertical sub-pattern and a second horizontal sub-pattern, and also extracts a second vertical sub-pattern and a second horizontal sub-pattern. Vertical sub-pattern extractor 10a extracts sub-patterns and left diagonal sub-patterns.
, a horizontal sub-pattern extraction section 10b, a right diagonal sub-pattern extraction section 10C5, and a left diagonal sub-pattern extraction section 10d. Each of the extraction units 10a to 10d has a memory for storing a button.

The feature matrix extraction unit 11 extracts features from each of the vertical, horizontal, right diagonal, and left diagonal sub-patterns output from the sub-pattern extraction unit 10 to create a feature matrix, and provides it to the identification unit 12. It has a function.

The identification unit 12 has a dictionary memory that stores a standard character feature matrix (standard character mask) G(k) and the character name of the standard character having the feature matrix G(k>). By comparing the feature matrix F (k) extracted by the feature matrix extraction unit 11 with the feature matrix G (k>) in the dictionary memory, the characters in the circumscribed frame area from which the feature matrix F (k> was obtained) are It has the function of recognizing figures and outputting character names OUT.

Next, processes (I) to (X) of each functional block will be described regarding a feature extraction method using the character recognition device configured as described above and a method of performing character recognition from the extraction results.

(I> Character stamp generation processing When the optical signal IN of the character image written on the root slip is input to the photoelectric conversion unit 1, the optical signal IN is converted into 2
The digital signal of the value, that is, the character line part is converted to "1°" (this is called "black bit"), and the background part is converted to "0" (this is called "white bit J)". Digital signals of character images for lines are stored in the line buffer 2.

The character cutting section 3 cuts out a digital signal for one character (character slam) from the digital signal of the character image stored in the line buffer 2, and stores it in the button register 4. In this embodiment, the form format is specified in advance, and the address indicating the character position within the line buffer 2 is stored in the memory of the character cutting section 3. Therefore, the character extraction operation is executed by reading the contents of the hf:line buffer 2 specified by the address.

(II) Line Width Measurement Process The line width measurement unit 5 inputs the digital signal from the button register 4 and calculates, for example, the number Q of states in which all points in a 2×2 window are black bits, and the number of all black bits. Count the number A,
The line width WL is calculated according to the conventionally known equation (1).

WL=A/(A-Q> ・・・・・・(1)(
I[I] First sub-battle extraction process The first sub-battle extraction unit 6 performs the following process.

The vertical sub-butter extractor 6a scans the entire surface of the button register 4 with the vertical direction as the main scanning direction, and detects continuous black bits (black runs) on the vertical scanning line. Then, from among the detected black runs, a black run with a length p that satisfies the following equation (2) is extracted.

1≧N, WL (2> However, 9; Length of black run in the main scanning direction N: An arbitrary constant for each sub-pattern (for example, 2) The vertical sub-pattern extraction unit 6a is calculated using the formula (2) A black run that satisfies the equation (2) is regarded as a black run constituting a sub-pattern and is stored in an internal vertical sub-pattern memory.A black run that does not satisfy formula (2) is regarded as a white bit.

In addition, the horizontal sub-battle extracting unit 6b scans the button register 4 with the horizontal direction as the main scanning direction by the same operation as the vertical sub-battle extracting unit 6a, and extracts the formula (2>) from among the black runs on the horizontal scanning line. A satisfying black run is extracted.The extracted black run is regarded as a black run constituting a sub-pattern and is stored in an internal horizontal sub-pattern memory.

(1v> Character slant extraction processing The character slant extraction section 7 extracts the first vertical sub-pattern (auxiliary vertical sub-pattern) obtained from the sub-pattern extraction section 6 and the first
For each of the horizontal sub-battens (auxiliary horizontal sub-battens), extract the character line element component (this is referred to as "stroke j") of the sub-pattern.Next, from the coordinate values of both ends of each extracted stroke, calculate the slope of each stroke. calculate,
By averaging them, the average inclination θV of the vertical stroke is extracted from the first vertical sub-button, and the average inclination θh of the horizontal stroke is extracted from the first horizontal sub-button.

Next, a specific method for extracting the average slopes θV and θh will be explained.

First, horizontal scanning is performed over the entire surface of the first vertical sub-pattern, and points of change from white bits to black bits and from black bits to white bits are detected.

Then, the coordinates of both ends of the stroke are extracted from the relationship between the scanning line one line before and the coordinates of the change point in the current scanning line.

The coordinates of both ends of the extracted stroke are (VXSi.

VYSi) and (VXE i , VYE i ), the average inclination of the vertical stroke θ■ is calculated using equation (3).
Calculate. However, i-1,..., Pv, Pv
is the number of strokes extracted from the first vertical sub-button, and VYSi<VYEi.

θ■= =1 ・・・・・・(3) Here, VLGi is obtained from the following equation (4).

VLGi=HAX I, j VXEi-VXSi 1.
,: VYEi-VYSi i) Number of strokes, also H
XSj<HXEj.

θh= Equation (4) is an approximate expression in which the distance between two points is the sum of 1/2 of the smaller of the horizontal and vertical coordinate differences between the two points and the other one.

Further, the average slope θh of the horizontal stroke is extracted from the first horizontal sub-button as follows.

Vertical scanning is performed for the first horizontal sub-button, and coordinates of both ends of the horizontal stroke are extracted. The coordinates of both ends are (H
XSj, HYSj ) and (HXEj.

HYEj), the average coordinate of the horizontal stroke θh
Calculate using formula (wo). However, j=1.・・・・・・、
Ph, Ph is the value extracted from the horizontal sub-button, where H
LGj is obtained from the following equation (6).

) ILGj= MAX I I HXEJ-HXSJHvEj-hv
Note that when the number of strokes is O, the slope is set to 0. That is, when Pv=O, θV=O1 When ph=O, θh
=○.

(V) Character frame detection processing The character frame detection unit 8 scans the pattern of the button register 4 and detects the left end coordinate χM, right end coordinate Xr, upper end coordinate Yt, and lower end coordinate Yb of the pattern. The circumscribing frame, that is, the character frame is (xfJ, yt>, (X, Q, Yb), (X
This is a rectangular frame connecting the four points r, Yt) and (Xr, Yb).

In addition, after character frame detection, in order to normalize the feature amount,
Calculate the size of the required character frame. In other words, let WPh be the size of the character frame in the parallel direction (horizontal direction) to the X axis of the button register 4, and let wph=Xr-XJI↑1 be the size of the character frame in the perpendicular direction (vertical direction). WPv=Yt-Yb+1 is calculated as WPv. Furthermore, the size of the character frame in the diagonal right and 45° left diagonal directions is set to WP.
WPh-i-WPv WPr=WPjl = is calculated as r and WPJI. These calculation results are given to the division point determining section 9.

(VI) Division Point Determination Process The division point determination unit 9 determines division point coordinates on the X-axis and Y-axis for dividing the inside of the circumscribed character frame into NX×NY partial regions for each circumscribed frame. However, NX is the number of divisions in the X-axis direction, and NY is the number of divisions in the X-axis direction.

The numbers of divisions NX and NY are preferably set to arbitrary suitable values depending on the complexity of the characters. For example, when recognizing characters with a small number of strokes such as kanji and katakana, (2X
2) Divide the area within the circumscribing frame into a small number of partial areas of about 3×3.

When kanji are to be recognized, (4X4) to (8X8
) The area within the circumscribing frame is often divided into partial areas of approximately However, in this embodiment, regardless of the complexity of the recognition target, the area within the circumscribed frame is divided into equal parts, for example, into (4x4) parts.

Division coordinate DX(n) on the X axis and division coordinate D on the Y axis
Y(n) is determined by the following equations (7) and (8).

DX (n> NX...(7) DY (m) However, n=1.2.=・=-, NX-lm=1.2.・
..., NY-1 In this embodiment, for example, NX=NY=4F) Sub-pattern extraction process FIGS. a> is an example of a character pattern, and the same figure (b
) is an example of a vertical sub-pattern. An arrow P in FIG. 4(a) indicates a scanning path when vertical sub-battles are extracted.

The second sub-pattern extraction section 10 extracts a vertical sub-pattern extraction section 10a, a horizontal sub-pattern extraction section 10b, a right diagonal sub-pattern extraction section 10c, and a left diagonal sub-pattern extraction section based on the average slopes θ■ and θh obtained by the character slope extraction section 7. Extraction part 10
d, a direction approximately perpendicular to the X-axis direction (vertical direction) and a direction approximately parallel to the X-axis direction (horizontal direction) set on the button register 4, and 45° counterclockwise from the axis (45° diagonally to the right)
45° clockwise direction (45° diagonal direction to the left) as the main scanning direction, and scans the button register 4,
A second vertical sub-pattern, a second horizontal sub-pattern, a right diagonal sub-pattern, and a left diagonal sub-pattern corresponding to each main scanning direction are extracted.

First, the operation of the vertical sub-button extraction section 10a will be explained.

The vertical sub-pattern extraction unit 10a scans the character pattern shown in FIG. 4(a) based on the average slope θV obtained by the character slope extraction unit 7, and detects continuous black bits (black runs) on the scanning line. Then, a black run with a length of 9 that satisfies the following equation (9) is extracted from the detected black runs.

p≧N·WL (9) However, g: length N of the black run in the main scanning direction; arbitrary constant for each sub-pattern (for example, 2) The scanning path P is as follows. Vertical scanning starts from the top edge. The coordinates (x-, y.) of the scanning path P from the scanning start address (XaYT>) can be expressed by the following equation (10).

xl=xa yl=YT x j=Xa-i-θv x (y iY T )'j
i='5'j 1;1 (10) However, θV is a real number, the result of θvx (y 1 - YT) is rounded down to the decimal point, and all coordinates are integers.

The vertical sub-battle extracting unit 10a stores the black runs that satisfy equation (9) in a vertical sub-battle memory (not shown), regarding them as black runs constituting the second vertical sub-battan. (9
) Black runs that do not satisfy the formula are considered white bits.

Further, the horizontal sub-pattern extracting section 10b extracts a second horizontal sub-pattern using the average slope θh obtained by the character slope extracting section 7 by the same operation as the vertical sub-pattern extracting section 10a. Similarly, the right diagonal and left diagonal sub-pattern extraction units 10c and 10d scan the original pattern with the right diagonal and left diagonal directions as the main scanning directions, and select (9) from among the black runs on the scanning lines in the respective main scanning directions. ), and store the extracted black runs in right diagonal and left diagonal sub-pattern memories (not shown) as black runs constituting a sub-pattern.

In this embodiment, when extracting right diagonal and left diagonal sub-patterns, the inclination of characters is not considered.

(■) Character frame extraction processing character frame detection unit 8 determines the coordinates
j, Extract features from each button and create a feature matrix.

That is, the feature matrix extraction unit 11 extracts one circumscribed frame area from the target dividing point coordinates and the coordinates XJ) and Xr.

The area is divided into NXXNY partial areas based on Yt and Yb, and feature amounts representing the character dose of the subpattern in each partial area are extracted. Then, a feature matrix consisting of NXxNYx4 feature quantities extracted from each sub-pattern within one circumscribed frame area is extracted as a feature quantity matrix for the circumscribed frame area.

First, feature extraction from the horizontal sub-panel (H3P>) will be explained.

The feature matrix extraction unit 11 extracts target division point coordinates and coordinates
, Q, Xr, Yt, Yb, the area within the circumscribing frame is divided into NXxNY partial areas.The coordinates of the target dividing point and the coordinates XI, Xr, Yt, Yb are the dividing point coordinates),
Horizontal sub-pattern within the partial area for each partial area) (number of black bits of SP BH(i.

Count j〉.

The division area is determined from the division point coordinates as follows. First, set the division point coordinates DX (n>) on the X-axis on a straight line parallel to the X-axis passing through the midpoint of the upper and lower edges of the circumscribed frame of the character pattern. From this coordinate as a starting point, use the following equation (11 )～(1
Let the left side of the coordinate series obtained in 3) be a division boundary S as shown in FIG. 4(b).

x o = Dχ(n) y () = (Y T completed Y B ) / 2...
...(11) From (x□, y□), the downward coordinate series is x, = D
X (n) TI NT (θ■be(y i-y○)=0.5)yi
=yi 171 (12) However, i=1.2, 3. ...< (YT-YB)/2-YT) From (x o , y o ), the upward coordinate series is
x−=DX (n> ”−INT(θv×(y Oy J ) 10 −5)y
j=yj 1-1 (13) However, j=-1,-2,-3,...(YB-(
YT-YB), /2). However, IN"IN + indicates that the operation in parentheses is performed using a real value, and the result is rounded down to an integer value.

Similarly, the horizontal division boundary S is set using DY (m) and θh.

For each divided area divided as described above, count the number of black bits BH (i, j>) of the horizontal sub-pattern H8P.
H(i,j) is the number of black bits in the partial area at the i-th row and j-th column regarding one circumscribed frame area. Next, according to equation (14), the feature amount FH(
Calculate i, j>.

BH(i, j) FH(i, j>= wt,, -wph......(
14) However, i=1.2. ......, NXj=1.2.・
..., NY WL; line width wph, character width (=Xr-XJI-i-1), and
In the same manner as in the case of H8P, the number of black bits BV(i, j) of VSP, R8P, and LSP in the partial area of the i-th row and j-th column,
BR (i, j) and BL (i, j) are counted and the following formula (15
) to (17) for the partial region in the i-th row and j-th column.

R3P, LSP feature quantity FV (i, j), FR<i, j
>, calculate FL(i,j).

FL (i, j) = WL - WPJ (17) However, WPv: character height (=Yb - Yt bottom 1) WPr = WP,
l! = (WPv+WPh＞/2 or more,
VSP, H3P, H3P for each partial area of the area within the circumscribed frame
, LSP features are extracted, and a feature matrix F (k>(k=1.2
．． ...NXXNYX4) is obtained. The feature matrix extraction unit 11 extracts the feature matrix F (k) for each circumscribed frame area, and sends the extraction result to the identification unit 12 .

(IX) The identification processing identification unit 12 uses the extracted feature matrix F(k),
Feature matrix G(k
>, the feature matrix F (
k) Recognize the characters and figures in the area within the circumscribed frame obtained. In this recognition, the feature matrix F (
Find the distance between k) and G (k), and output the character name of the standard character of the feature matrix G(k> with the minimum distance (for example, character code>OUT specified in the JIS standard) as a recognition result. do.

As described above, this embodiment has the following advantages.

In this embodiment, the character pattern in the button register 4 is scanned in both the vertical and horizontal directions by the first sub-button extractor 6, and the sequence of black bits on the scanning line and the line width of the character pattern are Based on the relationship, a sub-pattern representing the line element component in the scanning direction is extracted. moreover,
The character slope extraction unit 7 calculates the slopes θV and θh of the sub-pattern from the slopes of the line elements included in the sub-pattern.

Then, in the second sub-battle extraction unit 10, the slope θV,
The character pattern is scanned in each direction based on θh to extract sub-patterns, and then features are extracted. Therefore, the extracted features are stable even for character patterns having an inclination. Therefore, there is no need to prepare a dictionary corresponding to the deformation of the character slope in the identification unit 12, and by reducing the dictionary capacity, it is possible to shorten the collation time and thereby increase the processing speed, and the hardware size is small. Character recognition with high recognition accuracy becomes possible.

Note that the present invention is not limited to the above embodiments, and various modifications are possible. Examples of such modifications include the following.

(i) In this example, in both horizontal and vertical directions,
The case of extracting the slope of a stroke has been explained. However, since handwritten characters generally have many characters that slope upward to the right, it is sufficient to extract only the slope of horizontal strokes.

Moreover, when the target is an italic character of a printed character, it is sufficient to extract the inclination of a vertical stroke. In this way, the tilt extraction direction can be selected as appropriate depending on the object to be read, and the configuration can be simplified.

(ii) The feature extraction method of the above embodiment can also be applied to a method of extracting features by scanning a character pattern, such as that described in Document 2 above. For example, in the method of Document 2, the number of intersections between a scanning line and a stroke is used as a feature, so if the stroke has an inclination, the number of intersections extracted will vary greatly when the scanning direction is fixed. Therefore, by extracting the slope using the method of the above embodiment and performing feature extraction according to the slope, substantially the same effect as that of the above embodiment can be obtained.

(iii) The functional blocks shown in FIG. 1 may be configured to be executed by program control using a computer, instead of being configured by individual circuits.

(Effects of the Invention) As described in detail above, according to the first invention, a character pattern is scanned in both the vertical direction and the horizontal direction, or in either one or both of the vertical and horizontal directions, so that the black bits on the scanning line are continuous. Based on the relationship between and the line width of the character pattern, an auxiliary sub-pattern representing the line element component in the scanning direction is extracted. and,
The slope of the sub-pattern is determined from the slope of one line element included in the sub-pattern. Next, the character pattern is scanned in each direction according to the inclination, and features such as the number of intersections between a scanning line and a stroke are extracted from the character pattern. Therefore, it is possible to extract stable features even when dealing with characters such as handwritten characters in which the horizontal line elements are tilted upward to the right, or italic characters in which the vertical line elements are tilted to the right such as in printed characters. can. Therefore, there is no need to prepare a dictionary that accommodates the deformation of the character slope, and character recognition with high processing speed and small hardware scale is possible with high recognition accuracy.

In the second invention, sub-patterns are extracted according to the extracted slopes, and then the features of the sub-patterns are extracted, so that substantially the same effect as the first invention can be obtained.

[Brief explanation of drawings]

FIG. 1 is a functional block diagram of a character recognition device using a feature extraction method showing an embodiment of the present invention, FIG. 2 is a diagram showing an example of italic characters, and FIGS. FIGS. 4(a) and 4(b) are diagrams for explaining the feature extraction method. FIGS. 4(a) and 4(b) are diagrams for explaining the sub-pattern extraction method. 1...Photoelectric conversion unit, 2...Line buffer, 3...Character cutting unit, 4...Bun register, 5...Line Width measurement section, 6...1st
sub-battle extraction section, 7...character slope extraction section,
8...Character frame detection unit, 9...Division point determination unit, 10...Second sub-button extraction unit, 11
... Feature matrix extraction section, 12... Song identification section.

Claims (1)

[Claims] 1. A feature extraction method in which a character pattern obtained by photoelectrically converting characters on a medium is scanned in a plurality of directions to extract features from the character pattern, comprising: scanning in both or one of the horizontal directions, and extracting an auxiliary sub-pattern representing a line element component in the scanning direction based on the relationship between the number of consecutive black pixels in the scanning direction and the line width of the character pattern; After extracting the slope of the auxiliary sub-pattern from the slope of the line elements included in the auxiliary sub-pattern, the character pattern is scanned in a plurality of directions based on the slope of the auxiliary sub-pattern to extract features from the character pattern. A feature extraction method characterized by. 2. A character pattern obtained by photoelectrically converting characters on a medium is scanned in multiple directions, and a sub-pattern representing a line element component in the scanning direction is created based on the relationship between the number of consecutive black pixels in the scanning direction and the line width. In a feature extraction method that divides the sub-pattern into a plurality of regions and extracts a feature representing the amount of line elements of the sub-pattern for each divided region, the character pattern is divided into vertical and horizontal directions. scan in both or either direction,
Based on the relationship between the number of consecutive black pixels in the scanning direction and the line width of the character pattern, an auxiliary sub-pattern representing a line element component in the scanning direction is extracted, and from the slope of the line element included in the auxiliary sub-pattern, A feature extraction method, comprising: extracting the slope of the sub-pattern, and then extracting the sub-pattern based on the slope of the auxiliary sub-pattern.