JPH0358288A - Segment extracting device and graphic recognizing device - Google Patents

Abstract

PURPOSE:To improve the degree of extraction for the structure of characters and graphics by scanning a source pattern in a first scanning direction, extracting a sub pattern made into a code, executing scanning in the second scanning direction and detecting the head run, tail run and discontinuous run. CONSTITUTION:A threshold valuse set part 12, sub pattern extraction part 14 and segment extraction part 16 are provided. The source pattern is canned in the first scanning direction and a white code and a black code are attached. Then, the sub pattern made into the code is extracted. The sub pattern made into the code is canned in the second scanning direction and it is decided based on the coordinates of the black and white change points whether the second black run is linked or not. When the head run and tail run of the second black run to be linked are detected, the discontinuous run of the black pattern is detected based on the black code. Thus, the discontinuous run to change the geometrical feature of the black pattern can be detected and the segment expressing the geometrical feature can be extracted faithfully rather than conventional operation.

Description

[Detailed Description of the Invention] (Industrial Application Field) The first invention of this application is a segment extraction device for extracting segments of characters and graphics, for example, in a device or system for recognizing and searching characters and graphics. to darken.

The second invention of this application relates to a figure recognition device that recognizes character figures using segments extracted from character figures. (Prior Art) Segments extracted from text and graphics have conventionally been used to recognize and search text and graphics. As a conventional technique for extracting this segment, for example, there is a technique disclosed in the literature engineer, Proceedings of the Spring National Conference of the Institute of Electronics, Information and Communication Engineers, 1988, D-399J.Hereinafter, the conventional technique of this literature engineer will be explained. In this conventional technique, first, the original black and white binary pattern is scanned, the number of times A that all pixels in a (2×2) window become black bits, and the number of previous black bits Q of the original pattern are counted, and the number is calculated according to the following formula. Obtain the average line width of the original pattern.

W=Q/ (Q- A) Then, all the line segments on the original pattern are moved in the vertical direction (V), the Kihira direction (opening), the left diagonal direction (L), and the right diagonal direction (R).
IIll@1 to classify and extract line segments in four directions.
Set th.

Next, the four directions of the eye, V, L, and sun are each set as the first scanning direction, and the original pattern is scanned in each of the first scanning directions to extract a black run whose length in the scanning direction is 12th or more. , obtain line segment subpatterns for each first scanning direction.

Next, the subpattern is scanned in a second scanning direction that is orthogonal to the first scanning direction when extracting the subpattern, and a cluster of connected black runs is detected in the first scanning direction, and the first of the cluster of connected black runs is detected. Extract the coordinates of the midpoint of the black run that appeared in and the coordinates of the midpoint of the last black run that appeared as segments of the character shape. (Problems to be Solved by the Invention) However, in the above-mentioned conventional technology, there are cases in which the extracted segments cannot sufficiently represent the structure of the input figure. This point will be explained with reference to FIGS. 10 and 11.

FIGS. 10 and 11 are diagrams showing examples of original patterns,
The original pattern ○P shown in FIG. 10 (A). is the original pattern of a curved figure with small curvature, the original pattern shown in Figure (B) ○PI
l is the original pattern of a figure that connects two straight lines with a direction close to the scanning direction during subpattern extraction, and Figure (C) is the original pattern of a figure that connects two straight lines in the same direction via a small figure. Moreover, FIG. 11 is a diagram showing horizontal sub-patterns and horizontal segments extracted from the original pattern, and the diagram (A) in FIG.
Kihira subpattern HSP. and horizontal segment day SG.
are the subpatterns and segments extracted from the original pattern ○P1, the horizontal subpatterns SPゎ and horizontal segments HSGI in figure (E3), are the subpatterns and segments extracted from the original pattern OPb, and the horizontal subpatterns HS in figure (C)
P. and horizontal segment SG. are subpatterns and segments extracted from the original pattern ○Pe.

As can be understood from FIGS. 10 and 11 (A), in the prior art, only one segment representing a straight line figure could be extracted from a curved figure with a small curvature.

In addition, as can be understood from FIGS. 10 and 11 (B), in the conventional technology, one line figure representing one straight line figure is extracted from a figure in which two straight lines having a direction close to the scanning direction of sub-pattern extraction are connected. Sometimes only segments could be extracted.

Furthermore, as can be understood from (C.) in FIGS. 10 and 11, in the prior art, two straight lines in the same direction are connected to each other via a small figure to form one straight line figure. Sometimes only segments could be extracted.

As mentioned above, in the conventional technology, the structure or geometrical characteristics of the input figure may be affected depending on the mutual positional relationships and connection angles of the line segments that make up the input figure. In some cases, it was not possible to extract a segment that faithfully represented 5I. If such segments are used to recognize and search for characters and figures, the features extracted from the segments will become unstable, making it difficult to identify the characters and figures, making it impossible to accurately identify them.

Furthermore, since accurate identification is performed using unstable features, the processing method for identification becomes complicated, resulting in a problem in that the processing speed of identification decreases.

Furthermore, in the conventional character/figure identification processing using segments, unclassified segments are sequentially searched to detect segments constituting the character/figure, which poses a problem in that the identification process takes a long time.

A first object of the present invention is to provide a segment extraction device that can extract segments that represent the structure of character figures more faithfully than before, in order to solve the above-mentioned conventional problems. A second object of the present invention is to provide a figure recognition device that identifies characters and figures by searching for segments classified by item, in order to solve the above-mentioned conventional problems.

(Means for Solving the Problems) In order to achieve this object, the segment extraction device of the first invention provides a black-and-white segment extraction device for extracting a plurality of sub-patterns according to a first scanning direction from a black-and-white binary original pattern. a threshold value setting unit that sets a threshold value and a continuous length threshold value for classifying a first black run in a first scanning direction of a sub-pattern according to a continuous length of the first black run; The first black run having a continuous length equal to or greater than the black-and-white threshold is extracted as the black hit part of the sub-pattern, and a white code is attached to the white bit part of the sub-pattern. For the black bit part, connect the black code classified according to the length of the continuous length}! A sub-pattern extractor extracts a sub-pattern based on the comparison result of the length threshold and the continuous length of the first black run, and scans each sub-pattern in a second scanning direction forming a predetermined angle with the first scanning direction, and extracts white on the scanning line. Determine whether or not the second black runs in the second scanning direction are connected based on the coordinates of the black change point when changing from a bit to a black bit and the coordinates of the white change point when changing from a black bit to a white bit. Then, the first run and the last run of the second black run to be connected are detected, and the discontinuous runs of the second black run to be connected are detected based on the black code, and the first run of the first black run to be connected is detected. and a segment extraction unit that extracts segments based on the position of the last run and the position of the discontinuous run.

Further, the figure recognition device of the second invention includes a black-and-white threshold for extracting a plurality of sub-patterns for each first scanning direction from a black-and-white binary original pattern, and a first black run in the first scanning direction of the sub-patterns. a threshold setting unit that sets a continuous length threshold for classification according to the continuous length of the first black run; and a continuous length that is equal to or greater than the black and white threshold when the original pattern is scanned in the first scanning direction. The first black run is extracted as the black bit part of the sub-pattern, and the mortar bit part of the sub-pattern is assigned a white code, and the black bit part of the sub-pattern is classified according to the continuous length. a sub-pattern extraction unit that attaches a code based on a comparison result of a continuous length threshold value and a continuous length of the first black run; and a second scanning direction that scans each sub-pattern at a predetermined angle with respect to the first scanning direction; Whether the second black run in the second scanning direction is connected based on the coordinates of the black change point when a white bit changes to a black bit on the scanning line and the white change point when a black bit changes to a white bit on the scanning line. The first run and the last run of the second black run to be connected are detected, and the discontinuous runs of the second black run to be connected are detected based on the black code, and the second black run to be connected is detected. a segment extraction unit that extracts segments based on the positions of the first and last runs of a run and the positions of discontinuous runs; The present invention is characterized by comprising a segment classification unit that performs classification based on the definition of a character figure, and a recognition unit that detects segments that constitute a character figure from among segments that are classified according to the definition of a character figure. (Operation) According to the segment extraction device of the first invention having such a configuration, the original pattern is scanned in the first scanning direction to extract the sub-pattern coded by adding a white code and a black code. Then, the coded sub-pattern is scanned in the second scanning direction, and it is determined whether or not the second black runs are connected based on the coordinates of the black and white change points, and the second black runs (hereinafter referred to as This connected mass of second black runs is called a black pattern)
First run and last run IF! ．． At the same time, a discontinuous run of the black pattern is detected based on the black code. The black pattern is a pattern representing a certain specific character figure.

By scanning the coded sub-pattern and examining changes in the black code of the black pattern, it is possible to detect discontinuous runs in which the glyph-like features of the black pattern change. In addition to the positions of the first and last runs of the black pattern, the positions of these discontinuous runs! By extracting black pattern segments using %, it is possible to extract segments that represent geometric features more faithfully than before. Further, according to the figure recognition device of the second invention, the segment is
Classification is performed according to the first scanning direction and also based on segment type information obtained from the black code. Then, the segments that make up the character figure are detected from among the segments that are classified according to the definition of the character figure. By detecting segments that make up a character figure from among the classified segments, the speed of detecting segments that make up a character figure can be improved.

(Examples) Examples of the first and second inventions will be described below with reference to the drawings. Note that the drawings are only shown schematically to the extent that these inventions can be understood, and therefore, the connection levers, input/output signals, dimensions, shapes, and arrangement positions of each component are limited to the illustrated examples. isn't it.

Figure 1 is a functional block diagram for explaining the structure of an embodiment of the first invention.

As shown in FIG. 1, the segment extraction device 10 of this embodiment includes a black-and-white threshold detector for extracting a plurality of sub-patterns for each first scanning direction from a black-and-white 2ia original pattern, and a first sub-pattern. @price fixing unit 12 that sets a continuous length threshold for classifying first black runs in the scanning direction according to the continuous length of the first black runs; Then, the first black run having a continuous length equal to or greater than the black and white threshold is extracted as the black bit part of the sub-pattern, and a white code is attached to the white bit part of the sub-pattern, and a white code is attached to the black bit part of the sub-pattern. A sub-pattern extraction unit 14 attaches black codes classified according to the continuous length based on the continuous length threshold and the comparison result of the continuous length of the first black run; Scanning is performed in a second scanning direction that forms an angle, and the second scan is performed based on the coordinates of the black transition point when a white bit changes to a black bit on the scanning line, and the white transition point when a black bit changes to a white bit on the scanning line. It is determined whether or not the second black runs in the scanning direction are connected, and the first and last runs of the connected second black runs (hereinafter referred to as black patterns) are detected, and the black code {based on this A segment extraction unit 16 detects discontinuous runs of the black pattern and extracts segments based on the positions of the first and last runs of the black pattern and the positions of the discontinuous runs. Roughly speaking, the segment extraction device M10 of this embodiment includes a pattern memory 18 that stores a black and white binary original pattern, a line width calculation unit 20 that calculates the line width regarding the original pattern, and stores sample patterns for each first scanning direction. Sub pattern memory 22
and a segment memory 24 for storing segments extracted from subpatterns in each first scanning direction. In addition, 26 in the figure indicates a recognition unit, and the recognition unit 26 detects segments constituting a predetermined character figure from among the segments stored in the segment memory 24, and identifies characters and figures existing in the original pattern. do. The segment extraction device 10 of this embodiment will be explained in more detail below.
I will explain about it.

(Pattern Memory) In this embodiment, for example, a black and white binary electrical signal from a photoelectric conversion unit is used as an original pattern, and this original pattern is stored in a pattern memory 1.
Store in 8. The photoelectric conversion section optically scans a recording medium such as a logic circuit diagram in which line figures are written, and outputs an original pattern obtained by photoelectrically converting optical information from this medium.

An X-Y coordinate system is virtually set on the pattern memory 18 so that the original pattern of pixel positions expressed by this coordinate system can be read out from the pattern memory 18.

(Line calculation unit) The line calculation unit 1 20 has a shift register configuration similar to a well-known filter circuit, and for example, all points in a (2×2) window are black bits with respect to the original pattern input in pixel units. The number of points Q and the total number of black bits A in the original pattern are counted, and the average line width W is calculated based on the following well-known approximation formula (1).

W=Q/ (Q-A) (1) In this embodiment, the line width for the original pattern is the average line width W calculated by the line calculation unit 20, or the line width is Calculation part 2
It is also possible to omit 0 and use a predetermined value predicted in an information medium such as a logic circuit diagram as the line width for the original pattern. In addition, instead of calculating the line width by scanning the original pattern memory, the original pattern from the photoelectric conversion unit is input pixel by pixel to the line width calculation unit 20, and the line width is calculated based on the original pattern from the photoelectric conversion unit. You can also do the calculations. (Threshold value setting unit) In this embodiment, the first scanning direction is vertical (Y-axis direction), horizontal (X-axis direction), right diagonal direction (direction intersecting the X-axis at an angle of 45 degrees to the right), and left Vertical, horizontal, right diagonal, and left diagonal subpatterns are extracted for each of the first scanning directions in four diagonal 45° directions (directions intersecting the X axis at 45° diagonal to the left). For this reason, 115 @ setting unit 12 is a black and white IS for vertical, horizontal, right diagonal and left diagonal sub-pattern extraction!
l value L VWz L h*z L rl# and . Set W. Each black and white threshold is set according to the following equation (2). Lvw=Nv' W L,l-=Nh ・W shirw"Nr ゜W shi.w=N.-w ...... (2) However, NV, Nh, N, and N1 can be arbitrarily set. The constant to be set, W, is the average line width.As is well known in the art (for example, "Line Segment Extraction of Logic Circuit Diagrams, 1985, Shinjijo Zendai D-3994"), the constants Nv, Nfi, N, and N1 By setting arbitrarily and suitably, all patterns in the original pattern can be classified and extracted as any sub-pattern.

For example, Nv, Nh, N-, and N1 may all have the same value, or may each have different values, or when N,=Nh=N+ and in the diagonal right direction and diagonal left direction, N,= It is also possible to set Nl = N2. The threshold setting unit 12 sets a continuous length threshold based on an input signal from an input unit (not shown), such as a keyboard. The continuous length threshold is also set for each subpattern in the same way as the black and white threshold. Continuous length threshold L vs. for vertical, horizontal, right diagonal, and left diagonal sub-patterns, respectively. , L rs and L +s! Set. ? Duration threshold @Lv. , Lhs, Lr
, l and shi. S can be set arbitrarily and suitably based on the geometric characteristics of the character figure obtained according to the definition of the character figure to be identified, such as the thickness of the line segments and the relative positional relationship of the line segments that constitute the character figure. These consecutive length lII values may be different values or may be the same value. Continuous length threshold Lv. , Lh-
, L and Lis, it is possible to detect a second black run, that is, a discontinuous run, in a portion where the geometrical features of the character figure change, for each sub-battle.

(Sub-pattern extraction unit) In this embodiment, the first scanning direction is four directions: vertical, horizontal, diagonal right, and diagonal left, so the sub-pattern extraction unit 14
It is composed of a vertical sub-pattern extraction section 14a, a horizontal sub-pattern extraction section 14b, a right diagonal sub-pattern extraction section 14c, and a left diagonal sub-pattern extraction section +4d. Each extraction part 1 4a
, I 4bs I 4c and +4d scan the original pattern in vertical, horizontal, right diagonal and left diagonal directions to extract coded vertical, horizontal, right diagonal and left diagonal sub-patterns.

The vertical sub-pattern extraction unit 14a scans the original pattern with the main scanning direction as the vertical direction, and calculates the number of black bits (continuous length) that are continuously detected on the scanning line in the first scanning direction.
1, and extract a group of consecutively detected black bits (first black run) of the number g that satisfies the following equation (2). β≧L vw (3) The vertical sub-pattern extraction unit 14a regards the first black run of continuous length l that satisfies equation (3) as the black bit part of the vertical sub-pattern, and extracts the first black run as the black bit part of the vertical sub-pattern. The pattern is stored in the pattern memory 22a. The white bits of the original pattern and the black bits of continuous length l that do not satisfy equation (3) are regarded as the white bit portion of the vertical sub-pattern and are stored as white bits in the vertical sub-pattern memory 22a. Then, the vertical sub-pattern extraction unit 14a generates a white code, for example, 00 (00 is a binary number) for the white bit portion of the vertical sub-pattern, and a continuous length l for the black bit portion (
A black code is attached according to the classification. The continuous length 2 and the continuous length threshold LV are compared, and for the first black run of continuous length β that satisfies the following equation (4a), the black code 01 is set, for example, (01
is a binary number), and a black code, for example, 10 (10 is a binary number) is assigned to the second black run of continuous length 2 that satisfies the following equation (4b).

L vw≦l≦Lvs --------(4a)L
vs≦β ...... (4b) In the following explanation, numbers shown with the word "1 code" such as black code, operation code, final operation code, etc. represent binary numbers, and are referred to as "゜゜code". Numbers without the word are assumed to represent decimal numbers. Similar to the vertical sub-pattern extraction section 14a,
Horizontal, right diagonal and left diagonal sub-pattern extraction parts 14b, 14
c and +4d also scan the original pattern with the vertical, left diagonal, and right diagonal directions as the main scanning directions, and l≧Lhw, A
The first black run of continuous length l that satisfies ≧ and g≧Llw is regarded as the black bit part of the horizontal, right diagonal, and left diagonal subpatterns, respectively, and is used as the black bit for the Kihira, right diagonal, and left diagonal subpatterns. The data is stored in memories 22b, 22c and 22d. The white bits of the original pattern and the black bits with a continuous length of 1 that do not satisfy equation (3) are regarded as the white bit portion of the vertical sub-pattern, and are stored as white bits in the horizontal, right diagonal, and left diagonal subpattern memories 22b, 22c, and 22d. .

And horizontal, right diagonal, and left diagonal sub-pattern extraction section 14b
, 14c and +4d are assigned a white code 00 for the white bit part of the horizontal, right diagonal and left diagonal sub patterns, and a black code classified by the length 1 of the continuous length l for the black bit part. . Therefore, each extraction section 14b, 14c and +
4d represents the continuous length l as the continuous length threshold L h** L
r S and , , for the first black run of continuous length l that satisfies the following formula (5a), the black code is 01, and for the first black run of continuous length l that satisfies the following formula (5b) A black code of 10 (10 is a binary number) is assigned to the black run. As a result,
Each encoded subpattern is obtained. Lhw≦℃≦Lhs L rw≦l≦L4 L sw≦1≦, S... (5a) hs≦l L4≦l L. ≦l (5b) FIG. 2 shows an example of a coded vertical sub-pattern.

Figure 2 shows an example of vertical subpatterns obtained with L, W = 3 and L, , = 6. In the figure, one mesh corresponds to a subpattern for one pixel, and a white mesh corresponds to the original pattern and subpattern. Pixels represented by white bits in the odor, meshes with the number O are pixels represented by black bits in the original pattern and white bits in the subpattern, and meshes with numbers 1 and 2 are represented by the original pattern and subpattern. It is a pixel that cannot be expressed by a black bit in the odor, and the X-axis and Y-axis are the X-axis and Y-axis set virtually on the sub-pattern memory. The pixels of the white mesh and the mesh with the number O are given the white code ○O, the pixels of the mesh with the number 1 are given the black code 01, and the pixels of the mesh with the number 2 are given the black code 10. ing.

(Sub-pattern storage unit) The sub-pattern memory 22 of this embodiment includes vertical, horizontal, right-diagonal, and left-diagonal sub-pattern memories 22a, 22b, and 22c that store vertical, horizontal, right-diagonal, and left-diagonal sub-patterns.
and 22d, and store sub-patterns coded for each first scanning direction. Each sub pattern memory 22a
An X-Y coordinate system similar to that of the pattern memory 18 is set on ~22d, and black runs constituting each sub-pattern are stored at pixel positions corresponding to pixel positions on the pattern memory 18. (Segment Extracting Unit) In this embodiment, a sub-pattern is scanned with the second scanning direction being, for example, a direction perpendicular to the first scanning direction, and segments corresponding to the sub-pattern are extracted. For this purpose, the segment extraction unit 16 includes a vertical segment extraction unit 16a that scans vertical sub-patterns with the horizontal direction as the main scanning direction and extracts vertical segments, and a vertical segment extraction unit 16a that scans the horizontal sub-patterns with the vertical direction as the main scanning direction and extracts horizontal segments. a segment extraction unit +6b; a right diagonal segment extraction unit 16c that scans a right diagonal sub-pattern with the left diagonal direction as the main scanning direction and extracts a right diagonal segment; and a left diagonal segment extractor +6d for extracting segments.

Hereinafter, segment extraction by the vertical segment extraction section 16a will be explained, but other segment extraction sections 16b and 16c
And +6d segment extraction is also performed by the vertical segment extraction unit 1.
Since the segment extraction is performed in the same manner as 6a, a description of the segment extraction by the other segment extraction units 16b, 16c and +6d will be omitted.

The vertical segment extraction unit 16a scans a vertical sub-pattern with the main scanning direction as the horizontal direction, and extracts the coordinates of a black bit when a white bit changes to a black bit on a scanning line in the main scanning direction and the coordinates of a black change point on the scanning line. The coordinates of the black bit when the black bit changes to the white bit are detected as the coordinates of the white change point. Then, a block of black bits (second black run) detected continuously on the scanning line is detected, and the second black run detected in the current scanning line (current) and the scanning line one line before the current one (previous The connection relationship with the black run detected in the row) is determined based on the coordinates of the black and white changing points in the previous row and the coordinates of the current black and white changing points. Based on this determination result, the second black run connected between the scanning lines is detected, and the first detected second black run (leading run) and the last detected second black run (first run) among the connected second black runs (black patterns) are detected. The position at of the detected second black run (last run) is detected. As the positions of the first run and the last run of these black patterns, for example, the coordinates of the midpoints of these black runs are used.

The connection relationship of the second black run is determined using the following relational expression (6a)
, (6b).

X8Vim-11≦XWVm...(6a)xwv
+m−u≧x8v, ・ (6b) However, XeV+m
-11 and XWVfm-11 are the X coordinates of the black change point and white change point detected in the previous row Y0-1, X8Y, and
WVM is the current Y. ．． Black detected in l? These are the coordinates of the color change point and the white color change point.

Relational expressions (6a) and (6b)! The change point of the previous line that satisfies (
X[lV(II1-11*Ym-+), (X
*V(M-., Y,-,) and the current change point (Xevm
, Y, '), (Xwv-, Ya) are the change points detected for the same black pattern in the vertical sub-pattern, and the change points (X[IV (m-11, Y II1
-1) and (X mV (m-11, Ym-1
) and the change point (X8v,...) in the previous row.
, Y. , , ) and (Xwv-, Y-) indicate that they are connected.

On the other hand, the cutting change point (
Xa■, Y1) and (×■,, Y,) are the change points detected for the black patterns that are not connected to each other in the vertical sub-pattern, and therefore (Xo1-+>, Y m-1
) and (
Xwv-. Y. ．． This indicates that it is not connected to the current second black run between 1) and 1).

Also, if the scanning line crosses one black pattern, considering the stone,
On the scanning line, a white bit, a black run, and a white bit are sequentially detected. Therefore, the black transition point and the white transition point l8. The detection can be regarded as the detection of a black pattern. Furthermore, when a white hit, a black run, and a white bit are sequentially detected on the scanning line, it is found that the black and white transition points included in the black run are the black and white transition points of a certain black pattern, and therefore the scanning Even when a line crosses multiple black patterns, the black and white transition points of each black pattern can be detected for each black pattern. Also, considering the case where one black pattern is scanned line-sequentially, a black pattern for which no black or white transition point was detected in the previous line, but for which a black or white transition point was detected in the current line, is scanned line-sequentially. This is a black pattern that newly appeared in , and the black run between the current black and white change points of this new black pattern becomes the first run of the new black pattern. A black pattern whose black and white changing points were detected in the previous line, but for which no black and white changing points were detected in the current line, is a black pattern that has disappeared in the current line, and the black pattern before this disappeared black pattern The black run of the black and white changing point group in the row becomes the last run of the disappeared black pattern.

Roughly speaking, when the vertical segment extraction unit 16a detects a black change point on a horizontal scanning line, it performs a logical sum operation on the black code of black bits from the black change point to the next detected white change point. This OR operation is performed for each black pattern.

Assuming that a white bit is detected after M black bits are detected consecutively at the black transition point, in this logical OR operation, first the black code of the black transition point and the first black bit connected to the black transition point are detected. The first OR operation is performed for each bit (each digit) of the black code of the black bit. The logical sum operation result is also expressed in binary numbers, and the logical sum operation result expressed in binary numbers is hereinafter referred to as an operation code. Next, an OR operation is performed for each digit of the operation code of the first OR operation and the black code of the second black bit connected to the black change point. For example, if the black code of the black transition point, the first black bit, and the second black bit connected to the black transition point are code 01, code 01, and code 10, the first operation code is code 01. , the second operation code is code 11. In this way, a logical OR operation is performed for each digit of the (m-1)th (2≦m≦M) operation code and the mth black bit that is connected to the black change point. When a white bit is detected after the M-th black bit, the M-th black bit is detected as a white change point, and the M-1th operation code is converted to the second bit containing the M-th black bit. It is saved as the final OR operation result of the black run, that is, the final OR operation result of a certain black pattern in the scanning line. The final OR operation result expressed in binary numbers is called the final operation code.

In this way, when a black transition point is detected on a scanning line, by repeating the logical sum operation until a white transition point is detected, the final calculation for each second black run detected on the scanning line (for each black pattern) is performed. Get the code. If the i-th bit (first digit) of the final operation code is 1, the second black run includes the black bit of the first black run having the i-th continuous length, and the i-th bit If is O, the second black run does not include the black bits of the first black run having the i-th continuous length.

Then, the vertical segment extraction unit 16a extracts equations (6a) and (6b) from the branched previous line and the current second black run.
), a discontinuous run detection process is performed when a second black run that satisfies is detected. Discontinuous run detection processing is not performed for the previous row and the current second black run that do not satisfy equations (6a) and (6b) t.

In the discontinuous run detection process, it is determined whether or not a discontinuous run has been detected based on the final operation code for each second black run detected on the previous and current scanning lines. The detection of this discontinuous run will be explained with reference to Table 1vr below. In Table 1, the final operation codes of the second black runs detected in the previous row and the current row are shown in binary numbers in the previous row column and the current column, respectively.

The previous row and the current second black run are expressed by equations (6a) and (6b)i
If they are satisfactorily connected to each other, the final operation codes of these previous rows and the current second black run are codes 01 and 10,
Alternatively, when codes 01 and 11, codes 10 and 01, or codes 11 and 01 are obtained, it is determined that these previous rows and the current second black run are connected but discontinuous.

These previous and current discontinuous second black runs are the previous and current discontinuous runs.

For example, the midpoint position of the black and white transition points is used as the position of the discontinuous run. To extract segments, use the positions of the previous line and/or the current discontinuous run. When a discontinuous run is detected, it is assumed that a change has occurred in the geometrical characteristics of the black pattern including these discontinuous runs, such as the black pattern being bent.

Also, when the previous row and the current second black run are connected, the final operation codes of the previous row and the current second black run are codes 01 and 01, or codes 10 and 10, or codes 10 and 10. 11, or codes 11 and 10, or codes 11 and 11, these second black runs are not discontinuous, so the change in the geometric characteristics of the black pattern including these second black runs is Assume that it never happened. As mentioned above, depending on whether the i-th digit of the final operation code is 1 or O, it is determined whether the second black run includes the black bit of the first black run having the i-th continuous length. You can know what.

Therefore, we focused on the black bit of the first black run (hereinafter referred to as the longest bit) that has the longest continuous length among the black bits included in the detected second black run. The longest bit that was not present in the previous row and the longest bit that was present in the previous row and is now gone can be found based on the final opcode. Table 1 is when you find the longest bit that appears in the current line but did not exist in the previous line, or the longest bit that existed in the previous line but has disappeared in the current line. This is when the previous line and the current final operation code are discontinuous combinations, and at this time, a change in geometrical characteristics is often observed. Then, the vertical segment extraction unit 16a extracts segments based on the positions of the detected first run and last run of the black pattern and the positions of discontinuous runs.

In this embodiment, when detecting the first run of a black pattern and then detecting the last run of the black pattern without detecting a discontinuous run, the positions at of the first run and the last run of the black pattern are respectively divided into segments. Detected as the start and end points.

In addition, when detecting the discontinuous run and the last run of the black pattern after detecting the first run of the black pattern, the black pattern is divided based on the position of the discontinuous run (hereinafter, the divided black pattern is referred to as divided pattern). ) are detected as the start and end points of the segment, respectively. Similar to the vertical segment extraction section 16a, horizontal, right diagonal, and left diagonal segment extraction sections 16b, 16c,
Similarly to the vertical segment extraction section 16a, +6d is also
The horizontal, right diagonal, and left diagonal subpatterns are scanned in the second scanning direction to detect horizontal, right diagonal, and left diagonal segments. Next, with reference to FIG. 2, vertical segment extraction V--As an example, segment extraction processing will be explained using one specific example. In FIG. 2, black pattern 1 is a pattern consisting of one point, two points, and division patterns ]a, 1b, and ]C surrounded by three-dot chain lines, and represents a line figure having a small curvature. Further, a black pattern 2 surrounded by a dotted line is a pattern that is not connected to the black pattern 1, and represents a line figure having a linear shape.

When the vertical segment extracting unit 16a finishes extracting the coded vertical sub-pattern, it starts processing for extracting vertical segments of the vertical sub-pattern.

The vertical segment extraction unit 16a scans the vertical sub-patterns line-by-line from the scanning start line. In this scanning, when black and white changing points are sequentially detected on the scanning line, these black and white changing points are set as a set of black and white changing points that form a certain black pattern, and the previous line Then, it is determined for each set whether the current black and white change points satisfy relational expressions (6a) and (6b), and a connected second black run is detected. For each second black run that is roughly connected, the previous line and the current final operation code are checked to detect discontinuous runs in the black pattern. In the case of a black pattern for which no discontinuous runs have been detected, the positions of the first run and the last run of the black pattern are detected as the start and end points of the segment, respectively, and the geometric structure of the black pattern is divided into one line. Represented by segments. In addition, in the case of a black pattern in which a discontinuous run is detected, the position of the first run and the last run of the divided pattern obtained by dividing the black pattern based on the position of the discontinuous run @ the position of the start point and end point of the segment, respectively. The geometric structure of the black pattern is expressed as multiple segments. In the vertical segment extraction from the sub-pattern shown in FIG.
However, one set of black and white changing points is detected in scanning line Y1. The black codes between the black and white change points in scanning line Y1 are all code 01, and therefore the final operation code of the black run between these black and white change points is code 01. No change point was detected up to the line one line before scanning line Y, and no change point was detected, so scanning line Y,
Since we detected the point where one phase changes between black and white for the first time, we can see that a black pattern has appeared. one set of black in scanning row Y1;
The white change point is saved as the change point of black pattern 1. In addition, the position of the second black run (corresponding to the first run of the divided pattern 1a) between the pair of black and white change points in the scanning line Y1 is stored in the segment memory 24 as the starting point coordinates of the vertical segment 1 of the black pattern 1. Store. The vertical segment extraction unit 16a further scans the scanning lines below scanning line Y.

The vertical segment extraction unit 16a detects two sets of black and white change points by scanning the scanning line Y2. In scanning line Y2, the closing equations (6a) and (6t)) t are used for the black and white changing points detected in the previous line (scanning line Y,) and the black and white changing points detected in the current line (scanning line Y2). A set of satisfying change points and a set of unsatisfied change points are detected.

The silk of the change point that was not detected in the previous row but was detected in the current row is saved as a set of change points of the newly appearing black pattern 2. Roughly the second black run (corresponding to the first run of black pattern 2) between the pair of black and white change points that newly appeared in this current situation! 8. Store in the segment memory 24 as the starting point coordinates of vertical segment 2 of black pattern 2. Hereinafter, for example, the midpoint coordinates of the black and white transition points will be used as the position of the second black run.

As the scan proceeds roughly, the vertical segment extraction unit 16a extracts the black of the second black run constituting the black pattern 1 in the scanning line Y3.
A set of white change points is detected, but the final operation code of the second black run of black pattern 1 of the current (scanning line Y3) becomes code 11, and the black of the previous line (scanning line one line before scanning line Y3) becomes code 11. Since the final operation code of the second black run of pattern 1 is code 01, it can be seen that the second black runs of black pattern 1 in the current and previous rows are connected but discontinuous. The second black run in the previous row and the current black pattern 1 is detected as a discontinuous run. Based on these discontinuous runs, the vertical segment extraction unit 16a sets the last run and the first run of the preceding and subsequent division patterns (corresponding to division patterns 1a and 1b) that are adjacent to the corners of the black pattern 1, and The position N of the run and the first run is stored in the segment memory 24 as the positions of the end and start points of the segments (corresponding to segments 1 and 3) of the previous and subsequent division patterns. For example, the discontinuous run in the previous row is set as the R trailing run of the previous division pattern, and the current discontinuous run is set as the first run in the division pattern in the next stage, and the positions of these last runs and first runs are set between the black and white change points. Use point coordinates.

When the vertical sub-pattern is roughly scanned, discontinuous runs are not detected in the scan from scan line Y3 to one line before scan line Y4, but in the scan of scan lines Y and l, the current (Kurobatano 1) is detected.
Final operation code or code 0 for black pattern 1 in scanning line Y4)
1, and the final operation code of black pattern 1 in the previous row (scanning row one line before scanning row Y4) becomes code 11, so that a discontinuous run of black pattern 1 is detected.

In the same manner as described above, the last run and first run of the previous and subsequent division patterns (corresponding to division patterns 1b and 1c) to be FaWied for black pattern 1 are set based on the detected discontinuous runs, and these last runs are set. and position IF of the first run
r, are stored in the segment memory 24 as the positions of the end and start points of the segments of the previous and subsequent division patterns (corresponding to segments 3 and 4).

When scanning the vertical sub-pattern roughly, a black and white change point was detected in the previous line (the line before scanning line Y), but it was detected in the current line (scanning line Ya). Black pattern in which black and white transition points are no longer detected, that is, black pattern 21
8. To detect. Which black and white changing points of the black pattern have disappeared or newly appeared can be determined by, for example, the black and white change points of each scanning line.
This can be determined by checking the change in the number of sets of white changing points, or by comparing the black pattern in which black and white changing points were detected in the current process with the black pattern in which black and white changing points were detected in the previous line. be able to. The vertical segment extracting unit 16a uses the midpoint coordinates of the black and white transition points of the previous black pattern 2 as the end point coordinates of segment 2 of the black pattern 2, that is, the black pattern 2 for which no black and white transition points are currently detected. Store in memory 24. Further, when the scan line Y8 is scanned forward, no phase change point between black and white is detected. Therefore, the black pattern (
In other words, it can be seen that the end point of black pattern 1 has been detected.

The vertical segment extraction part 16a is the black of the previous row,
The midpoint coordinates of the change points are stored in the segment memory 24 as the end point coordinates of segment 4 of black pattern 1. In this way, if there is a black pattern in the previous row or one black pattern, it is currently 1.
If the black and white change points are no longer detected for silk, then the formula (6
There is no need to determine connectivity based on a) and (6b) and detect discontinuous runs based on the final opcode.

FIGS. 3(A) to 3(D) are diagrams showing other examples of segments. Segments can be arbitrarily set based on the positions of the first run, last run, and discontinuous runs of the black pattern. For example, as shown in FIG. 3(A), at the start point, end point, and discontinuous run of a black band, the first black bit having the longest continuous length among the black bits constituting each of these second black runs The black bit (hereinafter referred to as the longest bit) constituting the run is detected, and the coordinates of the midpoint of the black/white transition point of the detected longest bit are set as the start and end point coordinates of the segment.

In this case, if there is one longest bit, the coordinates of the one longest bit, or if there are multiple longest bits, the coordinates of the midpoint of these black bits, the start point or end point coordinates of the segment. Suppose that Also, as shown in Figure 3 (B),
In the first run, last run, and discontinuous run of the black pattern,
Among the black bits constituting each of the second black runs, the black bits constituting the first black run having a continuous length of any suitable length other than the longest continuous length, for example, the black bit of code O] are detected. However, the coordinates of the midpoint of the detected black bit may be used as the start point or end point coordinates of the segment. In addition, as shown in FIG. 3(C), in the first run, the last run, and the discontinuous run of the black pattern, the coordinates of the white change point of each of these second black runs are calculated as the start point and end point coordinates of the segment. It is also possible to do this. In the case of FIG. 3(D), the segment extraction section is a segment extraction section that extracts segments as described below. In other words, for black patterns for which no discontinuous runs have been detected, the positions at of the first run and the last run of the black pattern are detected as the start and end points of the segment, respectively, and the geometric structure of the black pattern is divided into one line. Represented by segments. In addition, for a black pattern in which a discontinuous run is detected, the first run and last 8% of the black pattern are detected as the start and end points of the segment, respectively, and the position of the discontinuous run is detected as the breaking point of the segment. Then, the black pattern is expressed by one segment having a break point.

The starting point and ending point of the segment are the positions of the first run and the last run of the 11% black pattern.For example, the middle point coordinates of the black and white transition points of these first and last runs are taken as the position of the break point, and the position of the break point is the position of the discontinuous run. For example, let it be the midpoint coordinates of the black and white transition points of the current discontinuous run. If the format of the segment information shown in FIG. 3(D) is, for example, the coordinates of the start and end points of the segment, the number of break points, and the coordinates of the break points, when recognizing a character figure, the break points or It is possible to determine whether or not a segment includes a bending point.Furthermore, if a more detailed investigation is required, the bending point coordinates can be referred to, so a segment in the form of Figure 3 (A) to (C) can be determined. It has the same information with a smaller amount of data than . Although not shown in the figure, when one black pattern with discontinuous runs is expressed by multiple segments, the starting point and/or ending point coordinates of these segments are A flag indicating that the segment is an extracted segment may be attached, and an extraction order number attached to the extraction order of the black pattern may be added to the start point and/or end point coordinates of the segment. You can.

By using such flags and extraction order numbers, it is possible to determine the connectivity between segments, and this determination of the connectivity between segments can be used to recognize characters and figures. Furthermore, in the above-described embodiment, the starting point and ending point of each segment of the black pattern 1 are individually determined, or gA
The position of the previous discontinuous run in the previous and subsequent segments to be searched (or the position 1f% of the current discontinuous run)
, may be used as common coordinates for the end point of the preceding segment and the starting point of the subsequent segment. Figures 4 and 5 are diagrams showing examples of segment extraction from two figures having similar parts. Figures 4 and 5 (A) are original patterns of two similar shapes; Figures 4 (B), (C), (D) and (E)
represent a horizontal segment h, a vertical segment, a left diagonal segment g, and a right diagonal segment r extracted by the original pattern shown in FIG. 4(A), respectively;
Figures (B), (C), (D), and (E) are a horizontal segment h, a vertical segment V, a left diagonal segment l, and a right diagonal segment extracted from the original pattern shown in Figure 5 (A) by the apparatus of the embodiment, respectively. Segment r7a is shown. Note that the hatched portions in the figure indicate black bit portions.

As can be understood from these figures, segments different from straight parts can be extracted even from curved parts with small curvatures, so when recognizing character shapes, the recognition unit 26 detects many features from the extracted segments. Therefore, similar figures can be identified with high accuracy.

FIGS. 6(A) to 6(C) are diagrams showing examples of other segments extracted by the embodiment apparatus, and FIGS. The horizontal subpatterns corresponding to B) and (C) show the extracted horizontal segments. In FIGS. 6(A) to (C), the diagonally hatched portion represents the black bit portion to which black code 01 is attached, and the lattice hatched portion represents the black bit portion to which black code 10 is attached,
The dashed line represents the extracted segment.

As can be understood from Figures 6 and 11, in the case of a curve with a small curvature, in the case of a figure in which two straight lines are directly connected at an intersection angle close to 1800, two straight lines that extend in roughly the same direction Even if there are small figures in between, segments that more faithfully represent the structure of the figures can be extracted than before.

Figure 7 is a diagram for explaining the embodiment of the second invention, and is a functional block diagram schematically showing the overall configuration of the embodiment of the second invention. In addition, the constituent precepts corresponding to the constituent precepts of the embodiment of the first invention described above are indicated by the same reference numerals. In the following description, points that are different from the embodiment of the first invention described above will be mainly explained, and detailed explanations of the same points as the embodiment of the first invention will be omitted.

The figure recognition device according to the embodiment of the second invention includes a threshold setting section 12, a sub-pattern extraction section 14, a segment extraction section 16,
A segment classification unit 28 that classifies segments according to the first scanning direction and based on segment type information obtained from the one-black code; and a recognition unit 30 that detects the segment. Roughly speaking, the figure recognition device of this embodiment has a pattern memory 18.
, a line width calculation section 20, a sub-pattern memory 22, and a segment memory 24 for storing segments by classification. This example will be explained in more detail below. (Segment Extraction Unit) The segment extraction 8B16 of this embodiment detects the first run of the black pattern and then detects the last run of the black pattern without detecting discontinuous runs. The 11% of the last run is detected as the start and end points of the segment, respectively.

In addition, when detecting the first run of the black pattern and then detecting the discontinuous run and the last run of the black pattern, the black pattern (division pattern) is divided into I1 based on the depth of the discontinuous run.
The first and last runs of 11 are detected as the start and end points of the segment, respectively.

The positions of the first run and last run of the black pattern or divided pattern, and the discontinuous runs of the roughly black pattern (all of these black runs are second black runs) are shown in Figures 2 and 3 (A).
~Use the position of the second black run explained using (C). Furthermore, the segment extraction unit 16 of this embodiment outputs, as segment type information, for example, the final operation code of the final operation result at the start point or end point of the segment. Roughly speaking, in this embodiment, when one black pattern having discontinuous runs is extracted, the black pattern is the extracted segment with respect to the start point and/or end point coordinates of the extracted segment. A flag (connection flag) indicating this is attached.

The segment extraction unit 16 of this embodiment also includes vertical, horizontal, right diagonal, and left diagonal segment extraction units 16a, +6b, 1 for extracting segments for each first scanning direction.
6c and +6d, each extraction part 16a116b,
16c and +6d are provided with vertical, horizontal, right diagonal, and left diagonal segment registers (not shown). Next, a specific example of the segment extraction process will be described, taking the extraction of a vertical segment in reference to FIG. 2 as an example. Similarly to the vertical segment extraction section 16a, another segment extraction section J6
b, 16c, and +6d are also extracted, so these segment extraction units 16b, ! 6c and 1
The explanation of segment extraction in 6d will be omitted. When the vertical segment extraction unit 16a finishes extracting the coded vertical subpatterns, it starts processing for extracting vertical segments corresponding to the vertical subpatterns. In vertical segment extraction using the sub-pattern shown in Figure 2, no black and white change points are detected from the scan start line to the line one line before scan line Y1, but one set of black and white change points is detected in scan line Y. Detect points. It can be seen that black pattern 1 has appeared because no change point was detected up to the line one line before scan line Y1, and a pair of black and white change points was detected for the first time in scan line Y. Save a set of black and white changing points in scanning line Y+ as changing points of black pattern 1. Also, the position of the second black dot (corresponding to the first run of the divided pattern 1a) between a pair of black and white change points in the scanning line Y is not shown as the starting point coordinates of the vertical segment 1 of the black pattern 1. Further, the final operation code 01 of the current Y (which is the final operation code at the start point of the division pattern 1a) is stored in the vertical segment register as the segment type information of the vertical segment 1, together with the starting point coordinates.

The vertical segment extraction unit 16a detects two sets of black and white change points by scanning the scanning line Y2. In scanning line Y2, there is a change that satisfies relational expressions (6a) and (6b)% between the black and white changing points detected in the previous line (scanning line Yl) and the black and white changing points detected in the current line (scanning line Y2). Detect a set of points and a silk of unsatisfied change points.

A set of changing points not detected in the previous line but detected in the current line is saved as a set of changing points of the newly appearing black pattern 2. Furthermore, the position of the second black run (corresponding to the first run of black pattern 2) between the pair of black and white change points that newly appeared in the current state is taken as the starting point coordinates of vertical segment 2 of black pattern 2. The final operation code 10 of the current Y2 (the final operation code at the start point of the black pattern 2) is stored in the vertical segment register as the segment type information of the vertical segment 2, along with the start point coordinates. .

After roughly scanning r8, the vertical segment extractor 16a
is scan line Y3 and the current (scan line Y3) and previous line (scan line Y
It is detected that the second black run in the scanning line one line before 3) is a discontinuous run.

Once detected, the vertical segment extracting unit 16a sets the last run and the first run of the preceding and subsequent division patterns (corresponding to division patterns 1a and 1b) that are adjacent to the corners of the black pattern 1 based on the discontinuous runs, and sets these last runs. Let the positions of the run and the first run be the end and start points of the segments (corresponding to segments 1 and 3) of the previous and subsequent division patterns. Then, the position of the end point of segment 1, the connection flag indicating that there is a segment (segment 3) connected to segment 1, and the start point and segment type information of segment 1 that have not been previously stored in the vertical segment register. It is output to the segment classification unit 28. At the same time, the starting point position of segment 3 and a connection flag indicating that there is a segment (segment 1) connected to segment 3 are stored in the vertical segment register. When the vertical sub-pattern is roughly scanned, a discontinuous run is not detected in the scan from scan line Y3 to one line before scan line Y4, but a discontinuous run of black pattern 1 is detected in the scan of scan line Y4.

In the same manner as described above, the last run and the first run of the front and rear division patterns (corresponding to division patterns 1b and 1c) that are in direct contact with black pattern 1 are set based on the detected discontinuous runs, and these last runs are set. and the position at of the first run, and the positions of the end and start points of the segments (corresponding to segments 3 and 4) of the previous and subsequent division patterns.

and the position of the end point of segment 3 and the segment 3
The end point concatenation flag indicating that there is a segment (corresponding to segment 4) connected to the end point of , and the start point, segment type information, and start point concatenation flag of segment 3 previously stored in the vertical segment register. It is output to the segment classification unit 28. Along with this, the start point position of segment 4 and a link flag indicating that there is a segment (segment 3) connected to the start point of segment 4 are stored in the vertical segment register. In scanning Y5, black and white changing points were detected in the previous line (scanning line one line before scanning line Y), but black and white changing points were not detected in the current line (scanning line Y5), that is, a black pattern. Detect black pattern 2. The vertical segment extraction unit t6a sets the position of the second black run of black pattern 2 in the previous row as the current end point coordinates of segment 2 of black pattern 2, and stores the end point coordinates of segment 2 in the vertical segment register first. The coordinates of the starting point of segment 2 are output to the segment classification unit 28. When scanning line Y6 is further scanned, the 181 pairs of black and white change points are no longer detected. Therefore, the black pattern (
That is, it can be seen that the end point of black pattern 1) has been detected.

The vertical segment extraction unit 16a extracts the second black run position 1 of the black pattern 1 in the previous row! %, the end point coordinates of segment 4 of black pattern 1, the end point coordinates of this segment 4, the start point coordinates of segment 4 stored in the vertical segment register, and the segment (segment 3) connected to the start point of segment 4. ) is output to the segment classification unit 28 as a connected flag indicating the existence of the black pattern. If it becomes 1, then the formula (6a
) and (6b) are not required.

(Segment Classification Unit) The segment classification unit 28 of this embodiment uses, for example, the final operation code of the final operation result at the start point or end point of the segment as the segment type information.

When this segment type information is used, the pattern of the segment obtained from the segment type information is constructed by checking whether the number of each bit (each digit) of the segment information is "○" or "1". You can find out the first black run with the longest continuous length. For example, in FIG. 2, the segment type information in the scanning line one line before scanning line Y3 or scanning line Y4 of division pattern 1b is code 11.
Therefore, it can be seen that the division pattern 1b includes a first black run having a continuous length corresponding to the code 10. Therefore, based on the segment type information, the pattern of the segment from which the segment type information has been obtained can be classified according to the length in the first scanning direction.

In this example, there are two long segments and one short segment.
If the segment type information is code 10 or code 11, the segment in the segment type information is considered to be a long segment, and if the segment type information is code 01, the segment in the segment type information is considered to be a short segment. .

Furthermore, when the segment type information is code 01, the segment classification unit 28 of this embodiment calculates the distance MS between the starting point and the ending point of the segment for which the segment type information has been obtained, and classifies the segment based on this distance. do.
The classification threshold @T set to an arbitrary suitable value for this classification is compared with the distance MS.

By setting the threshold T in any suitable manner, the short segments can be, for example, segments of a long linear pattern extending in a direction intersecting the first scanning direction at a predetermined angle, or approximately in the first scanning direction. ; 8 Segments of a pattern of curved figures that undulate in the direction (hereinafter, these segments are referred to as short segments of long figures); It can be determined whether the segment is a pattern segment of a short straight line figure (hereinafter referred to as a short segment of a short figure). Therefore, short segments can be classified into short segments of long figures or short segments of short figures.

FIG. 8 is a diagram showing an example of segment extraction.

In the same figure, 1 to H15 are short wood flat segments with segment type information or code 01, 2I to 27 are horizontal segments whose segment type information is code 10 or 11, and 2I to 27 are horizontal segments whose segment type information is code 10 or 11. Horizontal segments with code 01, , 1 and 12 are vertical segments with segment type information code O], V21 to V
24, segment type information is code ○] or code 11
Indicates the vertical segment.

Further, the segment classification unit 28 sets a storage location in the segment memory 24 for storing the segments classified as described above.

Figure 9 is a diagram for explaining the storage location of segments. In the figure, HIA. H 2A, }-1 3A
and VIA, V2A indicate the start address of a storage area for storing horizontal and vertical segments in the segment memory 24; is the start address of the storage area for the short Kihira segment of a short figure whose segment type information is Koto 01, H2A is the start address of the storage area for the long horizontal segment whose segment type information is code 10 or 11, and H3A is the segment The type information is code 01 and the start address of the storage area for the short horizontal segment of a long figure is VIA, the segment type information is code F:0.
1 indicates the start address of the storage area for short vertical segments of short graphics, and v2A indicates the segment type information or code 10 or 11, indicating the start address of the storage area for long vertical segments.

The segment classification unit 28 determines the expected number of extracted segments for each segment in each first scanning direction (according to this, sets the start address of the storage area of each segment in each first scanning direction at arbitrary suitable intervals). The expected number of segments to be extracted is inputted to the segment classification unit 28 via an input unit (not shown), for example, and the segment classification unit 2
8 sets a storage area large enough to store the extracted segments for each segment in each first scanning direction based on the number of extracted segments. Although not shown, the same starting addresses as in the case of the wooden flat and vertical segments are set for the right diagonal and left diagonal segments as well.

The segment classification unit 28 extracts segment information (start point and end point coordinates, connection flag), segment type information, and information on which direction in the first scanning direction the segment from which the segment information was obtained belongs (segment direction).
Based on this, the segment storage location is set in sequence and the segment is stored in the set storage location. As a result, the segments are classified according to each first scanning direction and based on the segment type information, and are stored in the segment memory 2.
It is stored in 4. (Recognition Unit) The recognition unit 30 of this embodiment detects a segment having a positional relationship that satisfies the definition of a character/figure to be identified as a segment to be identified. For this detection, the recognition unit 30 searches for a segment in a range that satisfies the definition from among the segments classified according to the definition of the character/figure.

An example will be explained with reference to FIGS. 8 and 9. For example, a figure representing a horizontal capacitor can be defined as a figure in which two short vertical lines oppose each other in a predetermined positional relationship, and a long horizontal line connects near the center of each vertical line. It is assumed that This definition also includes other geometric characteristics such as the length and direction of the line segments that make up the capacitor.

In order to identify the segment of the capacitor defined in this way, the recognition unit 30 searches the storage area at the start address 1A to detect a short vertical segment that is one of the line segments that constitute the defined capacitor. A vertical segment V. that satisfies the condition of a line segment that constitutes a capacitor to be identified from among the segments of the storage area. Detect. Next, to find another segment that makes up the defined capacitor, what is the short vertical segment at a given position 11W defined as vertical segment vI1? The storage area of the head address VIA that is estimated to be stored is searched, and the vertical segment V1■ having a predetermined positional relationship with the vertical segment Vz is detected from among the segments in the storage area. Thereafter, a storage area that is presumed to store line segments that meet the definition is similarly searched, and segments with conditions that meet the definition of a capacitor are detected from among the segments in the storage area. As a result, the first address 2 that stores the long horizontal segment is
From the storage area of A, horizontal segment H25 and vertical segment V are connected to vertical segment V and meet the definition.
Detecting the wooden flat segment H28 connected to I2 and satisfying the definition, the horizontal capacitor segment t is selected from among the segments stored in the segment memory 24.
I separate.

In the embodiment described above, one continuous length threshold [? However, a plurality of arbitrarily suitable continuous length thresholds W may be set for one sub-pattern, for example, j continuous length thresholds for a vertical sub-pattern, vsl Lvs2,
・・・・・・ LV. ,%, for horizontal subpattern]
continuous length threshold Lhs, ,2,...Lh
-it, right diagonal subbakun (j consecutive length thresholds for this, rsl, Lr52,...L--+t, j consecutive length thresholds for left diagonal subbakkun, is+,
L ls2 %... L1,, may be set.

For example, if multiple continuous length thresholds are set for one subpattern,
sl % Lvs2 ,... When vsj % is set, L vw≦j2v<LVs+! The first length classified satisfying continuous length lv, L vs+≦I
The second classified continuous length lv that satisfies 2Vkuhi, 2
Length of the th,... vs (+-11 ≦u,
<L. The continuous length ℃V that satisfies j is determined to be the classified j-th length, and the continuous length lV that satisfies L vsJ≦A is determined to be the classified j-th +1-th length. In addition, the threshold Ja setting section is provided with a table that stores threshold information for determining the black-and-white threshold and continuous length ram for each segment extraction target such as a document, drawing, etc., and the threshold Ja setting section stores the threshold information from the information table based on the selection signal from the input section. may be selected, and the black-and-white threshold and continuous length ra@ may be set based on the selected am information. For example, a constant such as Nv for setting the black-and-white threshold 18 and a continuous length R value may be stored as @value information. (Effects of the Invention) According to the segment extracting device of the first invention having such a structure, an original pattern is scanned in the first scanning direction, and a sub-pattern coded with a white code and a black code is extracted. do. Then, the encoded sub-pattern is scanned in the second scanning direction, and it is determined whether or not the second black runs are connected based on the coordinates of the black and white change points, and the second black runs (hereinafter referred to as This connected mass of second black runs is called a black pattern)
The system detects the first run and last run of the black pattern, and also detects discontinuous runs of black patterns based on the black code. A black pattern is a pattern that represents a certain character figure. By scanning the coded sub-patterns and examining the changes in the black code of the black pattern, we can determine the geometrical vt of the black pattern.
It is also possible to detect discontinuous runs in which yi changes. By extracting segments of the black pattern using the positions of the first run and last run of the black pattern as well as the position percentage of these discontinuous runs, geometric characteristics can be extracted more faithfully than before!
Extract the segment representing i.

Therefore, the features extracted from segments are more stable than before, and the accuracy of character/figure identification can be improved. Furthermore, since characters and figures are identified based on stable features, the speed of identification processing can be made faster than before. Further, according to the figure recognition device of the second invention, the segment is
The first classification is based on the scanning direction, and the segments are divided based on the segment type information obtained from the black code. Then, the segments that make up the character figure are detected from among the segments that meet the definition 1 of the character figure. By detecting segments constituting a character figure from among the classified segments, the speed of detecting segments constituting a character figure can be improved. This makes it possible to speed up the character/figure identification process.

[Brief explanation of drawings]

FIG. 1 is an explanatory diagram of the embodiment of the first invention; FIG. 2 is a diagram showing an example of the position of a coded sub-pattern; FIGS. 3(A) to (D) are diagrams showing other examples of segments; 4(A)-(E) and FIG. 5(A)-(E) are diagrams showing examples of segments extracted by the embodiment apparatus in figures having similar shapes, and FIG. 6(A)-(C). ) is a diagram showing another example of segments extracted by the embodiment device, FIG. 7 is an explanatory diagram of the embodiment of the second invention, FIG. 8 is a diagram showing an example of segment extraction, and FIG. 9 is a diagram showing segment storage. 10(A) to 10(C) are diagrams showing examples of original patterns, and FIGS. 11(A) to 11(C) are diagrams showing examples of conventional horizontal sub-patterns and horizontal segments. DESCRIPTION OF SYMBOLS 10... Segment extraction device 12... Threshold value setting part, 14... Sub pattern extraction part 16... Segment extraction part 28... Segment classification part 30... Recognition part.

Claims (9)

[Claims] (1) A black-and-white threshold for extracting a plurality of sub-patterns for each first scanning direction from a black-and-white binary original pattern, and a first black run in the first scanning direction of the sub-pattern as the first black run. a threshold setting unit that sets a continuous length threshold for classification according to the continuous length of the pattern; and a first black pattern that scans the original pattern in a first scanning direction and has a continuous length that is equal to or greater than the black and white threshold. The run is extracted as a black bit part of the sub-pattern, a white code is attached to the white bit part of the sub-pattern, and a black code classified according to the continuous length is attached to the black bit part of the sub-pattern. a sub-pattern extraction unit that attaches a sub-pattern based on a comparison result of the continuous length threshold value and the continuous length of the first black run, and scans each sub-pattern in a second scanning direction forming a predetermined angle with the first scanning direction; Whether or not the second black runs in the second scanning direction are connected based on the coordinates of the black change point when a white bit changes to a black bit on the line and the white change point when a black bit changes to a white bit on the line. The first run and the last run of the second black run to be connected are detected, and the discontinuous runs of the second black run to be connected are detected based on the black code, and the second black run to be connected is detected. 1. A segment extraction device comprising: a segment extraction unit that extracts segments based on the positions of the first and last runs of black runs and the positions of discontinuous runs. (2) The threshold setting unit is a threshold setting unit that sets n (where n is a natural number) continuous length thresholds, and the sub-pattern extraction unit is configured to set n consecutive length thresholds (n is a natural number); is a natural number such that 1≦i≦n+1)
A sub-pattern extracting section adds a black code expressed in binary as 2^i^-^1 to the first black run having a length of For the black bits up to the change point, perform a logical sum operation for each bit of the black code to obtain a logical sum calculation result for each second black run, and then obtain a logical sum calculation result for the previous row and the current second black run. 2. The segment extraction device according to claim 1, wherein the segment extraction unit detects a discontinuous run based on . (3) When the segment extraction unit does not detect a discontinuous run of the connected second black runs, the positions of the first run and the last run of the connected second black runs are used as the start and end points of the segment, respectively. When a discontinuous run of the second connected black run is detected,
A claim characterized in that the segment extraction unit detects the positions of the first run and the last run of the connected second black runs divided based on the positions of the discontinuous runs as the positions of the start and end points of the segment, respectively. 3. The segment extraction device according to 1 or 2. (4) When the segment extraction unit does not detect a discontinuous run of the connected second black runs, the positions of the first run and the last run of the connected second black runs are set as the start point and end point of the segment, respectively. When a discontinuous run of the second connected black run is detected,
A segment in which the positions of the first run and the last run of the second connected black runs are detected as the start and end points of the segment, respectively, and the position of the discontinuous run is detected as the break point or bending point of the segment. The segment extraction device according to claim 1 or 2, characterized in that the segment extraction device is an extraction section. (5) A black-and-white threshold for extracting a plurality of sub-patterns in each first scanning direction from a black-and-white binary original pattern, and a first black run in the first scanning direction of the sub-pattern as the first black run. a threshold setting unit that sets a continuous length threshold for classification according to the continuous length of the pattern; and a first black pattern that scans the original pattern in a first scanning direction and has a continuous length that is equal to or greater than the black and white threshold. The run is extracted as a black bit part of the sub-pattern, a white code is attached to the white bit part of the sub-pattern, and a black code classified according to the continuous length is attached to the black bit part of the sub-pattern. a sub-pattern extraction unit that attaches a sub-pattern based on a comparison result of the continuous length threshold value and the continuous length of the first black run, and scans each sub-pattern in a second scanning direction forming a predetermined angle with the first scanning direction; Whether or not the second black runs in the second scanning direction are connected based on the coordinates of the black change point when a white bit changes to a black bit on the line and the white change point when a black bit changes to a white bit on the line. The first run and the last run of the second black run to be connected are detected, and the discontinuous runs of the second black run to be connected are detected based on the black code, and the second black run to be connected is detected. a segment extraction unit that extracts segments based on the positions of the first and last runs of black runs and the positions of discontinuous runs; A figure recognition device comprising: a segment classification unit that classifies based on segment type information; and a recognition unit that detects segments constituting a character figure from among segments classified according to a definition of the character figure. (6) The threshold setting unit is a threshold setting unit that sets n consecutive length thresholds (where n is a natural number), and the sub-pattern extraction unit is a is a natural number such that 1≦i≦n+1)
A sub-pattern extracting section adds a black code expressed in binary as 2^i^-^1 to the first black run having a length of For the black bits up to the change point, perform a logical sum operation for each bit of the black code to obtain a logical sum calculation result for each second black run, and then obtain a logical sum calculation result for the previous row and the current second black run. 6. The figure recognition device according to claim 5, wherein the segment extraction unit detects discontinuous runs based on . (7) When the segment extraction unit does not detect a discontinuous run of the connected second black runs, the positions of the first run and the last run of the connected second black runs are set as the start point and end point of the segment, respectively. When a discontinuous run of the second connected black run is detected,
A claim characterized in that the segment extraction unit detects the positions of the first run and the last run of the connected second black runs divided based on the positions of the discontinuous runs as the positions of the start and end points of the segment, respectively. 6. The figure recognition device according to 5 or 6. (8) The figure recognition device according to claim 6, wherein the segment classification unit is a segment classification unit that uses the logical sum operation result as segment type information. (9) The figure recognition device according to any one of claims 5 to 8, wherein the segment classification unit is a segment classification unit that classifies segments based on a distance between a starting point and an ending point of the segment. .