JPH07104955B2 - Line figure shaping method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Outline] The bending points are classified into different types, weights are applied to the opposing bending points so as to cancel each other out, and fine variability is eliminated by performing averaging processing in the vicinity. Shape

[Industrial application field]

The present invention relates to a method of shaping a line figure that has undergone thinning processing and has removed branch points and intersections, and particularly to a method of extracting bending points.

Shapes such as letters, blank maps, and design drawings are line shapes,
A huge memory capacity is required to store these in the memory of the computer. Therefore, a method of compressing and storing the data amount by a process called vectorization is adopted. This is a line figure, if it is a straight line, only the coordinate values of the end points,
If it is a polygonal line, it is a process expressed only by the coordinate values of the end point and the bending point.

For the vectorization process, the line figure width is reduced to 1 by the thinning process.
The present invention is a method for a thinned figure, although it may be performed after the line has a width of one pixel or directly on the original figure.

In the thinned figure, the pixel value of the line portion is 1, the surrounding area is 0, or the pixel value of the line portion is 0 and the surrounding area is 1.
Although only the former case will be described below, the same applies to the latter case. There are two types of definitions for the continuity of 1 representing a line part, and any 3 × 3 on the image
When the value of the pixel at the center in the vicinity of 1 is 1, there are 4 connections in which it is considered that only 1s in the four directions of up, down, left and right are continuous to the center 1, and 8 connections in which the connectivity is considered in the diagonal direction. For example, FIG. 2A shows a thin line in the definition of 4-connection, but not a thin line in the definition of 8-connection. When FIG. 2 (a) is thinned by 8 connections, it becomes FIG. 2 (b) or FIG. 2 (c). In the following description, the connectivity is described as 8 connections, but the same applies to the case of 4 connections.

The end point in the thin line is one of the neighboring patterns shown in FIG. 3, and a portion having any one of these patterns is searched for in the input image, and 1 is set as the pixel value of the center thereof, otherwise. It can be detected by constructing a logic that outputs 0. Defining the neighborhood pattern like this,
The process of giving 1 or 0 as the pixel value of the center of the part having the pattern and 0 or 1 as the pixel value of the center of the part having the other pattern is known as a process called logic filtering which is easy to implement in hardware. ing.

Similarly, the branch points in the thin line are given in the pattern shown in FIG. 4, and the intersections are given in the pattern shown in FIG.
By obtaining an image in which these are detected by logical filtering and taking an exclusive OR with the original image, the branch point and the intersection are removed from the original image, and the image input by the present invention can be obtained.

[Conventional technology]

As a method of detecting the bending point, for each pixel on the line in the thinned binary image, the curvature is obtained from the positional relationship with the pixels on the line in the vicinity thereof, and the coordinates of the pixel at which the curvature sharply changes are defined as the bending point. There is a way to do it. For example, when the line figure is as shown in FIG. 6, if the angle θ obtained from the relationship between the pixel P and the pixel on the line at the distance 3 is greater than a certain value, then P is the bending point. .

[Problems to be solved by the invention]

However, in this method, since it is necessary to refer to pixel values separated by a certain distance, it is necessary to randomly access the memory, and it is difficult to realize hardware.

The present invention is an algorithm that can be easily implemented in hardware, and can detect a bending point at high speed.

[Means for solving problems]

In the present invention, the bending point patterns are classified into different types, positive and negative opposing weights that cancel each other are added to the pixels of the pair of bending points, which have a point-symmetrical pattern relationship, and the weighted images are averaged. Thus, the bending points of the point symmetric pattern having opposing weights appearing in the neighborhood in the straight line portion are not detected because their values cancel each other out, and only the bending points where the curvature largely changes are detected without being cancelled. To do

This will be described below with reference to the drawings. FIG. 7 is a pattern of bending points. Is the inflection point of 135 ° and is the inflection point of 90 °.

FIGS. 8 (1) and 8 (2) show examples of straight line figures. The graphic is quantized on a square lattice to be stored in the memory, and the straight line becomes a broken line when viewed in pixel units (the bending point pattern in FIG. 7 appears). In FIG. 8 (1), the slope is steeper than that of (2), and the steeper the slope, the shorter the repetition of 1.

However, in the examples of FIGS. 8 (1) and 8 (2), only the end points of the respective straight lines are important, and the inflection points in the middle have no meaning. That is, although the straight line portion in FIGS. 8 (1) and 8 (2) includes a large number of bending points when viewed on a pixel-by-pixel basis, it looks straight to the human eye especially when each point is minute and numerous. Therefore, it is not necessary to detect these bending points. The bending points in the middle of both straight lines (1) and (2) are composed of a pair of patterns having a point-symmetrical pattern relationship. Therefore, these straight lines can be expressed only by the coordinates of both end points by not detecting them as bending points having a point symmetric pattern relationship on the way and omitting the straight line part on the way.

FIG. 9 is an example of broken lines. In FIG. 9, it is natural to view the pixel indicated by as a broken line having a bending point. The present invention does not detect the bending point on the straight line shown in FIGS. 8 (1) and 8 (2), but detects the bending point as shown in FIG.

In order to detect the bending point appearing at the bent line bending portion of FIG. 9 by distinguishing it from the bending point appearing on the straight line, the present invention classifies the patterns of the bending points and has a paired bending with a point symmetrical pattern. Give positive and negative weight to each other,
The bending points of different types of point-symmetric patterns are made to have different weights. For example, the following weighting is performed. In the pattern of the inflection points shown in FIGS. 7A and 7B, 81 is set as the value of the central pixel and −81 is set for
Add 162 for, and -162 for.

FIG. 10 (1) is the result of weighting the straight line in FIG. 8 (1), and FIG. 10 (2) is the result of weighting the broken line in FIG. In FIG. 8 (2), two 11's and 1's continue from the lower left (first row), and then go up one step diagonally to the upper right to continue two 11's and 1's (second row). Is -81 at the lower left corner of Fig. 10 (1).
Corresponds to the rightmost 1 in the first column above (the rightmost 1 is the 7th
It corresponds to 1 in the center of the pattern in the figure, so the weight is -8
1. ), 81 next to FIG. 10 (1) is 1 at the left end of the second row.
(The left end corresponds to the center of the pattern in FIG. 7, and therefore the weight is 81.). The same applies hereinafter.

Also, in FIG. 9 corresponds to the center of the pattern in FIG. 7,
Therefore, the weight is 162.

The image thus obtained is subjected to an averaging process only at the portion having the value 1 of the thinned binary image, using the thinned binary image which is the original image as a mask, to obtain (1) and (2) in FIG.

11 (1) and 11 (2) are obtained by the following averaging process. In the 3 × 3 area F1 of FIG. 10 (1), the sum of the pixel values at the upper right corner is −81, and the value of the remaining pixels is 0, so the sum of the pixel values in F1 is −81. When divided by the number of pixels in
As shown in Fig. 12 (1), this is the center P1 of F1.
Pixel value. -9 at the lower left of FIG. 11 (1) is obtained in this way. Similarly, for the area F2, the sum of the pixel values in the area is 81−81 = 0, and thus the pixel value of the center P2 of F2 is 0. This corresponds to 0 on the diagonally upper right side of -9 at the lower left corner of Fig. 11 (1). This averaging process is
8 (1) in FIG. 8 (1).
When 0 is allocated to the position where the pixel value is 0 (meaning masking with the thinned binary image which is the original image),
FIG. 11 (1) is obtained from FIG. (1). For area F3 in Fig. 10 (2), the sum of the pixel values in that area is-
81 + 162-81 = 0, so F3 as shown in Fig. 12 (2)
The pixel value of the center P3 of is 0. Two 18 in Fig. 11 (2)
The 0 in the middle one row of is thus obtained. Thereafter, averaging is performed in the same manner and masking is performed in FIG. 9 to obtain FIGS. 10 (2) to 11 (2). Like this (with mask)
The averaging process is easy to implement in hardware and is currently widely used.

Further, averaging processing similar to the above is performed on FIGS. 11 (1) and 11 (2) to obtain FIGS. 13 (1) and 13 (2).

The pixel values in (1) and (2) of FIG. 13 are distributed from -2 to 4, and the pixel value of 0 is -2 and 2 or less, and the pixel value is 0 for other pixels. By the straightening process such that the pixel value 1 is assigned, only 1 is applied to the pixel at the true bending point.
14 (1) and (2) are obtained. Such binarization processing is easy to implement in hardware and is currently widely used.

By applying the degeneracy processing to (1) and (2) in FIG.
Figure 15 (1) and (2) are obtained. 8 (1) and FIG. 9 are input, this process has no meaning.
When an image as shown in FIG. 16 (1) is input, the image corresponding to FIG. 14 becomes as shown in FIG. 16 (2), and noise is generated at the bending point. The degeneracy process is a process in which a continuous pixel value 1 is represented by a pixel value 1 at one point, and can also be configured by logical filtering. By performing the degeneracy process on FIG. 16 (2), FIG. 16 (3) is obtained.

FIGS. 17 (1) and 17 (2) are images in which the end points are extracted by logical filtering using the end point patterns shown in FIGS. 8 (1) and 9 to 3.

Figure 18 (1) and Figure 17 (1) and Figure 15 (2) and Figure 17 (2) are ORed to obtain Figures 18 (1) and (2), respectively. . That is, only the end points are detected from FIG. 8 (1), and the end points and bending points are extracted from FIG.

The weights 81, -81, 162, and -162 given earlier are averaged twice, and the value of 9 is squared so that the value remains. It is determined by adopting twice the number to represent the type. Since the base number is set to 81, the value used for binarization is determined to be the range where the resulting value is considered to be a straight line and is -2 or more and 2 or less. Even if you use multiples of the numbers shown above for the weights and Iki values, each number will be 162 so that only positive numbers can be used.
The same process can be performed by adding.

When the pixel value of the line portion is represented by 0 and the periphery is represented by 1, the first weighting pattern in which 0 and 1 are inverted is used, and the result of binarization is obtained in which 0 and 1 are inverted. In this way, a pattern in which 0 and 1 are inverted may be used as the pattern in the degeneracy process.

When extending the range of the averaging process, the binarization value may be determined in the same way as above.

In the case of four connections, the bending point pattern shown in FIG. 19 may be used. (The correspondence between ~ and weights is the same as for 8-connection.) Also, it is necessary to use 4-connection for degeneration and end point detection.

[Action]

In this way, weighting is performed according to the type of bending point, and it is possible to extract only the bending point of the broken line while avoiding the extraction of the bending point from the bending portion that occurs in the diagonal straight line.

〔Example〕

 FIG. 1 shows the configuration of an apparatus according to an embodiment of the present invention.

In the line figure shaping circuit block diagram of FIG. 1, a is an image memory in which a thinned binary image as shown in FIGS. 8 (1) and 9 is stored.

b to e are logic filtering circuits, and FIG. 7 to FIG. 7 are logic filtering circuits b, FIG.
When the same d is detected, the same is detected by the same e, 1 is output, otherwise 0 is output.

f to i are density conversion circuits, and when the output of b to e is 1.
Weights of 81, -81, 162, and -162 are added.

Reference numerals j to 1 denote inter-pixel arithmetic circuits that create an image by adding the outputs of f to i. After processing 1, Fig. 10 (1),
(2) is obtained.

m and n are averaging filtering circuits, and outputs (1) and (2) in FIG. 13 are obtained as their outputs.

o is a binarization circuit, and p is a degeneration circuit. o and p
As the processing result of (1) and (2) in FIG.

q is an end point detection circuit by logical filtering using the pattern of FIG. 3, and detects only end points from the thinned binary image in the image memory of a.

r is a delay circuit, and the delay time is set so as to match the delays from b to e to the delays of q and r.

s is an inter-pixel arithmetic circuit, which takes the logical sum of the output of o and the output of r.

An image memory t stores the processing result.

Since this processing can be realized by a pipeline method, the image memory t may be omitted.

〔The invention's effect〕

As is clear from the above description, according to the present invention, it is possible to accurately detect a bending point, and since it is easy to implement hardware, it is possible to shape a line figure at high speed.

[Brief description of drawings]

 FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a diagram for explaining the definition of a thin line, FIG. 3 is a pattern diagram of 8-connected end points, and FIG. 4 is a pattern diagram of 8-connected branch points. Fig. 5 is a pattern diagram of 8-connected intersections, Fig. 6 is a diagram for explaining a conventional method, Fig. 7 is a pattern diagram of 8-connected bending points, Fig. 8 is an example of 8-connected straight lines, Fig. 9 shows an example of 8-connected polygonal lines, Fig. 10 shows an example of weighting, Fig. 11 shows an example of averaging, Fig. 12 is an explanatory diagram of averaging process, and Fig. 13 shows second averaging. , Fig. 14 is an example of binarization, Fig. 15 is an example of degeneration, Fig. 16 is an example of the case where the degeneracy effect is effective, Fig. 17 is an example of end point detection, Fig. 18 Is an example of the processing result, and FIG. 19 is a pattern diagram of 4-connected bending points.

Claims (1)

[Claims] 1. A weight that opposes a bending point of a thinned binary line image according to a pattern and has a positive and negative value that cancels each other between pixels of a pair of bending points having a point-symmetrical pattern relationship. , And different weights are applied between inflection points having different types of point symmetry patterns, and for this weighted image, the input thinning 2
The averaging process in the vicinity of the pixel of interest is repeatedly performed twice on the entire image using the value image as a mask, and the result is binarized by the threshold value determined so that only the pixel at the bending point is detected from the previous weight. , And extracting a true inflection point from the thinned binary image by further performing a degeneracy process in which a plurality of pixels having continuous pixel values are represented by a single pixel having a pixel value of 1 for the result. A method of shaping a line figure.