JPH02176982A - Graphic feature extracting method - Google Patents

Abstract

PURPOSE:To obtain a feature part without being affected by the small unevenness of a graphic outline by executing computation from the data of three points having data intervals in order to obtain the data of a part which express the feature of a graphic. CONSTITUTION:The coordinates d- and d+, which are separated from a coordinate do forward and backward only by the number of S, are obtained. When a length l of a perpendicular drawn down from a point, which are expressed by the do, to a straight line passing through two points being expressed by the d- and d+, is larger than a reference length lst of the perpendicular set in advance, this coordinate do is defined as the candidate feature point of the graphic. When an angle theta made by a vector direction from the d- toward do and a vector directed from the do to the d+ is larger than a reference angle thetas set in advance, this coordinate do is defined as the candidate feature point of the graphic. Then, the central candidate feature point in a series of the candidate feature points, which have successive adjacent relationship along the sequence of the coordinates, and the candidate feature point not to have the adjacent candidate feature points along the sequence of the coordinates are defined as the regular feature point. Thus, the data of the part to express the feature of the graphic can be efficiently extracted.

Description

[Detailed description of the invention] [Industrial application field]

The present invention relates to a graphic feature extraction method for extracting graphic features and compressing the amount of data using image processing techniques and simple numerical operations. Note that in the following figures, the same reference numerals indicate the same or corresponding parts.

[Conventional technology]

Conventionally, the method of extracting the parts that represent the features of a figure using contour line data and compressing the amount of data is to trace the contour line using Freeman's chain code and approximate it to a straight line. There is a way to vectorize it. 5(a) to 5(d) are diagrams for explaining the outline of this method, and FIG. 6 shows a specific example of a contour line vectorized by this method. FIG. 5(a) shows coordinate string data of the outline of a figure, that is, a data string in which the coordinates of each pixel indicating the outline of the figure are stored in order along the outline. In the same figure, a1 to an+1 are sequentially closed loops (contour lines) a
The coordinates of a series of pixels representing . However, the total number of pixels is n, and therefore an+1-at. Similarly, bl and b2 indicate part of the coordinates of a series of pixels representing the closed loop b. Figure (b) shows the direction code (also called chain code) used when tracing pixel coordinates along the contour line. Indicates that a direction code of 0 to 7 is assigned. The code of the tracking direction from the target pixel to the adjacent pixel is determined depending on which of the direction codes O to 7 the coordinates of the pixel in the tracking direction on the contour line adjacent to the target pixel corresponds to. Figure (C) shows the chain code string obtained in this way, and acl, which follows the coordinate a1 of the starting point of contour tracing for loop a, indicates the direction code from coordinate a1 to adjacent coordinate a2, and similarly ac2 indicates a direction code from coordinate a2 to adjacent coordinate a3. Below, ac3 to acn are sequentially changed to coordinates a3 to an
This is the direction code obtained for . Also for loop b, bl is the coordinate of the starting point of contour tracing,
bcl~bc3- are direction codes obtained in the same way as loop a. In this way, the contour of the loop a of the figure can be expressed by the position (coordinates) al of the starting point of contour tracing and the code strings in n directions. In this case, Fig. 5 (
Compared to the data in a), the data in the same figure (C) represented by the chain code requires less data.However, as it is, the number of data will not decrease, so in order to further reduce it, Examine changes in adjacent direction codes in the chain code data string, and as shown in (d) in the same figure, change the part of the direction code column in (C) to the direction code,
It is replaced with a vector data string consisting of a set of the code and the number of consecutive direction codes having the same value (that is, the length of the direction vector).

[Problem to be solved by the invention]

However, the conventional methods described above have the following problems. For example, if there is a slight unevenness, such as a hand-drawn outline of a figure, the vector will always be divided at that point. Even in such a case, such a substantially linear contour may be divided into a plurality of vectors, and efficient data compression may not be possible. Therefore, it is an object of the present invention to provide a figure feature extraction method that solves the problems of the conventional techniques and can efficiently extract data representing the features of a figure from outline coordinate string data of the figure. be.

[Means to solve the problem]

In order to solve the above-mentioned problems, the method of the present invention is a method of extracting feature points included in a contour line in a binarized image of an r-target figure, the method comprising: After obtaining the data for the column, on this coordinate column, distance a predetermined number of coordinates (data interval S, etc.) before and after the target coordinate (d., hereinafter referred to as the first coordinate). A total of two coordinates (
d+, d-, etc., hereinafter referred to as the second and third coordinates), and calculate the length of the perpendicular line (p, etc.) drawn from the first coordinate to the straight line connecting the second and third coordinates. , predetermined length (41!
st, etc.), or the angle (such as θ) between the vector from the third coordinate to the first coordinate and the vector from the first coordinate to the second coordinate is a predetermined angle ( θst, etc.), the first coordinates are set as candidate feature points, and all the candidate feature points are determined while sequentially selecting the angular coordinates of the coordinate string as the first coordinates; Among the feature points, a central candidate feature point among a series of candidate feature points that are sequentially adjacent to each other along the coordinate string;
And candidate feature points that do not have an adjacent candidate feature point along the coordinate string are set to 1 so that they are regular feature points (vector data, etc.).

[For use]

When considering two coordinate data d-d+ that are separated by the set number of data S before and after do with respect to target coordinate data do in the outline coordinate string data of a figure, 1) Expressed as d+, d- A straight line passing through the two points d. When the length 2 of the perpendicular line drawn from the point represented by is larger than the preset reference length fst of the perpendicular line, this coordinate data do is extracted as data representing the feature of the figure, and vector data is created. do. 2) When the angle θ between the vector from d- to do and the vector from do to d+ is larger than a preset reference angle θst, this coordinate data do is extracted as data representing the feature of the figure. Well, create a vector data sequence.

【Example】

The present invention will be explained in detail below based on FIGS. 1 to 4. FIG. 1 is a diagram showing the principle of applying the present invention to contour lines, FIG. 2 is a diagram showing an example of feature parts extracted from the contour line, and FIG. 3 is a diagram showing an example of the format of contour line coordinate string data. FIG. 4 is a flowchart showing the procedure of the present invention. As shown in Fig. 3, the symbol 31 indicating the number of the closed loop is placed first in the data of the contour coordinate string, then the coordinate string data is arranged in order for each pair of X, y coordinates, and finally, the symbol 31 indicating the closed loop number is placed. It is assumed that the last contour line coordinate data is included, and that this data is arranged one-dimensionally by the number N of closed loops. However, although the coordinate system in this case is a rectangular coordinate system in this embodiment, the present invention is not limited to a rectangular coordinate system. FIG. 1 shows part of such a contour. The present invention is based on the coordinate d+, which is located a predetermined data interval S (in this example, the number of coordinates is 11) in front of the target coordinate do on this contour line.
, and the coordinate d-, which is also located behind the target coordinate do by the same data interval S, and the coordinate d is placed on a straight line connecting the coordinates d+ and d-. When the length l of the perpendicular line drawn from is longer than the preset length 1st, the coordinate point do is set as a candidate for a feature point (referred to as a candidate feature point). Also, instead of the perpendicular length 2, the angle θ between the vector from the coordinate d- to do and the vector from the coordinate do to d+
is larger than a predetermined value θst, a method may be used in which the coordinate do is used as a candidate feature point. Next, FIG. 4 will be explained with reference to FIGS. 1 and 3. First, in step S1 in FIG. 4(a), outline coordinate string data is created as shown in FIG. Next, in step S2, the data interval S and standard line length lst in FIG. Set the number (order) j of the target coordinate data do in the contour line coordinate string data to 1.
Initialize to . Then, in step S3, the third coordinate data do in the contour coordinate string data of loop number i passes through the coordinate data d+ and d- of two points separated from each other by the data interval S set in step S2. Find the equation of a straight line by calculation. Next, the length l of a perpendicular line drawn from point do to the straight line obtained in step 3 is calculated (step 4). This perpendicular length l is the standard line length 1is set in step S2.
Compare with t! If is greater than or equal to 1st, point do is considered a candidate feature point, its coordinate data is left as is, and l is 1st.
If it is less than 0, the coordinate data of do is set to "0" and a new data string is created (step 35). The method for setting the standard line length 1st at this time is, for example, when trying to extract data of a part where the contour curves at a right angle from the contour line coordinate data string, l s t # S /J"r + number It is sufficient to set a value of about
, i
=N, that is, for all loops,
An example of coordinate data of a contour line consisting of coordinate data of a candidate feature point and coordinate data of 0° is obtained. Next, the data group of candidate feature points is compressed by the process shown in FIG. 4 0)) to obtain regular feature points. That is, in step S6, in step 5, the coordinate data of the candidate feature point to be extracted is left as is, and the others are set to "0"', but in order to further reduce the amount of data and create a vector data sequence, From among the data strings that did not become , only the coordinate data of the center point in the coordinate data string (referred to as vector data) sandwiched between "0" before and after will be retained, and the other data will be removed. For example, it is possible to obtain vector data as the coordinates of the representative feature point of the right-angled part of the figure.The process in step S6 is performed in the same way as in FIG. 4(a) for the parameters j-1 to nN and i-1 to N. , that is, the pixels within the loop, and all loops. The diagonal lines in Figure 2 indicate the pixels left as candidate feature points in this way, and the blank pixels represent the center pixel, that is, the representative of this feature. The coordinates (vector data) to Next, another method to extract the parts that represent the features of a shape is to
In contrast to steps 33, 34, and 35 described above, which only used the perpendicular length 2, the angle θ between the vectors d−, do and the vector stone π7o is greater than or equal to the preset reference angle θst. Whether to leave the coordinate data of point do “0”
”There is also a way to do that.

【Effect of the invention】

According to the present invention, after obtaining the coordinate string data for the pixel string representing the contour line, on this coordinate string, a total of two data points are set at a predetermined data interval S before and after the target coordinate do. coordinate d+,
Find d-, and connect the coordinates d- and d- to the straight line.
The length 2 of the perpendicular line drawn from o is the predetermined length! Is it longer than St?
Or an angle θ formed by a vector from the coordinate d- to the coordinate do and a vector from the coordinate do to the coordinate d+.
is larger than a predetermined angle θst, the corresponding coordinate do is set as a candidate feature point, and all the candidate feature points are determined while sequentially selecting each coordinate in the coordinate string as the coordinate do, and the candidate feature points are Among them, a central candidate feature point among a series of candidate feature points that are sequentially adjacent to each other along the coordinate string;
Since the candidate feature points that do not have any adjacent candidate feature points along the coordinate sequence are treated as regular feature points (vector data), in order to obtain the data of the part representing the feature of the figure, we need to change the data interval. Since the calculation is performed from the data of three existing points, it is possible to obtain the characteristic portion without being affected by small irregularities in the contour of the figure, and it is possible to compress the data by leaving only the characteristic portion. Furthermore, since vectorization is not performed while directly tracking the binarized image, but by using contour coordinate string data, the program can be made with fewer decisions and processing time can be shortened.

[BRIEF DESCRIPTION OF THE DRAWINGS] Fig. 1 is a diagram explaining the principle of an embodiment of the present invention, Fig. 2 is a diagram showing characteristic parts extracted from the contour line, and Fig. 3
Similarly, FIG. 4 is a flowchart explaining the present invention in detail, and FIG.
6 are explanatory diagrams of a conventional contour tracing method. do, d juice, d-: coordinate, S: data interval, 2: perpendicular length, θ: circumferential angle i, st: standard type line length, θst: standard angle. Figure 2 Figure 4 (b) Figure 5 O Figure 6

Claims (1)

[Claims] 1) A method for extracting feature points included in a contour line in a binarized image of a target figure, the method comprising: obtaining coordinate string data for a pixel string representing the contour line; On this coordinate string, find a total of two coordinates (hereinafter referred to as second and third coordinates, respectively) with a predetermined number of coordinates before and after the target coordinate (hereinafter referred to as first coordinate), A perpendicular line drawn from the first coordinate to a straight line connecting the second and third coordinates is longer than a predetermined length, or a vector is directed from the third coordinate to the first coordinate;
When the angle formed by the vector from the first coordinate to the second coordinate is larger than a predetermined angle, the first coordinate is set as a candidate feature point, and each coordinate in the coordinate string is sequentially Find all the candidate feature points while selecting the first coordinates, and among the candidate feature points, a central candidate feature point among a series of candidate feature points that are sequentially adjacent to each other along the coordinate string;
and a graphic feature extraction method characterized in that a candidate feature point that does not have an adjacent candidate feature point along the coordinate string is set as a regular feature point.