JP2000132692A - Method for extracting feature point of curve and recording medium recording the method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make efficiently executable an approximation in a curve approximation such as spline interpolation by detecting the appearance of feature points by means of the increase/decrease of the intersections of a search line with the curve and detecting the feature point from the positions of the intersections of the search lines before and after the detected search line with the curve. SOLUTION: Search is executed by successively shifting the search lines from L=1 in order to extract the point of a projecting part in the curve which is made into a picture and the point which is not smooth on the curve as the feature points (S1 and S10), the number of intersections of the search line with the curve and their positions are detected (S4-S6), the increased/decreased intersections are extracted from the relation of the intersection positions of the curve with the before and after search lines (S8) concerning the search line where the number of intersections are increased/decreased (S7) and the extracted intersections are made to be the feature points (S9).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for extracting characteristic points of a curve as preprocessing for approximating experimental data and contour lines of an image with a smooth curve.

[0002]

2. Description of the Related Art A method of approximating a curve by a function is, for example, as follows.
As described in the separate volume “Mathematics in the Internet Age”, Part 3, Chapter 5, Spline (by Fujiichi Yoshimoto),
There are methods using plane interpolation and B-Spline.

[0003]

When approximating a curve by Spline interpolation or B-Spline, nodes are determined in the curve, and Spline interpolation is performed between the nodes. It is efficient to collect nodes at convex portions of the curve, which are characteristic points of the curve, and at portions where the curve is not smooth. However, it has been difficult to extract characteristic points of a curve that can be nodes.

As a method of extracting the contour line of an image, the image is enlarged to four neighbors (front, rear, right and left, and eight neighbors including oblique) with each pixel as a center, and pixels within the enlarged area that are not included in the extraction area are extracted. The applicant of the present application filed a method on the same day as a line pixel, and sequentially searching for the contour pixels by searching the vicinity thereof by raster scan with reference to the contour pixels. According to this method, a method of approximating the extracted contour line pixels by Spline interpolation or the like can be considered. However, even if a plurality of contour lines are extracted, they are mistakenly approximated as one contour line. There is fear.

An object of the present invention is to solve the above problems,
An object of the present invention is to provide a method for efficiently extracting characteristic points of a curve and a recording medium on which the method is recorded.

[0006]

SUMMARY OF THE INVENTION According to the present invention, a search is performed on a line-by-line basis to extract a convex point or a non-smooth point on a curve, which is a characteristic point of a curve in an image. The appearance of a feature point is detected by increasing or decreasing the intersection with the curve, and the feature point is detected from the position of the intersection of the curve with the search line before and after the detected search line. Features media.

[0007] This is a method for extracting a point of a convex portion of an imaged curve or a point that is not smooth on the curve as a feature point. The image is searched line by line, and the search line and the curve are compared. The number of intersections and their positions are detected, and for search lines in which the number of intersections has increased or decreased, intersections that have increased or decreased have been extracted from the relationship between the intersection positions of the curves in the preceding and following search lines, and the extracted intersections have been identified as feature points. And extracting a characteristic point of a curve.

[0008] In addition, the method of the present invention is characterized in that the search line is set to one of vertical, horizontal and diagonal search directions.

[0009] A method for extracting characteristic points of a curve, wherein the search line has an arbitrary line width and search interval.

[0010] Also, as a program for causing a computer to execute the processing procedure in the method for extracting characteristic points of a curve, a recording medium which records the characteristic point extraction method of a curve, which is readable by the computer.

Further, according to the present invention, when a contour line of a curve is detected as a connection of ordered pixels, a start point and an end point of the ordering are extracted as feature points, and the ordered contour line is further extracted. When a plurality of are extracted, the distance D between each pixel is calculated according to the order, and two points larger than a certain threshold value P are assigned end point and start point signs and detected as feature points. A vector is generated from a pixel that is Nth pixels before and after a certain pixel, and when the angle and magnitude of the generated vector are larger than a certain threshold value, the pixel is extracted as a feature point. Features.

In a curve where the contour of the imaged curve is detected as a sequence of ordered pixels,
A feature point extracting method for a curve, wherein a starting point and an ending point of ordering are extracted as feature points.

In the curve, a distance D between pixels is calculated in the order described above, and two points at which the distance D is greater than a certain threshold value P are extracted as characteristic points. Extraction method.

Further, for a certain pixel A on the curve having the ordered pixels, N
The BA pixel and the AC vector between the pixels A, B, and C are generated by setting the first pixel as B and the Nth pixel in the already assigned order as C, and starting the BA vector up to the pixel A. The translated vector is defined as BA ′, and the BA ′
A feature point extraction method for a curve, comprising calculating an angle θ between a vector and the AC vector, and extracting the pixel A as a feature point when the angle θ is larger than a certain threshold Q.

Further, a B′C vector is generated by subtracting a BA ′ vector from the AC vector, and a pixel A is extracted as a feature point when the size X of the B′C vector is larger than a threshold value R. A feature point extraction method for a characteristic curve.

Further, a recording medium which records a method for extracting characteristic points of a curve, characterized in that it is recorded as a program readable by the computer as a program for causing a computer to execute the processing procedure in the method for extracting characteristic points of a curve.

[0017]

DESCRIPTION OF THE PREFERRED EMBODIMENTS (First Embodiment) FIG. 1 shows a feature point extraction algorithm according to an embodiment of the present invention, which is a method for extracting a feature point from an intersection between a search line and a curve.

The characteristic points of the curve as shown in FIG. 2 are indicated by circles in FIG. FIG. 3 is an enlarged view of a connected portion of pixels forming a curve including such a feature point. The search line is one line as shown in FIG. 4, and the search line is moved one line at a time in the search direction. In the present embodiment, a new intersection with the curve or the last intersection point is extracted as a feature point in the search using the search line. Less than,
The procedure of FIG. 1 will be described in detail.

In FIG. 1, the number of search steps L of a search line is L =
1 is initialized (step S1). This search stage number L
Is sequentially increased from the initial value 1 every time the search line is moved by one line (step S10).

In moving the search line, it is checked whether or not the current search line is within the search range (step S2), and when the current search line exceeds the search range, the search is terminated (step S3).

When the current search line is within the search range, it is checked whether or not the pixels within the search line include pixels forming a curve by moving the search line (step S).
4). That is, it is checked whether or not the search line crosses a part of the curve. If there is no intersection with the curve in this check, the number of search steps L is incremented by 1, that is, the search line is shifted by one step (step S10).

When a pixel in the search line and a pixel on the curve intersect, the number S (L) of intersections between the search line and the curve is calculated. At this time, an intersection where pixels are continuously detected on the search line is regarded as one point, such as an intersection area indicated by a circle in FIG. 5 (step S5). The number S of this intersection
Following the calculation of (L), the position T (L) of the intersection is detected (step S6).

Next, it is checked whether or not the number of intersections S (L) obtained in step S5 matches the number of intersections S (L-1) one line before (step S7). If there is a match in this check, that is, if the number of intersections is the same, the search line is shifted by one step without extracting this intersection as a feature point (step S10).

When the number of intersections increases or decreases, the current position T (L) of the intersection and the intersection position T (L-
The relationship between the intersections in 1) is checked, pixels that are not related are detected (step S8), and the detected pixels are extracted as feature points (step S9), and step S10 is performed.
And the search is continued until the search line has searched all over the image.

The above method will be described using a specific example. At the search line position in FIG. 6, the number of intersections increases from 0 to 1, and in FIG. 7, the number of intersections increases from 1 to 2. Therefore, the detection pixels in FIG. 6 are extracted as feature points.

In the case of FIG. 8, the number of intersections increases from 2 to 3, and in FIG. 9, the number of intersections increases from 3 to 4. Therefore, the pixel at the central intersection in FIG. 8 is extracted as a feature point.

Determining a feature point from a plurality of intersections is as follows.
It is determined from the relationship between intersections on the preceding and following search lines. For example, the distance between the pixel of the previous search line and the pixel of the current search line is calculated, and the closest one is associated. Then, a pixel that is not associated is determined as a feature point.

In this embodiment, a vertical or horizontal search line is used as the search direction. However, as shown in FIG. 10, the width of the search line may be increased (doubled in the figure). As shown in FIG. 11, a diagonal search line may be used as the search direction, or the number of pixels (search interval) that the search line advances may be arbitrarily set.

(Second Embodiment) In this embodiment, in a case where the contour of an imaged curve is detected as a connection of ordered pixels, spline interpolation or the like is performed on the ordered curve. , Among the pixels that make up the curve, the first and last pixels are important for interpolation,
The first and last two pixels are set as feature points.

In this embodiment, when a plurality of curves are approximated by spline interpolation or the like, the division points of the curves are important. In order to extract a division point, a distance D is calculated between the I-th pixel and the (I + 1) -th pixel. If the distance D is larger than a certain threshold value P, the two points are set as division points. Then, two points are extracted as feature points.

Further, in the present embodiment, a vector is generated from a pixel which is N pixels before and after a certain pixel, and when the angle or magnitude of the generated vector is larger than a certain threshold, the pixel is extracted as a feature point.

FIG. 12 shows a contour extraction method for sequentially detecting contour pixels from pixels constituting a curve and detecting the angle of a vector connecting the pixels in which the feature points of the extracted contour are ordered. It is an algorithm that extracts from large and small.

As shown in FIG. 13, when there is a curve composed of pixels that have already been ordered in the order of the arrows, as shown in FIG. In this case, the pixel before (No. 5) is B, and the pixel after N (No. 5 in the figure) is A. Then, from the pixels A, B, and C, FIG.
Generate a BA vector and an AC vector, as in
Vector B translated from the starting point of the vector to pixel A
Generate A '. Further, as shown in FIG.
The angle θ between the vector and the AC vector is calculated, and if the angle θ is larger than a certain threshold Q, the M-th point A is extracted as a feature point. Hereinafter, the procedure of FIG. 12 will be described in detail.

In FIG. 12, the number M = 1 of the pixel from which the feature point is to be extracted is initialized (step S21). This number M is sequentially increased from the initial value 1 each time the extraction determination processing is performed for one pixel (step S30).

It is checked whether or not the value obtained by adding the numerical value N arbitrarily set to the number M is larger than the last number (step S22). If the value is larger, the feature point extraction is terminated (step S23). If the value of M + N is smaller than the last number, it is checked whether M−N <0 (step S24). These checks are performed on the pixel A as shown in FIG.
, The presence of pixels B and C is used as an extraction condition.

As another embodiment, when the contour is separated into a plurality of parts, the separation point is extracted by the above-described feature point extraction, but the contour between the start point and the end point of the separation point is extracted. Consider a line segment and extract feature points with the line segment. That is, when M + N is larger than the order of the end points of the separation points, the feature point extraction is moved to the next line segment. If M−N is smaller than the start point of the separation point, M is moved to the next point.

When the pixels B and C exist for the pixel A, the M-th pixel A, the M-N-th pixel B, and M + N
Each coordinate of the pixel C is detected (step S25).
From these coordinates, as shown in FIG. 15, a BA vector and an AC vector are generated, and a vector BA ′ in which the start point of the BA vector is translated to the pixel A is generated (step S).
26).

Thereafter, as shown in FIG. 16, the angle θ between the BA ′ vector and the AC vector is calculated (step S27),
It is checked whether or not the angle θ is larger than a certain threshold Q (step (S28). If the angle θ is larger than a certain threshold Q, the M-th point A is extracted as a feature point (step S29).

In this embodiment, the angle θ is used.
An index representing the correlation between two vectors, such as an inner product, may be used. In the present embodiment, M is increased by one, but the amount of increase is arbitrary.

As another embodiment, a feature point can be extracted from the magnitude of a vector. This method
As shown in FIG. 17, the AB ′ vector is subtracted from the AC vector to generate a B′C vector, the size X of the B′C vector is calculated. Point A is extracted as a feature point.

The algorithm for extracting a feature point from the magnitude of this vector is as shown in FIG. FIG.
The difference from the above is that steps S31 to S33 are performed instead of steps S26 to S29.

The vectors BA, AC, and BA 'are generated (step S26). Thereafter, the AB' vector is subtracted from the AC vector to generate a B'C vector, and the size X of the B'C vector is calculated. (Step (S31), this value X
It is checked whether or not is larger than a certain threshold R (step (S32)). If it is larger, the M-th pixel A is extracted as a feature point (step S33).

According to the present invention, data of a curve to be approximated is stored, and the processing procedure or algorithm of the feature point extracting method of the present invention shown in the flowcharts of FIGS. And a computer-readable recording medium such as a floppy disk, a memory card, an MO,
It can be recorded on a CD or DVD and distributed.

[0044]

As described above, according to the present invention, the appearance of a feature point is detected by increasing or decreasing the intersection between the search line and the curve.
Since the feature point is detected from the position of the intersection between the search line before and after the detected search line and the curve, the Sp
In order to perform curve approximation such as line interpolation, approximation can be performed efficiently.

Further, according to the present invention, when a contour of a curve is detected as a connection of ordered pixels, a start point and an end point of ordering are extracted as feature points, and further, the ordered contour is extracted. When a plurality of are extracted, the distance D between each pixel is calculated according to the order, and two points larger than a certain threshold value P are assigned end point and start point signs and detected as feature points. Since a vector is generated from a pixel which is Nth anteroposterior from a certain pixel and the generated vector is extracted as a feature point when the angle or magnitude of the generated vector is larger than a certain threshold value, curve approximation such as Spline interpolation is performed. In addition, the approximation can be performed efficiently.

[Brief description of the drawings]

FIG. 1 is a feature point extraction algorithm using a search line according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram of characteristic points of a curve.

FIG. 3 is an enlarged view of a curve at a pixel level.

FIG. 4 is an example of a search line and a search direction.

FIG. 5 is an explanatory diagram of an intersection area between a search line and a curve.

FIG. 6 is an example of an intersection between a search line and a curve.

FIG. 7 is an example of an intersection between a search line and a curve.

FIG. 8 is an example in which a search line intersects a curve at a plurality of locations.

FIG. 9 is an example in which a search line intersects a curve at a plurality of locations.

FIG. 10 is an example in which the width of a search line is doubled.

FIG. 11 is an example in which the search direction of a search line is oblique.

FIG. 12 shows a feature point extraction algorithm based on a vector angle according to the embodiment of the present invention.

FIG. 13 is an example of a curve as a connection of ordered pixels.

FIG. 14 is a positional relationship diagram of pixels A, B, and C for explaining feature point extraction based on vector angles.

FIG. 15 is a vector relation diagram between pixels A, B, and C.

FIG. 16 is an explanatory diagram of a vector angle θ of pixels A, B, and C.

FIG. 17 is an explanatory diagram of a vector size B′C of pixels A, B, and C.

FIG. 18 is a feature point extraction algorithm based on the magnitude of a vector according to the embodiment of the present invention.

Claims (9)

[Claims] 1. A method for extracting a point of a convex portion of an imaged curve or a point which is not smooth on the curve as a feature point, wherein the image is searched line by line, and the search line and the curve The number of intersections and the position of the intersection are detected, and for the search line in which the number of intersections has increased or decreased, intersections that have been increased or decreased are extracted from the relationship between the intersection positions of the curves on the preceding and following search lines, and the extracted intersections are characterized. A method for extracting characteristic points of a curve, wherein the characteristic points are points. 2. The method according to claim 1, wherein the search line is set to one of vertical, horizontal, and oblique search directions. 3. The method according to claim 1, wherein the search line has an arbitrary line width and a search interval. 4. In a curve in which the contour of an imaged curve is detected as a series of ordered pixels, a start point and an end point of the ordering are extracted as feature points. Feature point extraction method. 5. The method according to claim 4, wherein a distance D between pixels in the curve is calculated in the order, and two points at which the distance D is larger than a certain threshold value P are extracted as feature points. The feature point extraction method of the described curve. 6. For a certain pixel A on the curve having the ordered pixels, the N-th pixel in the already assigned order is set to B, and the N-th pixel in the already assigned order is N-th pixel. Let C be the BA vector between pixels A, B, and C and A
A C vector is generated, and a vector obtained by translating the start point of the BA vector to the pixel A is defined as BA ′, an angle θ between the BA ′ vector and the AC vector is calculated, and when the angle θ is larger than a certain threshold Q, The method according to claim 4, wherein the pixel A is extracted as a feature point. 7. A method for generating a B′C vector by subtracting a BA ′ vector from the AC vector, and extracting a pixel A as a feature point when the size X of the B′C vector is larger than a threshold value R. 7. The method for extracting characteristic points of a curve according to claim 6, wherein: 8. A program for causing a computer to execute a processing procedure in the method for extracting characteristic points of a curve according to claim 1, wherein the program is recorded so as to be readable by the computer. A recording medium that records a method for extracting characteristic points of a curve. 9. A program for causing a computer to execute the processing procedure in the method for extracting characteristic points of a curve according to claim 4, wherein the program is recorded so as to be readable by the computer. A recording medium that records a method for extracting characteristic points of a curve.