JPH03261803A - Detection of edge position - Google Patents

Abstract

PURPOSE:To achieve instrumentation with high accuracy by determining the position of an edge from values of means for performing first order differentiation of image data and the position of the center of gravity in the range including the edge of the data subjected to the first order differentiation. CONSTITUTION:A linear measuring line 16 is set at an optional position of an image 15 from a video camera 14. By measuring edges that cross the line 16 various kinds of size of an object is detected. Supposing that the data of luminance on the measuring line is expressed by f(x), the difference according to formula I is regarded to represent the first order differentiation. If f'(x) satisfies L>¦f'(x)¦ for a level L determined by the result of the difference calculation, it is neglected regarded as a noise. A continuous part with the same sign exceeding the level L is within the range of formula I to formula III. The cumulative distribution of the first order derivative is obtained for each range, and only the range not smaller than a certain threshold value, it is judged that there is included an edge. If this condition is satisfied only for the range C, a position Mo where Y/2 of the cumulative distribution Y of the primary differential values in the range C is given is obtained, and is represented by a formula V from a formula IV. The coordinates of the center of gravity in the range are expressed by a formula VI from which a formula VII is held.

Description

DETAILED DESCRIPTION OF THE INVENTION "Field of Industrial Application" The present invention relates to an automatic edge position measurement technique in a digital image data processing apparatus.

"Conventional Technology" Automatic inspection and measurement using digital image data processing equipment is an important production technology. There are many data processing techniques used in digital image data processing devices, but detecting edge positions can be said to be one of the most important.

For example, edge position detection is the basis of digital image measurement, such as measuring the width of an object using the distance between edges or measuring the inclination of an object using two edges.

Conventionally, as an edge position detection method, a threshold value 7 is set as shown in Fig. 2, and the output value of a pixel is divided into whether it is larger or smaller than the threshold value (hereinafter referred to as binarization). A common method was to regard boundary 8 as the edge of the object.

``The problem that the invention seeks to solve However, this method had the following problems.

(1) A large error 9 occurs when the amount of light irradiated onto the target object changes due to aging deterioration of the illuminator or the like. (Figure 3) (2) The determination of the threshold value is delicate and troublesome.

Further, when edges with different brightness values coexist as shown in FIG. 4, it is necessary to have two threshold values 1 and 12 and perform measurement twice.

(3) The detected edges are ±17 as shown in Figure 5.
It has an error 13 of 2 pixel spacing, that is, accuracy less than the pixel spacing cannot be expected.

The present invention aims to solve the problems of the prior art as described above.

"Means for Solving the Problem" Therefore, the present invention adopts a method of detecting the peak of the first-order differential of image data in order to detect edge positions without binarizing. This is because the position where the brightness change is the steepest is generally considered to be the edge. Peak detection is performed as follows. First, first-order differentiation is applied to the image data, and the position (denoted as M) gives the center of gravity (denoted as G) within a range that sufficiently includes the edges in the data subjected to the first-order differentiation.
In the cumulative distribution of first-order differential data in this range, find the position (M) that gives 1/2 of the final cumulative value, and from the found values M, M, find the position (let it be M) that gives the maximum value of the first-order differential. Estimate M.

In addition, the data range for determining M and M is determined as continuous data whose first-order differential values have the same sign and whose absolute value exceeds a certain threshold. The presence or absence of an edge within the specified range is determined on the condition that the final cumulative value exceeds a certain threshold.

"Effect" In the methods described above, the difference in the amount of light to the object does not directly affect the measurement. Furthermore, the threshold values set for determining the range and the presence or absence of edges may be rough, and there is no need to change them for each object.

Furthermore, since it is based on differential processing, there is no problem even if edges with different brightness values coexist. The biggest advantage is that edge detection can be performed with accuracy less than the pixel interval.

"Example" Hereinafter, an example of the present invention shown in the drawings will be described. 6th
The figure shows a linear measurement line 16 at an arbitrary position with respect to an image 15 from a video camera 14 that can obtain a two-dimensional image.
This is an example of a device that measures various dimensions of an object by setting the measurement line and detecting edges that cross the measurement line. Assume that the luminance on a certain measurement line is as shown in FIG.

Assuming that this data is represented by f (x), f-(x>=f
(x+1>-f (x-1>) is the first-order differential.

FIG. 8 shows the results obtained using this formula.

If the level indicated by the broken line 4 in FIG. 8 is L, then L>l
f - (x) that satisfies f'' (x)I is ignored as noise. Extracting continuous parts with the same sign beyond L, A = (x l x = 106) B = (x l x = 108) There are three ranges: C = (xl 111≦X≦117).Find the cumulative distribution of the first-order differential value for each range, and calculate the final cumulative value 5.f-(x>・Σf-(x )
・Σf−(x)・Below a certain threshold x(B
It is determined that an edge is included only in the range that is on This is the cumulative distribution of the derivative values.

Assuming Y=11°(X), find the position M that gives Y/2x(C) in the cumulative distribution.In this case, M is the cumulative distribution expressed by the function F(x>), then F(xo)≦ Y/
Calculate as Y/2F(Xo) using Xo such that 2 < F (Xo +1 >). Also, calculate the position of the center of gravity in the range.

Position M that gives the peak of the first-order differential value. is obtained from M-MO=3 (IVI-M, ) using Pearson's empirical formula, which is said to apply well when the frequency distribution graph is close to the symmetrical shape.

Incidentally, when the measurement line 16 is oblique to the pixel array as shown in FIG. 10, two-dimensional first-order differentiation is performed. 10th
The pixels crossed by the measurement line (Pl, P2
, ......, PN ), and when the square pixel's edge is 1, and the length of the measurement line that crosses that pixel is N (P, ), each pixel is For the first derivative of , 1! By adding weights of (P,), calculations can be made in the same way as when measuring along the pixel array.

"Effects of the Invention" According to the edge position detection method of the present invention, problems associated with threshold determination, which were problems with detection methods based on binarization, can be solved.
Problems such as tediousness, error generation due to light intensity fluctuations, and accuracy limited by pixel spacing can be solved all at once.

As a result, inspection and measurement using the digital image data processing device will become more accurate and reliable, and it is believed that this will greatly contribute to improving productivity.

[Brief explanation of drawings]

FIG. 1 is a diagram showing the principle of this invention, in which (a) shows the original image data, (b) shows the data after first differentiation, and (C) shows the data after first differentiation.
represents the cumulative distribution. FIG. 2 is a diagram for explaining the conventional method, where (a) shows the original image data, and (b) shows the 2
Represents data after value processing. FIG. 3 is a diagram showing errors caused by variations in light amount, FIG. 4 is a diagram of data with two-stage edges, and FIG. 5 is a diagram showing the accuracy of binarization processing. Fig. 6 is a diagram showing an embodiment of the present invention, Fig. 7 is a diagram showing digitized original image data, Fig. 8 is a diagram showing data after first differentiation, and Fig. 9 is a diagram showing cumulative distribution. It is. Figure ↑0 is a diagram showing measurement lines set diagonally with respect to the pixel array. ↓・・・Positional coordinates of the center of gravity, 2・・・Positional coordinates that give the median of the cumulative distribution, 3
...Edge position, 44-...Threshold straightness for noise cutting,
5...Range including edges, 6...Final cumulative value, 7...Threshold value for binarization, 8...
・Edge position obtained by conventional method, 9...Error due to light intensity fluctuation, 10.10-...Image data 1, 12...Threshold value, 13... ...Error range, 14...Video camera, 15...Image data, 16...Measurement line

Claims (1)

[Claims] 1. A method for detecting an edge position of an object included in one-dimensional digital image data, comprising means for performing first-order differentiation of image data; and means for performing first-order differentiation of image data. means for determining a position 1 giving the center of gravity in a range 5 that sufficiently includes edges in the data, and a position 2 giving 1/2 of the final cumulative value 6 in the cumulative distribution of first-order differential data in the range; An edge position detection method characterized in that an edge position 3 is determined using a value obtained by determining an edge position 3. 2. The range is determined as continuous data whose primary differential values have the same sign and whose absolute value exceeds a certain threshold value 4, and the presence or absence of edges within the range thus determined is as follows: The edge position detection method according to the first item, characterized in that the determination is made on the condition that the final cumulative value exceeds a certain threshold value.