JPH07220076A - Image processing device - Google Patents

Abstract

PURPOSE:To provide an image processing device capable of detecting a corresponding area at high speed and accurately. CONSTITUTION:Data representing relation between a correlation value by inspection and a true correlation value is stored in program memory 16 at every step. Relation between the correlation value and a reliability section can be recognized by such data. The maximum correlation value when template matching can be obtained at certain step width is detected, and the reliability section in accordance with the maximum correlation value is detected by referring to data in the program memory 16. The template matching can be obtained by decreasing the step width of the reliability section when it overflows from the reliability section set in advance, decreasing it when the former is smaller than the latter, and with such step width when they are equal. The corresponding area with the reference image of a template can be detected from an image for inspection.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing device,
In particular, the present invention relates to an image processing apparatus used in a product inspection system that detects a region corresponding to a reference image from an inspection image by template matching and recognizes whether a product having a fixed shape is a good product or a defective product.

[0002]

2. Description of the Related Art As one of image processing techniques, an inspection image (digital image) input from an image pickup device or the like is compared with a reference image registered in advance, and a reference image is selected from the inspection images. There is a template matching method for finding the corresponding region. In this method, as shown in FIG. 5, a part of the inspection image 2 (reference image) is registered in a template 1 prepared in advance, while an arbitrary area (search box) 3 is provided in the inspection image. Then, the template 1 is scanned in the search box 3 as shown by the arrow 4, and the degree of correlation between the reference image and the inspection image is examined, and the area having the highest degree of correlation is set as the corresponding area.

Here, template matching will be described with reference to FIG. In FIG. 6, the area of the search box 3 including the inspection image is M × N, and the area of the template 1 including the reference image is Mr × Nr. And
If the step size of template 1 is decided, template 1
Scans the inside of the search box 3 with a predetermined step width.

For example, when the step width is 1 pixel at a time, and the arbitrary pixel in the template 1 is (mr, nr), the scanning range is expressed by the formula 1.

[0005]

## EQU00001 ## | i-mr | .gtoreq.Mr / 2, | j-nr | .gtoreq.Nr / 2 (.gtoreq.domain) where the central pixel of template 1 is (mc, nc), the scanning step is mc + .alpha. , Nc + α (α = 1,
2, 3, ...). Further, when the step width is 4 pixels each, the scanning step is mc + β, nc + β (β =
1, 5, 9, ...).

In the template matching, as a calculation formula for obtaining the correlation value between the inspection image and the reference image in the template 1, for example, a calculation formula using a correlation function shown in Formula 2 is used.

[0007]

[Equation 2]

The correlation value of the equation 2 is -1≤correlation value≤1.
Takes the value of. In the above-mentioned method, the inspection image and the reference image are statistically examined, and even if the input inspection image is somewhat unstable such as changing due to disturbance, etc. You can ask.

[0009]

However, this method requires a long calculation time. Therefore, in order to calculate at high speed, a method as shown in FIG. 7 is generally adopted. First,
In step Sa shown in FIG. 7, a step width is set as an initial setting. Then, in step Sb, as shown in FIG.
As shown in (A), the correlation value is determined by template matching in which the template 1 is scanned within the search box 3 for the inspection image at a rough interval, that is, the template 1 is scanned with a certain step width instead of being moved pixel by pixel. Is calculated, and the region having the maximum correlation value is detected. If the template 1 has a size of 64 × 64 pixels, for example, if the step width is set to 32 pixels, the template 1 scans as shown in FIG. 8A. Then, if the detected area is, for example, the area 5 shown in FIG. 8B, a small (for example, 100 × 200 pixels) search box 6 is provided around the area 5.
The relationship between the position in the search box 6 and the correlation value obtained by the inspection is as shown in FIG. As can be seen from FIG. 9, the correlation value of area 5 is the maximum correlation value. And FIG.
In step Sc, the search box 6 is scanned by the template 1 with a step width for each pixel to check template matching, and an accurate corresponding area is detected in step Sd.

However, in this method, the accuracy of the calculation result of the template matching having a large step width performed first in step Sb finally affects the accuracy of the template matching, so that the step width of the first template matching is set. It could not be increased so much that the calculation time could not be shortened. Further, in the template matching process of step Sb, an accurate correlation value cannot be obtained due to the influence of a change in the input image due to a change with time of an illumination component attached to the image pickup device or a disturbance, and as a result, an accurate corresponding area is detected. There were things I couldn't do. As a result, there are cases where the product is judged to be defective even though it is not a defective product.

Therefore, a main object of the present invention is to provide an image processing apparatus capable of reducing the calculation time and detecting an accurate corresponding area.

[0012]

According to the present invention, a template including a reference image is moved on the inspection image by a predetermined step width and template matching is performed to check template matching among the inspection images. The first means for detecting a large area, and the template with the step width smaller than the step width in the first means move the template over the area to check the template matching to find the area having the largest correlation value with the reference image. In an image processing apparatus including a second means as an area corresponding to a reference image, a memory for storing data showing a relationship between a correlation value and a reliability interval corresponding to each step width, a step width in the first means Means for presetting a desired confidence interval, and a confidence interval corresponding to the maximum correlation value in the template matching of the first means. And a preset reliability interval, and a means for changing the step width when the reliability interval corresponding to the maximum correlation value is out of the preset reliability interval range. It is a processing device.

[0013]

The data indicating the relationship between the correlation value obtained experimentally or empirically and the corresponding reliability interval is stored in advance in the memory. Then, a relatively large step width in the first means and a desired reliability interval are set in advance. Template matching is performed by the first means using a preset step width, and the reliability interval corresponding to the maximum correlation value at that time is obtained by referring to the data in the memory. If the reliability interval corresponding to this maximum correlation value is out of the preset desired reliability interval, the step size is reduced and the template matching is performed again by the first means. This process is repeated until the reliability interval corresponding to the maximum correlation value is equal to or within the preset reliability interval, and the step width is determined. Then, if the reliability interval corresponding to the maximum correlation value is equal to the preset reliability interval, then the first means is used for rough template matching, and then the second means is used for detailed template matching to obtain a desired correspondence. Get the area. If the reliability interval corresponding to the maximum correlation value is within the preset reliability interval, the step width is increased, and then the template matching is performed by the first means and the second means to determine the desired corresponding area. To get

[0014]

According to the present invention, when there is an unstable element such as a disturbance or a temporal change of illumination and the reliability interval corresponding to the maximum correlation value is out of the preset reliability interval, Reduce the step width. On the other hand, when it is stable and the reliability interval corresponding to the maximum correlation value is within the preset reliability interval, the step width is increased.
In this way, the step width can be obtained dynamically. Therefore, even if the step width of template matching is made large at first, it can be set again to the optimum step width, so that it can be calculated at high speed without problems, and it is accurate without being affected by instability factors such as disturbance or aging of illumination. The corresponding area can be detected. Therefore, when the present invention is applied to the product inspection system, it is possible to more accurately judge whether the product is a good product or a defective product.

The above-mentioned objects, other objects, features and advantages of the present invention will become more apparent from the detailed description of the following embodiments made with reference to the drawings.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Referring to FIG. 1, an image processing apparatus 10 of this embodiment is used, for example, in a product inspection system and has a CPU.
Including 12. The CPU 12 has a program memory 16, a RAM 18, an image memory 20, and A / A via a bus 14.
The D conversion circuit 22 and the I / O interface 24 are connected. The program memory 16 stores a program for controlling the operation of the image processing apparatus 10 and the data of the graph shown in FIG. 2, and the RAM 18 stores variables of the program, data for performing calculations, calculation results, and the like. It The I / O interface 24 has a lighting device 2
6 is connected, and the imaging device 28 is connected to the A / D conversion circuit 22. Then, the inspection object 30 is arranged between the image pickup device 28 and the illumination device 26, the inspection object 30 is photographed by the image pickup device 28, and the video signal is inspected via the A / D conversion circuit 22. The image data of the image is stored in the image memory 20. In addition to the above, the image memory 20 also stores a template including a reference image for detecting a corresponding region from the inspection image, and the like.

The data of the graph shown in FIG. 2 is for estimating the correlation value of the population in a section, and the correlation value when template matching is actually observed on the horizontal axis (hereinafter, simply referred to as "correlation value by inspection"). And the vertical axis represents the correlation value obtained by experiment or experience (hereinafter simply referred to as “true correlation value”). Such data is obtained for each step width, for example, step width 1 to step width 5, as shown in FIGS. For example, the step widths 1, 2, 3, 4 and 5 are each 1
It has step widths of pixels, 4 pixels, 8 pixels, 16 pixels, and 32 pixels. Note that FIG.
Within the range (α) surrounded by the curves (A) to (E), a predetermined probability (for example, 95% in this embodiment) of all possible values of the true correlation value with respect to the correlation value by inspection. Includes true correlation values in the range. The range (α) of the true correlation value with respect to the correlation value obtained by the inspection is called the confidence interval. Therefore, the relationship between the correlation value by inspection and the reliability interval can be known from the data of the graph shown in FIG. For example, a graph of step width 2 is shown in FIG. 2B. When the correlation value by inspection is 0.85, the true correlation value is 95%.
The probability is 0.50 to 0.96.

The operation of such an image processing apparatus will be described with reference to FIG. First, in step S1, initial setting is performed. As initial setting, first (step S3)
The step width of the template and the desired reliability interval in the template matching executed in step 1 are set respectively. Next, in step S3, template matching is performed using the set step size to find the maximum correlation value, and the graph data stored in the program memory 16 is referenced to refer to the reliability interval corresponding to the maximum correlation value. To detect. In this embodiment, the inspection object 3
The inspection image of 0 is composed of, for example, 512 × 512 pixels, and the template is composed of, for example, 64 × 64 pixels. Next, in step S5, the reliability interval is compared with the reliability interval initially set in step S1. If the reliability interval of the maximum correlation value is out of the initially set reliability interval due to the influence of disturbance, the step width is reduced in step S7, and template matching is performed again in step S3. And step S
In step 5, the process is repeated until the reliability interval of the maximum correlation value falls within the initially set reliability interval or becomes equal to the initially set reliability interval. In step S5, when the reliability interval of the maximum correlation value falls within the initially set reliability interval, the step width is increased in step S9 and the process proceeds to step S11. If the two reliability intervals are equal, the process also goes to step S11.

Thereafter, processing is performed by two-step template matching as in the above-mentioned conventional method. That is, in step S11, rough template matching is performed using the step width set in the above process to detect the corresponding area having the maximum correlation value. Then, in step S13, a small search box is set with the corresponding area as the center, and template matching with a step width of one pixel is checked in the search box. Note that the search box is, for example, 100 vertical × 20 horizontal.
It consists of 0 pixels. In step S15, the corresponding area having the maximum correlation value in the template matching in step S13 is detected, and the corresponding area becomes the correct corresponding area to be detected, and the process ends. Then, by finding the corresponding area in this way, the position of the target inspection object 30 can be known, and the determination of the non-defective product / defective product in the product inspection system becomes more accurate.

A detailed description will be given with reference to FIG.
As shown in FIG. 4A, the correlation value obtained by the inspection is 0.70.
, The reliability interval A of the step width 4 is 0.06 to 0.
89, the reliability interval C of step width 3 is 0.28 to 0.8
6 and the desired initial confidence interval B is set to 0.10.
˜0.90 (width is 0.80). First, as a result of performing template matching by stepping 16 pixels with a step width of 4, if the maximum correlation value by inspection is 0.70, the reliability interval is A and takes a value of 0.06 to 0.89. Since this range is out of the range of the reliability interval B, the template matching in the step width 4 has low reliability. Therefore, the step width is set to the step width 3
When the template matching is performed by stepping by 8 pixels and the maximum correlation value is 0.7, the reliability interval is C and is 0.28 to 0.86, and the reliability interval B can be sufficiently secured. Increases the reliability of template matching. Therefore, template matching is performed using the step width 3 that satisfies the reliability interval B in this way.

Further, as can be seen by comparing the reliability interval B and the reliability interval C, the width of the reliability interval B still has a margin, so that the condition of being within the reliability interval B is smaller in correlation. Even a value can be met. Since such a low correlation value can be determined, reliable template matching can be seen regardless of some noise. For example, by setting the step width to 3, it is possible to satisfy the condition of being within the confidence interval B even when the correlation value is 0.67, and to accurately see template matching.

As shown in FIG. 4 (B), if the desired reliability interval is 0.05 to 0.90 as shown in B ', when the correlation value by inspection is 0.70, the step width is 4 , The condition of being within the confidence interval B ′ is satisfied. However, if the correlation value obtained by the inspection becomes 0.60 due to, for example, a change with time of the illumination component or some disturbance, the reliability interval will take a value of -0.26 to 0.85, and the reliability interval B '. This causes the template matching to be less reliable. Therefore, when the step width is changed to the step width 3 (stepping by 8 pixels at a time), the reliability interval becomes 0.10 to 0.83 even if the correlation value obtained by the inspection becomes 0.60, and the reliability interval B'is Satisfy the condition. Therefore, template matching can be seen with high reliability. In this way, since the step width can be dynamically changed, it is possible to stably detect the corresponding area even if the correlation value due to the inspection becomes small. Further, even if the step width is dynamically changed, it is not necessary to change the equation of the normalized correlation function suitable for image inspection, that is, the equation 2 itself, and therefore the practical value is high.

When shading occurs due to the influence of disturbance and uneven illumination occurs, the correlation value obtained by inspection becomes small, and when the illumination becomes dark or bright, the correlation value obtained by inspection becomes small.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a graph showing data for interval estimation of a correlation value of a population.

FIG. 3 is a flowchart showing the operation of this embodiment.

FIG. 4 is a graph showing a relationship between a correlation value and a confidence interval by inspection in the case of step widths 3 and 4.

FIG. 5 is an illustrative view showing a registration image and an inspection image for viewing template matching.

FIG. 6 is an illustrative view for explaining template matching.

FIG. 7 is a flowchart showing the operation of the conventional technique.

FIG. 8 is an illustrative view for explaining a case where template matching is viewed in two stages.

FIG. 9 is a graph showing the relationship between the position of the search box and the correlation value by inspection.

[Explanation of symbols]

 10 ... Image processing device 12 ... CPU 16 ... Program memory 18 ... RAM 20 ... Image memory 26 ... Illumination device 28 ... Imaging device 30 ... Inspection object

Claims (1)

[Claims] 1. A method for detecting a region having a large correlation value with the reference image in the inspection image by moving a template including the reference image with a predetermined step width on the inspection image and checking template matching. 1 means, and by moving the template in the step width smaller than the step width in the first means and observing template matching, the area having the largest correlation value with the reference image is selected from the areas. In an image processing apparatus including a second means as an area corresponding to a reference image, a memory for storing data indicating a relationship between the correlation value and a reliability interval corresponding to the step width, the first means. Means for presetting a step width and a desired reliability interval, maximum correlation value when template matching of the first means is observed Means for comparing a corresponding reliability interval with the preset reliability interval, and the step when the reliability interval corresponding to the maximum correlation value is out of the range of the preset reliability interval. An image processing apparatus comprising means for changing the width.