JPH0310107A - Inspecting method utilizing gradation pattern matching - Google Patents

Abstract

PURPOSE:To facilitate high speed processing which can cope with the correction and change for the contents of inspection readily and to make it possible to perform accu rate judgment even if the density difference is not clear by judging the shape of an object by comparing correlation coefficient with a lower limit value. CONSTITUTION:At first, the density images of a good object and an object under inspection are picked up with a camera 2. The density values of many picture elements constituting the images are detected and written into memories 3a and 3b. One or more reference images are determined in the gradation images of the good object. The object image corresponding to the reference image is determined in the gradation images of the object under inspection 1. Then, a correlation coefficient is obtained from the density values of the corresponding picture elements in the object image and the reference image with a controller 3. The degree of agreement is judged based on the correlation coefficient. At this time, the degree of agreement is judged not simply the number of picture element of the object image and the reference image from, but from the correlation of the density values of the picture elements. Therefore, said degree of agreement represents the degree of agreement of the shapes. When the judgment is performed in this way, the correction and the change in contents of inspection can be readily processed.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an inspection method using gray pattern matching for inspecting the appearance of a product by image processing.

[Conventional technology and its issues]

Conventionally, a feature extraction method is generally used when performing an external appearance inspection of a product using image processing. This method captures and recognizes partial characteristics, but since it is necessary to decide on a discrimination algorithm for each visual inspection item, it takes a lot of time to decide on the algorithm, and it is difficult to change the inspection contents. There are also drawbacks that are difficult to address.

Furthermore, a pattern matching method using binarized image processing is known. In this method, pattern matching is performed on a binarized target image using nxm binarized reference images. Since it has to be detected, it takes time to process, and if the difference in density between images is not clear, there is a problem that the degree of matching will vary.

There is also a small area area determination method as an inspection method using another binarized image processing method. In this method, a plurality of rectangular target images are defined in the target gray image, and a base (1λ range) of the number of pixels of the target image is set, and the number of pixels of either a or light that the target image occupies is This method judges pass/fail based on whether the numerical value is within the standard range.In this case, the advantage is that the judgment is easy, but since it is a simple area judgment, the area may be affected by unevenness in the shape. If there is no difference, erroneous determination is likely to occur.Furthermore, if the density difference in the image is not clear, there is also a problem that the area difference appears too large.

As mentioned above, the inspection method using the binarized image processing method has four
Since only two types of discrimination are possible, it is difficult to distinguish minute defects, and it is unsatisfactory for visual inspection.

Therefore, an object of the present invention is to provide an inspection method using a-light pattern matching that can easily accommodate corrections and changes in inspection contents, facilitates high-speed processing, and allows accurate judgment even when density differences are not clear. It is about providing.

[Means to solve the problem]

” - In order to achieve the above object, the method of the present invention includes the steps of dividing a grayscale image of a good object into a large number of pixels and detecting the density value of each pixel, and is divided into the same number of pixels as the a-image of the non-defective object, and the density value of each pixel is detected.
a step of determining a rectangular reference image range of more than 100 points, and a step of determining a rectangular target image range of approximately 1L in the grayscale image of the inspection object, which corresponds to each reference image and has the same range as the temporary reference image; The process of calculating the correlation coefficient based on the density value of each pixel corresponding to each base image and the target image, and the process of setting the lower limit value of the correlation coefficient for each image range. This method includes the step of determining the degree of coincidence between each reference image and the target image by 1'11 by comparing the correlation coefficient with a lower limit value.

(Operation) That is, the present invention achieves the above object by combining the /!:4i43 pattern matching method and the small area area determination method.

First, gray scale images of the non-defective object and the inspection target are photographed using a camera or the like, and the density values of a large number of pixels constituting these four gray scale images are detected and written into a memory or the like. Next, one or more reference images are determined in the black/* image of the non-defective object, and a target image corresponding to the reference image in the tP5 light image of the inspection object is determined.

Then, a correlation coefficient is determined from the density value of each corresponding pixel of the target image and the base (1 East image), and the degree of matching is determined based on the correlation coefficient. Here, the difference from the conventional small area area determination method is Rather than simply identifying political changes based on the number of gray pixels between the target image and the base image, the degree of coincidence is determined based on the correlation between the density values of each pixel. The degree is not the degree of coincidence of the areas of the a-image images,
It represents the degree of conformity of shapes.

If you judge it like this? There is no need to determine a specific algorithm for each object, and it is possible to easily modify or change the inspection contents. Furthermore, since there is no need to determine the degree of coincidence between pixels one by one, high-speed processing is facilitated. Furthermore, even if the difference in density between images is not clear, it is determined based on the correlation of density values, so even minute defects can be determined precisely.

〔Example〕

The drawings show a specific example in which the inspection method according to the present invention is applied to the appearance inspection of a product.

In FIG. 1, l represents an object such as an electronic component, and an imager 2 such as a CCD camera is installed to photograph this object 1. A lens 2a and an image element 2b are built inside the imager 2, and the reflected light from the object 1 is reflected by the lens 2.
The light is focused at point a and imaged by image sensor 2b. Image sensor 2
For example, b is composed of 300,000 matrix elements,
Each element also outputs an electrical signal depending on the amount of light it receives, and this electrical signal is input to the controller 3. The controller 3 is made up of, for example, a microcomputer, a Birdronok circuit, etc., and has built-in memories 3a and 3b. An input bag W4 and a monitor 5 are connected to the controller 3, which reads a signal from the imager 2 in response to a start signal from the input device 4, converts this signal into a density value, and stores it in a predetermined memory 3a, 3b. Write. The human-powered device 4 is also capable of outputting an image range designation signal, and the operator can arbitrarily designate a reference image range or a target image range in the grayscale image of the object 1 while viewing the image on the monitor 5. The controller 3 compares and calculates the density value of the reference image specified by the human-powered device 4 and the density value of the target image, and outputs a quality determination signal indicating whether the target object to be inspected is a good product. .

Next, the operation of the visual inspection apparatus having the above configuration will be explained with reference to FIG.

First, photograph a good object (Step 1)
0). FIG. 3 is an example of a grayscale image of a good object taken by the imager 2. The image signal input from the imager 2 is converted into a density value, and the density value corresponding to each pixel is written into the reference image memory 3a (step 11).

Next, the object to be inspected is photographed (Step! 2)
, the density value of each pixel of the grayscale image (see FIG. 4) is written into the target image memory 3b (step 13). Note that FIG. 4 shows an example in which the upper right corner of the object to be inspected is missing.

Next, one or more rectangular reference image ranges A-D are specified in characteristic parts of the grayscale image of the non-defective object (step 14). The reference images A to D are not limited to four locations, and may have different sizes or may overlap each other. FIG. 5 is an enlarged view of image A at the right corner of the specified reference images A-D, and the actual grayscale image of the object is not clearly divided into dark area AI and bright area A2. A portion A3 of intermediate brightness exists at the boundary between the dark area A1 and the bright area At. FIG. 6 shows an example of the density value of each pixel of the reference image A shown in FIG.
Although it is made up of 5×4 pixels for the sake of explanation, it is actually made up of more pixels than this. Among each pixel, dark area A,
The density value of bright area A2 is 50, the density value of bright area A2 is 150, and the density value of intermediate area A is 50.
The density value of , is 100.

Next, specify the target image range a%d corresponding to the reference image A-D in the gray scale image of the target to be inspected (
Step 15). Figure 7 shows designated target images a - d.
This is an enlarged view of a target image a in which a defect exists, and includes a dark part a1, a bright part a2, and a part a of intermediate brightness. FIG. 8 shows an example of the density value of each pixel of the target image a shown in FIG. 7, which is composed of the same <5×4 pixels as the reference image A.

Next, the correlation coefficient r between the density value of each pixel of the reference image A and the density value of each pixel of the target image a is calculated using the following formula (
Step 16).

In the above equation, T (xa, y-) is the density value of the reference image, T' Cx, , y, ) is the density value of the target image, T
is the average density of the reference image, T is the average density of the target image, and T and T can be calculated using the following formula: 〒=ΣT
(x, , y, ) / N...
(2) T゛ −ΣT (x,,y,,)/N
-(3) (However, N - number of target pixels) The correlation coefficient r takes a range of -1≦r≦1, where r=1 indicates a complete match between the target image and the reference image, and the smaller r is, the better the match is. indicates that the value is low.

On the other hand, a correlation coefficient lower limit value r aim for determining the quality of the object to be inspected is set in advance in the memory, and the correlation coefficient r calculated by equation (1) and the lower limit value r + ai1 are compared ( Step 17). This lower limit value rat can be set as appropriate depending on the number of pixels of the image, the range of the image, the detection accuracy, etc. If r≧rail, the degree of coincidence between the target image and the reference image is high, so it is assumed that the object to be inspected is a good product and the rOKJ signal is output.Step 18), r
If <r aim, the degree of coincidence between the target image and the reference image is low, so the inspection target is determined to be a defective product and an rNGJ signal is output (step 19).

In the above case, the degree of matching was determined for one location among the reference images A-D and target images a-d, which were specified at multiple locations, but if the degree of matching for images at other locations was determined in the same way, This is desirable because it allows not only the presence or absence of defects in the object to be inspected, but also the location of the defects.

In addition, in the above explanation, in order to determine the degree of individual matching between the target image and the reference image, the correlation coefficient " was calculated by (1) 2, but actually calculating the correlation coefficient r would take a lot of time. It is not possible to quickly determine whether the product is good or bad.

Therefore, the following methods 1 to 2 can be used to darken the process.

■ Calculation method that simplifies the correlation coefficient calculation procedure and makes the sign of the deviation value constant (1) In 2, T (x, , yn ) and T' (xa ,
y, ), and after obtaining the average density T and T, it is necessary to perform the product sum of the deviation values again, which makes the procedure complicated, and furthermore, the sign of the deviation value changes. Therefore, if you try to perform product-sum using hardware, the circuit will become complicated, which will hinder simplification and speeding up.

Therefore, when calculating the correlation coefficient ``with equations (1) 2 to (3), the density values of the reference image and the target image have a constant value. .

When t(x
, , y, ), t" (Xa +V++)
shall be.

t(xll, y, 1) = T(x,, y,) −to
・=(41t' (XM, ys) -T' Cx
n, y-) L6 = (5) This 1. When formula (1) is rewritten using 1, it becomes, and the correlation coefficient r can be found from this 2.

In equation (6), T(x-+4'a) and T'(X+
Since the products and sums of deviation values can be sequentially performed while taking in + +yn ) and the sign is also constant, it is possible to perform dark processing and high-speed processing.

Also, since to and the reference image are determined in advance, Σ
t (x-+3'++) and Σ[1(X, +7s
) l'' can be calculated in advance, which is even more convenient.
The reason for setting it to + is to reduce invalid data and contribute to the reduction of traffic congestion and speed-up.

■A method to simplify the calculation by cutting off the gray scale of the reference image and target image.Usually, the image gray scale is set to 256.
The maximum density is often set to 256 gradations. It is necessary to perform these 256 gradations, or 8-bit operations,
When software processing is performed, the data length for product-sum calculation becomes long, which becomes an obstacle to high-speed processing. Furthermore, when performing hardware processing, an enormous amount of hardware circuitry is required.

In order to avoid such problems, high-speed processing can be achieved by cutting off the lower density values of the reference image and the target image to reduce the IIl tone value.

First, the work of calculating the square root of correlation coefficient calculation formula (1) is omitted. In other words, N N , but is the process of opening the denominator f both software and hardware? j! 11 processing, the denominator and numerator are squared as shown in the following equation. In other words, (-Σt-t'-t-
t'n2 N N , and the value of r is -]≦r≦1, but r2 is 0≦
It is necessary to keep in mind that if r2≦1 and the correlation coefficient r=0.7, for example, then r'=0.49, which means that it takes the value of a quadratic function.

Note that although the negative correlation is unclear in equation (8), the numerator (−
The sign of the correlation coefficient can be determined by checking whether the sign of Σt-t'-t·L') is positive or negative.

■ A method to facilitate the setting of standard patterns by specifying the coordinate transformation code for each window. It is necessary to provide reference images A to D at positions a% and d of the image, respectively. However, when determining the correlation between the reference image B in Fig. 3 and the target image S in Fig. 4,
Since the reference images A and B are left-right symmetrical, the target image
By converting the coordinates, the reference image can be shared.

As the coordinate transformation patterns described above, it is sufficient to prepare four patterns as shown below.

Swapping of X coordinates Naka mode ■ Swapping of X coordinates Naka mode ■ Swapping of x and X coordinates Rumor mode ■ No conversion → mode ■ Adding the coordinate transformation modes I to ■ in this way improves processing efficiency very much.

Incidentally, 1, the correlation coefficient r is as shown in equation (6), and what is affected by coordinate transformation is ΣL (Xn +Vn)
Since only L' (Xa +Va) is involved, there is no need to change the processing for the rest regardless of coordinate transformation, and the processing procedure is simple.

〔Effect of the invention〕

As is clear from the above explanation, according to the present invention, the shape of the object can be determined by comparing the correlation coefficient with the lower limit value, so there is no need for a special algorithm for the determination, and changes in the inspection content can be made. It also becomes easier.

Furthermore, since the present invention determines the degree of shape conformity between the reference image and the target image, rather than simply the degree of conformity in area, it is possible to easily determine even if there are irregularities but no difference in area. Then, since subtle differences in shading are determined by density values,
Even if the contrast of the image is not clear, it is possible to accurately determine whether the image is good or bad.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an apparatus for carrying out the inspection method according to the present invention, FIG. 2 is a flowchart explaining its operation, FIG. 3 is a grayscale image of an example of a non-defective object, and FIG. The figure is a grayscale image diagram of an example of the object to be inspected, Figure 5 is an enlarged view of the reference image, Figure 6 is a density value diagram of the reference image, and Figure 7 is a diagram of the density value of the reference image.
The figure is an enlarged view of the target image, and FIG. 8 is a density value diagram of the target image. l...Target, 2...Camera, 3...Controller, 4...Input device, A...Reference image, a...Target image.

Claims (1)

[Scope of Claims] A step of dividing a grayscale image of a non-defective object into a large number of pixels and detecting the density value of each pixel; a step of dividing the inspection target into pixels and detecting the density value of each pixel; a step of defining one or more rectangular reference image ranges in the grayscale image of the non-defective object; a step of determining a rectangular target image range that corresponds to the image and has the same range as the reference image; a step of calculating a correlation coefficient based on the density value of each corresponding pixel between each reference image and the target image; A step of setting a lower limit value of the correlation coefficient for each range, a step of determining the degree of matching between each reference image and the target image by comparing the calculated correlation coefficient and the lower limit value, and a grayscale pattern matching that includes the following steps. Inspection method by.