JPH10123064A - Visual inspection - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform visual inspection depending on the size of a defect to be detected or the irregularities by comparing every pixel visually between a product to be inspected and an acceptable work. SOLUTION: At first, many acceptable works of same type as a product to be inspected are collected and the images thereof are picked up individually by means of a CCD camera. Average gray level and standard deviation are then calculated for the image data 1a-1n of many acceptable works subjected to multilevel gray level processing thus determining the standard image 2 and the standard deviation image 3 of the acceptable work. Subsequently, the image of the product to be inspected is picked up to be superposed on the image of the acceptable work by a mechanical method or an image processing method to obtain an inspection image 4. A differential image between the inspection image 4 and the standard image 2 is then obtained for every pixel and the ratio of the differential image to the standard deviation image 3 is also determined thus obtaining an inspection ratio image 5 indicative of distribution of the ratio. Finally, a decision value is determined in units of window defined by sectioning the inspection ratio image 5 into many predetermined number of sections of appropriate size and presence of a defect is determined by comparing the decision value with a standard decision value.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for inspecting the appearance of a product by image processing, and more particularly to a method for detecting an abnormal uneven portion to be determined as a defect on a rough surface such as a casting. It relates to an effective appearance inspection method.

[0002]

2. Description of the Related Art Many methods have been developed for detecting defects on the surface of a product by image processing. JP-A 1-1439
In step 38, the image processing device performs statistical processing of the grayscale value of the entire image from the image captured from the CCD camera for the regular pattern, and determines the defect based on the difference in the statistical value of the grayscale value between the non-defective product and the defective product. A method of doing so is disclosed. Further, as shown in Japanese Patent Publication No. 6-72780, a large number of non-defective products are binarized with a certain threshold value, the number of pixels is subjected to statistical processing, and those out of the distribution of the statistical values are regarded as defective products. There is a way to judge.

[0003]

In the case of a casting having a rough surface and a different surface condition for each product, information on the density value of a defect which is an abnormal unevenness is based on ordinary surface roughness. It is difficult to detect a defect with high reliability by statistical processing of a simple gray value or statistical processing of the number of binarized pixels by a threshold value as in the above two conventional examples.

[0004]

SUMMARY OF THE INVENTION According to the present invention, there is provided a standard deviation image in which a plurality of non-defective workpieces are imaged to obtain a grayscale value, and a standard image obtained by averaging the grayscale values for each pixel in which the same portion is imaged and a standard deviation are obtained. For the inspection image obtained by imaging the product to be inspected and calculating and displaying the gray value, the difference from the standard image is calculated for each pixel that has imaged the same place as above, and then divided by the standard deviation image to obtain the ratio. Then, this numerical distribution is displayed as an inspection ratio image, the inspection ratio image is divided into a plurality of windows, and only those exceeding the set threshold value among the numerical values of the pixels in the window are summed up as a judgment value. There is provided a means for determining whether a defect is present by comparing the determined value with respect to the work.

[0005]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described with reference to FIG. First, data is collected from non-defective work of the same type as the product to be inspected. First, many non-defective work pieces are collected and individually imaged by a CCD camera. When taking an image, a non-defective work is always positioned at the same position by a mechanical method, or the image is moved and rotated based on the characteristic outer shape information of the product by an image processing method. Make sure that good work overlaps at the same position. The image data 1 (1a, 1b,..., 1n) of many non-defective workpieces that have been imaged and subjected to the multi-value gray-scale processing are subjected to gray-scale values for the same pixel as in the teaching process shown in FIG. The average value and the standard deviation value are calculated, and the standard image 2 and the standard deviation image 3 of the non-defective work are obtained. The standard image 2 represents the average roughness of the surface of a non-defective product as a gray value, and the standard deviation image 3 shows a portion where the variation in roughness is likely to occur with a numerical value based on the gray value.

Next, an inspection process based on the above image data will be described. In the same manner as described above, the inspection target product is imaged by a mechanical method or an image processing method so as to overlap the image of the non-defective work, and the inspection image 4 is obtained. As shown in the inspection process of FIG. 1B, a difference image between the inspection image 4 and the standard image 2 is obtained for each pixel, and a standard deviation image 3 is obtained for each pixel.
By dividing the difference image by, the ratio with the standard deviation image is obtained. Here, an image displaying the distribution of the ratio values is referred to as an inspection ratio image 5. According to the inspection ratio image 5,
An image is obtained in consideration of the standard deviation of each pixel. As a result, the influence of the portion where the surface roughness state tends to vary is corrected, and the roughness distribution of the product to be inspected is represented by a numerical value (hereinafter referred to as a correction value).
Can be expressed as

The presence or absence of a defect is determined by dividing the inspection ratio image into a large number of windows of a predetermined appropriate size, obtaining a judgment value for each window, and comparing the judgment value with a reference judgment value.
The judgment value is the total value of the correction values for the pixels in the window whose correction value exceeds the set threshold value. For example, if the absolute value of the threshold value is 1, if the absolute value of the correction value of the pixel is less than 1, it is 0, and if it is 1 or more, the correction values are summed as they are.

On the other hand, the standard image 2 and the standard deviation image 3
Is determined using the standard image 2 and the standard deviation image 3 for each of the non-defective work pieces 1 for which is determined, and the judgment value is determined for each window in the same manner as described above. From the obtained judgment values, a reference judgment value is appropriately selected and used. As the reference determination value, a value indicating the value closest to the defective product for each window among the determination values of the non-defective work can be used to perform a detailed and highly reliable inspection. From this, if the judgment value of the inspection product is larger than the judgment value of the reference non-defective work for each window, it is judged that there is a defect in the window. The size of the window may be appropriately determined based on the size of the defect area to be inspected with respect to the size of the pixel.

[0009]

EXAMPLE An example in which a defect having abnormal irregularities in appearance is detected using the above method will be described. In this example, one pixel is 0.2 mm,
The size of the window is 64 pixels x 64 pixels, and the shade is 25
The standard image and the standard deviation image are 6 good gradations.
The number was obtained by imaging. As for the shading value, the deeper the concave portion, the larger the positive value, and the higher the convex portion, the larger the negative value.

FIG. 4 summarizes the judgment values of one window of the inspected work. The top No. 1 work has a defect, and the 19 works after the No. 2 work are non-defective work in which a standard image and a standard deviation image having no defect are obtained. FIG. 2 and FIG.
The inspection ratio image of the window of the same part of 1 work and No. 2 work is shown. In the window shown in FIG. 2, there is a concave portion determined as a defect by visual inspection. There is no defective surface in the window shown in FIG. 3 by visual inspection. In FIG. 2, there is a pixel group portion having a correction value of 10 or more near the upper left center,
This portion is a portion visually determined to be an abnormal concave portion.

The rows in FIG. 4 show changes in the judgment value when the threshold value is changed. The numerical value described after the symbol sig indicates the threshold. For example, sig0 at the left end is a threshold value of 0, that is, a value obtained by adding all the values of the inspection ratio image in the window, and sig1 is 0 when the absolute value of the inspection ratio image is less than 1 and the pixel of 1 or more has the absolute value. In addition, this is a case where a judgment value is used. In addition, sig1-3 has an inspection ratio image value of -1 to 3
Is 0, and in other cases, the judgment value is obtained by adding the absolute value. The S / N ratio at the bottom indicates the ease of detecting a defect. If this value is high, the difference between a good product and a defective product is large, indicating that detection is easy,
It is represented by the following equation. 20 × log (judgment value of defective product / maximum judgment value of non-defective products) From the results of FIG. 4, sig3, sig4, or sig2-3
It was found that the use of the threshold value determined as described above makes it possible to determine the presence or absence of a defect with high reliability.

[0012]

The present invention has the following effects because it uses the means described above. Since the appearance of the inspection target product and the appearance of the non-defective work are compared for each pixel, information on the appearance details can be collected. Further, since this information is reprocessed for each certain area to determine the presence or absence of a defect, the information of the details can be averaged and determined.
There is little risk of determining noise as a defect. In addition, the size of the area can be set as appropriate, and the reprocessing judgment criterion for defect judgment can be set for each area, so that the defect can be detected in accordance with the size of the defect to be detected or the degree of irregularity.

[Brief description of the drawings]

FIG. 1 is an algorithm diagram showing an inspection method.

FIG. 2 is an example of an inspection ratio image having a defective portion.

FIG. 3 is an example of an inspection ratio image having no defect portion.

FIG. 4 is a measurement result example showing a judgment value when a threshold value is changed.

[Explanation of symbols]

 1 Contrast display image of good work 2 Standard image 3 Standard deviation image 4 Inspection image 5 Inspection ratio image

Claims (1)

[Claims] An image of a plurality of non-defective products is obtained to obtain a gray value, a standard image obtained by averaging the gray values for each pixel of the same portion, and a standard deviation image obtained by obtaining a standard deviation, and an image of a product to be inspected is obtained. For the inspection image displayed by calculating and displaying the gray value, the difference from the standard image is calculated for each pixel that has captured the same location as described above, then divided by the standard deviation image to obtain the ratio, and this numerical distribution is displayed. The inspection ratio image is divided into a plurality of windows, and among the numerical values of the pixels in the window,
Only those exceeding the set threshold are summed up as a judgment value,
An appearance inspection method characterized in that it is determined whether or not there is a defect by comparing with a judgment value separately obtained for a non-defective work.