LT6337B - METHOD FOR CHECKING PLATE OBJECTS - Google Patents

Abstract

A method for the detection of defects on surfaces of single wafer or any other single flat object, is characterized by the steps of: taking digital image of the inspected portion of the surface; generating a first processed image by plurality of applications of themorphologic operator Erode or Dilate on all image points of the image; generating second processed image by further one or more applications of the operator used in step (b) on the image generated in previous step; generating third processed image by calculating the difference or the quotient of the brightness values on each of the image points of the first and second processed image; filtering of image ranges with brightness values outside of threshold range.

Description

Technical field

The invention relates to a method for detecting defects on a single plate or another individual flat object.

In different industries, flat products are tested for defects by imaging. In the semiconductor product and solar cell industry, these products are, among others, wafers. The plates are discs of semiconductor, glass, sheet or ceramic material. The plates are inspected in whole or at least in large part. This is called a macro check.

Similar goals need to be addressed across industries. In the panel panel industry, screens must be inspected for defects during production. Partially defect detection uses image acquisition techniques to obtain a complete image. When testing printed circuit boards in the electronics industry, a series of defects are detected by optical means.

All these uses collectively mean that there is a need for rapid inspection of a large number of objects that are usually of the same type. Such objects are printed circuit boards, plates, solar cells, screens and the like. What they also have in common is that sensors are used to generate large images of objects. Images can be generated by optical imaging systems as well as point-to-point sensors depending on the type of defect detected. Optical imaging systems are, for example, linear antenna grid cameras. Spot sensors are, for example, detectors for measuring the reflections of optical rays, microwaves or acoustic waves. Magnetic sensors can also be used.

State of the art

Plaques or other surfaces are known to be scanned and compared primarily to defective or nearly defective image templates. The new image can be generated from the differences between the actual image and the image template. Such an image can be generated as a binary image. This means that there is either a defect in the pixel or no pixel. Surface defects such as scratches, dust, or manufacturing defects will be clearly visible in such a binary image, such as black defects on a white background. Defects can be very small, with a size range of 5 microns per pixel.

Surfaces that occur in practice have many more defects and interferences that can be harmless. The observer then has the problem of identifying defects of interest, such as scratches, from most different defects.

DE 10 2012 101 242 A1 describes a method for inspecting plates and other planar objects suitable for detecting defects on a plate. A series of images are taken for many objects of the same type. The image template is generated from such images. Contrary to prior art, a great image template in the form of a "golden image" is not necessary. Defective plates are not required.

The essence of the invention

It is an object of the present invention to provide a method of the above type which enables defects to be detected on an unstructured individual object without the need for an image template.

According to the invention, this object is achieved by a process comprising the steps of:

obtaining a digital image of the surface area under investigation;

generating a first processed image using a plurality of morphological Erode or Dilate operations on all pixels;

generating a second processed image by applying one or more operations used in step (b) on the images generated in step (b);

generating a third processed image by calculating the difference or coefficient of luminance values on each of the first and second processed pixels;

image ranges with luminance values for boundary range filtering.

Morphological operations on erosion or extension are well known as basic digital image processing operations. The Minkowski sum of the image formed by the expansion operation using a structural element such as a circle or a rectangle. Each pixel of the image is expanded to the shape of a structural pixel. Dual erosion operation results in skipping all pixels where the environment does not completely fill the element used. In a simple example, the area is slightly increased during the expansion operation and slightly reduced during the erosion operation, or in other words, the extension enlarges the area of the pixels that are brighter than their surroundings, while erosion increases the area of the pixels that is darker than their surroundings.

Unexpectedly, defects are more clearly evident from interference when the morphological operations are repeated several times with twice the same repetition rate. Because defects are exposed to much more than interference, the difference calculation enables interference filtering. The differences in the resulting images provide defective edges that can be easily filtered from interference. The threshold can be applied to an image with contours generated by difference estimation, where the binary image is generated wherever effects can be seen whose luminance values exceed the thresholds. Thus, the initial image interference is filtered from the signal. Alternatively, the quotient may be calculated rather than based on the difference between the values obtained.

In particular, an erosion operation may be applied if the luminance values of the image range are above the cut-off range, and the widening operation applies if the luminance values of the image range are below the cut-off range. Thus, the erosion operation is applied to dark defects and the extension operation is applied to light defects. Defects in the resulting image can be viewed as outlines.

Preferably, the morphological operation Close should be applied to all pixels in the image. The operation has the effect that all non-zero values are merged into areas within the processing area. The study allows to combine possible defects, which have been separated into separate, unrelated structures, forming a boundary image. In addition, the number of possible defective areas is reduced. Areas that were previously partitioned but actually belong to a common area are merged again.

It is desirable that the generated image be further processed in the contours generated around the merged areas. The outlines are defined around all pixels of the image that are not zero. The contours form coordinates that completely define the defect area. The contour region accurately surrounds the region that was generated during the morphological operation with the highest repetition rate.

For further processing, the size of the defect may be determined by the contour of the surrounded area and the number of applications to generate the second processed image. Preferably, the size of the defect could be determined by subtracting the corresponding width or height from the area surrounded by the width or height contour corresponding to the pixel width multiplied by twice the sum of operations applied to generate the second processed image and defining the area if both differences acquire a value greater than zero. Any area that is not defined as a defect will form a majority of interferences that are below the set threshold and therefore cannot be considered a defect.

Preferably, the outline is defined by a polygon. The polygon can be in the form of a binary image and used for further processing.

Further modifications of the invention are a matter of dependent claims. Embodiments of the invention will be described in more detail with reference to the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. shows a sectional view of an unstructured plate in the form of a raw image.

Fig. Figure 1 is a sectional view of the image after removing the total luminance change.

Fig. Fig. 2 is a sectional view of the image. after application of erosion operation n times.

Fig. Fig. 2 is a sectional view of the image. after application of erosion operation n + x times.

Fig. shows the difference of the image obtained by subtracting the image slice according to Fig. 3. from the image section in Figure 4.

Fig. shows the binary image that was created by determining all the slices of the image according to Fig. 5. pixels that exceed the threshold with a value of 1 and the rest of the pixels with a value of 0.

Fig. Figure 6 shows a sectional view of Figure 6 after applying the closure operation.

Fig. FIG. with an outline around the area in Figure 7.

Fig. shows a section of the image in Fig. 8, minimizing areas in the outline.

Fig. shown in Fig. 1. detail with one defect.

Fig. shown in Fig. 2. detail with defect in Figure 10.

Fig. illustrates the effect of erosion operation on pixels.

Fig. illustrates the effect of the erosion operation based on the defect in Figure 10.

Fig. illustrates the effect of application n of erosion operation based on the defect in Figure 10.

Fig. illustrates the effect of subsequent applications of two erosion operations based on the defect in Figure 14.

Fig. illustrated in Fig. 15 of the images. and Fig. 14. the result of the difference calculation.

Fig. illustrates the effect of operation closure, based on Fig. 16. view.

DESCRIPTION OF EMBODIMENT The figure shows a section of an unstructured plate in the form of a raw image. Various defects can be recognized in image section 10, such as defects 12, 14 and 16. All images in Figures 1-9 were cropped, i.e. dark spots are shown as light spots and vice versa. One of the defects, defect 12, is shown magnified in Figure 10. The raw image in section 10 was taken with a basic color camera on a co-inspector for macro inspection. Images can be processed and evaluated individually for each color channel. Video section 10 represents a single channel image.

The raw image in Figures 1 and 10 may have general color changes. Such changes are slight changes in luminance along large portions of the image, such as those caused by the imaging system. Figure 10 shows the differences in dash fluctuations for 18 and 20 different widths.

Prior to actual image processing, such general color changes are eliminated by known methods. The image is generated on a smooth background 22 as shown in Figures 2 and 11. The extension morphological operation is then applied to each 10 pixels of the image section. A structural element, also called a kernel, is used in an operation. In this embodiment and for a better understanding of the invention, a particularly simple structural element 24 (Fig. 12) is used, which consists of a 3x3 pixel array. The extension operation with such a structural element 24 causes each luminance value on an image point, such as image point 26, to be surrounded by a dark image point with a 3x3 matrix, also to be considered a dark value. In this embodiment, the maximum value underneath the kernel will be written on the image point according to the anchor position in the expansion operation. The lowest value is used in the erosion operation. In this embodiment, this results in a range 28 that is extended to range 26 prior to processing. The effect of operation on defect 12 is shown in Fig. 13. It can be seen that the defect 12 retains its shape but substantially larger. For illustration purposes, the original defect 12 is filled with a different pattern than the edge 30 that was added, but the luminance values are actually the same in the ranges 12 and 30 after treatment.

The extension operation is performed several times, namely n times. Defect 12 "grows" at the edges with each repetition. This is shown schematically in Fig. 14. with n =

3. The result of applying operation n times on image section 10 is shown in Fig. 3. Defects are more clearly recognized due to the wide variety of extension applications.

The extension operation is repeated in the next step of the method, i.e. x times. This causes a further increase in defects. Figure 15. illustrates the effect of two more operation repetitions. Defect 32 is shown after n operations and defect 34 after n + x operations with x = 2. It is understood that the examples are for illustration only and that the defects, the amount of repetitions, and the kernel may differ from the example in actual practice. Figure 4 shows a section view of the image after 10 applications of the n + x extension operation.

After the extension operations have been applied n times and n + x times, the images thus generated are subtracted from the sub-item. In an alternative embodiment, the coefficient is formed. The operation application n times the image generated by the operation generated by the operation n + x times. Only the edges around the defect remain in the difference image. This can be seen in Figure 5. Figure 16 illustrates the difference between defect 34 and defect 32 in more detail. The difference corresponds to a dark edge 36 with a light inner area 38.

The luminance values of the image slice 10 shown in Fig. 5 result from the various luminance values of the original image. The recognition of structures is sufficient to detect the defect. Therefore, the image obtained in Figure 5 is further processed only by maintaining values on such pixels where the luminance value exceeds the threshold value. The luminance value for such pixels is 1 and for the remaining pixels is 0. This results in a binary image where defect structures can be well recognized. Such an image is shown in Fig. 6. Interferences are effectively suppressed in this way.

The binary image in Figure 6 is further processed by applying a morphological operation to the parenthesis. The operation closure corresponds to the consistent application of expansion and erosion operations. Areas are filled with such operations, and in surrounded areas, values of 0 are set to 1 when the common structures are related. Figure 7 the resulting image is shown. Schematic representation in Figure 16 indicates that the area 38 inside the structure is filled.

The outlines are then defined around all the pixels in the resulting image, which is not zero. The contours exactly surround the area created after the extension operation is applied n-x times, i.e. application with the highest number of reps. The result is the resulting image, which is shown in Fig. 8. The contour corresponds to the edge 40 of Figure 42 in Figure 17.

The area 42 inside the contour 40 is too large due to different operations. But the operations are well known. It is therefore easy to reduce the defects to their original size. In this example, n + x = 3 + 2 = 5 times was used to expand the image with 3x3 wide structural elements of the pixels. Therefore, area 42 must be reduced at all edges by 3x5 image width. In other words: width and height are multiplied by double structural elements and multiplied by the maximum number of operations. From Figure 17. the structures shown in Figs. 10 and 11 can be determined. During image processing, interference was completely eliminated and defects are available as merged areas in the form of a binary image. Such an image is shown in Fig. 9. A defect occurs if the width and / or height of the area at non-zero luminance values in Figure 8 is greater than twice the structure element multiplied by the maximum number of applications of the operation. The resulting defect areas can be represented by polygons and stored in this form.

In the resulting image Figure 9. defects 12, 14 and 16 are clearly shown as binary regions on a white background. They can be easily identified and further evaluated. The calculated reference image or gold image of a perfect plate is not required. All image processing steps can be performed directly on a single plate image or on a cross section of the image.

An earlier embodiment has been described using light areas on a dark background. However, it is possible to recognize defects that look like dark areas on a light background in the same way. In this case, the extension operation is replaced by an erosion operation for repeated operations. It is also possible to use operations with differently formed structural elements, such as crosses, circles and the like, without departing from the scope of the present invention. The foregoing example merely illustrates the method of the present invention, but is not intended to limit the scope and scope of the invention.

Claims (7)

DEFINITION OF INVENTION 1. A method of detecting defects on an individual plate or other individual flat object, comprising the following steps: (a) obtaining a digital image of the area under investigation; (b) generating a first processed image by applying a plurality of morphological Erode or Dilate operations on all pixels; (c) generating a second processed image using an additional one or more operations used in step (b) on the images generated in step (b); (d) generating a third processed image by calculating a difference or coefficient of luminance values on each of the first and second processed pixels; (e) filtering image ranges with luminance values outside the cut-off range. 2. The method of claim 1, further comprising applying an erosion operation if the luminance values of the image range are above the threshold range, and applying an expanding operation if the luminance values of the image range are below the threshold range. 3. A method according to any one of the preceding claims, characterized in that the morphological operation applies to all pixels of the image. 4. The method of claim 3, wherein the contours are generated around the connected areas. 5. The method of claim 4, wherein the size of the defect is determined from the contoured area and the number of operations applied to generate the second processed image. 6. The method of claim 5, wherein the size of the defect is determined by subtracting the width or height, respectively, from the width or height of the contoured area, corresponding to the pixel width multiplied by twice the number of operation applications used to generate the second processed image. defines a surrounded area as a defect if both differences get a value greater than zero. 7. The method of claim 5 or 6, wherein the contour is defined by a polygon.