JP5609433B2 - Inspection method for cylindrical containers - Google Patents

Description

The present invention relates to a method for inspecting a printed pattern such as characters and figures printed on a side surface of a cylindrical container.

An image obtained by imaging a printing pattern on the side surface of the container by rotating the cylindrical container has errors such as tilt, deviation, and expansion / contraction in the rotation direction that occur during printing or inspection by the printing press. Therefore, in the method of inspecting based on the image read by rotating the cylindrical container, when the template matching with the reference image is performed as it is for the read image, it is judged that the product is non-defective due to the error. May end up. In particular, there is a problem that errors generated by rotating the cylindrical container are large and high-precision inspection cannot be performed.

Therefore, in Patent Document 1, template matching using a plurality of templates is used to correct the inclination of the inspection image and the expansion and contraction in the rotation direction, and the corrected inspection image and the reference image are compared.

JP 2000-105199 A

  However, the method of correcting the expansion / contraction in the rotation direction of Patent Document 1 can correct macro expansion / contraction of the inspection image and cannot completely correct local expansion / contraction.

  For example, no matter how high-precision the rotation mechanism that rotates the cylindrical container is, the cylindrical container is not an ideal cylinder, so it extends in some parts of the captured inspection image, but some parts Then, it may be shrunk.

  The present invention has been made in view of the above-described problems, and the object of the present invention is to provide a highly accurate inspection method capable of inspecting a cylindrical container having local expansion and contraction of an inspection image. It is.

As means for solving the above problems, the invention according to claim 1
A method for inspecting a cylindrical container that uses a template matching to inspect printed patterns such as characters and figures printed on the side of a cylindrical container,
A reference image setting step for setting a template by storing in advance a reference image used in template matching;
An image capturing process for rotating the cylindrical container to be inspected and capturing an image of its side surface;
An inspection image storage step of constructing and storing an inspection image from the imaged side surface;
An image comparison process for comparing the inspection image and the reference image,
In the reference image setting step and the inspection image storage step , the inspection image and the reference image are stored as image data divided into a plurality of blocks,
In the inspection image storage step, an image for one rotation of the side surface of the cylindrical container is stored as an inspection image,
In the previous image comparison process, in the block where the pattern exists, the template matching is performed in the area set for each block and the position-corrected inspection image is used in the block area, and in the block where the pattern does not exist, Without performing position correction, using the inspection images as they are, the images are combined and reconstructed as a new inspection image, and the pattern is inspected by comparing the new inspection image with the reference image. This is a method for inspecting a cylindrical container.

  According to the present invention, since the local expansion and contraction of the inspection image is corrected and then compared with the reference image, the printed pattern printed on the cylindrical container can be inspected with high accuracy without erroneous detection.

The conceptual diagram which shows an example of the apparatus structure which implements the test | inspection method of this invention. (A) is an example of a side image of a cylindrical container taken by the line camera of the present invention, (b) is an example of a template, and (c) is an image obtained by rotating (a) by 90 degrees. (A) is an example of a photographed image, and (b) is an example of an image extracted from (a). (A) is an example showing a situation where the reference image is divided into blocks, and (b) is an example showing a situation where the inspection image is divided into blocks. (A) is the block 2 portion of the reference image, (b) the area within the solid rectangle is the block 2 portion of the inspection image, and the area within the dotted rectangle is the region having the highest similarity to the block 2 portion of the reference image Indicates. (A) is the example which shows the state which corrected the coordinate of the block for every block about the test | inspection image, and raised the similarity most to a reference | standard image, (b) is an example which shows the test | inspection image after the correction | amendment. In the present invention, an example of an explanatory diagram showing that a difference image is obtained by calculating a corrected inspection image and a reference image. In the present invention, an example of an explanatory diagram showing that a difference image is binarized and white defects and black defects are extracted. In this invention, the example which shows the condition of the reference | standard image, test | inspection image, and position-corrected test | inspection image about a block with no pattern (plain). It is an example which shows the condition of the reference | standard image, test | inspection image, and position-corrected test | inspection image about the block in which a pattern exists, (a) is an example in case there are few divisions and one block is large, (b) is the number of divisions There are many cases where one block is small.

FIG. 1 is a conceptual diagram showing an example of an apparatus configuration for carrying out the inspection method of the present invention. The cylindrical container 1 is irradiated with illumination 2 and rotated by a rotating means (not shown). The line camera 3 images the rotating cylindrical container 1 at a predetermined timing and transfers the acquired image data to the image input means 4.

The image input means 4 is connected to a bus 10 together with a hard disk 5, a keyboard / mouse 6, a CPU 7, a memory 8, and a monitor 9.

  FIG. 2A is a side image of the cylindrical container 1 taken by the line camera 3. In order to reliably extract the inspection region, an image of two or more rotations of the cylindrical container 1 is acquired.

  And it collates with the template (FIG.2 (b)) preset in the memory 8, and cuts out the inspection area | region for 1 rotation. For this purpose, FIG. 2A is rotated 90 degrees to obtain FIG. 2C, and template matching is performed in FIG.

At this time, two candidates (FIG. 2 (
Candidate 1 and candidate 2) shown in the figure of c) are found. By referring to the coordinate information of the found candidate, the candidate that can extract the examination region, that is, candidate 1 can be selected.

  As shown in FIG. 3A, if the coordinates of the top left vertex of the selected candidate 1 are (P, Q), (P−Δp, Q−Δq) is set as the start coordinate, and ΔX and Y directions are set in the X direction. If the region of ΔY is cut out, FIG. 3B, which is an inspection region for one rotation, can be extracted.

Next, as shown in FIG. 4, the reference image and the inspection image are divided into a plurality of blocks (M × N). For the sake of explanation, each block is named block 1, block 2,..., Block (M + N).

Block 1, block 2,..., Block (M + N) of the reference image serves as a template for correcting local expansion / contraction of the inspection image. Hereinafter, the position correction of the inspection image will be described.

  First, since there is no pattern in block 1 of the inspection image, this block is used as it is and is arranged in block 1 of the corrected inspection image.

  Next, since the pattern 2 exists in the block 2 of the inspection image, local position correction is performed. This method will be described.

  FIG. 5A shows the block 2 portion of the reference image, and the area within the solid rectangle in FIG. 5B is the block 2 of the inspection image. With respect to each block of the inspection image, ± α and ± β position correction regions are provided, and the reference image block 2 (FIG. 5 (FIG. 5 (A)) is shifted by ± α in the X direction and ± β in the Y direction around the block 2 of the inspection image. The degree of similarity with a)) is calculated sequentially. As a method for calculating the similarity, a known normalized correlation coefficient may be used. And let the area | region in the dotted-line rectangle of FIG.5 (b) be an area | region with the highest similarity degree with Fig.5 (a).

  Therefore, the area within the dotted rectangle in FIG. 5B is arranged in the block 2 of the corrected inspection image.

  Thereafter, the same operation is similarly performed for the block 3, the block 4,..., The block (N + M). That is, the block in which the inspection image pattern does not exist is used as it is and is arranged in the corresponding block of the corrected inspection image. (FIG. 6 (a)) For the block in which the pattern of the inspection image exists, the similarity is calculated sequentially with the corresponding block of the reference image while shifting by ± α in the X direction and ± β in the Y direction. The area is cut out and placed in the corresponding block of the corrected inspection image.

  In this way, the corrected inspection image is reconstructed as shown in FIG.

  Subsequently, as shown in FIG. 7, the reference image is subtracted from the corrected inspection image, and then 128 is added to each pixel. This calculation result is called a difference image.

  Subsequently, as shown in FIG. 8, a binarization process is performed by providing an appropriate threshold for the difference image. Assuming that A (> 0) and B (> 0), the threshold for white defect extraction is (128 + A), and the threshold for black defect extraction is (128-B). Then, a portion having a higher luminance than (128 + A) is extracted as a white defect, and a portion having a lower luminance than (128−B) is extracted as a black defect.

Here, for a block having no pattern, the reason for using the block as it is without using the procedure of sequentially calculating the similarity with the reference image while shifting ± α in the X direction and ± β in the Y direction. Will be described. In general, a normalized correlation coefficient is used as the similarity in template matching, and for example, Equation 1 is used.

here,
f [i, j]: luminance value of corrected inspection image at coordinates (i, j) <f>: luminance average of corrected inspection image s [i, j]: coordinates (i, j) of reference inspection image ) Brightness value <s>: average brightness of reference image

  The block 1 in FIG. 4B or the like is a block having no pattern and may be an extreme example. However, if the luminances of all the pixels in the block 1 are equal, f [i, j] − < Both the value of f> and the value of s [i, j]-<s> are 0, and the similarity cannot be calculated.

  Actually, it may be rare that the luminance of all pixels is equal, but if template matching is applied to a plain area where no pattern exists, f [i, j] − <f> ≈0, s [ i, j] − <s> ≈0, and the similarity value diverges, which may lead to an unexpected result. One example is shown in FIG.

In FIG. 9, even in a plain block where no pattern exists, the similarity is sequentially calculated by shifting ± α in the X direction and ± β in the Y direction as in the block in which the pattern is present, and the similarity is maximized. The result of having arrange | positioned the part to the corresponding block of a position-corrected test | inspection image is shown.

  Even though no pattern exists in the block i, the position-corrected inspection image includes the pattern in the adjacent block (i + 1). This means that if the pattern in the block (i + 1) is slightly included, the similarity of Equation 1 is higher, but template matching is applied to an image that does not have a pattern in the first place. Then, you may get into such an unintended result.

  In order to correct local expansion and contraction, it is effective to finely divide the acquired inspection image, that is, to increase the number of blocks and to perform position correction processing for each block. A plain block that does not contain will come out.

Conversely, in order to prevent the appearance of plain blocks, it is possible to divide the blocks so that each block is large. However, if the block is large, local expansion and contraction may not be possible. is there. This example will be described with reference to FIG. FIG. 10 is a diagram showing how the local position correction effect is affected depending on whether the area per block is large or small. Focusing on the difference image ((Position-corrected inspection image) − (Reference image) +128) in the upper row where (A) the number of divisions is small and one block is large, there are many pixels close to the luminance value 128 on the left. It can be seen that this was performed well. On the other hand, pixels that are smaller (darker) than 128 and pixels that are larger (brighter) than 128 appear in pairs on the right side, and it can be seen that matching is not performed well. This is a local expansion and contraction phenomenon, and this is a major cause of determining a non-defective product as a defective product.

In FIG. 10B, when the number of divisions is large and one block is small, (A) the number of divisions is small and the number of blocks is larger than when one block is large, and position correction is performed for each block. In the difference image, there are many pixels closer to 128 than in the case of (A). In other words, by finely dividing, increasing the number of blocks, and performing position correction for each block,
The effect of suppressing local expansion and contraction can be confirmed.

As described above, in order to prevent erroneous detection due to local expansion and contraction phenomenon, it is necessary to finely divide and perform position correction for each block, and to care for plain blocks due to increase in the number of blocks (position correction) It is effective to use in combination without using

DESCRIPTION OF SYMBOLS 1 Cylindrical container 2 Illumination 3 Line camera 4 Image input means 5 Hard disk 6 Keyboard / mouse 7 CPU
8 Memory 9 Monitor 10 Bus

Claims (1)

A method for inspecting a cylindrical container that uses a template matching to inspect printed patterns such as characters and figures printed on the side of a cylindrical container,
A reference image setting step for setting a template by storing in advance a reference image used in template matching;
An image capturing process for rotating the cylindrical container to be inspected and capturing an image of its side surface;
An inspection image storage step of constructing and storing an inspection image from the imaged side surface;
An image comparison process for comparing the inspection image and the reference image,
In the reference image setting step and the inspection image storage step , the inspection image and the reference image are stored as image data divided into a plurality of blocks,
In the inspection image storage step, an image for one rotation of the side surface of the cylindrical container is stored as an inspection image,
In the previous image comparison process, in the block where the pattern exists, the template matching is performed in the area set for each block and the position-corrected inspection image is used in the block area, and in the block where the pattern does not exist, Using the inspection image as it is without performing position correction, the inspection image is combined and reconstructed as a new inspection image, and the pattern is inspected by comparing the new inspection image with the reference image. The inspection method of the cylindrical container characterized by the above-mentioned.