Machine vision-based method for detecting surface blisters on ceramic tiles

Abstract

The method for detecting the bulge on the surface of ceramic tile based on machine vision first extracts the red channel image of the ceramic tile and carries out the pretreatment of Gaussian filtering to improve uneven illumination and inhibit noise; second, the method of adaptive wavelet transform and linear median filtering is adopted to enhance the contrast of the bulge and the background area; finally, the information of the bulge area is obtained by using the method of binarization and morphology. The method for detecting the bulge on the surface of ceramic tile based on machine vision has high automation degree and can solve the problems of high labor intensity, strong subjectivity and low detection efficiency in manual detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ceramic tile surface defect detection, and particularly relates to a ceramic tile surface bump detection method based on machine vision. BACKGROUND

[0002] Ceramic tile surface defect detection is an important step in the production process. Due to different production conditions such as production process and production environment, ceramic tile surfaces may have defects such as bumps, cracks, missing corners, missing glaze, and color difference. At present, the defect detection of this production process still adopts the method of manual detection, and manual detection has the problems of strong subjectivity, high labor intensity, and low detection efficiency. In order to improve the detection efficiency and accuracy and promote industrial upgrading, it is crucial to realize the automatic detection of ceramic tile defects.

[0003] In recent years, in the defect detection of industrialized production, such as the surface defect detection of steel pipes, metals, and ceramic tiles, the method of machine vision is applied more and more commonly. For example:

[0004] Document [1]: Zhang Jianjun, Luo Jing. Surface crack edge detection algorithm based on improved Sobel operator [J]. Journal of Hefei University of Technology (Natural Science Edition), 2011, 34(6): 845-847. The improved Sobel operator is used to detect the surface cracks of steel pipes;

[0005] Document [2]: Wang Qingchen, Jing Junfeng. Metal surface transverse crack detection using machine vision [J]. Journal of Electronic Measurement and Instrument, 2018, 32(11): 71-77. Image blocking and self-defined filtering method are used to detect metal surface cracks;

[0006] Document [3]: Guo Meng, Hu Liaolin, Zhao Jiangtao. Ceramic bowl surface defect detection method based on Kirsch and Canny operators [J]. Acta Optica Sinica, 2016, 36(09): 27-33. Kirsch and Canny operators are used to detect defects on the surface of ceramic bowls;

[0007] Document [4]: Li Qiang, Zeng Shuguang, Zheng Sheng, etc. Ceramic tile surface crack detection method based on machine vision [J]. Progress in Lasers and Optoelectronics, 2020, 57(08): 51-57. Principal component analysis method is used to detect ceramic tile surface cracks.

[0008] The technical solutions in the above documents have certain effect on the surface crack detection of industrial products, but for ceramic tiles with a three-dimensional surface and a large number of patterns, bumps are one of the most complex and difficult defects to detect, and there are few reports on bump detection at home and abroad. SUMMARY

[0009] Aiming at the technical problem that the automatic detection of the surface bump defect of the ceramic tile is difficult due to the high and low relief of the ceramic tile surface and the interference of the complex patterns, the application provides a ceramic tile surface bump detection method based on machine vision, which has high automation and can solve the problems of high labor intensity, strong subjectivity and low detection efficiency in manual detection.

[0010] The technical scheme adopted by the application is:

[0011] The ceramic tile surface bump detection method based on machine vision comprises the following steps:

[0012] Step 1: Extract the red channel image of the ceramic tile to remove the interference of uneven illumination;

[0013] Step 2: Perform Gaussian filtering on the red channel image of the ceramic tile to suppress noise;

[0014] Step 3: Use the method of adaptive wavelet transform (AWT) and linear median filtering to enhance the contrast between the bump and the background area.

[0015] Step 4: Use the method of binarization and morphology to obtain the information of the bump area.

[0016] In the step 2, the Gaussian filtering is specifically as follows:

[0017] Each pixel of the image is weighted and averaged by a template to reduce the gray difference between adjacent pixels, and the Gaussian filtering window size is 3x3; the row mean N(i) and the column mean M(j) of the preprocessed image are compared with the total mean K to obtain the ceramic tile area W1 to be detected, and the calculation formula is as follows:

[0018]

[0019]

[0020]

[0021] Wherein: W0(i,j) is the pixel value of the red channel of W0; m and n respectively represent the number of rows and columns of the pixel matrix; i and j respectively represent the index of the pixel, and the gray matrix of the area to be detected is W1, and the size is m1x n1, m1 and n1 respectively represent the number of rows and columns of the gray matrix.

[0022] The step 3 comprises the following steps:

[0023] S3.1: Use the adaptive wavelet transform method to decompose W1, which can obtain the information of different scales of W1, select the scale containing the bump information and perform wavelet reconstruction to obtain the reconstructed image W2.

[0024] Let the original signal be P0(x) and the low-pass filter be V(l). After one filtering, the signal P0(x) can obtain a low-frequency component and a high-frequency component. The high-frequency component represents the detailed information of the image and is also known as the wavelet surface or wavelet coefficient. The low-frequency component represents the approximate information of the image and can be further filtered and decomposed.

[0025] Let the low-frequency component after n-fold filtering be P. n (x), the high-frequency component obtained by the nth decomposition is G. n (x) has the following mathematical description:

[0026]

[0027] G n (x)=P n-1 (x)-P n (x)n=1,2,…,N;

[0028] Where n is the number of decompositions, also known as the scale.

[0029] The àtrous wavelet transform decomposes a signal into multiple wavelet surfaces and an approximate signal. The wavelet surfaces represent different details, while the approximate signal represents background information. To reconstruct the signal, all wavelet surfaces and the approximate signal are superimposed, as shown in the following formula:

[0030]

[0031] Extending the àtrous wavelet transform into a two-dimensional image P(x,y) results in a decomposed image with the same dimensions as the original image, allowing for the extraction of high-frequency components G at different scales. n (x,y) enables feature separation of images with different characteristics. Similarly, the reconstructed image P'(x,y) can be obtained by superimposing the decomposed wavelet coefficients with background information.

[0032]

[0033] The àtrous wavelet algorithm also has the following advantages: ① It is simple to operate; ② It does not involve interpolation or sampling in the calculation, which is beneficial for obtaining the detailed texture features of the image; ③ The transformation process can be achieved through filtering, making it easy to program.

[0034] S3.2: The bulge information in the reconstructed image W2 appears as anomaly regions. This bulge information is removed using horizontal median filtering. A one-dimensional horizontal median filter is applied to the reconstructed image W2 to obtain the filtered image W3. Median filtering replaces the pixel value of a point in a digital image with the median value of all points in its neighborhood, thereby eliminating isolated noise points.

[0035] S3.3: Difference the image W2 and the image W3 to obtain the image W4, that is, subtract the reconstructed image W2 from the background image W3 after removing the bulge information. The purpose of the image difference is to improve the contrast between the bulge and the background area.

[0036] The step 4 comprises the following steps:

[0037] S4.1: Binaryzation the image W4 by using the gray threshold T1 to obtain the binaryzation image W5, realize the segmentation of the bulge, and the calculation formula is as follows:

[0038]

[0039] W4(i,j) represents the pixel value of W4; abs(W4(i,j)) represents the absolute value of the pixel value of W4;

[0040]

[0041] W5(i,j) represents the pixel value of W5;

[0042] S4.2: The binaryzation image W5 is further subjected to connected domain extraction and morphological filtering to obtain the image W6; finally, the number of the bulges is counted and the area thereof is calculated according to the image W6. The connected region is a pixel set composed of adjacent pixels with the same pixel value, and the connected domain extraction is to find the connected region in the binaryzation image, and then a unique identification is given to each connected domain, and then the required connected domain is retained according to the requirement, and the rest is set to zero. The method uses the closing operation in the morphological filtering to fill the small holes, cracks and the like of the extracted connected domain edge to make it more complete, but the total position and shape are unchanged.

[0043] The ceramic tile surface bulge detection method based on machine vision has the following beneficial effects:

[0044] 1) The method of the present application performs adaptive wavelet transform on the image, and uses horizontal median filtering, image difference, morphological filtering and other processing methods to realize the separation of the bulge and the background, and effectively eliminates the complex background interference problem mainly in the stereoscopic pattern.

[0045] 2) The method of the present application has high detection accuracy for the ceramic tile bulge, and the false detection rate and the missed detection rate are relatively low.

[0046] 3) The method of the present application has high automation degree, and can solve the problems of high labor intensity, strong subjectivity and low detection efficiency in manual detection. BRIEF DESCRIPTION OF DRAWINGS

[0047] Fig. 1(a) is a typical ceramic tile bulge image.

[0048] Fig. 1(b) is a typical ceramic tile bulge image surface gray distribution graph;

[0049] Figure 2 The flow chart of the ceramic tile surface bump detection algorithm proposed in the present application.

[0050] Fig. 3(a) is a ceramic tile original image W0, wherein the bump part is marked with a red rectangular frame and highlighted in the upper right corner;

[0051] Fig. 3(b) is the image W1 after pre-processing of W0.

[0052] Fig. 3(c) is the reconstructed image W2 after selecting a suitable scale of W1.

[0053] Fig. 3(d) is the result W3 after median filtering of W2.

[0054] Fig. 3(e) is the result W4 of the difference between W2 and W3.

[0055] Fig. 3(f) is the result image W5 after binarization of W4.

[0056] Fig. 3(g) is the result image W6 after connected domain extraction and morphological filtering of W5. DETAILED DESCRIPTION

[0057] The present application proposes a ceramic tile surface bump detection method combining machine vision and digital image processing technology, which adopts adaptive wavelet transform (AWT) to detect ceramic tile surface bumps. First, the red channel image of the ceramic tile is extracted to remove the interference of uneven illumination; second, the red channel image of the ceramic tile is subjected to Gaussian filtering to suppress noise; then, adaptive wavelet transform (AWT) and linear median filtering are used to enhance the contrast between the bump and the background area; finally, the information of the bump area is obtained by using binarization and morphological methods, and the specific implementation is as follows:

[0058] The ceramic tile has two parts of tile ridge and tile groove, wherein the tile ridge is convex and the tile groove is concave, having the characteristics of ups and downs, and the picture collected by the camera will be affected by uneven illumination, appearing bright and dark strips. A typical ceramic tile image is shown in Fig. 1(a), wherein the area indicated by the arrow is the bump. The surface gray scale image of the ceramic tile shown in Fig. 1(a) is shown in Fig. 1(b), and it can be seen from Fig. 1(b) that the brightness of the ceramic tile surface indeed has large fluctuations. The ceramic tile bump mainly appears in the pattern area, with low contrast, which is easy to be confused with the pattern. The pattern on the surface of the ceramic tile is a triangular claw type, which is distributed at 120°.

[0059] Due to the noise interference and uneven illumination in the image acquisition and transmission process, the original image needs to be preprocessed before the bulge detection. The red channel image can effectively remove the interference of uneven illumination, and the Gaussian filter can remove the noise interference formed in the image acquisition and transmission process. The flow chart of the ceramic tile surface bulge detection algorithm based on adaptive wavelet transform is shown in Figure 2 The specific steps are as follows:

[0060] Step 1: input image:

[0061] The original image of the ceramic tile photographed by the camera is W0. The original image of the ceramic tile W0 is shown in Fig. 3(a). W0 is a color image with a size of m x n x 3, containing red, green and blue channel components.

[0062] Step 2: image preprocessing:

[0063] The ceramic tile image is easily disturbed by stray light and other noise in the acquisition and transmission process, and the surface of the ceramic tile is easily unevenly illuminated, so the original image needs to be preprocessed before detection. First, the red channel image of W0 can effectively remove the interference of uneven illumination; second, the Gaussian filter can remove noise to a certain extent and the bulge information is relatively complete. The Gaussian filter is a linear smoothing filter used to remove noise and smooth the image. Its principle is to perform weighted average on each pixel of the image through a template, reduce the gray difference between adjacent pixels, and the size of the Gaussian filter window in the algorithm is 3 x 3. The row mean N(i) and column mean M(j) of the preprocessed image are compared with the total mean K to obtain the ceramic tile region to be detected W1, as shown in Fig. 3(b). The calculation formula is as follows:

[0064]

[0065]

[0066]

[0067] Wherein, W0(i,j) is the red channel pixel value of W0; and the gray matrix of the region to be detected is W1, with a size of m1 x n1.

[0068] Step 3: use the method of adaptive wavelet transform to decompose W1, and obtain the information of different scales of W1, select the scale containing the bulge information and perform wavelet reconstruction to obtain the reconstructed image W2, as shown in Fig. 3(c).

[0069] Step4: The bulge information in W2 is shown as an abnormal point area, and the horizontal median filter can erase the bulge information. The one-dimensional horizontal median filter is used to process W2 to obtain the filtered image W3, as shown in Fig. 3(d).

[0070] Step5: The image W4 is obtained by differentiating W2 and W3, as shown in Fig. 3(e), and the differentiation can improve the contrast between the bulge and the background area.

[0071] Step6: The binary image W5 is obtained by using the gray threshold T1 to binarize the image W4, as shown in Fig. 3(f), and the segmentation of the bulge is realized, and the calculation formula is as follows:

[0072]

[0073]

[0074] Step7: The connected domain extraction and morphological filtering are used to process the binary image W5 to obtain the image W6, as shown in Fig. 3(g). Finally, the number of bulges is counted and the area thereof is calculated according to the image W6. It is known that the size of the ceramic tile is 300mm x 400mm, and an approximately elliptical bulge is detected by the algorithm, the major axis thereof is about 5mm, the minor axis thereof is about 4mm, and the area thereof is about 15.71mm 2 .

Claims

1. A method for detecting surface blisters on ceramic tiles based on machine vision, characterized by The method comprises the following steps: Step 1: extracting a red channel image of the ceramic tile to remove the interference of uneven illumination; Step 2: performing Gaussian filtering on the red channel image of the ceramic tile to suppress noise; Step 3: adopting an adaptive wavelet transform and linear median filtering method to enhance the contrast between the bulges and the background region; Step 4: obtaining the information of the bulge region by using a binarization and morphological method; The step 3 comprises the following steps: S3.1: using the method of adaptive wavelet transform to decompose the region of ceramic tile to be detected , information of different scales can be obtained , select the scale containing the bulge information and perform wavelet reconstruction to obtain the reconstructed image ; Let the original signal be , the low-pass filter be , and the signal be filtered once to obtain a low-frequency component and a high-frequency component, wherein the high-frequency component represents the detail information of the image, also known as the wavelet surface or wavelet coefficient, and the low-frequency component represents the approximate information of the image and can be further filtered and decomposed; Let n the low frequency component after the n-th filtering be , the high frequency component after the n-th decomposition be n , and the low frequency component after the n-th decomposition be , which can be mathematically described as follows: ； ； wherein n is the order of decomposition, also known as the scale; The àtrous wavelet transform decomposes a signal into a plurality of wavelet faces and an approximate signal, the plurality of wavelet faces representing different detail information, and the approximate signal representing background information; if the signal is to be reconstructed, all the wavelet faces and the approximate signal are superimposed, as shown in the following formula: ； Extension of à trous wavelet transform to two-dimensional image In the method, the decomposed image is consistent with the original image in size, and high-frequency components at different scales can be extracted , feature separation of images with different features is realized; and the reconstructed image is obtained by superimposing the wavelet coefficients after decomposition and background information ； S3.2: Reconstructing image The middle bump information is expressed as an abnormal point region, and the bump information is removed by horizontal median filtering; the reconstructed image is subjected to one-dimensional horizontal median filtering to obtain a filtered image ; S3.3: subtracting the image from the image to obtain a difference image i.e. reconstructing the image from the background image with the bulge information removed ; the purpose of the image differencing is to improve the contrast between the bulge and the background region; The step 4 comprises the following steps: S4.1: Utilize gray scale threshold The image is binarized to obtain a binarized image The image is binarized to obtain a binarized image The bulge is segmented, and the calculation formula is as follows: ； represents a pixel value; represents the absolute value of a pixel value; wherein ; ; represents the pixel value of S4.2: Rebinarize the image Using connected component extraction, morphological filtering to get the image Finally, count the number of blobs and calculate their area. Statistical bubble number and calculate its area.

2. The method for detecting the surface bump of ceramic tile based on machine vision according to claim 1, characterized in that: In the step 2, Gaussian filtering is specifically as follows: A weighted average is applied to each pixel of the image using a template to reduce the grayscale difference between adjacent pixels. The Gaussian filter window size is 3×3. The row mean of the preprocessed image is then used. Column mean Compared with the overall mean In comparison, the area of ​​the ceramic tile to be tested was obtained. The calculation formula is as follows: wherein: is a pixel value of a red channel of the pixel; , respectively represent the number of rows and columns of the pixel matrix; , respectively represent the index of the pixel, and let the gray scale matrix of the region to be detected be , the size of which is , , respectively represent the number of rows and columns of the gray scale matrix.

Images