Chip defect detection method based on non-local mean filtering and adaptive difference

Abstract

The invention provides a non-local mean filtering and adaptive difference-based chip defect detection method, which comprises the following steps of: firstly, collecting image data of a chip to be detected through an industrial camera, and preprocessing to improve the image quality; and then, calculating a local entropy of each pixel to dynamically adjust a smoothing parameter of non-local mean filtering, and carrying out image denoising through the non-local mean filtering. Thirdly, a background image is obtained through a Gaussian filter, and a difference image is obtained through difference operation; and dynamically adjusting a difference threshold based on local frequency information, carrying out binarization processing on the difference image, and extracting a defect area. And finally, marking and classifying the detected defects through connected region analysis and contour detection. According to the method, by combining the self-adaptive non-local mean filtering of the local entropy and the self-adaptive difference threshold value method based on the local frequency, the denoising effect is improved, and the accuracy and robustness of defect detection are remarkably enhanced. According to the method, the limitation of a traditional chip defect detection method in a complex scene is overcome, and effective technical support is provided for chip quality control in actual industrial production.

Description

Technical Field

[0001] The invention belongs to the field of image detection, and in particular relates to a chip defect detection method based on non-local mean filtering and adaptive difference. Background Art

[0002] With the rapid development of semiconductor technology, chip manufacturing processes have become more and more complex, and the quality control requirements for chips have become higher and higher. Chip defect detection is one of the key links to ensure product quality. Although traditional chip defect detection methods, such as optical detection, electron microscope detection, and X-ray detection, can meet the detection needs to a certain extent, these methods often have problems such as high cost, slow detection speed, and strong dependence on operator experience. Therefore, automated defect detection methods based on image processing have gradually become mainstream. Image processing-based chip defect detection methods can efficiently and automatically detect various defects on the chip surface through computer vision and image processing technology. These methods usually include multiple steps such as image acquisition, preprocessing, feature extraction, defect detection, and classification. The application of image processing technology in chip defect detection not only improves detection efficiency, but also reduces manual intervention, making large-scale production and quality control more feasible.

[0003] However, in actual industrial production environments, due to factors such as changing lighting conditions, noise interference, image resolution limitations, etc., existing image processing-based defect detection methods still have deficiencies in precision and accuracy. Especially in high-noise environments, the noise in the image will seriously affect the effect of defect detection, leading to false detection and missed detection. Summary of the invention

[0004] The purpose of the present invention is to provide a chip defect detection method based on non-local mean filtering and adaptive difference, aiming to improve the accuracy and robustness of chip defect detection. By introducing local entropy to dynamically adjust the smoothing parameters of non-local mean filtering, more image details can be retained while effectively denoising. In addition, the local frequency information is used to dynamically adjust the difference threshold to improve the accuracy and robustness of defect detection. The technical solutions adopted are as follows: A chip defect detection method based on non-local mean filtering and adaptive difference includes the following steps: Collect image data of the chip to be tested; then perform preprocessing operations on the chip image; Obtain the denoised image: Calculate the local entropy of each pixel to dynamically adjust the smoothing parameters of the non-local mean filter; Use the adaptive smoothing parameters to perform non-local mean filtering to obtain the denoised image; Use Gaussian filter to obtain background image, and obtain difference image through difference operation; The differential threshold is dynamically adjusted based on the local frequency information, the differential image is binarized, and the defect area is extracted; Finally, the detected defects are marked and classified through connected region analysis and contour detection.

[0005] Preferably, an industrial camera is used to collect image data of the chip to be tested.

[0006] Preferably, preliminary denoising, brightness and contrast adjustment are performed on the collected images in sequence.

[0007] Preferably, the local entropy is calculated by a grayscale histogram within a neighborhood window.

[0008] Preferably, an adaptive smoothing parameter is used to perform non-local mean filtering, and image denoising is performed by calculating the weight between each pixel and other pixels within its search window.

[0009] Preferably, a Gaussian filter is used to filter the denoised image to obtain a background image.

[0010] Preferably, a difference operation is performed between the denoised image and the background image to obtain a difference image.

[0011] Preferably, the difference threshold is dynamically adjusted using local frequency information.

[0012] Preferably, the difference image is binarized to extract the defect area; after the binarization, the defect area is marked as white and the background area is marked as black.

[0013] Preferably, the detected defects are marked and classified by connected region analysis and contour detection; the connected region analysis is used to identify connected defect regions, and the contour detection is used to extract the boundaries of the defects, and further classify and mark the defects.

[0014] Compared with the prior art, the advantages of the present invention are: 1. Improve the non-local mean filtering method to effectively remove noise and retain image details.

[0015] By calculating the local entropy of each pixel, the smoothing parameters of the non-local mean filter are dynamically adjusted, so that the filtering process can better adapt to the noise characteristics of different regions. The non-local mean filter can retain the details and structural information in the image while removing noise, avoiding the loss of important features caused by over-smoothing.

[0016] 2. Improve the differential threshold method to improve the accuracy and robustness of defect detection.

[0017] The differential threshold is dynamically adjusted using local frequency information, so that the differential operation can better adapt to the noise characteristics of different regions, more accurately identify defective areas, reduce false detection and missed detection, and improve detection reliability.

[0018] In summary, by introducing adaptive smoothing parameter non-local mean filtering and adaptive differential threshold method based on local frequency, the precision and accuracy of chip defect detection are significantly improved. Compared with the prior art, the present invention has higher robustness, faster detection speed and lower manual intervention requirements, providing strong technical support for quality control in the chip manufacturing process. In addition, the method of the present invention also has wide applicability and flexibility, is suitable for a variety of application scenarios, and provides a new solution for image processing and defect detection in chips and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is an overall flow chart of the chip defect detection method based on non-local mean filtering and adaptive difference.

[0020] Figure 2 It is a schematic diagram of the improved non-local mean filtering method.

[0021] Figure 3 It is a schematic diagram of defect extraction based on improved differential threshold calculation. DETAILED DESCRIPTION

[0022] The chip defect detection method based on non-local mean filtering and adaptive difference of the present invention will be described in more detail below in conjunction with a schematic diagram, wherein a preferred embodiment of the present invention is shown, and it should be understood that those skilled in the art can modify the present invention described herein while still achieving the advantageous effects of the present invention. Therefore, the following description should be understood as being widely known to those skilled in the art, and not as a limitation of the present invention.

[0023] like Figure 1 As shown, the chip defect detection method based on non-local mean filtering and adaptive difference includes the following steps: Step 1: Data acquisition: Use an industrial camera to collect image data of the chip to be tested to ensure the clarity and integrity of the image for subsequent processing.

[0024] Step 2: Image preprocessing: Perform preliminary denoising, brightness and contrast adjustment on the collected images.

[0025] Preferably, the median filter method is used to preliminarily remove the noise in the image while retaining the edges and details of the image. The histogram stretching method is used to adjust the brightness of the image to ensure uniform brightness of the image and avoid false detection caused by uneven lighting. The linear transformation method is used to adjust the contrast of the image to enhance the visual effect of the image.

[0026] Step 3: Calculate local entropy: Calculate the local entropy of each pixel to dynamically adjust the smoothing parameters of the non-local mean filter to reflect the information complexity of the local area of ​​the image.

[0027] Specifically, the neighborhood window size is defined, and then the grayscale histogram within each pixel neighborhood window is calculated, the local entropy is calculated based on the grayscale histogram, and the local entropy value is stored in the corresponding pixel position.

[0028] Step 4: Calculate the adaptive smoothing parameter. The smoothing parameter is dynamically adjusted according to the local entropy calculated in the previous step, so that the non-local mean filter can better adapt to the noise characteristics of different regions.

[0029] Step 5: Non-local mean filtering: Use adaptive smoothing parameters to perform non-local mean filtering to obtain a denoised image.

[0030] Specifically, for each pixel, define a search window and a similarity window. Calculate the similarity (weight) between each pixel in the search window and the center pixel, and then calculate the new value of the center pixel by weighted average. Repeat the above steps until all pixels are processed.

[0031] Non-local means filtering can effectively remove noise while preserving image details and structural information.

[0032] Step 6: Obtain background image. Use a Gaussian filter to filter the denoised image to obtain a background image. The background image is used for subsequent differential operations to highlight defect areas.

[0033] Step 7: Difference operation. Perform a difference operation on the denoised image and the background image to obtain a difference image. The difference image reflects the area in the image that is different from the background, that is, the potential defect area.

[0034] Step 8: Calculate the adaptive differential threshold based on the local frequency.

[0035] The difference threshold is dynamically adjusted based on the local frequency information, the difference image is binarized, and the defect area is extracted.

[0036] Step 9: Binarization processing.

[0037] The difference image is binarized to extract the defect area. After the binarization, the defect area is marked as white and the background area is marked as black.

[0038] Step 10: Defect extraction and marking.

[0039] The detected defects are marked and classified through connected region analysis and contour detection.

[0040] Specifically, connected region analysis is used to identify connected defect regions, and contour detection is used to extract defect boundaries to further classify and mark defects.

[0041] Figure 2 Schematic diagram of the improved non-local mean filtering method of the present invention, i.e., steps 3 to 5, the specific steps include: Step A: local entropy calculation.

[0042] The local entropy of each pixel is calculated. The entropy value reflects the information complexity of the local area of ​​the image and is used for the subsequent dynamic adjustment of the smoothing parameters of the non-local mean filter.

[0043] Specifically, define the neighborhood window size, for each pixel in the image , select a defined neighborhood window around the pixel point, and then calculate the histogram of the pixel grayscale value in the neighborhood window.

[0044] Among them, in the grayscale histogram, the horizontal axis is the grayscale level and the vertical axis is the number of pixels.

[0045] Next, use the grayscale histogram to calculate the local entropy The local entropy is calculated by the following formula: (1) in, is the number of gray levels, Indicates grayscale The probability of occurrence is determined by calculating the ratio of the number of pixels of each gray level in the neighborhood window to the total number of pixels.

[0046] Specifically, a gray level in the neighborhood window The number of pixels is , the total number of pixels in the neighborhood window ,but: (2) Considering that the number of pixels of most gray levels is 0, only the probability of non-zero gray levels is calculated.

[0047] Then, the calculated local entropy Stored in a new matrix corresponding to the pixels at the same position in the original image.

[0048] Step B: Calculate the adaptive smoothing parameters.

[0049] The smoothing parameters are adjusted dynamically so that the non-local means filter can better adapt to the noise characteristics of different regions.

[0050] Smoothing parameter Dynamic adjustment based on the local entropy calculated in the previous step: (3) in, and It is user-defined parameters and can be adjusted according to actual needs; is the maximum value in the local entropy matrix.

[0051] Step C: non-local mean filtering.

[0052] Non-local means filtering is performed using adaptive smoothing parameters to effectively remove noise while maintaining image details and structural information.

[0053] Specifically, for each pixel in the image , define a search window, which needs to be much larger than the neighborhood window; A small area centered on the current pixel is defined as a similarity window, which is used to represent the local structure around the current pixel and is consistent with the size of the neighborhood window.

[0054] Every pixel in the search window , search for its similar window and the current pixel The similarity between similar windows is determined by comparing the pixel gray values ​​within two similar windows: (4) in, and Represents pixel points and Grayscale value of similarity window.

[0055] Then, update the pixels by weighted averaging Gray value: (5) in, Represents the set of all pixels within the search window.

[0056] Repeat the above steps until every pixel in the image has been processed.

[0057] Figure 3 This is a schematic diagram of defect extraction based on improved differential threshold calculation of the present invention, namely steps 6 to 10.

[0058] As shown in the figure, in order to effectively separate potential defects from the original image, it is first necessary to construct a clear and defect-free background image as a reference. Preferably, a Gaussian filter is used to filter the denoised image to obtain a background image. This provides a background reference for subsequent differential operations to highlight the defect area.

[0059] Specifically, a Gaussian filter kernel is created based on the selected filter parameters. Each element in the kernel is calculated using a Gaussian function, and the sum of all elements is normalized to ensure that the overall brightness of the filtered image remains unchanged. The created Gaussian filter kernel is slid pixel by pixel on the denoised image. For each pixel position, the pixel value in the neighborhood covered by the filter kernel is multiplied by the corresponding weight in the kernel, and the sum is calculated to obtain a new pixel value.

[0060] After Gaussian filtering, the entire image becomes smoother, reducing noise and small texture changes, thus obtaining a more ideal background image. This background image will serve as the basis for subsequent differential operations to better highlight those defect areas that are significantly different from the background.

[0061] The difference operation obtains the difference image by performing pixel-by-pixel subtraction between the denoised image and the background image: (6) in, represents the pixel value of the difference image at position (x, y), Indicates that the denoised image is at position The pixel value at Indicates that the background image is at position The pixel value at .

[0062] The difference values ​​of all pixels are combined into a new image, the difference image. The bright areas (high pixel values) in the difference image represent areas that are significantly different from the background image, which may be potential defects. The dark areas (low pixel values) in the difference image represent areas that are similar to the background image, which are usually normal background areas. By observing the difference image, the areas that are significantly different from the background image can be intuitively obtained, thus providing a basis for further defect detection and classification.

[0063] Then, for each pixel in the difference image , calculate its local frequency , i.e., the gray value change frequency within the neighborhood window. Preferably, the local frequency is obtained by calculating the frequency spectrum mean after obtaining the frequency spectrum by Fourier transform method.

[0064] Next, calculate the average value of all pixels in the neighborhood window of each pixel in the difference image and standard deviation .

[0065] (7) in, represents the size of the neighborhood window, and Relative to the center point The offset of

[0066] The difference threshold is dynamically adjusted according to the local frequency information, the difference image is binarized, and the defect area is extracted: (8) in, and It is an adjustable parameter.

[0067] For each pixel in the difference image , if its pixel value is greater than the difference threshold , then set its pixel value to white. Otherwise, set it to black.

[0068] Through the above steps, a binary image is generated, in which the defect area is marked as white and the background area is marked as black.

[0069] Furthermore, defects are extracted and marked.

[0070] First, each independent defect region in the binary image is identified through connected component analysis, and a unique identifier is assigned to each region.

[0071] Next, the Canny edge detection algorithm is used to extract the boundary contour of each defect area to determine the exact shape of the defect. Defects are classified according to their geometric properties (area, perimeter, aspect ratio, etc.).

[0072] Finally, on the original or processed image, a bounding box is drawn around the defect area and the type and number of the defect are annotated on the image.

[0073] The above is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any technician in the relevant technical field, without departing from the scope of the technical solution of the present invention, makes any form of equivalent replacement or modification to the technical solution and technical content disclosed in the present invention, which does not depart from the content of the technical solution of the present invention and still falls within the protection scope of the present invention.

Claims

1. A chip defect detection method based on non-local mean filtering and adaptive difference, characterized in that: The following steps are involved: Collect image data of the chip to be tested; then perform preprocessing operations on the chip image; Obtain the denoised image: Calculate the local entropy of each pixel to dynamically adjust the smoothing parameters of the non-local mean filter; Use the adaptive smoothing parameters to perform non-local mean filtering to obtain the denoised image; Use Gaussian filter to obtain background image, and obtain difference image through difference operation; The differential threshold is dynamically adjusted based on the local frequency information, the differential image is binarized, and the defect area is extracted; Finally, the detected defects are marked and classified through connected region analysis and contour detection.

2. The chip defect detection method based on non-local mean filtering and adaptive difference according to claim 1, characterized in that: Use industrial cameras to collect image data of the chip under test.

3. The chip defect detection method based on non-local mean filtering and adaptive difference according to claim 1, characterized in that: The collected images are subjected to preliminary denoising, brightness and contrast adjustment in turn.

4. The chip defect detection method based on non-local mean filtering and adaptive difference according to claim 1, characterized in that: The local entropy is calculated by the grayscale histogram within the neighborhood window.

5. The chip defect detection method based on non-local mean filtering and adaptive difference according to claim 4, characterized in that: Non-local means filtering is performed using adaptive smoothing parameters to denoise the image by calculating the weight between each pixel and other pixels within its search window.

6. The chip defect detection method based on non-local mean filtering and adaptive difference according to claim 1, characterized in that: The denoised image is filtered using a Gaussian filter to obtain the background image.

7. The chip defect detection method based on non-local mean filtering and adaptive difference according to claim 1, characterized in that: The denoised image is differentially computed from the background image to obtain a difference image.

8. The chip defect detection method based on non-local mean filtering and adaptive difference according to claim 1, characterized in that: The difference threshold is dynamically adjusted using local frequency information.

9. The chip defect detection method based on non-local mean filtering and adaptive difference according to claim 1, characterized in that: The difference image is binarized to extract the defect area; after the binarization, the defect area is marked as white and the background area is marked as black.

10. The chip defect detection method based on non-local mean filtering and adaptive difference according to claim 1, characterized in that: The detected defects are marked and classified through connected region analysis and contour detection; connected region analysis is used to identify connected defect regions, and contour detection is used to extract the boundaries of defects, and further classify and mark the defects.

Images