Self-adaptive multi-depth-of-field sub-pixel layer defect detection method and device

Abstract

The invention discloses a self-adaptive multi-depth-of-field sub-pixel layer defect detection method, which comprises the following steps of: acquiring pictures of a sub-pixel layer through an image acquisition module to obtain different depth-of-field images; the collected image is preprocessed based on an image preprocessing module, specifically, multiple filters are adopted to carry out noise reduction and smoothing processing on the input image, so that noise interference is eliminated, and the image quality is improved; a defect detection module is adopted to accurately extract edge features in images with different depth of field; calculating multi-dimensional feature information of the image through a feature calculation module; and designing defect evaluation indexes of human eye visual perception based on the obtained multi-dimensional feature information, and accurately positioning the clearest layer of the defect by adaptively adjusting the weight of each evaluation index and combining the depth information of the multi-depth-of-field image so as to determine the optimal distribution position of the defect in the sub-pixel layer. Meanwhile, a contrastive analysis mechanism consistent with human visual perception is introduced, and it is ensured that the detection result is highly consistent with human visual perception.

Description

Technical Field

[0001] The present invention relates to the field of high-precision defect detection in the electronic screen manufacturing industry, specifically an adaptive multi-depth-of-field sub-pixel layer defect detection method, which belongs to the specific application of traditional machine vision algorithms in industrial manufacturing, and particularly focuses on the detection requirements of the sub-pixel layer. With the rapid development of electronic display technology, especially the wide application of high-resolution LCD and OLED displays, the requirements for screen display quality are getting higher and higher. In the production process of display screens, the defect detection of the sub-pixel layer is a key link to ensure product quality and improve production efficiency. Background Art

[0002] Currently, traditional defect detection methods mainly rely on single-depth-of-field shooting technology, that is, a camera with a fixed depth of field is used to shoot the screen, and then machine vision algorithms are used for defect detection. However, this method has obvious limitations. Due to the complex physical structure of the sub-pixel layer, defects may appear in different depth layers, and single-depth-of-field shooting is difficult to comprehensively capture all defects, resulting in incomplete detection or false detection. In addition, existing algorithms lack targeted optimization for pictures with different depths of field, and the detection accuracy and efficiency are limited.

[0003] To overcome these technical problems, the present invention begins to explore multi-depth-of-field shooting technology. By shooting at different depth layers, defect information can be captured more comprehensively. However, how to effectively adapt to and process multi-depth-of-field pictures is still an urgent problem to be solved. Existing multi-depth-of-field detection schemes often use a unified algorithm to process all depth-of-field pictures, and fail to make full use of the unique information of each depth of field, resulting in the detection effect not meeting the expectations. Summary of the Invention

[0004] Aiming at the deficiencies of the existing technology, the present invention provides an adaptive multi-depth-of-field sub-pixel layer defect detection method. Through multi-depth-of-field image acquisition and an adaptive detection algorithm, this method can automatically identify and determine the clearest layer position of defects in the sub-pixel layer, thereby realizing accurate positioning and efficient detection of defects. The present invention significantly improves the accuracy and efficiency of defect detection, provides an innovative solution for product quality control and production efficiency improvement in the electronic screen manufacturing industry; in actual production, this technology can be widely applied to the manufacturing process of electronic screens, including but not limited to the defect detection of sub-pixel layers of LCD, OLED and other displays. Through comprehensive and accurate defect detection, this technology can effectively improve product quality, reduce the defective rate, optimize the production process, and reduce production costs. Therefore, the present invention has important application value in the electronic screen manufacturing industry, and provides strong technical support for improving the intelligent and automated level of industrial manufacturing.

[0005] To achieve the above object, the present invention provides the following technical solutions: An adaptive multi-depth-of-field sub-pixel layer defect detection method, comprising the following steps: Step S1: Collect pictures of the sub-pixel layer through an image acquisition module to obtain images with different depths of field; Step S2: Preprocess the collected images based on an image preprocessing module, specifically using a variety of filters to perform noise reduction and smoothing processing on the input images to eliminate noise interference and improve image quality; Step S3: Use a defect detection module to accurately extract edge features in images with different depths of field; Step S4: Calculate the multi-dimensional feature information of the image through a feature calculation module; Step S5: Design a defect evaluation index of human eye visual perception based on the above-obtained multi-dimensional feature information. By adaptively adjusting the weights of each evaluation index and combining the depth information of multi-depth-of-field images, accurately locate the clearest layer where the defect is located, thereby determining the optimal distribution position of the defect in the sub-pixel layer. At the same time, introduce a comparative analysis mechanism consistent with human eye visual perception to ensure that the detection results are highly consistent with human visual perception.

[0006] As a further solution of the present invention, in step S1, the image acquisition module uses a 12-megapixel black-and-white camera equipped with a telecentric lens to take pictures of 5 different depths of field in the sub-pixel layer.

[0007] As a further solution of the present invention, in step S2: specifically, by introducing a Fourier transform filter, a guided filter, a Gaussian filter, a median filter, and a bilateral filter, and adaptively selecting the optimal filter combination to effectively retain image details while removing noise.

[0008] As a further solution of the present invention, in step S3: use a multi-scale edge detection algorithm, combined with the Canny operator, the Sobel operator, the Prewitt operator, and the Laplacian operator, to extract edge features in images with different depths of field. Among them, the Canny operator is used to detect high-contrast edges, the Sobel operator and the Prewitt operator enhance the directionality of the edges, and the Laplacian operator is used to detect the sharpness of the edges. After applying the Gaussian filter to smooth the image, calculate the differences in the horizontal and vertical directions to enhance the directionality and clarity of the edges.

[0009] As a further solution of the present invention, in step S4, the multi-dimensional feature information of the image includes standard deviation, gradient magnitude, and Laplacian response, and the texture complexity of the image region is characterized by the standard deviation: , where, N is the number of overall data, μ is the overall average value, is the idata points; the gradient amplitude reflects the strength of the image edge: Horizontal gradient: ; Vertical Gradient: ; Gradient Magnitude: ; in, Gx : Indicates the horizontal gradient of the image, reflecting the rate of change of the brightness of the image in the horizontal direction. Gy : Indicates the vertical gradient of the image, reflecting the rate of change of the brightness of the image in the vertical direction. G : represents the gradient amplitude, that is, the length of the gradient vector, which comprehensively reflects the degree of brightness change of the image in the horizontal and vertical directions at that point; The Laplace response is used to detect the sharpness of the image.

[0010] in: L ( x , y ): indicates that at coordinates (x, y) ( x , y ), which is used to measure the edge strength of the point; f(x+1,y): represents the x The gray value after moving one pixel to the right in the direction; f(x−1,y): represents the x The gray value after shifting one pixel to the left in the direction; f(x,y+1): indicates the y The gray value after moving up one pixel in the direction; f(x,y−1): represents the y The gray value after moving down one pixel in the direction; 4f(x,y): represents the current pixel (x,y) ( x , y ) is four times the grayscale value at .

[0011] As a further solution of the present invention, in step S5, the defect evaluation index of human visual perception is designed based on the obtained multi-dimensional feature information, specifically: the edge sharpness is evaluated by calculating the gradient amplitude and Laplace response of the edge. The larger the gradient amplitude and Laplace value, the more drastic the brightness change near the point, and the edge is more obvious and sharper. Conversely, the brightness change near the point is gentle and the edge is not obvious enough. The standard deviation is used to measure the distribution of pixel values in the image. A large standard deviation indicates that the pixel values in the image change greatly, indicating that there are more details or textures in the image. A small standard deviation indicates that the pixel values change more slowly and the image appears more uniform or blurred.

[0012] As a further solution of the present invention, in step S5, by adaptively adjusting the weights of each evaluation index and combining the depth information of multi-depth-of-field images, the clearest layer where the defect is located is accurately positioned, so as to determine the optimal distribution position of the defect in the sub-pixel layer. At the same time, a contrast analysis mechanism consistent with human visual perception is introduced to ensure that the detection result has a high consistency with human visual perception. The specific process includes: proposing a human-eye perception-based weight adjustment formula as: . Among them, is the weight of the i th evaluation index, α and β are adaptive parameters, S i is the multi-dimensional feature information extracted by the feature calculation module, D i is the depth information of the multi-depth-of-field image.

[0013] The present invention also provides a sub-pixel layer defect detection device with adaptive multi-depth-of-field, including an image acquisition module for acquiring pictures of the sub-pixel layer to obtain images with different depths of field; an image preprocessing module for denoising and smoothing the input image by using a variety of filters to eliminate noise interference and improve the image quality; a defect detection module for accurately extracting edge features in images with different depths of field; a feature calculation module for calculating the multi-dimensional feature information of the image; an adaptive evaluation algorithm module for accurately positioning the clearest layer where the defect is located by adaptively adjusting the weights of each evaluation index and combining the depth information of the multi-depth-of-field image, so as to determine the optimal distribution position of the defect in the sub-pixel layer. At the same time, a contrast analysis mechanism consistent with human visual perception is introduced to ensure that the detection result has a high consistency with human visual perception.

[0014] The present invention has the following beneficial effects: 1) Improve the defect detection accuracy: The present invention combines the multi-depth-of-field shooting technology with the adaptive algorithm to comprehensively capture the defect details of the sub-pixel layer, and accurately position the clearest layer where the defect is located through the adaptive evaluation algorithm. This method is based on human visual perception, ensuring that the detection result has a high consistency with human visual perception, and significantly improving the accuracy and reliability of defect detection.

[0015] 2) Improve the detection efficiency: The present invention realizes the fast and accurate detection of the sub-pixel layer picture through multi-depth-of-field shooting and the adaptive detection algorithm. Compared with the traditional single-depth-of-field shooting method, this technology can automatically complete defect detection and evaluation, greatly reducing the need for manual detection and significantly improving the detection efficiency, providing efficient technical support for product quality control in the electronic screen manufacturing industry.

[0016] 3) Enhance robustness and adaptability: The adaptive algorithm of the present invention can automatically adjust the detection parameters and the weights of evaluation metrics according to the depth information and feature changes of the image, adapting to changes in different depth of field and image conditions. This method is not only applicable to standard sub-pixel layer defect detection, but also shows strong adaptability in different manufacturing environments and product types, with wide applicability and robustness.

[0017] To more clearly elaborate on the structural features and effects of the present invention, the following will combine the drawings with specific embodiments to describe the present invention in detail. Description of the Drawings

[0018] Figure 1 It is a schematic diagram of the hierarchical shooting of the camera mentioned in the present invention.

[0019] Figures 2a - 2e respectively are Figure 1 Schematic diagrams of the imaging effects of the CG layer (color filter layer), DITO layer (double-layer ITO conductive layer), CF layer (common electrode layer), Bottom Pol layer (lower polarizer layer), and Top Diffuser layer (diffusion layer) at different depth of field positions.

[0020] Figure 3 It is the overall algorithm flow chart mentioned in the present invention.

[0021] Figures 4a - 4c It is the image preprocessing process diagram mentioned in the present invention.

[0022] Figure 5a It shows the original image of the 1.4 layer (Top Diffuser layer), including various potential defect schematic diagrams.

[0023] Figure 5b It shows the Figure 5a defect contour schematic diagram extracted by the algorithm.

[0024] Figure 6 It is the schematic diagram of the comparison between the clearest layer and the AI visual area mentioned in the present invention. Detailed Embodiments

[0025] The following will further illustrate the present invention in combination with the drawings and relevant knowledge, and describe it clearly and completely. Obviously, the described applications are only a part of the embodiments of the present invention, rather than all the embodiments.

[0026] An adaptive multi-depth-of-field sub-pixel layer defect detection method of the present invention can automatically identify and determine the clearest layer position of defects in the sub-pixel layer through multi-depth-of-field image acquisition and an adaptive detection algorithm, so as to achieve precise positioning and efficient detection of defects. The present invention significantly improves the accuracy and efficiency of defect detection, and provides an innovative solution for product quality control and production efficiency improvement in the electronic screen manufacturing industry. The core idea of the present invention is to achieve comprehensive and precise detection of the sub-pixel layer image through multi-depth-of-field shooting technology combined with an adaptive detection algorithm.

[0027] The present invention provides an adaptive multi-depth-of-field sub-pixel layer defect detection method, including the following steps: Step S1: Collect pictures of the sub-pixel layer through an image acquisition module to obtain images with different depths of field; specifically: use a 12-megapixel black-and-white camera equipped with a telecentric lens to take 5 pictures with different depths of field on the sub-pixel layer (as Figure 1 shown), and the image effects (as Figures 2a - 2e shown) cover different depth regions, so as to ensure comprehensive capture of defect details; in Figure 1 , label 1 represents the first layer for shooting the CG layer, label 2 represents the second layer for shooting the DITO layer, label 3 represents the third layer for shooting the CF layer, label 4 represents the fourth layer for shooting the Bottom POL layer, label 5 represents the fifth layer for shooting the Top Diffuser layer, label 6 is the upper polarizer + color filter layer, and label 7 is the thin film transistor + lower polarizer layer. And in Figure 2a is a schematic diagram of the shooting effect of CG, Figure 2b is a schematic diagram of the shooting effect of DITO, Figure 2c is a schematic diagram of the shooting effect of CF, Figure 2d is a schematic diagram of the shooting effect of Bottom Pol, Figure 2e is a schematic diagram of the shooting effect of Top Diffuser.

[0028] In the present invention, a 12-megapixel black-and-white camera is used with a telecentric lens. The telecentric lens can reduce image distortion caused by the viewing angle problem, ensure that the collected images have high precision and accuracy, and provide high-quality original data for subsequent analysis; and take 5 pictures with different depths of field on the sub-pixel layer. By setting different depths of field, the captured images can cover different depth regions of the sub-pixel layer to comprehensively capture the possible defect details in the sub-pixel layer and avoid missing defects at different depth positions.

[0029] Step S2: Preprocess the acquired image based on the image preprocessing module. Specifically, multiple filters are used to perform noise reduction and smoothing on the input image to eliminate noise interference and improve image quality. Specifically, it is the adaptive detection algorithm processing stage (this module is the core part of this solution). First, in the image preprocessing module, multiple filters are used to perform noise reduction and smoothing on the input image to eliminate noise interference and improve image quality. This module introduces Fourier transform filters, guided filters, Gaussian filters, median filters, and bilateral filters, etc. By adaptively selecting the optimal filter combination, noise interference can be removed while effectively retaining image details, improving image quality, and providing a clear image basis for subsequent defect detection. This module is the core part of image preprocessing. By adaptively selecting the filter combination, it can automatically adjust the parameters and weights of the filters according to the characteristics and noise distribution of different images to achieve the best noise reduction and smoothing effects.

[0030] Step S3: Precisely extract the edge features in images with different depths of field using the defect detection module. Specifically, in the defect detection module, a multi-scale edge detection algorithm is adopted, combined with multiple edge detection operators such as the Canny operator, Sobel operator, Prewitt operator, and Laplacian operator, to extract the edge features in the image. This module realizes the precise extraction of edge features in images with different depths of field by adaptively adjusting the edge detection parameters and combining the depth information of the sub-pixel layer, thus providing high-quality edge information for subsequent defect detection.

[0031] In the present invention, by combining multiple operators, the edge features in the image can be more comprehensively extracted. Further, this module can automatically adjust the parameters of the operators, such as thresholds, convolution kernel sizes, etc., according to the characteristics and edge distributions of images with different depths of field by adaptively adjusting the edge detection parameters and combining the depth information of the sub-pixel layer, so as to achieve the precise extraction of edge features in images with different depths of field. This can provide high-quality edge information for subsequent defect detection and accurately locate the edge positions of defects.

[0032] Step S4: Calculate the multi-dimensional feature information of the image through the feature calculation module. Specifically, in the feature calculation module, the multi-dimensional feature information of the image is extracted, including standard deviation, gradient magnitude, Laplacian response, texture features, etc. These features can comprehensively characterize the local characteristics of the image and provide rich criteria for defect detection.

[0033] Specifically, in the feature calculation module, multi-dimensional feature information of the image is extracted, including standard deviation, gradient magnitude, Laplacian response, texture features, etc. The standard deviation can reflect the degree of dispersion of the image gray values, the gradient magnitude can represent the change rate of the pixels in the image, the Laplacian response is sensitive to features such as edges and corners in the image, and the texture features can describe the local texture structure of the image. These features can comprehensively characterize the local characteristics of the image, providing rich criteria for defect detection; by calculating these multi-dimensional feature information, the defect features in the image can be described from different angles. For example, defects may cause large changes in the local gray values of the image, resulting in an increase in the standard deviation and gradient magnitude; or they may cause changes in the texture structure of the image, and these changes can be detected through texture features.

[0034] Step S5: Design a defect evaluation index based on human eye visual perception based on the above-obtained multi-dimensional feature information. By adaptively adjusting the weights of each evaluation index and combining the depth information of the multi-depth images, accurately locate the clearest layer where the defect is located, so as to determine the optimal distribution position of the defect in the sub-pixel layer. At the same time, introduce a comparative analysis mechanism consistent with human eye visual perception to ensure that the detection result has a high degree of consistency with human visual perception. Specifically: In the adaptive evaluation algorithm system, a set of defect evaluation indexes based on human eye visual perception are designed, focusing on key factors such as edge sharpness, clarity, contrast, and uniformity. This evaluation system can accurately locate the clearest layer where the defect is located by adaptively adjusting the weights of each evaluation index and combining the depth information of the multi-depth images, so as to determine the optimal distribution position of the defect in the sub-pixel layer. At the same time, the evaluation algorithm introduces a comparative analysis mechanism consistent with human eye visual perception to ensure that the detection result has a high degree of consistency with human visual perception, thereby improving the reliability and practicality of the detection result. All in all, the proposed method significantly improves the accuracy and efficiency of defect detection, providing reliable technical support for quality control in the electronic screen manufacturing industry.

[0035] In the present invention, in the adaptive evaluation algorithm system, a set of defect evaluation indexes based on human eye visual perception are designed, focusing on key factors such as edge sharpness, clarity, contrast, and uniformity. Edge sharpness can reflect the clarity of the defect edge, clarity represents the overall clarity of the image, contrast can reflect the gray difference between different regions in the image, and uniformity describes the uniformity of the image gray distribution. By comprehensively considering these factors, the severity and position of the defect can be evaluated more accurately.

[0036] By adaptively adjusting the weights of various evaluation indicators and combining the depth information of multi-depth images, this evaluation system can automatically adjust the importance of each indicator according to the characteristics and defect distributions of different images. For example, for some defects with obvious edge features, the weight of edge sharpness may be relatively high; while for some defects that affect the overall clarity of the image, the weight of clarity will increase accordingly. This can accurately locate the clearest layer where the defect is located, thereby determining the optimal distribution position of the defect in the sub-pixel layer.

[0037] Furthermore, the evaluation algorithm introduces a comparative analysis mechanism consistent with human visual perception, and conducts a comparative analysis of the detection results with human visual perception. By simulating the human eye's perception process of images, the detected defects are further verified and adjusted to ensure that the detection results are highly consistent with human visual perception. This can improve the reliability and practicality of the detection results, make the detection results more in line with the actual situation, and facilitate operators to make judgments and handle them.

[0038] The method of the present invention significantly improves the accuracy and efficiency of defect detection in the sub-pixel layer of electronic screens through a series of steps such as image acquisition, preprocessing, edge feature extraction, feature calculation, and defect evaluation and positioning, providing reliable technical support for quality control in the electronic screen manufacturing industry.

[0039] Further preferably, referring to Figures 4a - 4c As shown, the image preprocessing module. The main purpose of the image preprocessing module is to perform noise reduction and smoothing processing on the input multi-depth image while retaining the detail information of the image. This module uses a variety of filters to process the image, including but not limited to the following: Fourier transform filter, guided filter, Gaussian filter, median filter, and bilateral filter, etc. In the adaptive image preprocessing module of the present invention, in the image preprocessing stage, an adaptive combination of a variety of filters (such as Fourier transform filter, guided filter, Gaussian filter, etc.) is used to effectively remove noise and improve the image quality while retaining the image detail information, providing high-quality input for subsequent defect detection. Referring to Figure 4a As shown, it is the original image of the Top Diffuser layer, Figure 4b is the process diagram, Figure 4c is the preprocessed picture.

[0040] In the present invention, the Fourier transform filter: can transform an image from the spatial domain to the frequency domain. In the frequency domain, it is convenient to distinguish the low-frequency components (such as the overall structure and shape) and high-frequency components (such as edges, textures, and noises) of the image. By designing different frequency-domain filters, such as a low-pass filter that can filter out high-frequency noises, and a high-pass filter that can enhance details such as edges, the noise reduction and detail enhancement of the image can be achieved. The guided filter: is an edge-preserving filter that can use a guiding image to guide the filtering process. While smoothing the image, it can well preserve the edge and detail information of the image, and is particularly suitable for preprocessing images with complex textures and edge structures. The Gaussian filter: filters the image using the Gaussian function, which can effectively remove Gaussian noises in the image, smooth the image, and to a certain extent, preserve the edge information of the image. Because of the characteristics of the Gaussian function, while smoothing the noises, it will not overly blur the details of the image. The median filter: belongs to a non-linear filter that replaces the value of each pixel point with the median of the pixel values in its neighborhood. This filter is very effective for removing impulse noises such as salt-and-pepper noises, and can well preserve the edges and sharpness of the image, avoiding the edge blurring problem that may occur in the denoising process of linear filters. The bilateral filter: combines the spatial proximity and pixel value similarity of the image for filtering. It not only considers the spatial distance between pixels, but also the difference in pixel values. Therefore, it can remove noises while well preserving the edges and details of the image, so that the edge part of the image will not be overly smoothed.

[0041] The advantages of the adaptive combination of multiple filters in the present invention: Different filters have different processing effects on different types of noises and image details. For example, the Gaussian filter has a good effect on Gaussian noises, but may blur details; the median filter is effective for salt-and-pepper noises and can preserve edges, but has a weak processing ability for Gaussian noises. By adaptively combining these filters, according to the specific characteristics and noise conditions of the image, the most suitable filter or filter combination can be automatically selected for processing, so as to effectively remove noises while maximizing the preservation of the detail information of the image, providing a high-quality input image for subsequent defect detection. For example, for an image with a lot of noises and rich edge details, a combination of the median filter and the bilateral filter may be adaptively selected; for an image containing Gaussian noises and requiring the preservation of certain low-frequency information, a combination of the Gaussian filter and the Fourier transform filter may be selected, etc.

[0042] Further preferably, referring to Figure 5a and Figure 5b As shown, the goal of the multi-scale defect edge detection module is to extract the edge features in the image for subsequent defect detection. This algorithm is implemented by combining multiple edge detection operators and filtering difference techniques.

[0043] The following is a detailed description of the algorithm and its advantages: Apply the Canny operator, Sobel operator, Prewitt operator, and Laplacian operator at different scales. The Canny operator is used to detect high-contrast edges, the Sobel operator and Prewitt operator enhance the directionality of edges, and the Laplacian operator is used to detect the sharpness of edges. Through multi-scale processing, edge features of different sizes and directions are captured. The filtering difference technique is to calculate the differences in the horizontal and vertical directions after applying a Gaussian filter to smooth the image, so as to enhance the directionality and clarity of edges. In addition, a bilateral filter is used for further smoothing while retaining edge details. The advantages of this method are: 1) Complementary: The Canny operator performs excellently in images with less noise, while the Sobel operator and Prewitt operator are more robust in images with more noise. The Laplacian operator is sensitive to the sharpness of edges and is suitable for detecting high-contrast edges. Combining these operators can comprehensively capture different types of edge features. 2) Accuracy: The combination of multiple operators improves the accuracy of edge detection and reduces the errors that may occur under specific conditions with a single operator. 3) Robustness: Under different depth-of-field and lighting conditions, multi-scale processing can maintain the stability of detection.

[0044] In the present invention, it should be noted that different operators have their own advantages under different image conditions. By combining them, various types of edge features can be comprehensively captured, and edges can be better detected regardless of the image noise level; multiple operators complement and verify each other, reducing the errors generated by a single operator due to its own limitations under specific conditions, thereby improving the accuracy of edge detection and making the detection results closer to the real edges; and multi-scale processing can capture more stable edge features by analyzing the image at different scales under different depth-of-field and lighting conditions, reducing the impact of environmental changes on the detection results and maintaining the stability of detection.

[0045] In the present invention, the process of calculating the multi-dimensional feature information of an image by the feature calculation module specifically includes: The main task of the feature calculation module is to extract the multi-dimensional feature information of the image, including standard deviation, gradient magnitude, Laplacian response, etc.

[0046] 1) The standard deviation characterizes the texture complexity of the image region:

[0047] Among them, N is the number of overall data, μ is the overall average value, is the i th data point.

[0048] 2) The gradient magnitude reflects the intensity of the image edge Horizontal gradient:

[0049] Vertical gradient:

[0050] Gradient magnitude:

[0051] Wherein, Gx : represents the gradient of the image in the horizontal direction (x-axis direction), reflecting the brightness change rate of the image in the horizontal direction. Gy : represents the gradient of the image in the vertical direction (y-axis direction), reflecting the brightness change rate of the image in the vertical direction. G : represents the gradient magnitude, that is, the length of the gradient vector, comprehensively reflecting the brightness change degree of the image in the horizontal and vertical directions at this point.

[0052] 3) The Laplacian response is used to detect the sharpness of the image

[0053] Wherein: L ( x , y ): represents the Laplacian response value at the coordinate (x, y) ( x , y ), used to measure the edge strength at this point.

[0054] f(x + 1, y): represents the gray value after shifting one pixel to the right in the x x direction.

[0055] f(x - 1, y): represents the gray value after shifting one pixel to the left in the x x direction.

[0056] f(x, y + 1): represents the gray value after shifting one pixel up in the y y direction.

[0057] f(x, y - 1): represents the gray value after shifting one pixel down in the y y direction.

[0058] 4f(x, y): represents four times the gray value at the current pixel (x, y) ( x , y ).

[0059] In the present invention, the adaptive evaluation algorithm module is the core of the entire method, which is used to evaluate and classify defects in different sub-pixel layers. This module designs a set of defect evaluation metrics based on human visual perception, with key considerations given to factors such as edge sharpness and image contrast. The edge sharpness is evaluated by calculating the gradient magnitude and Laplacian response of the edge. The larger the gradient magnitude and Laplacian value, the more drastic the brightness change near that point, indicating a more obvious and sharp edge. Conversely, if the brightness change near that point is gentle, the edge is not obvious enough. The standard deviation can be used to measure the distribution of pixel values in the image. A large standard deviation indicates a large variation in pixel values in the image, meaning there are more details or textures in the image. A small standard deviation indicates a relatively gentle change in pixel values, and the image appears more uniform or blurred. In the multi-dimensional feature extraction and adaptive evaluation system of the present invention, multi-dimensional feature information of the image (such as standard deviation, gradient magnitude, Laplacian response, texture features, etc.) is extracted, and a set of adaptive evaluation algorithms based on human visual perception is designed. This algorithm accurately locates the clearest layer where the defect is located by adaptively adjusting the weights of each evaluation metric and combining the depth information of multi-depth images, ensuring that the detection results are highly consistent with human visual perception. In the adaptive evaluation mechanism based on human visual perception of the present invention, a comparative analysis mechanism consistent with human visual perception is introduced during the defect evaluation stage, with key factors such as edge sharpness, clarity, contrast, and uniformity being considered, ensuring the reliability and practicality of the detection results.

[0060] Furthermore, it should be noted that this algorithm can accurately locate the clearest layer where the defect is located by adaptively adjusting the weights of each evaluation metric and combining the depth information of multi-depth images. This means that the optimal distribution position of the defect in the sub-pixel layer can be accurately determined, avoiding misjudgment or omission of defects caused by different image depths. Additionally, a comparative analysis mechanism consistent with human visual perception is introduced, fully considering key factors such as edge sharpness, clarity, contrast, and uniformity, making the detection results highly consistent with human visual perception, thereby enhancing the reliability and practicality of the detection results and better meeting the requirements for defect detection in actual application scenarios.

[0061] In the present invention, referring to Figure 6 as shown, the schematic diagram of the analysis of the consistency between the clearest layer selected by this module and AI. Specifically, the evaluation algorithm realizes the accurate identification and classification of defects in different sub-pixel layers by adaptively adjusting the weights of each evaluation metric and combining the depth information of multi-depth images. First, the weight adjustment formula based on human perception is proposed as: . Among them, is the weight of the i th evaluation metric, α and β are adaptive parameters, S iare the feature values extracted by the feature calculation module (such as standard deviation, gradient magnitude, Laplacian response, etc.), D i is the depth information of the multi-depth-of-field image, with values of 0.2, 0.5, 0.8, 1.1, 1.4, corresponding to the CG layer, DITO layer, CF layer, Bottom Pol layer, and TopDiffuser layer respectively. Specifically, when the edge sharpness of the image is relatively high (i.e., the gradient magnitude is large) in a certain layer, and the image contrast is relatively high (i.e., the standard deviation is large), then the weight value of this layer will also be large. By comparing the weight values of the 5 layers, the optimal distribution position of the defect in the sub-pixel layer can be determined. And determine which depth-of-field layer is the clearest.

[0062] The present invention combines multi-depth-of-field shooting with an adaptive algorithm. By using the multi-depth-of-field shooting technology, sub-pixel layer images with different depths of field are obtained, and combined with the adaptive algorithm, precise positioning and detection of defects are realized. This method can automatically determine the position of the defect in the clearest layer, significantly improving the accuracy and efficiency of detection.

[0063] In addition, in the present invention, multi-scale edge detection is combined with depth information. In the defect detection stage, a multi-scale edge detection algorithm is used, combined with various edge detection operators (such as Canny, Sobel, Prewitt, etc.), and depth information is introduced to adaptively adjust the detection parameters to achieve precise extraction of edge features in images with different depths of field.

[0064] The following provides specific embodiments of the present invention Embodiment 1, a method for detecting sub-pixel layer defects with adaptive multi-depth of field, referring to Figure 3 the overall algorithm flow chart shown; specifically, images of the sub-pixel layer are collected by an image acquisition module to obtain images with different depths of field. Figure 1 shows a schematic diagram of the camera taking multi-depth-of-field pictures of the sub-pixel layer of an electronic screen, respectively showing the imaging effects of the CG layer (color filter layer), DITO layer (double-layer ITO conductive layer), CF layer (common electrode layer), Bottom Pol layer (lower polarizer layer), and TopDiffuser layer (diffusion layer) at different depths of field. Through the multi-depth-of-field shooting technology, the optical characteristics and potential defects of each layer can be clearly captured. Specifically, different depths of field correspond to the depth positions of different sub-pixel layers, and can comprehensively cover all key levels of the screen structure. This schematic diagram intuitively shows the application of multi-depth-of-field shooting in sub-pixel layer defect detection, providing multi-dimensional image data support for the subsequent adaptive algorithm, thereby realizing precise positioning and classification of defects. This multi-depth-of-field layered shooting method can effectively solve the limitations of traditional single-depth-of-field detection methods, significantly improving the comprehensiveness and accuracy of defect detection.

[0065] Based on the image preprocessing module, the collected images are preprocessed. Specifically, multiple filters are used to denoise and smooth the input images to eliminate noise interference and improve image quality. Refer to Figures 2a - 2e As shown, it demonstrates the imaging effect of the product under five-layer depth of field shooting, intuitively presenting the optical characteristics and imaging quality of the sub-pixel layer at different depth positions. Through the multi-depth of field shooting technology, the imaging differences of the CG layer, DITO layer, CF layer, Bottom Pol layer, and TopDiffuser layer at different depth positions can be clearly observed. The imaging effect of each layer reflects its optical characteristics and potential defects. For example, the clarity of the CG layer, the edge features of the DITO layer, the color uniformity of the CF layer, the structural integrity of the Bottom Pol layer, and the light diffusion characteristics of the Top Diffuser layer. These multi-depth of field images provide rich image data support for the subsequent adaptive defect detection algorithm, enabling the algorithm to accurately locate and classify the defects in the sub-pixel layer by combining depth information and image features. This figure intuitively demonstrates the application value of the multi-depth of field shooting technology in sub-pixel layer defect detection, providing strong technical support for quality control in the electronic screen manufacturing industry.

[0066] Refer to Figure 4 as shown, which demonstrates the schematic diagram of the image preprocessing process, showing the combined application of multiple filters and their processing effects on the image. In the image preprocessing stage, the system uses multiple filters (such as Fourier transform filter, guided filter, Gaussian filter, median filter, and bilateral filter, etc.). By adaptively selecting the optimal filter combination, the input multi-depth of field images are denoised and smoothed. Specifically, aiming at the characteristic differences of different depth of field images, we first analyze the noise type, texture features, and edge information of each layer of the image. For example, the Top Diffuser layer is mainly affected by the complex background texture, so a combination of median filter and guided filter is used to better remove the background texture and reduce noise. In addition to the complex background texture, the CF layer also has Gaussian noise, so a combination of Gaussian filter and bilateral filter is selected to effectively remove the noise while retaining the image details, providing a high-quality image basis for subsequent defect detection.

[0067] These filters work together to effectively remove the noise interference in the image and retain the detailed information of the image, thereby improving the overall quality of the image and the accuracy of subsequent defect detection. Figures 4a - 4c It intuitively demonstrates the application effects of different filters in the preprocessing process, reflecting the important role of the preprocessing module in optimizing image quality and providing high-quality input for subsequent defect detection.

[0068] Refer to Figure 5a 、 Figure 5b As shown, it demonstrates the results of defect extraction, where Figure 5aThe original image of 1.4 layers is shown, containing various potential defects. Figure 5b The defect contours extracted by the algorithm are clearly shown. The method of the present invention uses the multi-depth-of-field shooting technique to obtain image information at different depth-of-field positions, and combines an adaptive algorithm to preprocess the images, detect edges and extract features. Through these steps, the system can accurately identify and locate defects and capture the detailed contours of the defects.

[0069] Refer to Figure 6 As shown, the comparison results between the clearest layer and the AI visual inspection area are presented. Through 10 pictures, it intuitively shows the high consistency between the defect detection results of the algorithm of the present invention and the results of manual visual inspection, proving the reliability of the algorithm of the present invention in terms of defect detection accuracy.

[0070] The technical principle of the present invention has been described above in combination with specific embodiments, which are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. Those skilled in the art can easily think of other specific embodiments of the present invention without creative labor, and these embodiments will fall within the protection scope of the present invention.

Claims

1. An adaptive multi-depth-of-field sub-pixel layer defect detection method, characterized in that The following steps are involved: Step S1: The image acquisition module is used to acquire images of the sub-pixel layer to obtain images with different depths of field; Step S2: preprocessing the collected images with different depths of field based on the image preprocessing module, specifically using a variety of filters to perform noise reduction and smoothing on the input images with different depths of field to eliminate noise interference and improve image quality; Step S3: using a defect detection module to accurately extract edge features from pre-processed images of different depths of field; Step S4: Calculating multi-dimensional feature information of the image through a feature calculation module based on the extracted edge features; Step S5: Based on the multi-dimensional feature information obtained above, defect evaluation indicators of human visual perception are designed. By adaptively adjusting the weights of each evaluation indicator and combining the depth information of the multi-depth-of-field image, the clearest layer where the defect is located is accurately located, thereby determining the optimal distribution position of the defect in the sub-pixel layer. At the same time, a comparative analysis mechanism consistent with human visual perception is introduced to ensure that the detection results are highly consistent with human visual perception.

2. The sub-pixel layer defect detection method with adaptive multi-depth of field according to claim 1, wherein, In step S1, the image acquisition module uses a 12-megapixel black-and-white camera configured with a telecentric lens to take pictures with five different depths of field at the sub-pixel layer.

3. The sub-pixel layer defect detection method with adaptive multi-depth of field according to claim 2, characterized in that In step S2: specifically, by introducing Fourier transform filter, guided filter, Gaussian filter, median filter and bilateral filter, the optimal filter combination is adaptively selected to effectively retain image details while removing noise.

4. The sub-pixel layer defect detection method with adaptive multi-depth of field according to claim 3, characterized in that In step S3: a multi-scale edge detection algorithm is used to extract edge features in images of different depths of field by combining the Canny operator, the Sobel operator, the Prewitt operator and the Laplacian operator, wherein the Canny operator is used to detect high-contrast edges, the Sobel operator and the Prewitt operator enhance the directionality of the edge, and the Laplacian operator is used to detect the sharpness of the edge. After applying a Gaussian filter to smooth the image, the horizontal and vertical differences are calculated to enhance the directionality and clarity of the edge.

5. The sub-pixel layer defect detection method with adaptive multi-depth of field according to claim 4, wherein In the step S4, the multi-dimensional feature information of the image, including the standard deviation, gradient magnitude, and Laplacian response, characterizes the texture complexity of the image region through the standard deviation: , where N is the number of the overall data, μ is the overall average value, is the i th data point; the intensity of the image edge is reflected by the gradient magnitude: Horizontal gradient: ; Vertical gradient: ; Gradient magnitude: ; Among them, Gx represents the gradient of the image in the horizontal direction, reflecting the brightness change rate of the image in the horizontal direction, Gy represents the gradient of the image in the vertical direction, reflecting the brightness change rate of the image in the vertical direction, G represents the gradient magnitude, that is, the length of the gradient vector, comprehensively reflecting the brightness change degree of the image in the horizontal and vertical directions at this point; The Laplace response is used to detect the sharpness of the image; ； Wherein: L ( x , y ): represents the Laplacian response value at the coordinate (x, y)( x , y ), which is used to measure the edge strength of this point; f(x + 1, y): represents the gray value after shifting one pixel to the right in the x x direction; f(x−1,y): represents the grayscale value after shifting one pixel to the left in the x x direction; f(x,y+1): represents the gray value after shifting one pixel upward in the y y direction; f(x,y−1): represents the gray value after shifting one pixel downward in the y y direction; 4f(x,y): represents four times the gray value at the current pixel (x,y)( x , y ).

6. The method for detecting sub-pixel layer defects with adaptive multi-depth of field according to claim 5, characterized in that, In the step S5, the defect evaluation index of human visual perception is designed based on the obtained multi-dimensional feature information. Specifically, the edge sharpness is evaluated by calculating the gradient amplitude and Laplace response of the edge. The larger the gradient amplitude and the Laplace value, the more drastic the brightness change near the point, and the edge is more obvious and sharper. On the contrary, the brightness change near the point is gentle and the edge is not obvious enough. The standard deviation is used to measure the distribution of pixel values in the image. A large standard deviation indicates that the pixel values in the image change greatly, indicating that there are more details or textures in the image. A small standard deviation indicates that the pixel values change more slowly, and the image appears more uniform or blurred.

7. The method for detecting sub-pixel layer defects with adaptive multi-depth of field according to claim 6, characterized in that, In the step S5, by adaptively adjusting the weights of each evaluation index, combining the depth information of the multi-depth-of-field image, accurately locating the clearest layer where the defect is located, so as to determine the optimal distribution position of the defect in the sub-pixel layer, and at the same time introducing a comparative analysis mechanism consistent with human visual perception to ensure that the detection result has a high degree of consistency with human visual perception. The specific process includes: proposing a weight adjustment formula based on human eye perception as: , where is the weight of the i th evaluation index, α and β are adaptive parameters, S i is the multi-dimensional feature information extracted by the feature calculation module, D i is the depth information of the multi-depth-of-field image.

8. An adaptive multi-depth-of-field sub-pixel layer defect detection device, characterized in that It includes an image acquisition module, which is used to collect pictures at the sub-pixel layer to obtain images with different depths of field; an image preprocessing module, which uses multiple filters to reduce noise and smooth the input image to eliminate noise interference and improve image quality; Defect detection module, used to accurately extract edge features in images of different depths of field; A feature calculation module for calculating multi-dimensional feature information of an image; An adaptive evaluation algorithm module that, by adaptively adjusting the weights of each evaluation index, combines the depth information of multi-depth images to accurately locate the clearest layer where the defect is located, thereby determining the optimal distribution position of the defect in the sub-pixel layer. At the same time, a comparative analysis mechanism consistent with human visual perception is introduced to ensure that the detection result has a high degree of consistency with human visual perception.

Images