Screen arc edge region defect detection system and method based on area array camera mobile device

Abstract

The application discloses a screen arc edge area defect detection system and method based on a surface array camera mobile device, which can be applied to comprehensive defect detection of electronic device screens such as mobile phones, tablets and computers, and has excellent detection performance in an arc edge area (BM area). Through the combination of a surface array camera, a multi-angle shooting scheme and a dust removal light source, and the cooperation of an innovative algorithm process, high-precision defect detection of the screen arc edge area can be realized, including automatic identification and positioning of defect types such as BMinkdelam, discolourdelam and bubble. The technology is particularly suitable for screen quality control and defect detection in the electronic manufacturing industry, and has a wide industry application prospect and high-efficiency detection capability.

Description

Technical Field

[0001] The present invention relates to the field of detection technology, and in particular to a system and method for detecting defects in arc edge areas of a screen of a mobile device based on an area array camera. Background Art

[0002] In smartphone manufacturing, screen defect detection is a critical step in ensuring product quality. Due to its unique curved surface structure, the curved edge region (BM) of mobile phone screens has always been a challenging area for defect detection. Traditional defect detection systems typically use line scan cameras, but these cameras have significant limitations when capturing the outer edges of curved edges. Due to the limitations of their imaging principles, line scan cameras are unable to effectively capture the outer edge details of curved edges, making it difficult for subsequent algorithms to accurately locate the curved edge, thus affecting the accuracy of defect detection.

[0003] In existing technologies, defect detection systems primarily use line scan cameras in conjunction with a single-angle strip light source for imaging. This solution performs adequately when detecting defects in the center area of ​​the screen, but has the following problems when processing the arc edge area: First, the line scan camera cannot fully present the geometric features of the outer edge of the arc edge, resulting in insufficient defect location accuracy; second, the single-angle light source illumination cannot fully reveal the defect characteristics of the arc edge area, especially for special defect types such as BMinkdelam, discolourdelam, and bubble, making it difficult to obtain clear imaging effects; finally, traditional systems lack an effective dust cleaning mechanism, and dust on the screen surface can interfere with defect detection, resulting in a high false detection rate. In view of the limitations of existing line scan cameras in defect detection in the arc edge area of ​​mobile phone screens, there is an urgent need for a solution that can more accurately capture the outer edge of the arc edge, improve defect location accuracy, and reduce false detection rates. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this paper proposes a screen edge defect detection system and method based on an area array camera for mobile devices. This system is suitable for comprehensive defect detection on screens of electronic devices such as mobile phones, tablets, and computers, with particularly excellent detection performance in the edge-of-screen (BM) region. By combining an area array camera, a multi-angle capture scheme, and a dust-clearing light source, along with an innovative algorithmic process, it achieves high-precision defect detection in the edge-of-screen region, including automatic identification and location of BMinkdelam, discolourdelam, and bubble defects. This technology is particularly well-suited for screen quality control and defect detection in the electronics manufacturing industry, offering broad industry application prospects and highly efficient detection capabilities.

[0005] The present invention proposes a screen arc edge area defect detection system based on an area array camera mobile device, comprising: a global exposure area array camera for acquiring screen images; a multi-angle strip light source module for realizing multimodal imaging of the arc edge area through synchronous trigger control; a dust cleaning light source for capturing dust images and generating a high signal-to-noise ratio dust distribution map; a spectral confocal unit for real-time measurement of the axial distance between the screen surface and the camera, and for ensuring the consistency of the imaging depth of field through closed-loop feedback control of the lifting slide; a shadowless light source for eliminating reflections and shadows on the screen surface through 360° uniform diffuse lighting; an algorithm system, including a product positioning and arc edge area extraction module, a dust cleaning module, and a 360° uniform diffuse lighting system. block, multi-scale defect detection module and feature calculation module; the product positioning and arc edge area extraction module is used to perform edge detection, threshold segmentation, closing operation and connected domain analysis on the input image to achieve accurate positioning of the arc edge area of ​​the screen; the dust cleaning module is used to extract the dust mask image and perform repair processing based on the dust image taken by the dust cleaning light source; the multi-scale defect detection module is used to perform multi-scale enhancement on the pre-processed image and combine multiple edge detection operators to achieve accurate identification of defects; the feature calculation module is used to extract the geometric, grayscale and gradient features of the defect area, construct a multi-dimensional feature vector and realize defect type discrimination.

[0006] In the present invention, the hardware system consists of a global exposure area array camera, a multi-angle strip light source module, a dust cleaning light source, a spectral confocal unit, and a shadowless light source. The global exposure area array camera can instantly capture a complete screen image, avoiding image distortion caused by scanning. The dust cleaning light source uses the strong scattering properties of dust particles on short-wavelength light to capture dust images and generate a high-signal-to-noise ratio dust distribution map, providing a basis for subsequent dust cleaning. The spectral confocal unit measures the axial distance between the screen surface and the camera in real time and controls the lifting slide through closed-loop feedback to ensure consistent imaging depth of field and image clarity. The shadowless light source uses 360° uniform diffuse illumination to effectively eliminate reflections and shadows on the screen surface, providing a distortion-free base image for manual review and algorithm preprocessing.

[0007] As a further solution of the present invention, it also includes a light source lifting assembly for adjusting the height of the light source; a lifting slide for adjusting the distance between the area array camera and the screen; the multi-angle strip light source module includes 0°, 45°, 90°, and 135° strip light sources, which illuminate the arc edge area of ​​the screen from different angles through synchronous trigger control, realize multimodal imaging, and fully capture the geometric features of the outer edge of the arc edge; the light source lifting assembly can flexibly adjust the height of the light source to adapt to the detection requirements of screens of different sizes and types; and the lifting slide is used to accurately adjust the distance between the area array camera and the screen.

[0008] As a further solution of the present invention, the product positioning and arc edge area extraction module uses the Sobel operator in combination with the horizontal template and the vertical template for edge detection, and uses Gaussian smoothing preprocessing to suppress imaging noise, and dynamically calculates the adaptive segmentation threshold of the gradient amplitude based on the Otsu algorithm. Specifically, the Sobel operator is used in combination with the horizontal and vertical gradient templates to detect the edges of the input image, and Gaussian smoothing preprocessing is used to suppress the imaging noise of the array camera, thereby enhancing the continuity of the grayscale mutation in the arc edge area. Based on the adaptive segmentation threshold of the gradient amplitude dynamically calculated by the Otsu algorithm, the optimal segmentation point is determined by maximizing the inter-class variance, effectively solving the problem of arc edge contrast differences under different lighting conditions. The edge detection results are closed to fill small breaks, and the connected domain analysis is performed in combination with geometric feature constraints such as area and aspect ratio to eliminate interference areas and achieve accurate positioning of the arc edge area of ​​the screen.

[0009] As a further solution of the present invention, the dust cleaning module adopts an improved sample block repair algorithm based on the Criminisi algorithm as the basic framework, introduces the gradient direction consistency constraint in the priority calculation, and uses the rotation-invariant SSIM similarity in the matching block search stage. Specifically, based on the dust image captured by the dust cleaning light source, the high scattering characteristics of dust particles to short-wavelength light are used to make the dust in the image form bright spots (areas with grayscale values ​​≥ 220), and combined with global threshold segmentation to extract a binary dust mask image. Based on the Criminisi algorithm as the basic framework, the sample block repair algorithm is improved, and the gradient direction consistency constraint is introduced in the priority calculation ( , where N is the direction of the iso-illuminance line), prioritizing areas with strong continuity with surrounding structures. During the matching block search phase, the rotationally invariant SSIM similarity is used to adaptively select the best matching block within regions with known textures, avoiding edge misalignment and texture breakage caused by traditional Euclidean distance. After restoration, bilateral filtering is performed on the processed area to smooth transitions between blocks while preserving edge sharpness, enabling accurate detection and removal of dust interference on the screen surface.

[0010] As a further solution of the present invention, the multiscale defect detection module uses the Difference of Gaussian (DG) algorithm to perform multiscale image enhancement, combining the Canny and Sobel operators for precise edge location. Canny dual thresholds are adaptively calculated based on the global gradient histogram distribution. The Difference of Gaussian (DoG) algorithm is used to perform multiscale enhancement on the preprocessed image. Two-level Gaussian kernels (σ1 = 3.0, σ2 = 9.0) are set to extract fine-grained and coarse-grained edge responses, respectively. A differential operation (DoG = G_σ1 - G_σ2) amplifies the local contrast of small defects (such as bubbles). The module combines the Canny and Sobel operators for precise edge location. Based on the global gradient histogram distribution, Canny dual thresholds (low threshold = 0.05 × maximum gradient value, high threshold = 0.15 × maximum gradient value) are adaptively calculated. Non-maximum suppression is used to eliminate false edges. Sobel horizontal and vertical gradient amplitudes are used to supplement the detection of weak edge defects (such as discolourdelam), achieving accurate identification of small defects in arc edge areas.

[0011] As a further solution of the present invention, the feature calculation module implements defect type discrimination based on a weighted voting classifier with a geometric weight of 0.4, a grayscale weight of 0.3, and a gradient weight of 0.3. Specifically, the module extracts the geometric features (aspect ratio, area), grayscale features (average grayscale, contrast), and gradient features (gradient amplitude, Laplace response) of the defect region to construct a multidimensional feature vector. Defect type discrimination is achieved using a weighted voting classifier (with a geometric weight of 0.4, a grayscale weight of 0.3, and a gradient weight of 0.3). Based on the characteristic differences between different defect types, it accurately distinguishes between defects such as BMinkdelam, discolourdelam, and bubble.

[0012] As a further solution of the present invention, the horizontal template is -1, 0, 1; -2, 0, 2; -1, 0, 1; and the vertical template is 1, 2, 1; 0, 0, 0; -1, -2, -1.

[0013] As a further solution of the present invention, the shadowless light source includes a diffuser, an LED array and a uniform light cavity; the spectral confocal unit includes a laser emitter, a confocal lens group and a high-sensitivity photoelectric sensor; the dust cleaning module includes a green light LED array and a dichroic prism combination, so that dust particles form bright scattering points on the screen surface, providing high-contrast input for dynamic mask generation.

[0014] The present invention also provides a detection method for a screen arc edge area defect detection system based on an area array camera mobile device, comprising the following steps:

[0015] Image acquisition: Use a global exposure area array camera with a multi-angle strip light source module to capture screen images; at the same time, use a dust-clearing light source to capture dust images;

[0016] Perform Sobel operator edge detection and Gaussian smoothing preprocessing on the acquired screen image, calculate the adaptive segmentation threshold based on the Otsu algorithm, perform closing operation and connected domain analysis on the edge detection results to achieve accurate positioning of the screen arc edge area;

[0017] Based on the dust image captured by the dust-clearing light source, a binary dust mask is extracted, and the dust is repaired using an improved sample block repair algorithm, and finally a bilateral filtering process is performed.

[0018] Perform Gaussian difference multi-scale enhancement on the pre-processed image, and combine the Canny operator and Sobel operator to accurately locate edges and extract defects;

[0019] The geometric, grayscale and gradient features of the defect area are extracted, a multidimensional feature vector is constructed, and the defect type is identified using a weighted voting classifier.

[0020] Beneficial effects of the present invention:

[0021] Abandoning traditional line scan cameras, the system adopts global exposure area array cameras and multi-angle strip light combination lighting technology. The multi-angle optical path collaborative design significantly enhances the imaging contrast of the arc edge boundary, clearly captures the complete geometric outline of the arc edge area, and completely solves the arc edge detection failure problem caused by image stitching misalignment and dynamic blur of line scan cameras, achieving a fundamental breakthrough in the imaging and positioning capabilities of the arc edge area; through the integrated dust cleaning and detection design, dedicated light sources and algorithms are used to automatically eliminate dust interference, combined with multi-angle synchronous imaging technology to achieve rapid acquisition of arc edge data, and support multiple curvature adaptation without manual intervention, reducing the tediousness and inconsistency of manual operations, and improving the automation and stability of the detection process; adopting a static imaging architecture without moving parts, greatly reducing hardware costs; through lightweight algorithm optimization to achieve low-power embedded operation, greatly improving system reliability and production line adaptability, reducing maintenance costs and deployment costs, and having significant industrial-grade application advantages. The modular imaging system and scalable algorithm framework can quickly adapt to the inspection needs of various materials such as LCD, OLED, and microcrystalline glass, and can be expanded to scenarios such as crack detection and reflection suppression. It still maintains high stability under harsh working conditions such as strong vibration, high humidity, and extreme temperatures, providing the 3C industry with a universal and highly robust defect detection solution.

[0022] In order to more clearly illustrate the structural features and effects of the present invention, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is the overall hardware design diagram of the present invention.

[0024] Figure 2 It is an enlarged view of the shadowless light source of the present invention.

[0025] Figure 3 This is an enlarged view of the spectral confocal unit of the present invention.

[0026] Figure 4 This is an enlarged view of the dust cleaning light source of the present invention.

[0027] Figure 5 This is a comparison diagram of arc edge area defect imaging effects of the line scan camera of the present invention and the area array camera.

[0028] Figure 6 It is the algorithm flow chart of the present invention.

[0029] Figure 7 This is a schematic diagram of product positioning and arc edge area extraction according to the present invention.

[0030] Figure 8 、 Figure 9 、 Figure 10 This is a comparison chart of the dust detection and repair effects of the dust cleaning light source of the present invention, where Figure 8 For the original picture, Figure 9 For dust pictures, Figure 10 This is the picture after cleaning and restoration.

[0031] Figure 11 、 Figure 12 、 Figure 13 This is a defect detection result diagram of the present invention. DETAILED DESCRIPTION

[0032] The present invention will be further described below with reference to the accompanying drawings and related knowledge, and described clearly and completely. Obviously, the described applications are only part of the embodiments of the present invention, rather than all of the embodiments.

[0033] In order to solve the shortcomings of the existing line scan camera in the detection of defects in the arc edge area of ​​the mobile phone screen, the present invention proposes a screen arc edge area defect detection system based on a mobile device with an area array camera. The system uses an area array camera machine with multiple strip lights, and shoots from multiple angles such as 0°, 45°, 90°, 135°, etc., which can fully capture the geometric features of the outer edge of the arc edge and provide high-quality image data for subsequent algorithm processing. At the same time, the system introduces a dust-cleaning light source, which can capture special dust images for use by the algorithm, thereby effectively removing dust interference on the screen surface and significantly reducing the false detection rate. At the algorithm level, the present invention has developed a set of processes specifically for arc edge area defect detection. The improved edge detection algorithm is used to achieve precise positioning of the arc edge area, and the dust-cleaning algorithm is combined to achieve efficient denoising processing, ultimately achieving high-precision defect detection. This solution not only solves the technical bottleneck of the traditional detection system in arc edge area defect detection, but also significantly improves the detection efficiency and the reliability of the detection results, providing a more efficient and accurate defect detection method for smartphone screen manufacturing. The details are as follows:

[0034] The core idea of ​​this invention is to build a high-precision, high-efficiency defect detection system-level solution through the bidirectional integration of hardware imaging architecture innovation and algorithm process collaborative design. Its core innovations are reflected in the following two aspects:

[0035] Hardware system (such as Figures 1-4 As shown in the figure: Abandoning the imaging mode of the traditional line scan camera that scans line by line, a global exposure area array camera is used in combination with a multi-angle strip light illumination system (0°, 45°, 90°, 135°). Through the multi-angle synchronous triggering imaging strategy, a panoramic high-resolution capture of the outer edge details of the arc edge area is achieved, which effectively overcomes the problem that the line scan camera cannot fully capture the continuous curvature boundary of the arc edge area (BM area) of the mobile phone screen (as shown in the figure). Figure 5 At the same time, the innovative dust cleaning auxiliary light source mode is introduced to utilize the strong scattering characteristics of dust particles on short-wavelength light to generate a high signal-to-noise ratio dust distribution map, providing accurate prior information for eliminating dust interference on the algorithm side.

[0036] Further, if Figures 1-4 As shown in the figure, this system completely revolutionizes the traditional imaging architecture, abandoning the inherent line-by-line scanning mode of line scan cameras and building a multi-dimensional visual acquisition system based on global exposure area array cameras. By integrating a multi-angle strip lighting system with 0°, 45°, 90°, and 135°, and coordinating with a multi-view synchronous trigger imaging mechanism, it achieves panoramic high-resolution capture of the outer edge of the screen's curved edge area. This design effectively overcomes the technical bottlenecks of traditional line scan cameras when processing the curved edge area (BM area) of mobile phone screens, such as inaccurate defect positioning and high missed detection rate due to the inability to fully present the continuous curvature boundary (see details). Figure 5Comparative analysis) lays a solid data foundation for subsequent high-precision detection.

[0037] The system also features an innovative dust-cleaning auxiliary light source module. Based on the strong scattering properties of dust particles against short-wavelength light, a precise combination of a green LED array and a beam-splitting prism creates a high-brightness scattered image on the screen surface, generating a high-signal-to-noise ratio dust distribution map. This provides critical prior information for precise denoising on the algorithm side, enabling deep synergy between hardware acquisition and algorithm processing, significantly improving the system's anti-interference capabilities and detection reliability.

[0038] In terms of algorithm design: the present invention has developed a set of detection processes that match the area array camera equipment. First, the input image is positioned by the edge detection operator, and the arc edge area is accurately positioned to solve the problem of the unstable screen position; secondly, based on the dust map provided by the dust cleaning light source, combined with the image restoration algorithm, the automatic dust cleaning of the screen area is realized, which significantly reduces the false detection problem caused by dust interference and avoids manual dust wiping operations; finally, a multi-scale defect detection algorithm is adopted, combined with a variety of edge detection operators such as Canny operator, Sobel operator, Gaussian difference and filter difference method, to accurately extract the defect features of the arc edge area of ​​the screen, and by calculating the multi-dimensional feature information of the defect (such as length, width, area, average grayscale, gradient amplitude, etc.), to achieve accurate distinction between BMinkdelam, discolourdelam, bubble and other defects. In short, the system significantly improves the efficiency and accuracy of defect detection in the arc edge area of ​​mobile phone screens through the innovative combination of equipment and algorithms;

[0039] like Figure 6-Figure 7The product positioning and arc edge extraction module of the present invention is shown. This module uses traditional image algorithms to accurately locate the screen's arc edge area. It first performs edge detection on the input image using the Sobel operator combined with horizontal and vertical gradient templates (horizontal templates: -1, 0, 1; -2, 0, 2; -1, 0, 1; vertical templates: 1, 2, 1; 0, 0, 0; -1, -2, -1). Gaussian smoothing preprocessing is then used to suppress area array camera imaging noise and enhance the continuity of grayscale mutations in the arc edge area. Subsequently, an adaptive segmentation threshold based on the gradient amplitude is dynamically calculated using the Otsu algorithm. The optimal segmentation point is determined by maximizing the inter-class variance, addressing the issue of arc edge contrast differences under different lighting conditions. Further, a closed-loop operation is performed on the edge detection results to fill in small breaks. Connected domain analysis, combined with geometric feature constraints such as area and aspect ratio, effectively eliminates interference from non-arc edge areas (such as screen scratches or reflective noise). This comprehensively improves the integrity and anti-interference capabilities of the arc edge's outer contour, ultimately achieving stable positioning of the screen's arc edge area in complex scenes. For product positioning and edge extraction, the algorithm employs a hybrid edge detection strategy. It first uses the Sobel operator combined with horizontal (-1, 0, 1; -2, 0, 2; -1, 0, 1) and vertical (1, 2, 1; 0, 0, 0; -1, -2, -1) gradient templates to extract edge features from the input image. Gaussian smoothing preprocessing suppresses area array camera imaging noise and enhances grayscale mutations in the edge region. Subsequently, an adaptive segmentation threshold based on the Otsu algorithm dynamically calculates the gradient amplitude, achieving accurate segmentation under varying lighting conditions by maximizing the inter-class variance. To address gaps in edge detection results, morphological closing operations are used to fill them in. Connected domain analysis, combined with geometric constraints such as area and aspect ratio, effectively eliminates interfering areas and achieves sub-pixel precision in the edge region of the screen, completely resolving the detection challenges associated with screen position offsets.

[0040] Reference Figures 8-10 As shown in the figure, the dust cleaning module of the present invention realizes accurate detection and elimination of dust interference on the screen surface based on hardware-algorithm collaborative design. First, a dust cleaning light source of a specific wavelength is used to shoot under dark field lighting conditions. The high scattering characteristics of dust particles to short-wavelength light are used to form highlight spots (areas with grayscale values ​​≥ 220) in the image. Combined with global threshold segmentation, a binary dust mask image is extracted. Then, an improved sample block repair algorithm is used. Based on the Criminisi algorithm as the basic framework, the repair efficiency is improved by optimizing the priority function. The priority calculation introduces the gradient direction consistency constraint ( , where N is the direction of the iso-illuminance line), give priority to filling the area with strong coherence with the surrounding structure, and at the same time use the rotation invariant SSIM similarity in the matching block search stage to adaptively select the best matching block in the known texture area to avoid edge dislocation and texture breakage caused by the traditional Euclidean distance; after the repair is completed, the processed area is bilaterally filtered to smooth the transition between blocks and retain edge sharpness. The dust cleaning processing module relies on the high-contrast dust image collected by the dust cleaning light source to build a hardware-algorithm collaborative intelligent dust cleaning mechanism. The algorithm first uses the strong scattering characteristics of dust to short-wavelength light, and extracts the binary dust mask map through global threshold segmentation; on this basis, it uses the Criminisi algorithm as a framework for optimization and upgrading, and introduces the gradient direction consistency constraint ( , N is the direction of the iso-illuminance line), prioritizing repair of areas with strong continuity with surrounding structures. The rotationally invariant SSIM similarity metric is used in the matching block search phase to avoid texture misalignment. Finally, bilateral filtering is used to achieve smooth transitions within the repaired area, preserving screen texture detail while fully automatically removing dust interference.

[0041] Reference Figure 11-13 As shown in the figure, this module achieves accurate identification of tiny defects in the arc edge area by integrating multi-scale edge detection and multi-dimensional feature analysis. First, the Gaussian difference (DoG) algorithm is used to perform multi-scale enhancement on the preprocessed image, and a two-level Gaussian kernel (σ1=3.0, σ2=9.0) is set to extract fine-grained and coarse-grained edge responses respectively. The local contrast of tiny defects (such as bubbles) is amplified through differential operation (DoG=G_σ1-G_σ2); then the Canny operator and the Sobel operator are combined to accurately locate the edge, and the Canny double threshold (low threshold = 0.05×maximum gradient value, high threshold = 0.15×maximum gradient value) is adaptively calculated based on the global gradient histogram distribution. The pseudo-edge is eliminated by non-maximum suppression, and the Sobel horizontal and vertical gradient amplitudes are used to supplement the detection of weak edge defects (such as discolourdelam). The multi-scale defect detection module employs a hierarchical feature extraction strategy. It first applies the Difference of Gaussian (DoG) algorithm to multi-scale enhancement of the preprocessed image. A two-level Gaussian kernel (σ1 = 3.0, σ2 = 9.0) is set to capture fine-grained and coarse-grained edge responses, respectively, effectively amplifying the local contrast of tiny defects. Furthermore, it integrates the advantages of the Canny and Sobel operators: Adaptive calculation of the Canny dual threshold (low threshold = 0.05 × maximum gradient value, high threshold = 0.15 × maximum gradient value) combined with non-maximum suppression enables precise edge location. The Sobel operator's horizontal and vertical gradient amplitudes are used to supplement the detection of weak edge defects. This composite detection strategy enables multi-level feature extraction for different defect types, including BMinkdelam, discolourdelam, and bubble.

[0042] The feature calculation module of the present invention further extracts multidimensional feature information of the defect area. This includes geometric features (aspect ratio, area), grayscale features (average grayscale, contrast), and gradient features (gradient amplitude, Laplace response). This module constructs a multidimensional feature vector and uses a weighted voting classifier (geometric weight 0.4, grayscale weight 0.3, and gradient weight 0.3) to distinguish defect types. For example, bubbles are distinguished by their high circularity (aspect ratio ≈ 1) and high gradient amplitude, while BMinkdelam exhibits low gradient variations and irregular contours.

[0043] 1) Average grayscale:

[0044] ;in, I ( x , y ) is the image at coordinates ( x , y ) is the grayscale value at . D A collection of pixels representing a defect area. N Indicates the total number of pixels in the defect area ( N =∣ D ∣). μ Defect represents the average grayscale of the defect area and characterizes the overall brightness characteristics of the area.

[0045] 2) Contrast

[0046] ;in, μ defect is the average grayscale of the defect area; μ background is the average grayscale of the background area,

[0047] 3) Gradient amplitude reflects the strength of the image edge

[0048] Horizontal gradient: ;

[0049] Vertical Gradient: ;

[0050] Gradient Magnitude: ;

[0051] in, Gx : Indicates the gradient of the image in the horizontal direction (x-axis direction), reflecting the rate of change of the brightness of the image in the horizontal direction. Gy : Indicates the gradient of the image in the vertical direction (y-axis direction), reflecting the rate of change of the brightness of the image in the vertical direction. G : Indicates 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.

[0052] 4) Laplace response is used to detect the sharpness of the image ;in: L ( x , y ): indicates the coordinate (x,y) ( x , y ), which is used to measure the edge strength at that point.

[0053] f(x+1,y): represents the x The grayscale value after shifting one pixel to the right in the vertical direction.

[0054] f(x-1,y): represents the value at x x The grayscale value after shifting one pixel to the left in the vertical direction.

[0055] f(x,y+1): represents the value at y y The grayscale value after moving up one pixel in the direction.

[0056] f(x,y-1): represents the value at y y The grayscale value after moving down one pixel in the direction.

[0057] 4f(x,y): represents the current pixel (x,y) ( x , y ) is four times the grayscale value at ).

[0058] The defect feature calculation and classification module builds an intelligent decision-making model that integrates multidimensional features. It extracts the geometric features (aspect ratio, area), grayscale features (average grayscale, contrast), and gradient features (gradient amplitude, Laplace response) of the defect area to construct a high-dimensional feature vector. Using a weighted voting classifier (geometric weight 0.4, grayscale weight 0.3, gradient weight 0.3), combined with a machine learning algorithm, the feature vector is trained and optimized to accurately classify various defects.

[0059] This invention addresses the inherent flaws of traditional line-scan cameras, which suffer from edge distortion, image stitching misalignment, and motion blur when capturing curved surfaces or curved edges due to the limitations of their imaging principles (relying on a moving platform for progressive scanning). In particular, these flaws prevent the camera from fully capturing the continuous curvature boundary of the curved edge region (BM) of a mobile phone screen, leading to inaccurate defect location and a high rate of missed detection. Through systematic innovations in hardware architecture and algorithms, this invention thoroughly resolves these technical bottlenecks, offering the following specific advantages:

[0060] 1) Fundamental breakthrough in arc edge imaging and positioning capabilities

[0061] Due to limitations in imaging principles, traditional line scan cameras experience severe edge blur and distortion when capturing images of arc edges (the naked eye cannot clearly discern the complete outline), preventing the algorithm from effectively locating the area to be inspected. This innovative approach employs a global exposure area array camera combined with multi-angle strip lighting (0°, 45°, 90°, and 135°). This multi-angle optical path collaborative design significantly enhances the imaging contrast of arc edges (the naked eye can clearly discern the complete outline), completely resolving the arc edge detection failure issue caused by image splicing misalignment and dynamic blur in line scan cameras (see attached for a comparison of the improved results). Figure 5 ).

[0062] 2) Full-process anti-interference and automation upgrade

[0063] Traditional detection solutions rely on manual dust removal and light source angle adjustment, resulting in cumbersome operations and poor consistency. This invention integrates dust removal and detection into an integrated design, using a dedicated light source and algorithm to automatically eliminate dust interference. Combined with multi-angle synchronous imaging technology, it enables rapid acquisition of arc edge data and supports adaptive adjustment of multiple curvatures, eliminating the need for manual intervention.

[0064] 3) Industrial-grade reliability and cost advantages

[0065] Traditional detection solutions rely on expensive motion control components and high-computing hardware, resulting in frequent maintenance and high deployment costs. This invention utilizes a static imaging architecture with no moving parts, significantly reducing hardware costs. It also achieves low-power embedded operation through lightweight algorithm optimization, significantly improving system reliability and production line compatibility, significantly outperforming traditional technologies.

[0066] 4) Technology scalability and industry applicability

[0067] Traditional inspection solutions, due to rigid hardware architectures and algorithms, are difficult to adapt to diverse screen materials and complex industrial environments. This invention, through a modular imaging system and scalable algorithm framework, is rapidly compatible with the inspection needs of various materials, including LCDs, OLEDs, and glass-ceramics, and can be expanded to cover scenarios such as crack detection and reflection suppression. Field tests have shown that the system maintains high stability under harsh operating conditions such as high vibration, high humidity, and extreme temperatures. This provides a universal, highly robust defect detection solution for the 3C industry.

[0068] Example 1, screen arc edge area defect detection system based on area array camera mobile device, refer to Figure 1 As shown, it includes: area array camera unit 1, light source lifting component 2, dust cleaning light source 3, lifting slide 4, shadowless light source 5, and spectral confocal unit 6. Each module is integrated with the slide linkage mechanism through a rigid bracket. The hardware synchronization signal triggers the multi-angle light source time-sharing exposure, realizing the automation of the whole process of "dust cleaning-multimodal imaging-confocal calibration"; refer to Figure 2A magnified image of the shadowless light source shows the physical structure of the shadowless light source, including the diffuser, LED array 8, and uniform light cavity 7. This light source eliminates reflections and shadows on the screen surface through 360° uniform diffuse illumination, providing a distortion-free base image for manual review and algorithm preprocessing. Figure 3 A magnified image of the spectral confocal unit shows the precise optical structure of the spectral confocal ranging module, comprising a laser emitter 9, a confocal lens assembly 10, and a highly sensitive photoelectric sensor 11. This unit measures the axial distance between the screen surface and the camera in real time, ensuring consistent depth of field through closed-loop feedback control of the lift stage. Figure 4 The coaxial optical path design of the dust cleaning light source module uses a green LED array 12 and a beam splitter prism 13 to make dust particles form bright scattering points on the screen surface, providing high-contrast input for dynamic mask generation.

[0069] Reference Figure 5 As shown in the figure, by comparing the imaging effects of line scan cameras and area array cameras on three typical defects, namely scratches, ink delamination, and bubbles, the technical advantages of the present invention can be intuitively verified: due to the limitations of the dynamic scanning imaging principle, traditional line scan cameras have serious boundary blur and image distortion in the arc edge area, resulting in failure to locate the defect area; while the present invention adopts a global exposure area array camera combined with a multi-angle light source, which can clearly capture the complete geometric contour of the arc edge area (such as the direction of the scratch, the sharpness of the delamination boundary, and the roundness characteristics of the bubble), significantly improving the recognizability and positioning accuracy of the inspected area, and providing a reliable visual benchmark for the automated classification and quantitative analysis of defects.

[0070] Reference Figure 7 As shown, this invention uses an innovative algorithm to precisely locate the arc edge region to be inspected in the screen's BM area. Even on complex curved surfaces and under highly reflective conditions, it can clearly extract the complete boundary contour (as shown in the red box in the attached figure). Compared to traditional methods, this solution eliminates the need for manual calibration or complex preprocessing, directly generating high-precision inspection masks. Furthermore, its positioning stability has been verified through multiple batches of field measurements, providing a reliable benchmark for defect detection.

[0071] Reference Figures 8-10 As shown, Figure 8 is the original screen image, Figure 9 The dust distribution (highlight white spot area) highlighted by the dust-clearing light source imaging is shown in the figure. Figure 10 This is the result after the dynamic dust cleaning algorithm repairs. By comparison, it can be seen that the traditional solution is difficult to distinguish between dust and real defects (such as Figure 9 The proposed method completely eliminates dense noise, while preserving screen texture and edge details. This technology significantly reduces false detection rates by automating the mask generation and repair process, eliminating the need for manual intervention.

[0072] The present invention utilizes a combined area array camera and multi-angle strip light illumination system: a global exposure area array camera is used in conjunction with multi-angle strip light source modules at 0°, 45°, 90°, and 135°, and multi-modal imaging of the arc edge area is achieved through synchronous trigger control. In other words, this system deeply integrates the global exposure area array camera with the multi-angle strip light source modules at 0°, 45°, 90°, and 135°. Through a synchronous trigger control mechanism, multi-modal imaging of the arc edge area of ​​the screen is achieved. The global exposure area array camera can instantly capture complete image information, avoiding the image distortion problems caused by traditional scanning methods; the multi-angle strip light source illuminates the arc edge area from different directions, giving each imaging perspective unique characteristics, thereby comprehensively and meticulously presenting the geometric shape and surface features of the arc edge area, providing rich and high-quality image data for subsequent defect detection.

[0073] Dynamic dust cleaning and image restoration adaptive module: Generates a high-brightness dust mask based on the dust cleaning light source, extracts the interference area through differential motion state, and adopts the Criminisi restoration algorithm with gradient direction constraints to adaptively match the screen texture to achieve seamless restoration and eliminate the reliance on manual dusting. In other words, this module is based on the special design of the dust cleaning light source, and uses the strong scattering characteristics of dust on light of specific wavelengths to generate a high-brightness dust mask image. Through differential motion state analysis technology, the dust interference area in the image is accurately extracted. In the restoration stage, the Criminisi restoration algorithm optimized by gradient direction constraints is adopted. This algorithm can perform adaptive matching according to the texture characteristics of the screen and fill the dust area in a seamless and natural way, so that the repaired image is perfectly integrated with the original screen texture, completely getting rid of the reliance on manual dusting, and effectively improving the detection efficiency and accuracy.

[0074] Multi-scale edge detection mechanism: This mechanism integrates Gaussian difference, Canny double-threshold segmentation, and Sobel gradient magnitude calculation methods, leveraging the complementary advantages of different edge detection operators to achieve multi-level feature extraction of defects in arc edge areas. This mechanism utilizes multiple edge detection methods, including Gaussian difference, Canny double-threshold segmentation, and Sobel gradient magnitude calculation, leveraging the complementary advantages of different operators in edge detection. The Gaussian difference algorithm enhances image edge information at different scales, highlighting the characteristics of minor defects. The Canny double-threshold segmentation uses adaptive threshold selection to accurately locate edges and reduce the interference of false edges. The Sobel gradient magnitude calculation quickly and accurately detects gradient changes in the image, capturing various types of edge information. This multi-scale, multi-level edge detection strategy enables comprehensive and in-depth extraction of defect features in arc edge areas, providing a reliable basis for subsequent defect classification.

[0075] The multi-dimensional defect feature classification adaptive decision model extracts defect geometric features (aspect ratio, area), grayscale features (contrast Δ, standard deviation σ), and gradient features (gradient amplitude, Laplace response) to construct a weighted voting classifier. This classifier can adaptively distinguish defect types such as BMinkdelam (low gradient), discolourdelam (low contrast), and bubble (high circularity). This model constructs a comprehensive and detailed feature vector by extracting multi-dimensional defect feature information, including geometric features (such as aspect ratio, area), grayscale features (such as contrast Δ, standard deviation σ), and gradient features (such as gradient amplitude, Laplace response). Based on this feature vector, a weighted voting classifier algorithm is used to adaptively distinguish different defect types. The classifier automatically adjusts weight parameters based on the feature differences between different defect types, accurately identifying defects such as BMinkdelam (low gradient features), discolourdelam (low contrast features), and bubble (high circularity features), significantly improving the accuracy and reliability of defect classification.

[0076] The technical principles of the present invention have been described above in conjunction with specific embodiments, which are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the principles of the present invention fall within the scope of protection of the present invention. Those skilled in the art will be able to conceive of other specific embodiments of the present invention without inventive effort, and such methods will fall within the scope of protection of the present invention.

Claims

1. A screen arc edge defect detection system for mobile devices based on area array cameras, characterized by: include: Global exposure area array camera, used to acquire screen images; The multi-angle strip light source module realizes multi-modal imaging of the arc edge area through synchronous trigger control, and also includes a light source lifting component for adjusting the height of the light source; a lifting slide for adjusting the distance between the area array camera and the screen; the multi-angle strip light source module includes 0°, 45°, 90°, and 135° strip light sources; a dust cleaning light source for capturing dust images and generating a high signal-to-noise ratio dust distribution map. The dust cleaning light source uses a green LED array and a dichroic prism combination to make dust particles form bright scattering points with a grayscale value ≥ 220 on the screen surface, providing high-contrast input for dynamic mask generation; a spectral confocal unit for real-time measurement of the axial distance between the screen surface and the camera, and ensures the consistency of imaging depth of field through closed-loop feedback control of the lifting slide. The spectral confocal unit includes a laser emitter, a confocal lens group, and a high-sensitivity photoelectric sensor; a shadowless light source that eliminates reflections and shadows on the screen surface through 360° uniform diffuse illumination. The shadowless light source includes a diffuser plate, LED array and uniform light cavity; algorithm system, including product positioning and arc edge area extraction module, dust cleaning module, multi-scale defect detection module and feature calculation module; the product positioning and arc edge area extraction module is used to perform edge detection, threshold segmentation, closed operation and connected domain analysis on the input image to achieve accurate positioning of the screen arc edge area. The product positioning and arc edge area extraction module uses the Sobel operator combined with horizontal and vertical templates for edge detection. The horizontal template is -1,0,1;-2,0,2;-1,0,1, and the vertical template is 1,2,1;0,0,0;-1, -2, -1, and Gaussian smoothing preprocessing is used to suppress imaging noise. The adaptive segmentation threshold of the gradient amplitude is dynamically calculated based on the Otsu algorithm. Small breaks in the edge detection results are filled by closing operations, and connected domain analysis is performed in combination with area and aspect ratio to eliminate interference areas. The dust cleaning module is used to extract the dust mask image and perform repair processing based on the dust image taken by the dust cleaning light source. The dust cleaning module adopts an improved sample block repair algorithm based on the Criminisi algorithm. The priority calculation introduces the gradient direction consistency constraint. The matching block search stage adopts the rotation invariant SSIM similarity. After the repair is completed, the processed area is subjected to bilateral filtering to smooth the transition between blocks and retain edge sharpness. The multi-scale defect detection module is used to perform multi-scale enhancement on the pre-processed image and combine multiple edge detection operators to achieve accurate identification of defects. The multi-scale defect detection module uses the Gaussian difference algorithm to perform multi-scale enhancement on the image, sets a two-level Gaussian kernel to extract fine-grained and coarse-grained edge responses respectively, amplifies the local contrast of small defects through differential operations, and combines Canny The operator and the Sobel operator are used to accurately locate edges. The Canny double threshold is adaptively calculated based on the global gradient histogram distribution, and pseudo-edges are eliminated through non-maximum suppression. The Sobel horizontal and vertical gradient amplitudes are used to supplement the detection of weak edge defects. The feature calculation module is used to extract the geometric, grayscale, and gradient features of the defect area, construct a multidimensional feature vector, and realize defect type discrimination.

2. The screen arc edge area defect detection system based on an area array camera mobile device according to claim 1, characterized in that: The product positioning and arc edge region extraction module uses the Sobel operator combined with horizontal and vertical templates for edge detection, and uses Gaussian smoothing preprocessing to suppress imaging noise, and dynamically calculates the adaptive segmentation threshold of the gradient amplitude based on the Otsu algorithm.

3. The screen arc edge defect detection system for a mobile device based on an area array camera according to claim 2, wherein: The feature calculation module realizes defect type discrimination based on a weighted voting classifier, with a geometric weight of 0.4, a grayscale weight of 0.3, and a gradient weight of 0.

3.

4. A detection method for a screen arc edge area defect detection system based on an area array camera mobile device as claimed in claim 1, characterized in that: The following steps are involved: Image acquisition: Use a global exposure area array camera with a multi-angle strip light source module to capture screen images; at the same time, use a dust-clearing light source to capture dust images; Perform Sobel operator edge detection and Gaussian smoothing preprocessing on the acquired screen image, calculate the adaptive segmentation threshold based on the Otsu algorithm, perform closing operation and connected domain analysis on the edge detection results to achieve accurate positioning of the screen arc edge area; Based on the dust image captured by the dust-clearing light source, a binary dust mask is extracted, and the dust is repaired using an improved sample block repair algorithm, and finally a bilateral filtering process is performed. Perform Gaussian difference multi-scale enhancement on the pre-processed image, and combine the Canny operator and Sobel operator to accurately locate edges and extract defects; The geometric, grayscale and gradient features of the defect area are extracted, a multidimensional feature vector is constructed, and the defect type is identified using a weighted voting classifier.

Images