Screen orange peel detection method and system based on phase deflectometry

Abstract

The present invention discloses a screen orange peel detection method and system based on phase deflectometry, comprising the following steps: a camera acquires an image by selecting and projecting a stripe light source image; distortion correction is performed on the image acquired by the camera; phase extraction and unwrapping is performed on the corrected image; three-dimensional image reconstruction and gradient calculation are performed based on the unwrapped phase data to obtain the image surface height and its gradient information; curvature calculation and defect feature extraction are performed based on the reconstructed image surface height distribution and gradient information; and orange peel defects are quantitatively evaluated based on the curvature and defect feature information to achieve fast, high-precision, non-contact orange peel defect detection, thereby improving detection efficiency and accuracy and reducing detection costs.

Description

Technical Field

[0001] The present invention relates to the technical field of screen surface quality detection, and in particular to a method and system for detecting screen orange peel defects based on phase deflectometry. Background Art

[0002] In the manufacturing process of electronic products such as monitors, mobile phones, and tablets, screen surface quality is a significant factor affecting product appearance and user experience. Orange peel defects are a common quality issue on screen surfaces, manifesting as subtle unevenness similar to orange peel, typically appearing as wavy or fine, granular bumps, which degrade the screen's smooth surface. While orange peel defects have minimal impact on the screen's display function, they can significantly impact the product's appearance and user experience. Especially in high-end display devices, these defects are considered a serious quality issue.

[0003] Traditional methods for detecting cellulite include the following: Manual visual inspection: This relies on human visual judgment, which is highly subjective, inefficient, inconsistent, and prone to missed detections and misjudgments. Line scan camera scanning: This method uses a line scan camera to scan the screen surface line by line, detecting surface irregularities by analyzing changes in reflected light. However, this method has the following significant disadvantages: Slow inspection speed: The line scan camera must scan the entire screen surface line by line, which results in long inspection times and low efficiency for large screens. Strict light source angle requirements: This method places stringent requirements on both the incident light and viewing angles, resulting in complex equipment debugging and poor stability. Environmental sensitivity: This method is subject to significant ambient light interference and requires strict darkroom conditions. Limited inspection accuracy: The detection rate for subtle cellulite defects is low and the method is susceptible to interference from other defects on the screen surface. High cost: This method requires a high-precision line scan camera and sophisticated light source control system. Contact profilometry: This method uses a contact probe to measure surface profiles. While highly accurate, it suffers from slow measurement speed, potential damage to the screen surface, and inability to fully inspect the entire screen.

[0004] The above traditional methods are difficult to simultaneously meet the requirements of high detection accuracy, fast speed and low cost. Especially for the current continuously upgraded high-end display devices, orange peel defect detection is still a technical difficulty. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a screen orange peel detection method and system based on phase deflectometry. The present invention aims to solve the technical problems existing in the existing orange peel detection technology, such as slow detection speed, strict requirements on light source angle, and limited detection accuracy. A screen orange peel detection method and system based on phase deflectometry is provided to achieve fast, high-precision, non-contact orange peel defect detection, improve detection efficiency and accuracy, and reduce detection costs.

[0006] The present invention proposes a screen orange peel detection method based on phase deflectometry, which uses a combination of structured light projection and phase analysis to detect orange peel defects by analyzing the deformation of the screen surface to the stripe pattern. The method includes the following steps:

[0007] Step S1: The camera collects images by selecting and projecting stripe light sources;

[0008] Step S2: performing distortion correction on the image captured by the camera;

[0009] Step S3: performing phase extraction and unwrapping on the corrected image;

[0010] Step S4: Based on the unwrapped phase data, perform three-dimensional image reconstruction and gradient calculation to obtain image surface height and gradient information;

[0011] Step S5: performing curvature calculation and defect feature extraction based on the reconstructed image surface height distribution and gradient information;

[0012] Step S6: Quantitatively evaluate the orange peel defect based on the curvature and defect feature information.

[0013] As a further solution of the present invention, in step S1, the stripe light source image selection includes using a multi-frequency sinusoidal stripe pattern, where the sinusoidal stripe pattern is expressed as: I(x, y) = I0[1+γcos(2πfx+Φ)]; wherein I0 is the average light intensity, γ is the modulation index, f is the stripe frequency, and Φ is the initial phase; and the projection method specifically includes using a multi-step phase shift method for fringe projection, specifically using a four-step phase shift method to project four stripe patterns with phases that differ by π / 2:

[0014] I1(x,y)=I0[1+γcos(2πfx)];

[0015] I2(x,y)=I0[1+γcos(2πfx+π / 2)];

[0016] I3(x,y)=I0[1+γcos(2πfx+π)];

[0017] I4(x,y)=I0[1+γcos(2πfx+3π / 2)].

[0018] In this invention, the stripe frequency is dynamically switched for different screen areas (such as high-curvature edges and flat central areas), with high-frequency stripes used at the edges (to improve detail resolution) and low-frequency stripes used in the center (to reduce unwrapping errors). Horizontal and vertical stripes: In actual detection, horizontal (along the x-direction) and vertical (along the y-direction) stripes are projected separately to obtain phase gradient information in the x- and y-directions. Four-step phase shifting uses one more image than three-step phase shifting, suppressing random noise (such as camera readout noise and ambient light fluctuations) through redundant data, improving phase calculation accuracy by 30% (root mean square error reduced from 0.1π to 0.07π). Multi-step phase shifting offsets ambient light (such as 500 lux background light in a workshop) by differentially capturing multiple images, more than doubling the signal-to-noise ratio (SNR) compared to single-stripe projection. The combination of multi-frequency sinusoidal stripes and four-step phase shifting is the core technology for high-precision orange peel detection using phase deflectometry. Multi-frequency solves the contradiction between global consistency and local high precision in phase unwrapping, making it suitable for complex curved surface detection. Four-step phase shifting suppresses noise through redundant images, improving the anti-interference capability of phase extraction. This design ensures the system's robust detection of different screen types (rigid / flexible) and different defect levels in industrial practice.

[0019] As a further solution of the present invention, the step S2 specifically includes: camera calibration: using a calibration plate to capture multiple images, obtain the correspondence between the image coordinates and the actual physical coordinates, and calculate the internal and external parameters and distortion coefficients of the camera;

[0020] Establish a correction model: According to the camera calibration results, establish a correction model that includes radial and tangential distortion. The radial distortion correction model: x d =x u ·(1+k1·r 2 +k2·r 4 +k3·r 6 );y d =y u ·(1+k1·r 2 +k2·r 4 +k3·r 6 ); where (x u， y u ) is the point without distortion, (x d， y d ) is the image point after distortion; r is the distance from the point to the optical axis, k 1， k2, k3 are radial distortion coefficients; tangential distortion correction model: x d =x u +[2P1·x u ·y u +2P2·(r 2 +2x u 2 )];

[0021] y d =y u +[P2·x u ·y u +2P1·(r 2 +2y u 2 )], where P 1, P2 is the tangential distortion parameter;

[0022] Image remapping: Using a correction model, the pixel coordinates of the distorted image are mapped to ideal, undistorted coordinates, generating a corrected image. Through high-precision internal and external parameter calculations and nonlinear distortion compensation, this method controls lens-induced geometric errors to the sub-pixel level, providing reliable image data for subsequent phase extraction, 3D reconstruction, and defect detection. The core value of this step lies in establishing a precise mapping between image pixels and the physical world, ensuring the geometric authenticity of the measurement results.

[0023] As a further solution of the present invention, step S3 specifically includes: phase extraction: calculating the wrapping phase of each pixel based on the four images of the four-step phase shift method: Phase unwrapping: Since the arctan function has a value range of [-π, π], the calculated phase has a jump of 2π. A multi-frequency heterodyne method is used for phase unwrapping to obtain a continuous absolute phase Φ(x, y). The present invention can perform phase unwrapping across the entire image range, avoiding the error accumulation problem that may occur with local unwrapping methods. The multi-frequency heterodyne method has good adaptability for complex surface topography and a large phase variation range, and can accurately calculate the absolute phase.

[0024] As a further embodiment of the present invention, step S4 specifically comprises: gradient calculation: calculating the surface height gradient according to the unwrapped phase data:; ; Among them, Φ x and Φ y are the phase distribution in the horizontal and vertical directions, respectively, f x and f y is the fringe frequency in the corresponding direction;

[0025] Surface reconstruction: Reconstruct the surface height distribution by integrating the gradient field: ;

[0026] The specific implementation uses the Fast Fourier Transform (FFT) method: ; Where Z(u,v) is the Fourier transform of the surface height, G x (u,v) and G y (u,v) are and The Fourier transform of the image is obtained, with (u, v) being the frequency domain coordinates. In this invention, a single calculation covers the entire screen, avoiding the time loss of point-by-point scanning. Through the rational design of fringe parameters, calibration procedures, and algorithm optimization, reliable conversion from phase resolution to 3D topography can be achieved.

[0027] As a further solution of the present invention, the step S5 specifically includes: curvature calculation: calculating Gaussian curvature and mean curvature according to the reconstructed surface height distribution: ; ;Where K is the Gaussian curvature and H is the mean curvature;

[0028] Feature Extraction: Bandpass filtering removes low-frequency background and high-frequency noise; local contrast enhancement enhances contrast in local areas, and histogram equalization optimizes grayscale distribution to improve image contrast. This method quantitatively describes surface curvature using Gaussian and mean curvatures, effectively distinguishing the micro-undulations of orange peel defects from the macroscopic morphology of normal surfaces. Bandpass filtering removes irrelevant frequencies, while CLAHE and histogram equalization enhance local contrast, making micron-level defects visually prominent.

[0029] As a further solution of the present invention, step S6 also includes defect feature fusion and defect assessment: specifically including adaptive feature weight algorithm, Fisher discriminant score: ;in: and are the means of the i-th feature in the foreground and background regions, respectively, and are the corresponding variances respectively;

[0030] Adaptive weights: ;

[0031] Image fusion and defect map generation:

[0032] Multi-feature fusion: ;

[0033] Among them: F i is the normalized feature map, w i is the feature weight;

[0034] Nonlinear enhancement: D enhanced (x,y)=(D(x,y)) γ ;

[0035] Where γ<1 is a nonlinear parameter used to enhance weak defect areas;

[0036] Calculation of cellulite assessment index:

[0037] Comprehensive scoring model: ;

[0038] Where: f i is the eigenvalue, Φ i Is a nonlinear mapping function: , w i is the feature weight, and The present invention combines Fisher discriminant with nonlinear model to realize automatic classification of defects and avoid the subjectivity of manual judgment.

[0039] As a further solution of the present invention, step S6 specifically includes: performing image enhancement processing on the obtained horizontal and vertical gradient maps and curvature maps to improve the visibility of orange peel defects; specifically including receiving the gradient map obtained by phase analysis and ; Calculate Gaussian curvature K and mean curvature H;

[0040] Basic enhancement: Apply multi-scale decomposition enhancement: ;

[0041] Frequency domain bandpass filtering: I2=BandpassFilter(I1);

[0042] Local contrast optimization:

[0043] Adaptive CLAHE: I3=AdaptiveCLAHE(I2)

[0044] Laplace-Gaussian enhancement: ; Feature fusion: Directional gradient and curvature fusion: ;

[0045] High-order statistical feature enhancement: I6=HigherOrderEnhancement(I5);

[0046] Final defect map:

[0047] Feature weight calculation: w1,w2,w3,w4=AdaptiveWeights(I3,I4,I5,I6);

[0048] Weighted fusion: D = w1I3 + w2I4 + w3I5 + w4I6;

[0049] Nonlinear enhancement: D final =D γ, γ = 0.7. The image enhancement processing presented in this paper builds a complete link from raw gradient / curvature data to a high-signal-to-noise ratio defect image through multi-scale decomposition, frequency-domain filtering, adaptive contrast enhancement, and multi-feature fusion. Targeting the high-frequency micro-undulations of orange peel defects, bandpass filtering and nonlinear enhancement improve the visual response of tiny defects by 5-10 times. Gradients locate defect edges, curvature describes local shape, and high-order statistics capture distribution anomalies. The weighted fusion of these three reduces the missed detection rate from 12% to 2%.

[0050] The present invention also provides a detection system that primarily includes a stripe light source, a camera, a computer processing unit, and a bracket fixture. The system utilizes a non-contact detection method, using a stripe light source to project a specific stripe pattern onto the screen surface to be tested. The camera captures the reflected image, and the computer processing unit analyzes the stripe deformation in the image to detect and evaluate orange peel defects. Specifically, the stripe light source projects the stripe pattern onto the screen surface; the camera captures the reflected image; the bracket fixture secures the stripe light source and camera; the computer processing unit performs image processing and analysis; and a calibration device is used to calibrate the system. The stripe light source utilizes a multi-frequency sinusoidal stripe pattern, including stripes in both horizontal and vertical directions.

[0051] The present invention has the following beneficial effects: Fast detection speed: using an area array camera and structured light projection method, full-screen data can be obtained in one imaging, and the detection time is shortened to seconds, which is 5-10 times more efficient than the traditional linear array scanning method. High detection accuracy: through phase deflection, nanometer-level surface unevenness changes can be detected, and the orange peel defect detection rate reaches more than 95%, which is much higher than the traditional method. Low light source requirements: The system has reduced requirements for ambient light and incident angle, simple debugging, strong adaptability, and greatly improved application flexibility on the production line. Non-contact detection: using optical methods, there is no need to contact the screen surface, avoiding secondary damage to the product.

[0052] Quantitative Assessment: Quantitatively assesses orange peel defects using characteristic parameters such as Gaussian curvature and mean curvature, providing objective and consistent results. Highly Adaptable: Applicable to a variety of screen surfaces, including glass, plastic, and metal. Highly Cost-Effective: The system boasts a relatively simple structure, long service life, and low maintenance, reducing total cost of ownership by approximately 30%. This invention achieves this through the synergistic effect of hardware innovation (area array imaging + structured light), algorithmic breakthroughs (phase resolution + multi-feature fusion), and engineering optimization (adaptive control + cost design), creating a high-speed, high-precision, and highly adaptable screen orange peel detection solution. Its core value lies in transforming laboratory-grade phase deflectometry into an industrial-grade online detection system, resolving the efficiency, accuracy, and cost constraints of traditional methods. This system promotes the intelligent upgrade of screen quality control from "spot-based qualitative inspection" to "full-scale quantitative inspection," offering significant industry-wide adoption and economic benefits.

[0053] 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

[0054] Figure 1 Flow chart of the detection method of the present invention;

[0055] Figure 2a and Figure 2b Schematic diagrams of horizontal and vertical stripes projection respectively;

[0056] Figure 3 This is the phase extraction and unwrapping flow chart;

[0057] Figure 4 Schematic diagram of surface reconstruction and curvature calculation;

[0058] Figure 5 This is the flow chart for orange peel defect feature extraction and evaluation. DETAILED DESCRIPTION

[0059] 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.

[0060] The present invention discloses a screen orange peel detection method based on phase deflectometry, which includes the following steps: equipment selection and construction, including camera selection, stripe light source selection and bracket construction; stripe light source image selection and projection, using a multi-step phase shift method to project the stripe pattern; camera calibration and geometric calibration, establishing a system coordinate transformation relationship; distortion correction, eliminating the influence of lens distortion on the image; phase extraction and unwrapping, obtaining continuous absolute phase information; three-dimensional reconstruction and gradient calculation, obtaining surface height and its gradient information; curvature calculation and feature extraction, calculating Gaussian curvature and average curvature, wherein the curvature calculation step includes calculating Gaussian curvature and average curvature, and performing image enhancement processing on the curvature map; defect detection and evaluation, quantitatively evaluating orange peel defects, wherein the defect detection and evaluation step includes using the OTSU algorithm for threshold segmentation and performing comprehensive scoring based on area ratio, contrast, curvature characteristics and texture characteristics, and three-dimensional reconstruction using the fast Fourier transform (FFT) method to perform surface height integral reconstruction.

[0061] Example 1, reference Figure 1-Figure 5 As shown, the method of combining structured light projection with phase analysis is used to detect orange peel defects by analyzing the deformation of the screen surface to the stripe pattern. The specific technical solution is as follows:

[0062] Step S1: The camera captures images by selecting and projecting a streak light source image. This specifically includes: Image Acquisition Process: Equipment Selection and Setup: Camera Selection: The MERS-502-79U3M camera was selected. This camera utilizes the Sony IMX250MZR CMOS polarization sensor, providing sharp 5-megapixel image output and capturing rich details, making it suitable for applications requiring high image quality. The accompanying lens has a 25mm focal length, which is appropriate for the workpiece size. Compared to wide-angle lenses, this lens offers less distortion, allowing for a more realistic representation of the object's shape and size, making it crucial for applications requiring precise measurement or shape analysis.

[0063] Stripe Light Source Selection: A programmable stripe light source with a resolution of 1920×1080 and a brightness of at least 3000 lumens is required to provide a clear and stable stripe pattern. This light source allows for flexible and practical control of stripe parameters. It utilizes a high-brightness constant-current backlight for increased brightness and supports multi-camera triggering for superior synchronization. Other equipment includes a glass plane mirror, computer processing unit, bracket fixture, and calibration plate.

[0064] Stripe light source image selection and projection:

[0065] Stripe Light Source Image Selection: A multi-frequency sinusoidal fringe pattern is used, including both horizontal and vertical stripes. The fringe frequency is designed based on the required detection accuracy. The sinusoidal fringe pattern can be expressed as: I(x,y)=I0[1+γcos(2πfx+Φ)]; where I0 is the average light intensity, γ is the modulation index, f is the fringe frequency, and Φ is the initial phase.

[0066] Projection method: Multi-step phase shift method is used for fringe projection. Usually, a four-step phase shift method is used to project four fringe patterns with phases that differ by π / 2:

[0067] I1(x,y)=I0[1+γcos(2πfx)];

[0068] I2(x,y)=I0[1+γcos(2πfx+π / 2)];

[0069] I3(x,y)=I0[1+γcos(2πfx+π)];

[0070] I4(x,y)=I0[1+γcos(2πfx+3π / 2)].

[0071] Step S2: performing distortion correction on the image captured by the camera; specifically including:

[0072] Camera calibration and geometric calibration:

[0073] Camera calibration: This is the process of using Zhang's calibration method to calibrate the camera's intrinsic parameters, including focal length, principal point coordinates, distortion coefficients, and other parameters, as well as external parameters (such as rotation matrix and translation vector), to establish the camera imaging model. This camera calibration uses a dot calibration plate, and the specific parameters are as follows:

[0074] ①Calibration plate model: HC7-100-5;

[0075] ②Size: 100mm×100mm;

[0076] ③Number of dots: 7×7;

[0077] ④Dot diameter: 5mm;

[0078] ⑤ Dot center distance: 10mm;

[0079] During the calibration process, the correspondence between the image coordinates of the dots on the calibration plate and the known world coordinates is extracted, and the intrinsic and extrinsic parameters of the camera are solved using a mathematical model.

[0080] Geometric calibration: Determine the geometric relationship between the camera and the projector, and establish the transformation relationship between the world coordinate system, the camera coordinate system, and the projector coordinate system. The purpose of geometric calibration is to establish an accurate mapping relationship between the camera imaging coordinates and the coordinates of the actual physical world (or measurement system). This step usually includes: Plane mapping correction: For a planar calibration plate, the image coordinates can be transformed into the actual coordinates of the plane where the calibration plate is located using the homography matrix. Let H be the homography from the world coordinates (calibration plate plane) to the image coordinates, then: ; Three-dimensional rigid transformation: If the camera coordinates need to be converted to the world coordinates of the actual measurement system, the external parameters are used: P=R*P+T; In Pmd measurement, geometric calibration can be used to determine the geometric relationship between the camera installation position and the calibration plate installation position to ensure the accuracy of subsequent measurements. Calibration results: Get the camera intrinsic parameter matrix And the distortion parameter vector D={k1,k2,k3,p1,p2}, where k1,k2,k3 are radial distortion parameters, p1,p2 are tangential distortion parameters, and the extrinsic parameter matrix [R|T] between the projector and the camera.

[0081] Distortion Correction: Using the distortion parameters obtained through calibration, the camera's captured images are corrected for distortion, removing image distortion caused by lens distortion. During the camera imaging process, the optical properties of the lens may cause image distortion, primarily radial and tangential. This distortion can cause straight lines in the image to appear curved, affecting both measurement and visual quality. Therefore, distortion correction is a crucial step in camera calibration. The steps for distortion correction are as follows: Camera Calibration: Using a calibration plate (such as a checkerboard), capture multiple images to determine the correspondence between image coordinates and actual physical coordinates, and calculate the camera's intrinsic and extrinsic parameters, as well as the distortion coefficients. Calibration Model Creation: Based on the calibration results, a correction model is constructed that accounts for radial and tangential distortion.

[0082] Radial distortion correction model: x d =x u ·(1+k1·r 2 +k2·r 4 +k3·r 6 );

[0083] y d =y u ·(1+k1·r 2 +k2·r 4 +k3·r 6 ); where (x u， y u ) is the point without distortion, (x d， y d ) is the image point after distortion; r is the distance from the point to the optical axis, k 1， k2, k3 are radial distortion coefficients.

[0084] Tangential distortion correction model:

[0085] x d =x u +[2P1·x u ·y u +2P2·(r 2 +2x u 2 )];

[0086] y d =y u +[P2·x u ·y u +2P1·(r 2 +2y u 2 )]; Image remapping: Using the correction model, the pixel coordinates of the distorted image are mapped to the ideal coordinates without distortion to generate the corrected image.

[0087] Step S3: performing phase extraction and unwrapping on the corrected image; specifically comprising:

[0088] Phase extraction and unwrapping: Phase extraction: Based on the four images of the four-step phase shift method, calculate the wrapped phase of each pixel: Phase unwrapping: Since the value range of the arctan function is [-π, π], the calculated phase has a 2π jump. The multi-frequency heterodyne method is used for phase unwrapping to obtain a continuous absolute phase Φ (x, y).

[0089] Step S4: Based on the unwrapped phase data, perform three-dimensional image reconstruction and gradient calculation to obtain image surface height and gradient information; specifically, the following steps are performed:

[0090] 3D reconstruction and gradient calculation: Gradient calculation: Calculate the surface height gradient based on the unwrapped phase data: ; ; Among them, Φ x and Φ y are the phase distribution in the horizontal and vertical directions, respectively, f x and f y is the fringe frequency in the corresponding direction. Surface reconstruction: Reconstruct the surface height distribution by gradient field integration: ; The specific implementation can use the fast Fourier transform (FFT) method: ; where Z(u,v) is the Fourier transform of the surface height, G x (u,v) and G y (u,v) are and The Fourier transform of , (u, v) is the frequency domain coordinate.

[0091] Step S5: Calculate curvature and extract defect features based on the reconstructed image surface height distribution and gradient information; specifically, the following steps are involved:

[0092] Curvature calculation: Based on the reconstructed surface height distribution, calculate the Gaussian curvature and mean curvature: ; ; Where K is the Gaussian curvature and H is the mean curvature.

[0093] Image Enhancement: The resulting horizontal and vertical gradient maps and curvature maps are enhanced to improve the visibility of cellulite defects. Bandpass Filtering: Removes low-frequency background and high-frequency noise, preserving the frequency components of cellulite defects. Local Contrast Enhancement (CLAHE): Enhances contrast in local areas. Histogram Equalization: Optimizes grayscale distribution to improve image contrast.

[0094] Key algorithms and mathematical models for defect enhancement involved in this invention

[0095] 1. Multi-scale enhancement strategy

[0096] Gaussian pyramid decomposition and Laplace reconstruction:

[0097] Gaussian pyramid generation:

[0098] G0(x,y)=I(x,y);

[0099] G i (x,y)=DOWN(G i-1 (x,y));

[0100] Laplace Pyramid:

[0101] L i (x,y)=G i (x,y)-UP(G i+1 (x,y));

[0102] Enhanced Refactoring: ;in, is the weight coefficient of each scale, usually smaller scale (high frequency) corresponds to Larger values ​​emphasize subtle variations in orange peel defects.

[0103] 2. Adaptive bandpass filtering

[0104] Frequency domain bandpass filter design: ;in: is the frequency domain distance function; D L and D Hare the low-frequency and high-frequency cutoff points respectively; n is the filter order, which controls the steepness of the filter;

[0105] Frequency domain filtering process:

[0106] F(u,v)=F{f(x,y)};

[0107] G(u,v)=F(u,v)•H(u,v);

[0108] g(x,y)=F -1 {G(u,v)};

[0109] 3. Enhanced local contrast algorithm (CLAHE improved version)

[0110] Adaptive parameter CLAHE based on local characteristics of the image:

[0111] Adaptive cropping limitations: ;

[0112] where σ local is the local area standard deviation, and σ0 is the standardization parameter.

[0113] Adaptive local histogram equalization: Divide the image into N×N sub-blocks; calculate the histogram of each sub-block: h k (i) Apply adaptive clipping constraints:

[0114] h ’ k (i)=min(h k (i), clip_limit); redistribute the clipped part: Evenly distribute To all gray levels: ; Calculate the cumulative distribution function (CDF): ; Normalized CDF: ; Transformation function: T k (i)=round(cdf ’ k (i)); Bilinear interpolation fusion of adjacent block mappings;

[0115] 4. Gaussian-Laplacian operator enhancement

[0116] LoG operator enhancement: ; Among them: G σ is a Gaussian function with standard deviation σ: ; is the Laplace operator: ;α is the enhancement strength coefficient;

[0117] Laplace-Gaussian (LoG) operator expression: ;

[0118] Gradient map and curvature map fusion enhancements include:

[0119] 1. Directional gradient weight fusion

[0120] Gradient direction consistency calculation: ;

[0121] where ε is a small constant that prevents division by zero.

[0122] Adaptive Weight Fusion:

[0123] w curv (x,y)=w0+(1-w0)•C grad (x,y);

[0124] ;

[0125] Where: K is the normalized Gaussian curvature; H is the normalized mean curvature; w1, w2, w3, w4 are the weight coefficients of each feature; w0 is the basic curvature weight.

[0126] 2. Multi-scale fusion in wavelet domain

[0127] Wavelet decomposition: W ψ f(j,k)=∫f(x)ψ j,k (x)dx;

[0128] where ψ j,k (x)=2 -j / 2 ψ(2 -j -k) is a wavelet function with scale j and translation k.

[0129] Adaptive wavelet coefficient fusion:

[0130] W fused (j,k)=λ(j,k)W grad (j,k)+(1-λ(j,k))W curv (j,k);

[0131] The weight function λ(j,k) is calculated based on the energy ratio: ;

[0132] Wavelet reconstruction: ;

[0133] The enhancement of orange peel defect characteristics in the present invention includes:

[0134] 1. Texture spectrum-based enhancement

[0135] Radial Frequency Analysis:

[0136] 1. Perform a two-dimensional Fourier transform on the image f(x,y): F(u,v)=F{f(x,y)}

[0137] 2. Calculate the amplitude spectrum:

[0138] 3. Radial energy calculation: , where C r is a circle with the origin as its center and radius r

[0139] 4. Characteristic frequency detection: r peak =argmax r E(r),r>r min;

[0140] Spectral enhancement filter: ;in: is the distance to the center of the frequency domain; r peak is the characteristic frequency; σ r is the bandwidth parameter; β is the enhancement coefficient;

[0141] Enhanced image:

[0142] F enhanced (u,v)=F(u,v)•H texture (u,v);

[0143] f enhanced (x,y)=F -1 {f enhanced (u,v)};

[0144] 2. High-order statistical feature enhancement

[0145] Local high-order moment calculation:

[0146] 1. Local zero mean: ,in is the mean value of pixels in the window centered at (x, y);

[0147] 2. Local variance:

[0148] 3. Local third moment (skewness):

[0149] 4. Local fourth-order moment (kurtosis):

[0150] High-order feature fusion enhancement:

[0151] ;

[0152] ;

[0153] Where: f'(x,y) is the normalized original image; α is the weight coefficient of skewness and kurtosis; γ is the enhancement strength coefficient;

[0154] Step S6: Quantitatively evaluate the orange peel defect based on the curvature and defect feature information, specifically including:

[0155] Feature fusion and defect assessment:

[0156] 1. Adaptive feature weight algorithm

[0157] Fisher discriminant score: ;

[0158] in: and are the means of the i-th feature in the foreground and background regions respectively; and are the corresponding variances respectively;

[0159] Adaptive weights: ;

[0160] 2. Image fusion and defect map generation

[0161] Multi-feature fusion: ;

[0162] Among them: F i is the normalized feature map, w i is the feature weight

[0163] Nonlinear enhancement:

[0164] D enhanced (x,y)=(D(x,y)) γ ;

[0165] Where γ<1 is a nonlinear parameter used to enhance weak defect areas.

[0166] 3. Calculation of cellulite evaluation index

[0167] Comprehensive scoring model: ;

[0168] Where: f i is the characteristic value (area ratio, contrast, curvature, etc.); Φ i Is a nonlinear mapping function: ;w i is the feature weight, and For mapping parameters

[0169] The complete enhancement process in the present invention includes:

[0170] Input processing: receiving the gradient map obtained by phase resolution and ; Calculate Gaussian curvature K and mean curvature H;

[0171] Basic enhancement: Apply multi-scale decomposition enhancement: ;

[0172] Frequency domain bandpass filtering:

[0173] I2=BandpassFilter(I1);

[0174] 3. Local contrast optimization: Adaptive CLAHE:

[0175] I3=AdaptiveCLAHE(I2);

[0176] Laplace-Gaussian enhancement: ;

[0177] 4. Feature Fusion:

[0178] Directional gradient and curvature fusion: ;

[0179] High-order statistical feature enhancement: I6=HigherOrderEnhancement(I5);

[0180] 5. Final defect map:

[0181] Feature weight calculation: w1,w2,w3,w4=AdaptiveWeights(I3,I4,I5,I6);

[0182] Weighted fusion: D = w1I3 + w2I4 + w3I5 + w4I6;

[0183] Nonlinear enhancement: D final =D γ ,γ=0.7;

[0184] This enhancement process, through multi-level feature extraction and fusion, effectively captures subtle textural variations in orange peel, significantly improving detection sensitivity and accuracy. Especially for subtle, imperceptible orange peel defects, the enhanced image clearly displays their characteristic morphology, providing a reliable basis for subsequent defect detection and assessment.

[0185] In Example 2, the present invention further provides a screen orange peel detection system based on phase deflectometry;

[0186] Includes the following components: Stripe light source: A programmable stripe light source is used with a resolution of 1920×1080 and a brightness of 4500 lumens, which can project stripe patterns of various frequencies and phases. Industrial camera: A HIKROBOTMV-CA5000-10GC industrial camera is selected with a resolution of 2448×2048, a frame rate of 40fps, and an industrial lens with a focal length of 16mm. Bracket system: A three-dimensional adjustable bracket built with aluminum profiles is used to fix the projector and camera to ensure system stability and precise positioning. Computing processing unit: An industrial computer equipped with an Intel Core i7 processor, 32GB of memory, and an NVIDIA RTX 3060 graphics card is used for image acquisition, processing, and analysis. Calibration plate: A 9×7 checkerboard calibration plate with a grid spacing of 10mm is used for system calibration.

[0187] The specific testing process is as follows:

[0188] 1. System calibration

[0189] (1) Camera intrinsic parameter calibration: Place the calibration plate at different positions and angles, take at least 15 calibration images, and use the calibrateCamera function in the OpenCV library to calculate the camera intrinsic parameter matrix and distortion coefficient.

[0190] (2) Geometric calibration: A calibration plate is placed in the inspection area. The calibration pattern is simultaneously captured by the camera and projected by the projector to establish the transformation relationship between the camera coordinate system and the projector coordinate system. The calibration results include: the camera intrinsic parameter matrix K; the distortion coefficient vector D = [k1, k2, p1, p2, k3]; the rotation matrix R and translation vector T between the projector and the camera. In this embodiment, a 9×7 checkerboard calibration plate (with a grid spacing of 10 mm) is placed at different positions and angles (at least 15 viewing angles), and the industrial camera captures the calibration images. Using Zhang's calibration method in OpenCV, we obtain the correspondence between image coordinates and world coordinates (the calibration plate plane) through corner detection (sub-pixel accuracy). We then calculate the camera's intrinsic parameter matrix K (including focal length and principal point coordinates) and distortion coefficients D = [k1, k2, p1, p2, k3] (radial and tangential distortion parameters). The intrinsic parameter matrix K is used for subsequent image-to-camera coordinate conversion, and the distortion coefficients D are used to correct image distortion, with the reprojection error controlled within 0.8 pixels. A conversion relationship is established between the camera and projector coordinate systems to ensure geometric consistency between fringe projection and image acquisition.

[0191] Stripe pattern design and projection: A sinusoidal stripe pattern with four phase shifts is designed. The stripe frequency is selected based on the required detection accuracy. For typical mobile phone screen inspection, the horizontal and vertical stripe frequency is set to 50 lines per screen. The four-step phase shifted stripe pattern is projected in sequence in the horizontal and vertical directions. The camera simultaneously captures the reflected images, acquiring four images in each direction, for a total of eight images.

[0192] Phase extraction and unwrapping, using a four-step phase shift algorithm to calculate the wrapped phase, calculate the wrapped phase in the horizontal and vertical directions; gradient calculation and surface reconstruction, calculate the surface gradient based on the unwrapped phase, and use the fast Fourier transform (FFT) method to reconstruct the surface height; curvature calculation and feature extraction, calculate Gaussian curvature and mean curvature, image enhancement processing, enhance the curvature map; defect detection and assessment, use OTSU thresholding to segment potential orange peel defect areas, extract features and evaluate the degree of orange peel defects.

[0193] In this embodiment, defect detection and assessment: quantitative grading and visualization, OTSU threshold segmentation and morphological processing;

[0194] Automatic thresholding: The OTSU algorithm is used to automatically calculate the optimal threshold for the enhanced curvature map, binarizing the image into defects (white) and background (black), with a segmentation accuracy of >95%.

[0195] Morphological operations: Opening (erosion followed by dilation) was performed using a 3×3 kernel to remove isolated noise points with an area less than 100 pixels and connect adjacent defect areas to ensure contour integrity (reducing the breakage rate from 20% to 5%).

[0196] Multi-feature fusion evaluation: Geometric features: Calculate the defect area ratio (defect area / full screen area), average curvature, maximum curvature, and quantify the defect size and degree of curvature.

[0197] Texture features: Energy (texture regularity) and entropy (texture complexity) are calculated through the gray-level co-occurrence matrix (GLCM) to capture the periodic micro-undulations of orange peel.

[0198] Comprehensive scoring model: weighted fusion area ratio (0.3), contrast (0.25), average curvature (0.25), texture entropy (0.2), with scores of 0-1 corresponding to defect levels (none / mild / moderate / severe), achieving objective grading (repeatability accuracy <0.05 points).

[0199] This invention automates the entire inspection process, from image acquisition to quantitative evaluation, through a closed loop of optical projection, phase resolution, 3D reconstruction, feature enhancement, and defect identification. Key technologies include: multi-step phase shifting and quality-guided unwrapping, which improve phase resolution accuracy and adapt to complex curved surfaces; FFT integration and curvature calculation, which rapidly reconstruct surface topography and capture micron-level fluctuations; and CLAHE and multi-feature fusion, which enhance defect visibility and support robust detection in industrial environments. This system achieves a 96.5% defect detection rate while maintaining an inspection speed of 100ms per screen, providing a highly efficient solution for automated quality control in screen production.

[0200] Example 3. The present invention also provides a multi-scale detection system. Specifically, for large-size screens (such as TV panels), the system adopts a multi-area splicing method for detection; for small-size screens (such as mobile phone panels), the system adopts a high-precision imaging mode for detection.

[0201] Among them, multi-area stitching detection: divide the large screen into several areas, perform fringe projection and phase calculation on each area separately; use the overlapping parts of the areas to perform stitching correction to eliminate stitching errors; and perform an overall analysis of the complete gradient map and curvature map after stitching.

[0202] High-precision imaging mode: uses a macro lens to improve spatial resolution; adopts a higher-frequency fringe pattern to increase phase sensitivity; introduces an additional noise suppression algorithm to improve the detection rate of tiny orange peel defects;

[0203] In this embodiment, the system can adapt to the detection requirements of screens of different sizes from 5 inches to 65 inches, and the detection accuracy can reach the micron level, meeting the orange peel defect detection requirements of various display devices.

[0204] 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 orange peel detection method based on phase deflectometry, characterized in that: The following steps are involved: Step S1: The camera collects images by selecting and projecting stripe light sources; Step S2: performing distortion correction on the image captured by the camera; Step S3: performing phase extraction and unwrapping on the corrected image; Step S4: Based on the unwrapped phase data, perform three-dimensional image reconstruction and gradient calculation to obtain image surface height and gradient information; Step S5: performing curvature calculation and defect feature extraction based on the surface height distribution and gradient information of the reconstructed image; Step S6: quantitatively evaluating the orange peel defect based on the curvature and defect feature information; It also includes defect feature fusion and defect assessment: specifically, adaptive feature weighting algorithm and Fisher discriminant score: ;in: and are the means of the i-th feature in the foreground and background regions, respectively, and are the corresponding variances respectively; Adaptive weights: ; Image fusion and defect map generation: Multi-feature fusion: ; Among them: F i is the normalized feature map, w i is the feature weight; Nonlinear enhancement: D enhanced (x,y)=(D(x,y)) γ ; Where γ is a nonlinear parameter, and when γ < 1, it is used to enhance weak defect areas; Calculation of cellulite assessment index: Comprehensive scoring model: ; Where: f i is the eigenvalue, Φ i Is a nonlinear mapping function: , w i is the feature weight, and For mapping parameters.

2. The screen orange peel detection method based on phase deflectometry according to claim 1, characterized in that: In step S1, the stripe light source image selection includes adopting a multi-frequency sinusoidal stripe pattern, and the sinusoidal stripe pattern is represented as: I(x,y)=I0[1+γcos(2πfx+Φ)]; where I0 is the average light intensity, γ is the modulation index, f is the fringe frequency, and Φ is the initial phase. The projection method specifically includes using a multi-step phase shift method for fringe projection, specifically using a four-step phase shift method to project four fringe patterns with phases that differ by π / 2: I1(x,y)=I0[1+γcos(2πfx)]; I2(x,y)=I0[1+γcos(2πfx+π / 2)]; I3(x,y)=I0[1+γcos(2πfx+π)]; I4(x,y)=I0[1+γcos(2πfx+3π / 2)].

3. The screen orange peel detection method based on phase deflectometry according to claim 2, characterized in that: The step S2 specifically includes: camera calibration: using a calibration plate to capture multiple images, obtaining the correspondence between image coordinates and actual physical coordinates, and calculating the camera's internal and external parameters and distortion coefficients; Establish a correction model: According to the camera calibration results, establish a correction model that includes radial and tangential distortion. The radial distortion correction model: x d =x u ·(1+k1·r 2 +k2·r 4 +k3·r 6 );y d =y u ·(1+k1·r 2 +k2·r 4 +k3·r 6 ); where (x u， y u ) is the point without distortion, (x d， y d ) is the distorted image point; r is the distance from the point to the optical axis, k 1， k2, k3 are radial distortion coefficients; tangential distortion correction model: x d =x u +[2P1·x u ·y u +2P2·(r 2 +2x u 2 )];y d =y u +[P2·x u ·y u +2P1·(r 2 +2y u 2 )], where P 1, P2 is the tangential distortion parameter; Image remapping: Using the correction model, the pixel coordinates of the distorted image are mapped to the ideal coordinates without distortion to generate the corrected image. image.

4. The screen orange peel detection method based on phase deflectometry according to claim 3, characterized in that: The step S3 specifically includes: Phase extraction: Calculating the wrapping phase of each pixel based on the four images of the four-step phase shift method: Phase unwrapping: Since the value range of the arctan function is [-π, π], the calculated phase has a 2π jump. The multi-frequency heterodyne method is used for phase unwrapping to obtain a continuous absolute phase Φ (x, y).

5. The screen orange peel detection method based on phase deflectometry according to claim 4, characterized in that: The step S4 specifically includes: gradient calculation: calculating the surface height gradient based on the unwrapped phase data: ; ; Among them, Φ x and Φ y are the phase distribution in the horizontal and vertical directions, respectively, f x and f y is the fringe frequency in the corresponding direction, π is the circumference of the circle, , Indicates the gradient of the surface height in the x and y directions, that is, the tilt angle, and z represents the height of the object being measured; Image 3D reconstruction: Reconstruct the image surface height distribution through gradient field integration: , 、 Integration variables in the x and y directions; The specific implementation uses the Fast Fourier Transform (FFT) method: ; where Z(u,v) is the Fourier transform of the surface height, G x (u,v) and G y (u,v) are and The Fourier transform of , (u, v) is the frequency domain coordinate.

6. The screen orange peel detection method based on phase deflectometry according to claim 5, characterized in that: The step S5 specifically includes: curvature calculation: calculating Gaussian curvature and mean curvature according to the height distribution of the reconstructed image surface: ； ； Among them, K is the Gaussian curvature and H is the mean curvature; Feature extraction: Remove low-frequency background and high-frequency noise through bandpass filtering; enhance the local area through local contrast enhancement Contrast, and optimize grayscale distribution and improve image contrast through histogram equalization.

7. The screen orange peel detection method based on phase deflectometry according to claim 6, characterized in that: The step S6 specifically includes: performing image enhancement processing on the obtained horizontal and vertical gradient images and curvature images to improve the visibility of orange peel defects; specifically includes receiving the gradient images obtained by phase analysis and ; Calculate Gaussian curvature K and mean curvature H; Basic enhancement: Apply multi-scale decomposition enhancement: ; Frequency domain bandpass filtering: I2=BandpassFilter(I1); Local contrast optimization: Adaptive CLAHE: I3=AdaptiveCLAHE(I2); Laplace-Gaussian enhancement: ; Feature Fusion: Directional gradient and curvature fusion: ; High-order statistical feature enhancement: I6=HigherOrderEnhancement(I5); Final defect map: Feature weight calculation: w1,w2,w3,w4=AdaptiveWeights(I3,I4,I5,I6); Weighted fusion: D = w1I3 + w2I4 + w3I5 + w4I6; Nonlinear enhancement: D final =D γ , γ=0.

7.

8. A detection system using the screen orange peel detection method based on phase deflectometry as claimed in claim 1, characterized in that: include: Stripe light source, using programmable stripe light source, with a resolution of 1920×1080 and a brightness of 4500 lumens, used to project stripe patterns of various frequencies and phases; The camera is used to collect the reflected image; the bracket fixing device is used to fix the stripe light source and the camera; and the computer processing unit is used for image processing and analysis.

9. The detection system according to claim 8, wherein: For large-size screens, a multi-area stitching method is used for inspection; for small-size screens, a high-precision imaging mode is used for inspection.

Images