System and method for detecting chromatic aberration of liquid crystal display screen

Abstract

The invention relates to the technical field of display screen detection, and discloses a liquid crystal display screen color difference detection system and method.The system comprises a multispectral imaging module, a curved surface deformation modeling module, a coupling tensor decomposition module, a biological excitation feature extraction module, a dynamic compensation control module and a technological parameter interface module; the method comprises the following steps of synchronously collecting a multispectral image and screen deformation data, constructing and decomposing a multidimensional tensor containing geometric constraints, extracting biological excitation visual features, and generating and applying voltage compensation through model prediction control so as to dynamically correct the color difference of the flexible screen. According to the method, through multispectral and geometry synchronous perception, multidimensional tensor constraint decomposition, biological excitation and deformable convolution feature extraction, time-varying model prediction control and dynamic weight adjustment of fused manufacturing parameters, high-precision capture and personalized compensation of the dynamic chromatic aberration of the flexible screen are realized. The defects of dynamic perception, quick response and individual adaptability are overcome.

Description

Technical Field

[0001] The present invention relates to the technical field of display screen detection, and in particular to a system and method for detecting color difference of a liquid crystal display screen. Background Art

[0002] Liquid crystal displays, especially flexible and curved ones, are prone to complex color shifts under dynamic deformation, severely impacting the visual experience and professional applications. Existing color shift detection and calibration technologies primarily target static flat screens, using RGB images or point-to-point measurement for correction. These methods struggle to adapt to the dynamic, nonlinear color shifts caused by physical deformations like bending and stretching on flexible displays, and are unable to accurately capture the complex relationship between screen deformation and color shift.

[0003] Furthermore, current color difference feature extraction technologies rarely incorporate the characteristics of human visual perception and the effects of physical screen deformation, resulting in significant discrepancies between the proposed features and actual perception. Compensation control strategies are often static or simple feedback-based, slow to respond to rapid dynamic color differences and struggling to cope with complex color difference patterns driven by multiple factors. Furthermore, individual differences in screen manufacturing process parameters significantly impact color difference performance, but existing technologies often ignore this prior information and employ a unified calibration model, resulting in a lack of targeted compensation.

[0004] Therefore, the present invention provides a system and method for detecting color difference of a liquid crystal display screen to address the deficiencies of the prior art. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a system and method for detecting color differences in liquid crystal displays, which solves the problem of difficulty in accurately perceiving and effectively compensating for the complex, time-varying color differences generated by flexible or curved liquid crystal displays during dynamic physical deformation.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a system for detecting color difference of a liquid crystal display screen, the system comprising the following modules:

[0007] a multispectral imaging module configured to synchronously collect multi-channel narrowband spectral image data of the screen through a microlens array and a narrowband filter set;

[0008] A surface deformation modeling module constructs a dynamic surface manifold model based on the spatial distribution of the multi-channel narrow-band spectral image;

[0009] A coupled tensor decomposition module organizes the multi-channel narrowband spectral image data into a four-dimensional tensor comprising a spatial resolution dimension, an RGB color channel dimension, and a spectral channel dimension, and performs a Tucker decomposition with covariant derivative constraints based on the geometric constraints of the manifold model;

[0010] a bio-inspired feature extraction module that generates a color difference feature map with a dual antagonistic characteristic based on the decomposition result of the four-dimensional tensor, wherein the dual antagonistic characteristic simulates the color response mechanism of retinal ganglion cells;

[0011] a dynamic compensation control module, which establishes a time-varying partial differential equation including a time derivative of curvature according to the color difference characteristic map, and generates a feedforward compensation signal of the screen pixel voltage based on a model predictive control algorithm;

[0012] The process parameter interface module injects the polarizer thickness tolerance and dielectric constant parameters in the screen manufacturing process into the surface manifold model parameter calculation process in real time, and synchronously adjusts the compensation amount optimization weight coefficient of the dynamic compensation control module. The optimized weight coefficient is used to control the amplitude and response speed of the feedforward compensation signal.

[0013] Preferably, the multispectral imaging module includes:

[0014] An array of microlenses arranged at a preset density, with lens pitch configured to match the pixel distribution characteristics of the screen;

[0015] A filter set consisting of multiple narrowband filters, where the wavelength range of each channel continuously covers the visible spectrum and the wavelength ranges of adjacent channels partially overlap;

[0016] The synchronous trigger device generates synchronous acquisition instructions based on the screen refresh signal, controlling the time synchronization accuracy of image acquisition of each channel to the microsecond level.

[0017] Preferably, the surface deformation modeling module includes:

[0018] An array of displacement sensors arranged at the edge of the screen is used to obtain real-time three-dimensional coordinate data of the surface;

[0019] a curvature calculation device for dynamically calculating local curvature parameters based on the spatial positional relationship between adjacent sampling points;

[0020] The manifold update device adaptively adjusts the model update frequency according to the curvature change rate, and enables the interpolation algorithm to supplement the sampling points in the area of ​​severe deformation.

[0021] Preferably, the interpolation algorithm of the manifold updating device includes:

[0022] Curvature trend prediction device based on historical deformation data;

[0023] A spatial interpolation device to add virtual sampling points in areas with severe deformation;

[0024] The smoothing constraint device limits the variation range of the second-order derivative of the curvature parameter after interpolation.

[0025] Preferably, the coupled tensor decomposition module includes:

[0026] A multidimensional data organization unit maps the spatial coordinates, color channels, and spectral dimensions of the spectral image into a tensor structure;

[0027] Geometric constraint loading unit, which converts the differential geometric parameters of the manifold model into regularized constraints for tensor decomposition;

[0028] The iterative optimization unit uses a constrained optimization algorithm to solve the core tensor and factor matrix that satisfies the joint minimization of geometric features and data reconstruction errors.

[0029] Preferably, the bioinspired feature extraction module includes:

[0030] A dual-antagonistic channel construction device generates a difference channel between long-wave and medium-wave spectra and a difference channel between short-wave and synthetic brightness;

[0031] a nonlinear brightness processing device for applying a hyperbolic function transformation to the synthesized brightness component and adaptively adjusting the nonlinear intensity;

[0032] The deformable convolution device dynamically adjusts the size and orientation characteristics of the convolution kernel according to the local curvature parameters.

[0033] Preferably, the deformable convolution device comprises:

[0034] The main direction detection device determines the direction characteristics of the convolution kernel through curvature parameter analysis;

[0035] Dynamic kernel size adjustment device, which adjusts the convolution kernel coverage in inverse proportion to the absolute value of the local curvature;

[0036] The directional filtering device performs an anisotropic filtering operation that matches the main direction of curvature.

[0037] Preferably, the dynamic compensation control module includes:

[0038] A time-varying model construction device uses tensor decomposition characteristics, manifold geometric parameters and their time derivatives as coefficient terms of differential equations;

[0039] A rolling optimization device solves the optimization problem with physical constraints in the prediction time domain to generate a compensation control sequence;

[0040] The signal conversion device converts the optimization result into a voltage gradient signal that matches the screen driving circuit.

[0041] Preferably, the process parameter interface module includes:

[0042] Process parameter mapping device, which establishes a nonlinear relationship model between manufacturing parameters and algorithm adjustment coefficients;

[0043] A dynamic weight adjustment device adjusts the amplitude weight and response speed weight in the optimization target according to real-time process parameters;

[0044] The abnormality handling device activates the compensation amount limiting mechanism and generates an abnormal state mark when the parameter exceeds the preset range.

[0045] The present invention also provides a method for detecting color difference of a liquid crystal display screen, the method comprising the following steps:

[0046] S1, synchronously collect multispectral image data through microlens array and narrowband filter set;

[0047] S2, updating the manifold geometry model based on real-time measurement data from the displacement sensor array;

[0048] S3, jointly constructing a multidimensional tensor from multispectral data and manifold parameters and performing constraint decomposition;

[0049] S4, generating a visual enhancement feature map through a bio-inspired feature extraction module;

[0050] S5. Solve the optimization problem with physical constraints in the rolling time domain to generate a compensation signal;

[0051] S6. Convert the compensation signal into a driving voltage and inject it into the screen control circuit.

[0052] The present invention provides a system and method for detecting color difference of a liquid crystal display screen. It has the following beneficial effects:

[0053] 1. This invention uses a microlens array and a narrowband filter set to synchronously capture multispectral image data, and combines this with real-time measurement data from a displacement sensor array to update a manifold geometry model, achieving high-precision simultaneous perception of screen color information and physical form. Compared to existing solutions that rely solely on a single RGB image or static calibration, this invention addresses the inability to accurately capture the complex color distribution of flexible or curved screens under dynamic deformation.

[0054] 2. This invention combines multispectral data with manifold parameters to construct a multidimensional tensor and performs a covariant derivative-constrained Tucker decomposition based on manifold geometric constraints. This processing approach can deeply explore the intrinsic coupling relationship between color difference and physical screen deformation. Compared with traditional methods that simply process color and geometric information independently or shallowly fuse them, this invention overcomes the shortcomings of traditional methods that have difficulty revealing the core mechanism of deformation-induced color difference and insufficient feature extraction.

[0055] 3. This invention employs a bio-inspired feature extraction mechanism to generate a dual-antagonistic feature map simulating the color response of retinal ganglion cells based on tensor decomposition results. This feature is then processed using deformable convolution coupled with local curvature parameters. This ensures that the extracted color difference features are more consistent with human visual perception and can adapt to the local curvature of the screen. Compared to existing approaches that use standard image processing algorithms to extract color difference features, this invention addresses the significant differences in human perception and poor adaptability to curved surfaces.

[0056] 4. Based on the extracted visual enhancement feature map, the present invention establishes a time-varying partial differential equation model containing the time derivative of curvature, and uses a model predictive control algorithm to solve the optimization problem with physical constraints in the rolling time domain to generate a feedforward compensation signal. This strategy of combining dynamic modeling with optimization control makes the compensation behavior predictable and adaptable to dynamic changes. Compared with the existing solutions that mostly use static lookup tables or simple PID feedback control, the present invention solves the shortcomings of compensation lag, poor response to rapid deformation, and easy oscillation.

[0057] 5. This invention incorporates a process parameter interface module that allows real-time injection of parameters such as polarizer thickness tolerance and dielectric constant during screen manufacturing into the manifold model calculation and weight coefficient adjustment for dynamic compensation control. This incorporation of manufacturing prior information allows for fine-tuning of compensation strategies based on the inherent physical properties of specific screens. Compared to existing approaches that fail to account for individual manufacturing variations and utilize a unified compensation model, this invention addresses the lack of universality and adaptability of these compensation schemes to specific process deviations. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a system architecture diagram of the present invention;

[0059] Figure 2 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0061] See also Figure 1 The embodiment of the present invention provides a system for detecting color difference of a liquid crystal display screen, the system comprising the following modules:

[0062] a multispectral imaging module configured to synchronously collect multi-channel narrowband spectral image data of the screen through a microlens array and a narrowband filter set;

[0063] In this embodiment, the multispectral imaging module utilizes an optically integrated design to achieve high-precision capture of the screen's spectral characteristics. The microlens array's lens spacing is customized based on the screen's pixel layout. The optical axis of each microlens is aligned with the geometric center of the corresponding pixel area, ensuring that the pixel areas covered by a single lens are spatially continuous and non-overlapping. The lens array substrate is made of low-thermal expansion optical glass, and a photolithographic process forms an aspheric lens structure, effectively correcting for field curvature and chromatic aberration.

[0064] The narrowband filter set adopts a stepped wavelength distribution design. The center wavelength of each filter is evenly spaced in the visible light range, and the half-height width is controlled within the range of 10-15nm. The passband ranges of adjacent filters are set to partially overlap. The width of the overlapping area is 20%-30% of the half-height width of the filter to ensure the continuity of the spectral data in the wavelength dimension. The filter coating adopts a hard oxide multilayer film structure. The average transmittance in the wavelength range of 400-700nm is greater than 85%, and the out-of-band rejection ratio is better than 10 4 Magnitude.

[0065] The synchronization trigger device includes a signal conditioning circuit and an FPGA timing controller. When the rising edge of the vertical synchronization signal from the screen driver circuit is detected, the trigger circuit generates a pulse signal with an adjustable delay. This delay compensates for the screen pixel response delay and optical transmission time. The timing controller uses phase-locked loop technology to control the synchronization error of the exposure start time of each channel image sensor to the sub-microsecond level, satisfying the formula:

[0066]

[0067] Where Δt is the maximum allowable synchronization error, N is the number of spectral channels, and f refresh is the screen refresh rate. This constraint ensures that image acquisition of all channels is completed within the screen refresh period without motion blur.

[0068] The optical parameter design of the microlens array follows the following relationship:

[0069]

[0070] Where d is the lens diameter, p is the screen pixel pitch, D is the image sensor pixel size, and M is the optical system magnification. This relationship ensures that the pixel area imaged by a single lens corresponds to an integer multiple of the sensor pixel, avoiding aliasing. The numerical aperture of the lens is optimized based on the maximum change in screen surface curvature to ensure image clarity is maintained even under maximum deformation conditions.

[0071] The wavelength selection function of the filter set is modeled using a Gaussian transmission curve:

[0072]

[0073] Among them, λ c is the center wavelength of the filter, σ is the standard deviation parameter, and its relationship with the filter half-maximum width FWHM is: This model is used to construct the channel response matrix in the subsequent spectral reconstruction algorithm.

[0074] The microlens array preferably adopts a hexagonal close-packed structure, which improves the fill factor by 13.4% compared to a rectangular arrangement. Each lens unit is equipped with an independent aperture structure, and the aperture is dynamically adjusted according to the screen's luminous characteristics to prevent crosstalk between adjacent channels while maintaining the signal-to-noise ratio. The filter set is mounted on a rotatable wheel, driven by a stepper motor to achieve rapid switching of filter channels, with switching time synchronized with the screen refresh cycle.

[0075] The image sensor utilizes a global shutter CMOS device, with each spectral channel corresponding to an independent sensor module. Each sensor module is coupled to the corresponding filter output via a fiber bundle. The fiber numerical aperture is matched to the microlens output light cone angle to maximize light energy utilization. The sensor gain is adaptively adjusted based on the spectral energy distribution of each channel, ensuring consistent dynamic range of the output signals across different wavelength channels.

[0076] A surface deformation modeling module constructs a dynamic surface manifold model based on the spatial distribution of the multi-channel narrow-band spectral image;

[0077] In this embodiment, the surface deformation modeling module is designed to accurately and in real time capture the dynamic geometry of a flexible or curved screen. The displacement sensor array, preferably a non-contact optical displacement sensor such as a laser Doppler sensor or a structured light scanning module, is distributed along the periphery of the screen. This deployment method can obtain the three-dimensional coordinate data of the edge contour of the screen without interfering with the central display area of ​​the screen. The sampling frequency of the sensor matches the highest deformation frequency that may occur on the screen to ensure that rapid deformation processes can be captured.

[0078] The curvature calculation device receives a three-dimensional coordinate data point set {P i =(x i ,y i ,z i )}. Among them, P i represents the i-th sampling point, (x i ,y i ,z i) are the coordinates in the three-dimensional Cartesian coordinate system. For any point on the screen surface, its local curvature characteristics can be approximated by analyzing the spatial position relationship of its neighboring sampling points. A common method is based on discrete differential geometry theory, which estimates the principal curvatures k1, k2 and Gaussian curvature K=k1k2 by constructing local triangular facets or fitting local quadratic surfaces. For example, for sampling point P i , whose Gaussian curvature K i It can be obtained through its first-order neighborhood point set N(P i ) is calculated. If the local quadratic surface is used to fit z=ax 2 +by 2 +cxy+dx+ey+f, then the Gaussian curvature and mean curvature H=(k1+k2) / 2 can be analytically obtained from the fitting coefficients a,b,c,d,e,f.

[0079] The core function of the manifold update device is to adaptively adjust the complexity and update rate of the model according to the dynamic characteristics of the screen deformation. The device first calculates the curvature change rate Here, K represents Gaussian curvature, and t represents time. When the rate is below the preset low threshold, it indicates that the screen is deforming slowly or is static. In this case, the frequency of model updates can be reduced to conserve computing resources. When the curvature change rate exceeds the preset high threshold, it indicates that the screen is deforming drastically. In this case, it is necessary not only to increase the frequency of model updates but also to increase the density of sampling points in areas of drastic deformation.

[0080] Enable the interpolation algorithm to supplement sampling points in the area with severe deformation. The specific steps are as follows:

[0081] First, the curvature trend prediction device based on historical deformation data uses time series analysis methods (such as Kalman filtering or ARIMA model) to predict the curvature change trend of key sampling points. It can determine in advance which areas may be prone to drastic deformation. represents the predicted value of Gaussian curvature at the future time t+Δt, K(t) represents the Gaussian curvature at the current time t, K(t-Δt) represents the Gaussian curvature at the past time t-Δt, f(·) represents the prediction function, and Δt represents the time step.

[0082] Secondly, the spatial interpolation device generates virtual sampling points through interpolation algorithms based on the position and curvature information of existing sampling points in the predicted or real-time detected areas of severe deformation. Common interpolation methods include radial basis function (RBF) interpolation or Kriging interpolation. For example, for a point to be interpolated P new , whose coordinate z new It can be obtained by its adjacent real sampling point P j The value of z jAnd the spatial distance and curvature relationship between them is weighted to obtain:

[0083] z new =∑ j w j z j ;

[0084] Among them, z new is the z coordinate (or curvature value) of the point to be interpolated, z j is the z coordinate (or curvature value) of the jth adjacent real sampling point, w j is the weight coefficient assigned to the jth neighboring true sampling point. Weight w j It is a function of distance and curvature, ensuring that the interpolation result can transition smoothly and reflect the local curvature characteristics.

[0085] Finally, the smoothness constraint imposes smoothness constraints on the interpolated surface model to avoid introducing unrealistic oscillations or sharp features during the interpolation process. This is usually achieved by minimizing the second-order derivative of the curvature or an energy function. For example, a regularization term can be introduced into the construction process of the manifold model, which penalizes excessive curvature changes:

[0086]

[0087] Among them, E smooth is the smoothing energy term, Ω represents the integral area of ​​the screen surface, K is the Gaussian curvature, s is the arc length parameter along the surface, represents the second derivative of Gaussian curvature along the arc length parameter, ||·|| 2 Represents the square of the 2-norm. By minimizing the total energy function containing this smoothing term, a surface model that can fit the sampled data while maintaining good smoothness can be obtained.

[0088] Through the collaborative work of the aforementioned devices and algorithms, the surface deformation modeling module dynamically constructs and updates a manifold model that accurately reflects the screen's actual three-dimensional form. This model not only provides precise positional information for every point on the screen but also incorporates important differential geometry properties (such as curvature and normal vectors). This information is crucial for subsequent color difference analysis and compensation, as the curvature of the screen directly affects the viewing angle and color performance of pixels. The model's adaptive update mechanism ensures a balance between computational efficiency and model accuracy under different deformation states.

[0089] A coupled tensor decomposition module organizes the multi-channel narrowband spectral image data into a four-dimensional tensor comprising a spatial resolution dimension, an RGB color channel dimension, and a spectral channel dimension, and performs a Tucker decomposition with covariant derivative constraints based on the geometric constraints of the manifold model;

[0090] In this embodiment, the coupled tensor decomposition module is responsible for deeply fusing the high-dimensional image data collected by the multispectral imaging module with the geometric model constructed by the surface deformation modeling module to extract the intrinsic structure of chromatic aberration related to the physical deformation of the screen.

[0091] The multidimensional data organization unit integrates the collected multi-channel narrow-band spectral image data into a high-order tensor Specifically, the tensor The data structure consists of four dimensions: the first dimension, I1, and the second dimension, I2, represent the spatial resolution of the image (e.g., rows and columns of pixels); the third dimension, I3, represents the RGB color channels (or more generally, color primitives); and the fourth dimension, I4, represents the number of narrowband spectral channels. This organized data structure fully preserves the spatial, color, and spectral correlations of the original information.

[0092] The core task of the geometric constraint loading unit is to convert the differential geometric parameters of the manifold model obtained by the surface deformation modeling module, such as the Riemann metric tensor g uv Or curvature tensor, into an effective constraint on the tensor decomposition process. This constraint is intended to make the factor matrix obtained by decomposition reflect the local geometric characteristics of the screen surface. A specific implementation method is to introduce covariant derivative constraints. On the manifold, covariant derivatives Describes the rate of change of the vector field Y along the vector field X direction, which takes into account the curvature of the manifold itself. Imposing this geometric constraint on the factor matrices in the tensor decomposition, such as the factor matrices related to the spatial dimensions, can make the element changes of these factor matrices consistent with the actual curvature of the screen surface. For example, a regularization term can be constructed that penalizes the gradient norm of the factor matrix on the manifold, that is, minimizes Among them, U is the factor matrix related to the spatial dimension, is a covariant derivative operator defined under the metric given by the manifold model, ||·|| F represents the Frobenius norm.

[0093] The iterative optimization unit uses a constrained optimization algorithm to solve the Tucker decomposition problem with geometric constraints. Tucker decomposition converts the original tensor Approximately represented as a core tensor With a series of factor matrices (For n=1,2,3,4) the product form is:

[0094]

[0095] Among them, × n Represents the product of the tensor along the nth mode, R n is the rank or number of components of the nth dimension.

[0096] The goal is to solve the core tensor and the factor matrix A (n) , so that the weighted sum of data reconstruction error and geometric constraint regularization term is minimized:

[0097]

[0098] in, represents the square of the Frobenius norm, λ is the regularization parameter used to balance the importance of data fitting terms and geometric constraints, S geam is a collection of indices into factor matrices that impose geometric constraints (usually factor matrices corresponding to the spatial dimensions).

[0099] In one possible implementation, an iterative algorithm such as Alternating-Least-Squares (ALS) or Alternating-Direction-Method-of-Multipliers (ADMM) can be used to solve the above optimization problem. In each iteration, one or a group of variables are optimized while the other variables are fixed. For example, in ALS, the core tensor and each factor matrix are updated alternately until convergence. Geometric constraints are implemented by introducing corresponding regularization terms in the subproblem of updating the relevant factor matrices.

[0100] The coupled tensor decomposition module reduces the complexity of the original high-dimensional data and extracts the main variation patterns, ensuring that they are physically and geometrically interpretable and closely related to the actual deformation state of the screen. The decomposed core tensor and factor matrix can then be used by the bio-inspired feature extraction module to generate more discriminative color difference features.

[0101] a bio-inspired feature extraction module that generates a color difference feature map with a dual antagonistic characteristic based on the decomposition result of the four-dimensional tensor, wherein the dual antagonistic characteristic simulates the color response mechanism of retinal ganglion cells;

[0102] In this embodiment, the bioinspired feature extraction module is designed to extract color difference features that are more sensitive to human visual perception from the factor matrix and core tensor output by the coupled tensor decomposition module. This module is derived from the efficient encoding mechanism of color and spatial information in the biological visual system.

[0103] The dual-antagonistic channel construction device simulates the color response characteristics of retinal ganglion cells, specifically the opponent color theory. Specifically, the device uses the factor matrix related to the spectral dimension obtained after tensor decomposition or reconstructed spectral data to generate at least two types of antagonistic color channels. A typical implementation method is to construct a red-green (RG) antagonistic channel and a blue-yellow (BY) antagonistic channel.

[0104] Red-green antagonist channel C RG The long-wavelength spectral response S L (λ) (e.g., corresponding to red perception) and the mid-band spectral response S M (λ) (for example, corresponding to green perception) is differentiated to obtain:

[0105] C RG (x,y)=α1∫S L (λ,x,y)w L (λ)dλ-α2∫S M (λ,x,y)w M (λ)dλ;

[0106] Among them, S L (λ,x,y) and S M (λ, x, y) represent the spectral intensity of long wave and medium wave at the spatial position (x, y), respectively, w L (λ) and w M (λ) is the corresponding spectral weighting function (e.g., simulating the spectral sensitivity curves of L cones and M cones), and α1 and α2 are balancing coefficients.

[0107] Similarly, the blue-yellow antagonist channel C BY The short-wave spectrum response S S The difference operation between the lightness component L(x,y) (λ) (corresponding to blue perception, for example) and the composite lightness component L(x,y) (usually a weighted combination of long-wave and medium-wave responses to simulate yellow perception) is:

[0108] C BY (x,y)=β1∫S S (λ,x,y)w S (λ)dλ-β2L(x,y);

[0109] Among them, S S (λ,x,y) is the shortwave spectrum intensity, w S (λ) is the short-wave spectral weighting function (simulating the spectral sensitivity curve of S cone cells), L(x,y)=γ1∫S L (λ,x,y)w L (λ)dλ+γ2∫S M (λ,x,y)w M (λ)dλ, β1, β2, γ1, and γ2 are the corresponding weighting and balancing coefficients. These antagonistic channels can effectively highlight the color contrast and color difference that the human eye is sensitive to.

[0110] The nonlinear luminance processing device performs a nonlinear transformation on the synthesized luminance component L(x,y) to simulate the adaptive response of the visual system to different luminance levels. One possible implementation method is to use an S-type function, such as the hyperbolic tangent function (tanh) or the sigmoid function, to perform a compression transformation on the luminance signal:

[0111] L NL (x,y)=Atanh(k·L(x,y)+b)+C;

[0112] Among them, L NL (x, y) represents the brightness after nonlinear processing, A controls the output range, k is the nonlinear strength (gain) parameter, b is the bias, and C is the overall offset. The nonlinear strength parameter k can be adaptively adjusted based on local image characteristics (such as local contrast or average brightness), enhancing details in low-contrast areas and avoiding saturation in high-brightness areas. This processing helps subsequent feature extraction remain robust to color variations under different lighting conditions.

[0113] The feature maps generated by the dual-antagonistic channel and the nonlinear luminance processing are further processed by the deformable convolution unit. The core idea is to make the convolution operation adaptable to the local geometric deformation of the screen surface. This unit receives the local curvature parameters (such as the principal curvatures k1, k2 and the principal curvature directions) provided by the surface deformation modeling module.

[0114] The main direction detection device analyzes the curvature tensor at each pixel position to determine the main curvature direction of the local surface. This can be achieved by performing eigendecomposition on the curvature tensor, where the eigenvectors indicate the main curvature directions.

[0115] The dynamic kernel size adjustment device adjusts the effective coverage range or sampling interval of the convolution kernel according to the absolute value of the local curvature. Generally speaking, in areas with large curvature (i.e., where the screen is more sharply curved), the visual correlation between pixels may be weakened due to changes in viewing angle. In this case, a smaller or more directional convolution kernel should be used; conversely, in flat areas with small curvature, a larger convolution kernel can be used to capture a wider range of contextual information. One possible adjustment strategy is to make the size of the convolution kernel proportional to the inverse of the absolute value of the curvature or to satisfy a certain preset nonlinear mapping relationship.

[0116] The directional filtering device adjusts the shape of the convolution kernel or the distribution of sampling points according to the local principal curvature direction determined by the main direction detection device, so that anisotropic filtering is performed along or perpendicular to the main direction. For example, an elliptical or long strip convolution kernel can be used, whose long axis is aligned with or orthogonal to the local principal curvature direction. Deformable convolution is achieved by adding a learnable offset to the sampling grid of the standard convolution. This offset can be generated by guiding the local curvature parameter, so that the convolution kernel can better adapt to the geometric deformation and posture changes of the object.

[0117] Specifically, for a standard convolution kernel operating on the input feature map X at position p0, its output Among them, R defines the sampling points of the receptive field. In deformable convolution, each sampling point p n Each time, an offset Δp is learned n , becomes These offsets Δp n It can be learned from the input features through an additional convolutional layer and can be modulated by the local curvature parameter so that the offset direction and magnitude are related to the local curvature properties of the screen.

[0118] Through the aforementioned dual-antagonistic processing, nonlinear brightness adaptation, and geometrically adaptive deformable convolution operations, the bio-inspired feature extraction module generates feature maps that are more sensitive and robust to screen color variations. These feature maps fully account for the effects of human visual characteristics and screen physical deformation, providing high-quality input for the subsequent motion compensation control module.

[0119] a dynamic compensation control module, which establishes a time-varying partial differential equation including a time derivative of curvature according to the color difference characteristic map, and generates a feedforward compensation signal of the screen pixel voltage based on a model predictive control algorithm;

[0120] In this embodiment, the dynamic compensation control module is designed to generate a pixel-level voltage compensation signal that can actively offset or reduce perceived color difference based on the color difference feature map extracted by the aforementioned module and the screen geometry. A model is established that can predict the dynamic evolution of color difference and optimize control based on this model.

[0121] The time-varying model building device is responsible for building a mathematical model to describe the dynamic behavior of the color difference characteristics C(x,y,t) over time and space. Generally, this model can be expressed as a time-varying partial differential equation (PDE). The construction of this PDE integrates information from multiple sources:

[0122] Features obtained by tensor decomposition (e.g., core tensor or factor matrices related to space and spectrum) can be used as coefficient terms in PDE, reflecting the intrinsic physical mechanism of chromatic aberration or the contribution of different factors.

[0123] Manifold geometry parameters, such as the local curvature K(x,y,t), directly affect the spatial differential operators (such as the Laplacian or gradient operators) in PDEs, because these operators are expressed differently on curved manifolds than on flat spaces.

[0124] Time derivative of curvature Or the screen deformation velocity field v(x, y, t) can be used as the convection term or time-varying coefficient in the PDE to describe the color difference changes caused by the dynamic deformation of the screen.

[0125] A possible PDE model form is as follows:

[0126]

[0127] Among them, C(x,y,t) is the color difference feature map at spatial position (x,y) and time t; Indicates the rate of change of color difference characteristics over time; is a spatial differential operator related to the local curvature K (e.g., the Laplace-Beltrami operator on a manifold), describing the diffusion or smoothing effect of chromatic aberration on the screen surface. Its coefficients may be derived from the results of tensor decomposition; f(v,C) is a convection term related to the deformation velocity field v and the chromatic aberration C, describing the change in chromatic aberration caused by screen motion; S(x,y,t) is a source term that may include chromatic aberration caused by inherent screen characteristics, environmental factors, or other unmodeled factors; U(x,y,t) is the control input, i.e., the effect of the pixel voltage compensation signal to be determined. The goal is to minimize C or approach zero by adjusting U.

[0128] This PDE model is time-varying because its coefficients (such as K, v) and possible boundary conditions will change with time.

[0129] The rolling optimization device uses the Model-Predictive-Control (MPC) algorithm to calculate the optimal compensation control sequence based on the above time-varying PDE model. MPC is an advanced control strategy that performs the following operations in each control cycle:

[0130] State estimation / prediction: Based on the color difference feature map C(x,y,t current ) and the screen geometry state, and use the PDE model to predict a finite time window in the future (called the prediction time domain T p )Evolution of internal chromatic aberration.

[0131] Optimization calculation: In a limited control time domain T c (T c ≤T p ), solve a constrained optimization problem to find an optimal set of control inputs U(x,y,t)(t current ≤t <t current +T c ). The optimization objective is usually to minimize some norm of the color difference in the prediction time domain (e.g., L1 or L2 norm) while penalizing the amplitude or rate of change of the control input to ensure smoothness and feasibility of the control.

[0132]

[0133] Where J(·) is the cost function, for example Q, R, S are the corresponding weight matrices. The optimization problem needs to consider physical constraints, which may include:

[0134] The upper and lower limits of the screen pixel voltage.

[0135] Limitation of the voltage change rate (related to the response characteristics of the drive circuit).

[0136] The upper limit of the compensation rate of change is indirectly determined by the motion parameters of the screen mechanical structure (such as maximum acceleration and maximum deformation rate).

[0137] The allowable voltage gradient amplitude boundary is dynamically adjusted based on the screen aging status.

[0138] Control implementation: The first element U(x,y,t current ) is applied to actual systems.

[0139] Rolling Horizon: In the next control cycle, the above steps are repeated, using new measurement information to update the state and re-perform prediction and optimization. This rolling implementation allows MPC to effectively handle model uncertainty and external disturbances. Because PDE models are typically distributed parameter systems, their discretization can lead to large-scale optimization problems. Therefore, efficient numerical optimization algorithms (such as interior point methods and sequential quadratic programming) and model reduction techniques may be required.

[0140] The signal conversion device is responsible for converting the discrete, digital compensation control sequence U(x,y,t) calculated by the rolling optimization device into an analog voltage gradient signal suitable for the screen drive circuit. This usually includes:

[0141] Digital-to-Analog Converter (DAC): Converts the digital compensation value to an analog voltage level.

[0142] Signal conditioning: This may include amplification, filtering (e.g., low-pass filtering to remove high-frequency noise from the control signal that could damage the screen or produce undesirable visual effects), and impedance matching to ensure that the compensation signal can be effectively injected into the screen’s pixel driver.

[0143] Gradient formation: Ensures that the generated voltage signal has a smooth gradient in space and time to avoid visual abrupt changes or artifacts, and meets the voltage change rate requirements of the driver circuit. Preferably, the device considers the specific characteristics of the screen driver circuit, such as response time, maximum drive voltage, and addressing mode, to ensure that the generated compensation signal can be accurately applied to the corresponding pixels.

[0144] Through the dynamic compensation control mechanism, the pixel voltage can be actively adjusted according to the real-time detected color difference and screen deformation status, thereby minimizing or eliminating the color difference problem caused by various factors such as screen bending, viewing angle changes, and manufacturing tolerances at the perception level.

[0145] A process parameter interface module, which injects the polarizer thickness tolerance and dielectric constant parameters used in the screen manufacturing process into the curved manifold model parameter calculation process in real time, and synchronously adjusts the compensation optimization weight coefficient of the dynamic compensation control module. The optimized weight coefficient is used to control the amplitude and response speed of the feedforward compensation signal;

[0146] In this embodiment, the process parameter interface module acts as a bridge between the physical manufacturing characteristics of the display and the color difference detection and compensation algorithm. Its core function is to utilize the specific process parameter data generated during the display manufacturing process to guide and optimize the establishment of the color difference model and the adjustment of the compensation strategy, thereby achieving more targeted and robust color difference control.

[0147] The process parameter mapping device is responsible for establishing a nonlinear relationship model between manufacturing parameters and the internal adjustment coefficients of the algorithm. The manufacturing parameters of the screen, such as the actual thickness of the polarizer, the dielectric constant distribution of the liquid crystal material, the discreteness of the electrical characteristics of the thin film transistor (TFT), etc., although there are nominal values ​​in the design, there will be a certain tolerance range in actual production. These tiny differences in physical parameters may affect the initial optical properties and subsequent deformation response of the screen. The device establishes these key manufacturing parameters (for example, the thickness of the polarizer d) through pre-experimental calibration or simulation analysis based on physical models. pol , dielectric constant∈ lc ) and the surface manifold model parameters (e.g., initial curvature K0, equivalent value of the elastic modulus of the material) or the model parameters in the dynamic compensation control module (e.g., certain coefficients in the PDE model).

[0148] One possible implementation is to use a look-up table (LUT) or a neural network model f map To represent this mapping:

[0149] P model =f map (M param1 ,M param2 ,…);

[0150] Among them, M parami is the manufacturing parameter of the ith input, P model These are internal parameters or coefficients of the algorithm model that need to be adjusted. For example, a small change in the thickness of the polarizer may slightly alter the initial curvature of the screen or its response to external stress. This information can be used to modify the initial conditions or material property parameters of the manifold model in the surface deformation modeling module.

[0151] The weight dynamic adjustment device dynamically adjusts the weight coefficients in the objective function of the model predictive control (MPC) optimization problem in the dynamic compensation control module based on real-time or batch process parameter input. The objective function of MPC is usually to minimize the weighted sum of color difference and control cost:

[0152] J(C,U)=w C ||C|| 2 +w U ||U|| 2 +w ΔU ||ΔU|| 2 ;

[0153] Among them, ||C|| 2 is the color difference penalty term, ||U|| 2 is the control input (compensation signal amplitude) penalty term, |ΔU|| 2 Is the penalty term for the control input change rate (response speed). Weight coefficient w C ,w U ,w ΔU It determines the trade-off between chromatic aberration elimination, compensation signal amplitude and response speed during the optimization process.

[0154] The process parameter interface module can adjust these weights based on the input manufacturing parameters. For example, if the polarizer uniformity of a batch of screens is poor, it may cause large initial color difference or be sensitive to deformation. In this case, the color difference penalty weight w can be appropriately increased. C , making the compensation algorithm more aggressive in eliminating color differences. If certain process parameters indicate that the screen is more sensitive to rapid voltage changes or prone to visual artifacts, the penalty weight w for the control input change rate can be increased. ΔU, sacrificing a certain response speed in exchange for a smoother compensation effect and visual comfort. This adjustment can be rule-based or based on the relationship between the manufacturing parameters and the optimal weights learned previously.

[0155] The exception handling system monitors input process parameters to ensure they are within a pre-set, reasonable range. If critical manufacturing parameters are detected to have exceeded their normal fluctuation thresholds (for example, polarizer thickness in a certain area far exceeding the upper tolerance limit, or a significant anomaly in the dielectric constant), this could indicate a potentially serious defect in the screen, or that conventional compensation algorithms may struggle to achieve the desired results or even have a negative impact.

[0156] In this case, the exception handling device will trigger the corresponding mechanism: first, it can start the compensation amount limitation mechanism, for example, limiting the amplitude of the compensation signal generated by the dynamic compensation control module to a safer, preset smaller range, or temporarily disabling part of the compensation function to prevent excessive or improper compensation from exacerbating the problem.

[0157] Second, it can generate an abnormal status identification or alarm signal to notify the operator or the upper-level quality control system so that further inspection, diagnosis or special treatment can be performed on that specific screen.

[0158] This mechanism ensures that even in extreme cases where there are large deviations in the manufacturing process, the color difference detection and compensation system can operate in a safe and controllable manner, avoiding potential damage to the screen or worse visual effects.

[0159] By integrating a process parameter interface module, the color difference detection and compensation system of the present invention goes beyond feedback control based solely on real-time optical measurement, incorporating prior knowledge and feedforward adjustments based on manufacturing process information. This enables the system to not only adapt to dynamic changes in screen usage but also provide personalized optimization for inherent manufacturing variations across batches and even individual screens, thereby improving the accuracy, robustness, and applicability of overall color difference compensation.

[0160] See also Figure 2 The present invention also provides a method for detecting color difference of a liquid crystal display screen, the method comprising the following steps:

[0161] S1, synchronously collect multispectral image data through microlens array and narrowband filter set;

[0162] In this step, the system uses a precisely arranged microlens array to perform detailed optical sampling of the screen surface, with each microlens corresponding to a small area on the screen. Working in conjunction with the microlens array is a set of narrowband filters that cover different spectral bands of visible light sequentially or simultaneously, allowing the system to capture image information of the screen in multiple specific narrow spectral channels. The key lies in "synchronous acquisition", which is to ensure that image data of all spectral channels is obtained in a very short time (usually synchronized with the screen refresh cycle or faster) to avoid mismatches in data of different channels due to changes in screen content or dynamic deformation. The collected raw data is a series of two-dimensional images in different narrow spectral bands.

[0163] S2, updating the manifold geometry model based on real-time measurement data from the displacement sensor array;

[0164] In this step, the system monitors the three-dimensional deformation of the screen surface in real time through an array of displacement sensors deployed on the edge or back of the screen. These sensors can be contact or non-contact, and they continuously measure the spatial coordinates of specific points on the screen. Based on these discrete measurement data, the system uses geometric modeling algorithms (for example, based on triangular meshes, spline surfaces or non-parametric methods) to construct and update a mathematical model that can accurately describe the current three-dimensional shape of the screen, namely the "manifold geometry model". This model not only records the macroscopic deformation of the screen, such as bending and twisting, but also calculates important geometric parameters such as the local curvature and normal vector of the surface. The update of the model is dynamic and can reflect any changes in the shape of the screen in real time.

[0165] S3, jointly constructing a multidimensional tensor from multispectral data and manifold parameters and performing constraint decomposition;

[0166] In this step, the multispectral image data collected in S1 is fused with the updated manifold geometric model parameters in S2. First, the multispectral image data is organized into a high-dimensional data structure, namely a "multidimensional tensor". This tensor contains at least spatial dimensions (width and height of the image), color dimensions (such as RGB channels, if applicable), and spectral dimensions (different narrowband channels). Then, "constrained decomposition" is performed on this high-dimensional tensor using the geometric information provided by the manifold model (such as local curvature, surface normals, etc.) as constraints. Tensor decomposition (such as Tucker decomposition or CP decomposition) is a technique that decomposes high-dimensional data into a set of lower-dimensional, more easily interpretable core tensors and factor matrices. By introducing geometric constraints, the decomposed factors can better reflect the changes in optical properties related to the physical deformation of the screen, thereby extracting the intrinsic patterns of chromatic aberration coupled with geometric deformation.

[0167] S4, generating a visual enhancement feature map through a bio-inspired feature extraction module;

[0168] In this step, the system uses the core tensor and factor matrix obtained from the tensor decomposition in S3 as input, and generates a "visual enhancement feature map" through a "biologically inspired feature extraction module" that simulates the information processing mechanism of the biological visual system. This module may include operations that simulate the color opposition mechanism in the retina (such as red-green and blue-yellow antagonistic channels), as well as convolution operations that simulate the responses of neurons in the visual cortex that are sensitive to direction and edges. In particular, these convolution operations may be "deformable", that is, the shape and direction of the convolution kernel will be dynamically adjusted according to the local curvature parameters obtained in S2 to better adapt to the curved shape of the screen surface. The final visual enhancement feature map can more effectively represent the color difference information actually perceived by the human eye.

[0169] S5. Solve the optimization problem with physical constraints in the rolling time domain to generate a compensation signal;

[0170] In this step, the system builds a dynamic model (such as a time-varying partial differential equation whose coefficients may be related to the time derivative of the screen curvature) that can predict future changes in color difference based on the visual enhancement feature map generated by S4. Then, advanced control algorithms such as model predictive control (MPC) are used to solve an optimization problem within a finite time window that rolls forward ("rolling time domain"). The goal of this optimization problem is to find a set of compensation signals that can minimize the predicted color difference over a period of time in the future. Importantly, this optimization process will fully consider various "physical constraints", such as the voltage range that the screen pixels can withstand, the limit on the voltage change rate, and the deformation constraints determined by the mechanical structure of the screen. After solving this constrained optimization problem, a series of optimal compensation control instructions will be obtained.

[0171] S6, converting the compensation signal into a driving voltage and injecting it into the screen control circuit;

[0172] In this step, the system converts the digital compensation signal optimized and generated in S5 (usually a compensation value for each pixel or specific area) into an actual analog drive voltage. This conversion process needs to ensure that the generated voltage signal is compatible with the existing drive circuit of the screen, and that the voltage changes smoothly and stably without introducing new visual interference or damaging the screen. After precise conversion and possible signal conditioning (such as amplification and filtering), these compensation voltage signals are accurately "injected" into the control circuit of the screen, usually superimposed on the original pixel drive signal, thereby adjusting the actual luminous state of each pixel in real time to offset or weaken the detected color difference, ultimately improving the display quality.

[0173] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A system for detecting color difference of a liquid crystal display screen, characterized in that: The system includes the following modules: a multispectral imaging module configured to synchronously collect multi-channel narrowband spectral image data of the screen through a microlens array and a narrowband filter set; A surface deformation modeling module constructs a dynamic surface manifold model based on the spatial distribution of the multi-channel narrow-band spectral image; A coupled tensor decomposition module organizes the multi-channel narrowband spectral image data into a four-dimensional tensor comprising a spatial resolution dimension, an RGB color channel dimension, and a spectral channel dimension, and performs a Tucker decomposition with covariant derivative constraints based on the geometric constraints of the manifold model; a bio-inspired feature extraction module that generates a color difference feature map with a dual antagonistic characteristic based on the decomposition result of the four-dimensional tensor, wherein the dual antagonistic characteristic simulates the color response mechanism of retinal ganglion cells; a dynamic compensation control module, which establishes a time-varying partial differential equation including a time derivative of curvature according to the color difference characteristic map, and generates a feedforward compensation signal of the screen pixel voltage based on a model predictive control algorithm; The process parameter interface module injects the polarizer thickness tolerance and dielectric constant parameters in the screen manufacturing process into the surface manifold model parameter calculation process in real time, and synchronously adjusts the compensation amount optimization weight coefficient of the dynamic compensation control module. The optimized weight coefficient is used to control the amplitude and response speed of the feedforward compensation signal.

2. The system for detecting color difference of a liquid crystal display according to claim 1, wherein: The multispectral imaging module includes: An array of microlenses arranged at a preset density, with lens pitch configured to match the pixel distribution characteristics of the screen; A filter set consisting of multiple narrowband filters, where the wavelength range of each channel continuously covers the visible spectrum and the wavelength ranges of adjacent channels partially overlap; The synchronous trigger device generates synchronous acquisition instructions based on the screen refresh signal, controlling the time synchronization accuracy of image acquisition of each channel to the microsecond level.

3. The system for detecting color difference of a liquid crystal display screen according to claim 1, wherein: The surface deformation modeling module includes: An array of displacement sensors arranged at the edge of the screen is used to obtain real-time three-dimensional coordinate data of the surface; a curvature calculation device for dynamically calculating local curvature parameters based on the spatial positional relationship between adjacent sampling points; The manifold update device adaptively adjusts the model update frequency according to the curvature change rate, and enables the interpolation algorithm to supplement the sampling points in the area of ​​severe deformation.

4. The system for detecting color difference of a liquid crystal display screen according to claim 3, wherein: The interpolation algorithm of the manifold updating device includes: Curvature trend prediction device based on historical deformation data; A spatial interpolation device to add virtual sampling points in areas with severe deformation; The smoothing constraint device limits the variation range of the second-order derivative of the curvature parameter after interpolation.

5. The system for detecting color difference of a liquid crystal display screen according to claim 1, wherein: The coupled tensor decomposition module includes: A multidimensional data organization unit maps the spatial coordinates, color channels, and spectral dimensions of the spectral image into a tensor structure; Geometric constraint loading unit, which converts the differential geometric parameters of the manifold model into regularized constraints for tensor decomposition; The iterative optimization unit uses a constrained optimization algorithm to solve the core tensor and factor matrix that satisfies the joint minimization of geometric features and data reconstruction errors.

6. The system for detecting color difference of a liquid crystal display screen according to claim 1, wherein: The bio-inspired feature extraction module includes: A dual-antagonistic channel construction device generates a difference channel between long-wave and medium-wave spectra and a difference channel between short-wave and synthetic brightness; a nonlinear brightness processing device for applying a hyperbolic function transformation to the synthesized brightness component and adaptively adjusting the nonlinear intensity; The deformable convolution device dynamically adjusts the size and orientation characteristics of the convolution kernel according to the local curvature parameters.

7. The system for detecting color difference of a liquid crystal display screen according to claim 6, wherein: The deformable convolution device comprises: The main direction detection device determines the direction characteristics of the convolution kernel through curvature parameter analysis; Dynamic kernel size adjustment device, which adjusts the convolution kernel coverage in inverse proportion to the absolute value of the local curvature; The directional filtering device performs an anisotropic filtering operation that matches the main direction of curvature.

8. The system for detecting color difference of a liquid crystal display screen according to claim 1, wherein: The dynamic compensation control module includes: A time-varying model construction device uses tensor decomposition characteristics, manifold geometric parameters and their time derivatives as coefficient terms of differential equations; A rolling optimization device solves the optimization problem with physical constraints in the prediction time domain to generate a compensation control sequence; The signal conversion device converts the optimization result into a voltage gradient signal that matches the screen driving circuit.

9. The system for detecting color difference of a liquid crystal display screen according to claim 1, wherein: The process parameter interface module includes: Process parameter mapping device, which establishes a nonlinear relationship model between manufacturing parameters and algorithm adjustment coefficients; A dynamic weight adjustment device adjusts the amplitude weight and response speed weight in the optimization target according to real-time process parameters; The abnormality handling device activates the compensation amount limiting mechanism and generates an abnormal state mark when the parameter exceeds the preset range.

10. A method for detecting color difference of a liquid crystal display screen, applied to the system according to any one of claims 1 to 9, characterized in that: The method comprises the following steps: S1, synchronously collect multispectral image data through microlens array and narrowband filter set; S2, updating the manifold geometry model based on real-time measurement data from the displacement sensor array; S3, jointly constructing a multidimensional tensor from multispectral data and manifold parameters and performing constraint decomposition; S4, generating a visual enhancement feature map through a bio-inspired feature extraction module; S5. Solve the optimization problem with physical constraints in the rolling time domain to generate a compensation signal; S6. Convert the compensation signal into a driving voltage and inject it into the screen control circuit.

Images