An LED display screen pixel correction method based on image data

Abstract

The present application relates to the technical field of image data processing or generation, and discloses an LED display screen pixel correction method based on image data, comprising: constructing a light energy transfer topology matrix representing signal space coupling characteristics between pixels according to sparse response image data; obtaining original coupled image data in a high gray full-field display mode, and based on the light energy transfer topology matrix and the gray observation value of non-saturated associated pixel points, reconstructing the super-range theoretical brightness value of saturated pixel points by solving a local inverse deduction equation; and performing linear weighted summation operation on the reconstructed data by using an inverse difference operator generated based on the matrix to generate decoupled intrinsic brightness data and a correction coefficient matrix. The present application uses signal space coupling characteristics as an information carrier, and under the constraint of single acquisition and saturated data truncation, realizes the reconstruction and correction of super-range high dynamic range data.

Description

Technical Field

[0001] This invention relates to a pixel correction method for LED displays based on image data, belonging to the general field of image data processing or generation technology. Background Technology

[0002] The current basic principle is to acquire pixel light intensity response data using acquisition equipment, and generate gain or bias compensation coefficients. Mainstream acquisition schemes rely on high-resolution imaging sensors in conjunction with optical lenses to capture full-screen or partitioned illuminated images, analyze pixel-level brightness information, and eliminate optical crosstalk between pixels. Existing technologies generally adopt deconvolution algorithms or iterative restoration strategies based on point spread functions to restore the intrinsic luminous intensity of pixels from images containing coupled noise. This kind of processing logic based on restoration before calculation theoretically improves the spatial resolution of data and provides input basis for the generation of correction coefficients.

[0003] As pixel pitch in displays shrinks and panel sizes increase, mainstream technologies face challenges. The diffraction limit and discrete sampling aliasing effects in optical imaging systems mean that the acquired image is essentially a convolution of the pixel array's light field and the optical system's impulse response, causing irreversible physical energy dispersion between pixels. In high-density arrays, this manifests as neighborhood crosstalk. Traditional deconvolution algorithms require highly complex iterative computations to handle this ill-conditioned inverse problem, are sensitive to the input image's signal-to-noise ratio, and amplify minute observation noise during iteration, leading to ringing effects or local artifacts in the solution, reducing the confidence of correction coefficients. Obtaining accurate point spread function models under the high-speed requirements of industrial production lines requires multiple time-division sparse samplings of the display, increasing time costs and introducing spatial alignment errors between multiple frames. Existing acquisition equipment, facing high dynamic range display modes, is limited by the sensor's full-well capacity, resulting in charge overflow truncation at the central bright pixel, affecting critical high-resolution images. The loss of high-frequency brightness information makes it difficult for existing algorithms to reconstruct truncated data in a single exposure, forcing production lines to adopt time-consuming multi-exposure fusion solutions. To address the physical dispersion of light energy, the industry is attempting to improve the physical packaging structure of devices to suppress crosstalk at the source. For example, the utility model patent with authorization announcement number CN220543943U discloses a core component for anti-light crosstalk LED displays applied to mini displays. The solution proposes to build a light-blocking structure of a metal composite layer or Bragg reflector layer between adjacent LED chips, attempting to cut off the lateral scattering path of light through physical barriers. However, such capital-intensive hardware improvement solutions not only increase the complexity and manufacturing cost of miniaturization processes, but are also difficult to promote on large-scale industrial production lines. Even if the physical structure isolates crosstalk to a certain extent, once the sensor at the acquisition end experiences charge overflow due to high dynamic range display, resulting in data truncation, a simple hardware barrier is powerless to recover the lost high-frequency brightness information.

[0004] Therefore, how to extract pixel intrinsic brightness information from observation data containing optical crosstalk and potential saturation truncation through a computationally efficient and numerically stable data processing mechanism has become the technical problem to be solved by this invention. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A pixel correction method for an LED display screen based on image data, comprising the following steps: Step 101: Acquire sparse response image data when the acquisition device loads sparse sampling signals onto the LED display screen, and calculate the signal energy distribution coefficient of each pixel data point center relative to its neighborhood position based on the sparse response image data, so as to construct a light energy transfer topology matrix characterizing the signal spatial coupling characteristics between pixels. Step 102: Obtain the original coupled image data of the LED display screen in the high grayscale full-field display mode, perform numerical saturation detection on the original coupled image data to locate saturated pixels whose grayscale values ​​have reached the upper limit of data quantization, and identify unsaturated related pixels located within the preset neighborhood of the saturated pixels. Step 103: Extract the specific spillover coefficient representing the signal energy mapped from saturated pixels to unsaturated associated pixels from the light energy transfer topology matrix. With the grayscale observation value of the unsaturated associated pixels as known quantities and the undetermined intrinsic brightness of the saturated pixels as unknown quantities, construct a local inverse deduction equation. Step 104: Solve the local inverse deduction equation to obtain the overrange theoretical brightness value of the saturated pixel in the untrunculated state, and use the overrange theoretical brightness value to replace the observed value of the corresponding saturated pixel in the original coupled image data to generate reconstructed observation data. Step 105: Generate an inverse differential operator based on the light energy transfer topology matrix, and use the inverse differential operator to perform a linear weighted summation operation on the reconstructed observation data to remove the signal superposition component of the neighboring pixels to the target pixel in order to generate decoupled intrinsic brightness data. Step 106: Compare the decoupled intrinsic brightness data with the preset reference brightness target value to generate a correction coefficient matrix for compensating for differences in spatial consistency of image data.

[0006] Preferably, step 101, which involves acquiring sparse response image data and constructing a light energy transfer topology matrix, specifically includes the following steps: Step 201, acquiring frequency modulation image data obtained by a single acquisition of an LED display screen presenting a preset spatial frequency test signal by an acquisition device; Step 202, dividing the frequency modulation image data into multiple local analysis windows, calculating the local modulation transfer rate of pixel grayscale values ​​within each local analysis window, whereby the local modulation transfer rate is used to characterize the signal amplitude attenuation degree of the preset spatial frequency in the imaging data; Step 203, based on a preset frequency domain and spatial domain dual mapping table, directly converting the local modulation transfer rate into the signal energy self-retention coefficient and signal energy distribution coefficient of the corresponding local region to instantiate the light energy transfer topology matrix.

[0007] Preferably, step 105 utilizes the inverse difference operator to perform a linear weighted summation operation on the reconstructed observation data, specifically including the following steps: Step 301, traverse each target pixel in the reconstructed observation data to determine the signal coupling source neighborhood set corresponding to the target pixel in the light energy transfer topology matrix; Step 302, obtain the reconstructed grayscale value of each neighboring pixel in the signal coupling source neighborhood set, and the signal contribution weight of each neighboring pixel to the target pixel as defined in the light energy transfer topology matrix; Step 303, calculate the sum of the products of the reconstructed grayscale value of each neighboring pixel and the corresponding signal contribution weight to obtain the total coupling interference; Step 304, subtract the total coupling interference from the current reconstructed grayscale value of the target pixel, and use the calculation result as the decoupled intrinsic brightness data.

[0008] Preferably, the process of calculating the decoupled intrinsic brightness data in steps 303 to 304 follows the following operational rules: Step 401: Perform the following mathematical operations: ,in, coordinates Decoupled intrinsic brightness data at the location; Reconstruct the observation data at this coordinate; The relative coordinate range of the neighborhood set of the signal coupling source; The relative positions defined in the light energy transfer topology matrix The signal contribution weight of the neighboring pixels to the center pixel.

[0009] Preferably, identifying unsaturated correlated pixels in step 102 specifically includes the following steps: Step 501, establishing a data search window centered on the saturated pixel and scanning all pixels within the data search window; Step 502, filtering out effective pixels whose grayscale values ​​are lower than the data quantization upper limit and higher than the background noise threshold; Step 503, determining whether each effective pixel is located within the effective signal diffusion radius of the saturated pixel based on the light energy transfer topology matrix; Step 504, marking the effective pixels located within the effective signal diffusion radius as unsaturated correlated pixels, and defining the set of these pixels as the observation anchor set of the saturated pixel.

[0010] Preferably, the optical energy transfer topology matrix is ​​a sparse matrix. The construction of the optical energy transfer topology matrix in step 101 further includes the following steps: Step 601, setting a signal energy truncation threshold, which characterizes the negligible boundary of the signal spatial coupling effect; Step 602, when calculating the signal energy distribution coefficient, forcing the coefficients whose calculation results are lower than the signal energy truncation threshold to zero; Step 603, retaining only the coefficients whose values ​​are higher than the signal energy truncation threshold and their position indices, generating an optical energy transfer topology matrix in compressed storage format to limit the computational neighborhood range of subsequent operations.

[0011] Preferably, the method is applied to an LED display screen containing red, green, and blue primary color sub-pixels. The inverse differential operator in step 105 is configured to handle data coupling between color channels, specifically including the following steps: Step 701, constructing independent light energy transfer topology matrices for the red, green, and blue channels respectively; Step 702, when performing linear weighted summation, not only is the spatial signal energy superposition of neighboring pixels of the same color removed, but also, according to the preset inter-channel signal response overlap coefficient, the signal crosstalk component of neighboring pixels of different colors to the target color channel is removed.

[0012] Preferably, the generation process of the correction coefficient matrix also incorporates time-domain evolution characteristics, and the method further includes the following steps: Step 801, obtaining multiple historical correction coefficient matrices generated by the LED display at different historical moments; Step 802, calculating the numerical decay rate index of the corresponding pixel coefficients between the currently generated correction coefficient matrix and the historical correction coefficient matrix; Step 803, when the numerical decay rate index exceeds a preset data anomaly threshold, before generating the final correction coefficient matrix, introducing a damping factor based on the data evolution trend to smooth and correct the current calculation results.

[0013] Preferably, in step 103, the construction of local inverse derivation equations follows the least squares criterion and includes the following steps: Step 901, for each saturated pixel, establish a system of equations containing multiple linear equations, wherein each linear equation expresses that the grayscale observation value of the unsaturated associated pixel is equal to the sum of its own intrinsic brightness and the overflow signal energy from the saturated pixel; Step 902, introduce a regularization term to constrain the solution space of the undetermined intrinsic brightness and prevent the solution from diverging due to observation noise; Step 903, solve the optimal solution of the overdetermined system of equations as the overrange theoretical brightness value of the saturated pixel.

[0014] Preferably, after generating the correction coefficient matrix in step 106, the following steps are also included: Step 1001, converting the correction coefficient matrix into a fixed-point format lookup table suitable for the LED display control system; Step 1002, sending the lookup table to the drive control unit of the LED display screen, so that the drive control unit can call the lookup table in real time to perform reverse pre-distortion processing on the pixel data grayscale value for each frame of image data input during the subsequent display process.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In LED display pixel correction, by constructing an energy transfer response operator and corresponding inverse decoupling matrix that characterize the light energy coupling relationship between display pixels, the traditional image correction process, which relies on iterative approximation or blind deconvolution nonlinear solution, is transformed into a deterministic linear spatial transformation process. Using sparse matrix linear weighted differential logic, the incoherent superposition energy of neighboring pixels is removed from the original observation data. This mathematically eliminates the interference of diffraction diffusion and sampling discreteness of optical imaging system on the intrinsic brightness resolution of pixels. Based on the topological mapping single-time calculation mechanism, the ringing effect and edge artifacts generated when the iterative algorithm processes high-frequency textures are avoided. The complexity of full-field data processing is reduced to the linear level. Real-time high-fidelity solution of pixel-level light intensity truth value of ultra-high density display array is realized using general-purpose computing hardware.

[0016] 2. To address the nonlinear truncation problem of image sensors under high-brightness driving conditions, a mechanism based on the eigenvalue derivation of bypass overflow energy is established. The energy distribution ratio is determined by using the light energy transfer response operator. The overflow energy received by neighboring pixels under optical diffusion is transformed from interference noise into an observation vector representing the intensity of the central saturated pixel. By constructing and solving a set of local linear equations with the neighboring unsaturated observation value as a known quantity and the central saturated eigenvalue as an unknown quantity, the sensor capacity limitation is overcome under the physical constraint of a single exposure. The mathematical reconstruction of the theoretical brightness data far exceeds the dynamic range of the acquisition device, ensuring the linearity of correction and color consistency of the display screen across the entire brightness domain.

[0017] 3. Utilizing the principle of frequency-spatial duality in Fourier optics, a method for constructing an energy transfer model based on spatial frequency attenuation characteristics is established. By analyzing the local modulation transfer rate changes of a specific spatial frequency test pattern during imaging, the frequency domain amplitude attenuation characteristics are directly mapped to the spatial energy diffusion coefficient. This replaces the traditional technique that requires multiple frames of time-division sparse scanning to map the diffusion function of measurement points. Based on frequency domain feature analysis and modeling, the energy transfer topology of the entire field of pixels can be analyzed in a single acquisition, shortening the identification time of the production line calibration system, avoiding time alignment errors introduced by multiple frame acquisitions, and improving the engineering efficiency and data confidence of calibration parameter generation. Attached Figure Description

[0018] Figure 1 This is a flowchart of the correction method for saturated reconstruction and inverse decoupling of the present invention; Figure 2 This is a comparison chart of the full grayscale brightness response characteristics before and after data reconstruction in this invention; Figure 3 This is a schematic diagram of the logical architecture and data interaction of the correction system of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] This invention provides a pixel correction method for LED displays based on image data, comprising four core processing stages: constructing a light energy transfer topology matrix, reconstructing saturated data, performing inverse differential decoupling operations, and generating a correction coefficient matrix. This method is applied to a correction system including an image acquisition device and a data processing unit. In the light energy transfer topology matrix construction stage, sparse response image data is acquired when the acquisition device applies a sparse sampling signal to the LED display. To improve detection efficiency and quantify frequency response characteristics, this sparse response image data is specifically frequency-modulated image data, i.e., data obtained by the acquisition device from a single acquisition of an LED display presenting a preset spatial frequency test signal. The processing unit divides this frequency-modulated image data into multiple local analysis windows and calculates the local modulation transfer rate of the pixel grayscale values ​​within each local analysis window. This local modulation transfer rate is used to characterize the signal amplitude attenuation degree of the preset spatial frequency in the imaging data. A modulation transfer function (MTF) and point spread function (PSF) correlation calibration procedure is executed: setting the sensor integration time. The refresh rate is an integer multiple of the display refresh rate to eliminate time-domain flicker interference; dark current noise is subtracted. To ensure the linear response range, the step loading normalized spatial frequency is used. The test pattern of the sinusoidal grating.

[0021] For each Pixel analysis window, calculate modulation depth ,in, This represents the maximum grayscale value of the pixels within the window. To minimize the value, synchronously activate the isolated pixel at the geometric center of the window and acquire it. The neighborhood light intensity distribution was fitted to a two-dimensional Gaussian surface using the Levenberg-Marquardt algorithm. ,in, Peak amplitude coefficient, The x-axis is the x-axis. The standard deviation along the y-axis is used to extract the feature variance. Establish a nonlinear mapping dataset Using a third-order polynomial Perform least squares regression to determine the coefficient of determination. Based on the fitting function Discretization generation step size is The lookup table (LUT) will display the frequency domain attenuation characteristics. Deterministic mapping to spatial energy distribution parameters The data is stored in the FPGA's on-chip BlockRAM. Based on a pre-defined frequency-domain and spatial domain dual mapping table, the local modulation transfer rate of each local analysis window is converted into the signal energy self-retention coefficient and signal energy distribution coefficient of the corresponding local region. The signal energy self-retention coefficient represents the proportion of pixel emission energy captured by the corresponding sensor pixel, and the signal energy distribution coefficient represents the proportion of energy diffused to the neighborhood. Based on the above coefficients of all pixels, a light energy transfer topology matrix is ​​instantiated. To optimize storage and computation efficiency, this light energy transfer topology matrix is ​​constructed as a sparse matrix. In specific implementation, a signal energy cutoff threshold is set, for example, 1% of the center energy, to characterize the negligible boundary of signal spatial coupling effect. When calculating the signal energy distribution coefficient, coefficients with calculation results below the signal energy cutoff threshold are set to zero, and only coefficients with values ​​above the signal energy cutoff threshold and their position indices are retained, generating a compressed signal energy distribution coefficient. The light energy transfer topology matrix is ​​stored in a compressed format. During the high grayscale data acquisition and reconstruction stage, the original coupled image data of the LED display screen in high grayscale full-field display mode is acquired, and numerical saturation detection is performed on this data. The detection logic scans the entire field of data to locate saturated pixels whose grayscale values ​​reach the upper limit of data quantization. For each saturated pixel, unsaturated associated pixels within its preset neighborhood are identified. The identification process specifically includes: establishing a data search window centered on the saturated pixel and scanning all pixels within the window; selecting effective pixels whose grayscale values ​​are lower than the upper limit of data quantization and higher than the background noise threshold; determining whether each effective pixel is within the effective signal diffusion radius of the saturated pixel based on the light energy transfer topology matrix; marking the effective pixels that meet the above conditions as unsaturated associated pixels, and defining the set of these pixels as the observation anchor set of the saturated pixel.

[0022] The brightness information of saturated pixels is reconstructed using a set of observation anchor points. Specific spillover coefficients representing the energy of the signal mapped from saturated pixels to each unsaturated associated pixel are extracted from the light energy transfer topology matrix. Following the least squares criterion, the grayscale observation values ​​of unsaturated associated pixels are known, while the undetermined intrinsic brightness of saturated pixels is unknown. A local inverse derivation equation is constructed, containing multiple linear equations. Each equation states that the grayscale observation value of an unsaturated associated pixel equals the sum of its own intrinsic brightness and the spillover signal energy from that saturated pixel. During the solution process, a regularization term is introduced to constrain the solution space of the undetermined intrinsic brightness, preventing solution divergence due to observation noise. A Tikhonov regularization objective functional is constructed. ,in For a local ill-conditioned convolution operator submatrix extracted from the global sparse topological matrix, To remove saturation points and obtain the effective observation vector, The regularization operator is dynamically tuned based on the Morozov bias principle. Establish constraints ,in , This represents the number of linear equations contained in the system of equations derived locally inversely. To read out the noise calibration values ​​from the sensor, a preprocessed conjugate gradient (PCG) method is deployed to solve the normal equations. Setting incomplete Cholesky decomposition as a preprocessor To improve the condition number, the iteration termination criterion is set as the relative residual. or maximum number of iterations Regarding the solution results Perform nonnegative constraint projection The system performs bit-depth inverse normalization, outputs high dynamic range reconstructed brightness values, solves the optimal solution of the overdetermined equations, obtains the overrange theoretical brightness values ​​of saturated pixels in the untrunculated state, and uses these overrange theoretical brightness values ​​to replace the observed values ​​of the corresponding saturated pixels in the original coupled image data to generate reconstructed observation data.

[0023] In the decoupling operation stage, an inverse differential operator is generated based on the optical energy transfer topology matrix. This operator is then used to perform a linear weighted summation operation on the reconstructed observation data to remove the signal superposition component of the target pixel from the neighboring pixels. The specific operation process is as follows: traverse each target pixel in the reconstructed observation data to determine the signal coupling source neighborhood set corresponding to the target pixel in the optical energy transfer topology matrix; obtain the reconstructed grayscale value of each neighboring pixel in this set, as well as the signal contribution weight of each neighboring pixel to the target pixel as defined in the optical energy transfer topology matrix; calculate the sum of the products of the reconstructed grayscale value of each neighboring pixel and the corresponding signal contribution weight to obtain the total coupling interference; subtract this total coupling interference from the current reconstructed grayscale value of the target pixel to obtain the decoupled intrinsic brightness data. This decoupling calculation follows the following mathematical expression: ,in, coordinates Decoupled intrinsic brightness data at the location; Reconstruct the observation data at this coordinate; The relative coordinate range of the neighborhood set of the signal coupling source; The relative positions defined in the light energy transfer topology matrix The signal contribution weight of the neighboring pixels to the center pixel.

[0024] For an LED display screen containing red, green, and blue primary color sub-pixels, independent light energy transfer topology matrices are constructed for each of the red, green, and blue color channels. The aforementioned inverse differential operator is configured to handle data coupling between color channels: when performing linear weighted summation, not only is the spatial signal energy superposition of neighboring pixels of the same color removed, but also, according to the preset inter-channel signal response overlap coefficient, the signal crosstalk component of neighboring pixels of different colors to the target color channel is removed. In the correction coefficient generation stage, the decoupled intrinsic brightness data is compared with the preset reference brightness target value to generate an initial correction coefficient matrix. To ensure the temporal stability of the correction parameters, this process is modified in conjunction with temporal evolution characteristics. Multiple historical correction coefficient matrices generated by the LED display screen at different historical moments are obtained, and the numerical decay rate index of the corresponding pixel coefficients between the currently generated correction coefficient matrix and the historical correction coefficient matrix is ​​calculated. When the numerical decay rate index exceeds the preset data anomaly threshold, a damping factor based on the data evolution trend is introduced to smooth and correct the current calculation result before generating the final correction coefficient matrix.

[0025] Example 1: In an application scenario of production line-level pixel brightness consistency correction for high-density Micro-LED displays with a P0.9 pitch, the correction system faces dual physical constraints: limited dynamic range of the imaging sensor and optical crosstalk between pixels. Under these conditions, the acquisition device images a high grayscale full-field display mode to obtain correction data covering the entire brightness range. This causes the high-brightness pixels in the center area of ​​the display to exhibit a saturated state with numerical truncation due to charge overflow in the imaging data. At the same time, the densely arranged pixels cause spatial coupling of signals between adjacent pixels. The system directly calls the light energy transfer topology matrix, which is pre-instantiated based on frequency-modulated image data. This matrix contains the signal energy distribution coefficients that quantize the light energy radiated by each pixel to its spatial neighborhood. The system performs numerical saturation detection on the original coupled image data. The system measures and locates saturated pixels and identifies unsaturated associated pixels within the effective radius of signal diffusion of saturated pixels. Using a specific spillover coefficient extracted from the light energy transfer topology matrix, the system establishes a local inverse deduction equation with the grayscale observation values ​​of unsaturated associated pixels as known constraints and the undetermined intrinsic brightness of saturated pixels as unknowns. This process transforms the neighborhood spillover light energy, originally considered interference noise, into an effective observation vector characterizing the intensity of the central saturated pixel. By solving an overdetermined system of equations with regularization constraints, the system calculates the over-range theoretical brightness value exceeding the sensor cutoff threshold, achieving numerical reconstruction of the truncated data. The system then uses an inverse difference operator generated based on the light energy transfer topology matrix to perform a linear weighted summation operation on the full-field image containing the reconstructed data. This operation is based on a mathematical expression... The total coupling interference generated by the neighborhood set of the signal coupling source is subtracted from the reconstructed gray value of the target pixel.

[0026] Example 2: To verify the practical engineering performance of the LED display pixel correction method based on image data of the present invention, this example constructs a complete experimental platform including a hardware acquisition end and an algorithm processing end. A high-density COB packaged Micro-LED display module with a physical pixel pitch of 0.9mm (P0.9) is used as the test object (DUT). This module has With a resolution and peak brightness of up to 1200 nits, the image acquisition device uses an industrial-grade luminance and colorimeter equipped with a 29-megapixel cooled CCD sensor. It features a dedicated macro lens to ensure one-to-many oversampling of DUT pixels on the imaging surface. The sampling rate is set to the value of each LED pixel on the corresponding sensor. The experiment was conducted in a dark room with constant temperature and humidity and no stray light to eliminate the interference of ambient light on weak dark signals. In order to simulate the signal fluctuations and equipment noise that may exist in real industrial sites, the experiment did not use an ideal pure power supply, but used a standard industrial power grid and introduced random electronic noise with a signal-to-noise ratio of 40dB to test the stability of the algorithm under non-ideal conditions.

[0027] The experiment verified the construction of the optical energy transfer topology matrix. A DUT (Distributed Under Test) was controlled to display a set of sinusoidal fringe test patterns with a preset spatial frequency of 0.5 (brightness changing every two pixel cycles). The acquisition device performed a single-exposure imaging of this pattern to obtain frequency-modulated image data. By analyzing the light penetration from the bright stripes to the adjacent dark stripe regions in this data, the local modulation transfer rate was calculated. The data showed that in the uncorrected original image, the brightness value at the center of the dark stripes was not theoretically zero, but reached 8.5% of the peak value of the bright stripes. This directly quantifies the degree of optical crosstalk between pixels. Based on this modulation transfer rate, the system instantiated the optical energy transfer topology matrix according to a preset frequency-spatial mapping table. To verify the accuracy of the matrix, isolated lit pixels were selected in the experiment, and their brightness in the neighborhood was measured. The actual light intensity distribution generated at the pixel location was compared with the corresponding signal energy distribution coefficient in the matrix. The results showed that the correlation coefficient between the crosstalk distribution predicted by the matrix and the measured value reached 0.98, proving that the mathematical model can accurately characterize the diffusion characteristics of the physical light field. The experiment entered the core stage of high grayscale data reconstruction and decoupling verification. In order to build a convincing control system, three processing schemes were set up: the first group was the original control group, that is, no processing was done on the acquired saturated image data, and conventional correction was performed directly; the second group was the traditional decrosstalk group, which only used the deconvolution algorithm based on the fixed point spread function (PSF) for decoupling, but did not include the saturation reconstruction mechanism; the third group was the experimental group of this invention, which adopted a complete technical solution including saturation reconstruction and inverse differential decoupling.

[0028] The experimental control unit (DUT) was illuminated at 100% high grayscale (grayscale level 255). Due to the high brightness characteristics of Micro-LEDs, the exposure time of the acquisition device was set to be sufficient to capture low grayscale details. As a result, approximately 15% of the bright pixel areas in the full white field image experienced sensor saturation truncation, with their grayscale values ​​clamped at the quantization upper limit, such as 65535. Compared to the original control group, due to the loss of bright information, the corrected display exhibited significant non-linear normalization errors at extremely high brightness, leading to a loss of image details and noticeable differences at the boundaries between bright and dark areas. The halo effect, while somewhat reduced in traditional decrosstalk groups, suffers from ringing artifacts at the edges of saturated regions due to the input saturated data violating the assumptions of the deconvolution algorithm, and fails to recover truncated highlight values. In contrast, the experimental group of this invention performs numerical saturation detection, locates all saturated pixels, and reconstructs the center brightness by solving local inverse deduction equations using the observed values ​​of its neighboring unsaturated pixels. Table 1 shows a comparison of brightness values ​​obtained by different schemes for the same group of typical saturated pixels. This data selects one pixel from the center of the display screen. The highlighted area, with the center point physically overexposed.

[0029] Table 1: Comparison of Saturated Pixel Brightness Reconstruction Effects

[0030] Referring to Table 1, for the center point where saturation occurs, the original observed value is truncated at 65535, failing to reflect the true brightness. The experimental group of this invention, through reverse deduction, resolved an over-range value of 71248, which highly matches the true physical brightness (calibrated by a high-range luminance meter), with a relative error controlled within 1.8%. After reconstruction, the experimental group of this invention used an inverse differential operator to perform decoupling operations. To verify the decoupling effect, the contrast change of a black background with white dots pattern on the display screen before and after processing was measured. In the original image, due to optical crosstalk, the area around the white dots... The black background brightness was improved, but the contrast ratio was only 3000:1. After the decoupling process of this invention, the white point energy was refocused back to the center pixel, and the crosstalk noise of the surrounding background was removed. The measured contrast ratio was improved to 8500:1, and the white point edges were sharp, without the ringing effect common in traditional algorithms. Finally, based on the above processing, correction coefficients were generated and sent to the control system. The corrected display screen was then scanned for linearity across the entire grayscale level. The data showed that, within the grayscale range from 0 to 255, the measured brightness response curve of the experimental group of this invention achieved a good fit with the ideal gamma 2.2 curve. It is superior to the original control group's 0.92 and the traditional de-crosstalk group's 0.95, especially in the high grayscale range of 220-255.

[0031] Example 3: This example combines Figures 1 to 3This document describes a pixel correction method for an LED display screen based on image data, as follows: Figure 1 As shown, the core processing flow begins with a two-way parallel data input mechanism: the left link inputs sparse response image data, constructs a light energy transfer topology matrix in step 101, calculates the signal energy distribution coefficient of each pixel, and generates an inverse differential operator based on this topology matrix for subsequent use; the right link simultaneously inputs the original coupled image data in high grayscale full-field display mode, performs saturation detection and associated point identification in step 102, and locates saturated pixels and unsaturated associated pixels. The two data streams converge in step 103, where specific overflow coefficients are extracted and associated pixel information is provided, including information on unsaturated associated pixels. The grayscale observation value is a known quantity, and the undetermined intrinsic brightness of the saturated pixel is an unknown quantity. A local inverse deduction equation is constructed. In step 104, the equation is solved to obtain the theoretical brightness value of the overrange. This value is then used to replace the original saturated observation value to generate reconstructed observation data. The process proceeds to step 105, where the aforementioned inverse difference operator is used to perform a linear weighted summation operation on the reconstructed observation data to remove the superimposed components of the neighboring signals and generate decoupled intrinsic brightness data. Finally, in step 106, the decoupled intrinsic brightness data is compared with the reference brightness target value to generate the final LED display correction coefficient matrix.

[0032] like Figure 2 As shown, the horizontal axis covers an 8-bit grayscale range from 0 to 240 and above, and the vertical axis represents the physical brightness intensity from 0 to 1,200 nits. The figure clearly depicts three contrasting curves: the ideal gamma 2.2 curve, indicated by a dashed line, represents the theoretical standard display response reference; the original control group's measured curve, indicated by a dotted line, shows that without reconstruction processing, as the grayscale level increases, limited by the dynamic range of the acquisition device, the brightness response exhibits significant nonlinear saturation and numerical truncation at 900 nits, resulting in a flat-top shape; while the experimental group's measured curve, indicated by a solid line, shows that after processing with this method, the display's brightness response curve breaks through the saturation bottleneck of hardware acquisition, restoring a smooth and continuously increasing physical response characteristic across the entire grayscale range and in high-brightness areas; as... Figure 3 As shown, the physical display domain includes a high-density LED display screen with inter-pixel optical energy coupling. The light signals emitted by the screen enter the optical sensing domain through a spatial propagation path. In the optical sensing domain, a high-precision brightness acquisition device is responsible for capturing images and monitoring the numerical saturation truncation phenomenon in real time. The generated sparse sampling signals and image data streams are transmitted to the digital computing and control domain. The digital computing and control domain, as the core processing unit, integrates four serial processing modules, which sequentially execute tasks such as optical energy transfer topology construction, saturation data inverse reconstruction, inverse differential decoupling, and correction coefficient lookup table generation. Finally, the generated correction coefficients are sent back to the drive control unit of the physical display domain, thus forming a closed-loop automated correction control loop.

[0033] Example 4: To further verify the stability of the method of the present invention under extremely low grayscale conditions and to eliminate the potential technical black box regarding the low grayscale nonlinear response compensation mechanism, this example focuses on the display performance of LED displays in the 0 to 5% grayscale range for specific testing and parameter calibration. This range is typically limited by the micro-current driving nonlinearity of Micro-LED devices and the low grayscale quantization accuracy of the driver chip, making it prone to color shift, dead zones, or jump phenomena. This experiment verifies the ability of the present method to achieve smoothing and linear correction in the low grayscale range by combining time-domain evolution characteristics with inverse differential operators. The experiment selected... Using the same hardware platform as in Example 2, namely a Micro-LED display module with a P0.9 pitch, a low grayscale linearity calibration procedure was constructed to address the nonlinear response characteristics under low grayscale conditions. The display screen was controlled to gradually increase from 0% to 5% grayscale in 0.5% grayscale steps, and full-screen brightness data was collected at each step. In the uncalibrated state, the measured data showed that due to the minimum pulse width limitation of the PWM drive, the display screen exhibited a significant brightness dead zone in the grayscale range of 0.5% to 1.5%, meaning that the actual brightness failed to increase with the increase of the grayscale command, resulting in the loss of low grayscale details.

[0034] To address this issue, this invention introduces dynamic compensation logic based on time-domain evolution characteristics. The system retrieves multiple correction coefficient matrices generated during the display's 1000-hour operating history and calculates the numerical decay rate index for the corresponding low grayscale range. Analysis reveals that the nonlinearity of the low grayscale response is positively correlated with the device's aging time. Based on this pattern, when generating the current correction coefficient matrix, the system introduces a dynamic bias term based on operating time, specifically designed to compensate for the drive dead-zone voltage under low grayscale conditions. (in For pixels with a preset lower limit of the linear region (e.g., 2% grayscale), the corrected driving value It is no longer determined solely by linear gain, but follows an adaptive mapping relationship: ,in, The linear gain coefficient, To run for a period of time The dynamically adjusted zero-grayscale bias is used to raise the drive level to the linear turn-on region of the device. To verify the effectiveness of this compensation mechanism, an experiment was conducted to compare the display effects of the no-dynamic-bias group and the dynamic-bias group at 1% grayscale. In the no-dynamic-bias group, due to the lack of compensation for dead-zone voltage, the image showed obvious black patches, and some pixels failed to light up. In the dynamic-bias group, the system automatically injected the calculated zero-grayscale bias based on the current device aging state. The bias amount shifts the operating point of all pixels to the linear region of microcurrent.

[0035] Furthermore, addressing the more sensitive issue of color drift at low grayscale levels, this embodiment also verifies the decoupling capability of the inverse differential operator between multi-color channels. At low grayscale, the difference in the on-state voltage of red, green, and blue sub-pixels leads to white balance deviation. Tests measured color temperature changes within a 1% to 5% grayscale range. In the control group without inter-channel decoupling, as grayscale decreased, the color temperature drastically drifted from the standard 6500K to above 8500K, resulting in a noticeable bluish tint. However, after enabling the inverse differential operator based on the inter-channel signal response overlap coefficient, the system not only eliminated spatial crosstalk in the same-color neighborhood but also compensated for electrical crosstalk between different-color channels in real time according to a preset coupling coefficient. The results show that the color temperature deviation of the display screen was controlled within the entire low grayscale test range. Within this range, ensure accurate color reproduction in low-light environments.

[0036] Example 5: To ensure that the pixel correction method can stably perform the expected decoupling and reconstruction performance on Micro-LED displays with different production batches, different driver chip models, and different physical aging stages, this example formulates and discloses a standardized offline calibration and data filling procedure. Through a set of controlled engineering experiments, the key signal energy distribution coefficients in the light energy transfer topology matrix are obtained and filled, and the reference values ​​of specific overflow coefficients used for saturation reconstruction are determined. The procedure requires that in a dark room environment, a luminance and colorimeter controlled by a displacement stage be used to perform point-by-point scanning or sparse sampling scanning on the target display. For the selected sampling pixel, it is controlled to be lit at maximum brightness, while keeping all pixels in its neighborhood in an off state. The acquisition device records the light intensity distribution data in this state, and the image processing algorithm is used to extract the energy response peak of the center pixel on the sensor target surface.

[0037] The energy attenuation curve of the peak spreading to the surrounding neighborhood is calculated, the signal crosstalk ratio of the central pixel to the relatively positioned pixels in the neighborhood is calculated, and these ratios are used as initial signal energy distribution coefficients to fill the corresponding positions in the light energy transfer topology matrix. For sub-pixels of different colors, the above scanning and calculation process needs to be performed separately to construct independent red, green, and blue three-channel topology matrices. In addition, the procedure also includes the calibration of a specific overflow coefficient: in the above single-point lighting test, when the central pixel is driven to saturation brightness, the grayscale response values ​​of the unsaturated pixels in the neighborhood are recorded simultaneously. The mapping relationship between the central saturation brightness and the overflow grayscale in the neighborhood is established through linear regression analysis, thereby determining the overflow coefficient used for subsequent inverse derivation equations; and considering the equipment differences and environmental fluctuations that may be encountered when deploying this correction method in actual industrial sites, this... The embodiment further constructs a pre-deployment calibration / debugging procedure. After the system is first started or key hardware is replaced, an automated baseline calibration process is executed. The display screen is controlled to display specific test patterns, such as checkerboard or grayscale gradient maps. The current imaging data is acquired using the acquisition device. The system automatically analyzes specific feature points in the image to calibrate the geometric alignment parameters between the acquisition device and the display screen, ensuring a one-to-one correspondence of pixel coordinates. The system performs dark current noise calibration, that is, acquiring multiple frames of images in a completely black state and taking the average value to quantify and subtract the background noise and thermal noise of the acquisition system, and determine the field benchmark for the background noise threshold. The system fine-tunes the signal energy truncation threshold in the light energy transfer topology matrix according to the actual lighting conditions and acquisition distance on site, so as to achieve the best balance between calculation accuracy and processing speed suitable for the current working conditions.

[0038] Example 6: To ensure the measurement accuracy and stability of the pixel correction system under different operating environments, this example constructs and discloses a standardized field baseline calibration procedure, which is designed as a pre-deployment step when the system is first deployed or when environmental conditions change. The procedure requires on-site calibration of the optical geometric parameters of the image acquisition device, and controls the display screen to display a test image containing an array of specific geometric feature points, such as a checkerboard or an array of equally spaced dots. After the acquisition device acquires the image, the system automatically extracts the pixel coordinates of the feature points and establishes a precise mapping relationship from the physical coordinate system of the display screen to the pixel coordinate system of the sensor by calculating the relative positional deviation between the feature points.

[0039] The procedure involves quantitatively calibrating ambient light interference and system noise floor. While maintaining constant ambient lighting conditions, the display screen is turned off completely. Multiple frames of dark field images are continuously captured by the acquisition device, and their average value is calculated to obtain a background noise distribution map of ambient light and dark current superposition. Based on this distribution map, the system determines the background noise threshold for each pixel location, serving as a benchmark for subsequent valid signal determination. Secondly, considering the brightness drift characteristics of Micro-LED devices with temperature variations, the procedure also includes a calibration step for thermal effect compensation parameters. The display screen is controlled to be fully illuminated in white and continuously operated for a predetermined time, such as 30 minutes. During this period, brightness data is collected at regular intervals, and the brightness decay curve over time is recorded. Based on this curve, the system fits a thermal effect compensation coefficient, which is used to dynamically correct the correction coefficient during long-term operation.

[0040] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A pixel correction method for an LED display screen based on image data, characterized in that, Includes the following steps: Step 101: Acquire sparse response image data when the acquisition device loads sparse sampling signals onto the LED display screen, and calculate the signal energy distribution coefficient of each pixel data point center relative to its neighborhood position based on the sparse response image data, so as to construct a light energy transfer topology matrix characterizing the signal spatial coupling characteristics between pixels. Step 102: Obtain the original coupled image data of the LED display screen in the high grayscale full-field display mode, perform numerical saturation detection on the original coupled image data to locate saturated pixels whose grayscale values ​​have reached the upper limit of data quantization, and identify unsaturated related pixels located within the preset neighborhood of the saturated pixels. Step 103: Extract the specific spillover coefficient representing the signal energy mapped from saturated pixels to unsaturated associated pixels from the light energy transfer topology matrix. With the grayscale observation value of the unsaturated associated pixels as known quantities and the undetermined intrinsic brightness of the saturated pixels as unknown quantities, construct a local inverse deduction equation. Step 104: Solve the local inverse deduction equation to obtain the overrange theoretical brightness value of the saturated pixel in the untrunculated state, and use the overrange theoretical brightness value to replace the observed value of the corresponding saturated pixel in the original coupled image data to generate reconstructed observation data. Step 105: Generate an inverse differential operator based on the light energy transfer topology matrix, and use the inverse differential operator to perform a linear weighted summation operation on the reconstructed observation data to remove the signal superposition component of the neighboring pixels to the target pixel in order to generate decoupled intrinsic brightness data. Step 106: Compare the decoupled intrinsic brightness data with the preset reference brightness target value to generate a correction coefficient matrix for compensating for differences in spatial consistency of image data.

2. The LED display pixel correction method based on image data according to claim 1, characterized in that, Step 101 involves acquiring sparse response image data and constructing a light energy transfer topology matrix, specifically including the following steps: Step 201, acquiring frequency modulation image data obtained by a single acquisition of an LED display screen presenting a preset spatial frequency test signal by an acquisition device; Step 202, dividing the frequency modulation image data into multiple local analysis windows, calculating the local modulation transfer rate of pixel grayscale values ​​within each local analysis window, which is used to characterize the signal amplitude attenuation degree of the preset spatial frequency in the imaging data; Step 203, based on a preset frequency domain and spatial domain dual mapping table, directly converting the local modulation transfer rate into the signal energy self-retention coefficient and signal energy distribution coefficient of the corresponding local region to instantiate the light energy transfer topology matrix.

3. The LED display pixel correction method based on image data according to claim 1, characterized in that, Step 105 uses the inverse difference operator to perform a linear weighted summation operation on the reconstructed observation data, specifically including the following steps: Step 301, traverse each target pixel in the reconstructed observation data and determine the signal coupling source neighborhood set corresponding to the target pixel in the light energy transfer topology matrix; Step 302, obtain the reconstructed grayscale value of each neighboring pixel in the signal coupling source neighborhood set, and the signal contribution weight of each neighboring pixel to the target pixel as defined in the light energy transfer topology matrix; Step 303, calculate the sum of the products of the reconstructed grayscale value of each neighboring pixel and the corresponding signal contribution weight to obtain the total coupling interference; Step 304, subtract the total coupling interference from the current reconstructed grayscale value of the target pixel, and use the calculation result as the decoupled intrinsic brightness data.

4. The LED display pixel correction method based on image data according to claim 3, characterized in that, The process of calculating decoupled intrinsic brightness data in steps 303 to 304 follows the following operational rules: Step 401: Perform the following mathematical operations: ,in, coordinates Decoupled intrinsic brightness data at the location; Reconstruct the observation data at this coordinate; The relative coordinate range of the neighborhood set of the signal coupling source; The relative positions defined in the light energy transfer topology matrix The signal contribution weight of the neighboring pixels to the center pixel.

5. The LED display pixel correction method based on image data according to claim 1, characterized in that, Step 102, identifying unsaturated correlated pixels, specifically includes the following steps: Step 501, establishing a data search window centered on the saturated pixel and scanning all pixels within the data search window; Step 502, filtering out valid pixels whose grayscale values ​​are lower than the data quantization upper limit and higher than the background noise threshold; Step 503, determining whether each valid pixel is located within the effective signal diffusion radius of the saturated pixel based on the light energy transfer topology matrix; Step 504, marking the valid pixels located within the effective signal diffusion radius as unsaturated correlated pixels, and defining the set of these pixels as the observation anchor set of the saturated pixel.

6. The LED display pixel correction method based on image data according to claim 1, characterized in that, The optical energy transfer topology matrix is ​​a sparse matrix. The construction of the optical energy transfer topology matrix in step 101 also includes the following steps: Step 601, setting a signal energy truncation threshold, which represents the negligible boundary of the signal spatial coupling effect; Step 602, when calculating the signal energy distribution coefficient, forcing the coefficients whose calculation results are lower than the signal energy truncation threshold to zero; Step 603, only retaining the coefficients whose values ​​are higher than the signal energy truncation threshold and their position indices to generate an optical energy transfer topology matrix in compressed storage format, so as to limit the computational neighborhood range of subsequent operations.

7. The LED display pixel correction method based on image data according to claim 1, characterized in that, The method is applied to an LED display screen containing red, green, and blue primary color sub-pixels. The inverse differential operator in step 105 is configured to handle data coupling between color channels, specifically including the following steps: Step 701, constructing independent light energy transfer topology matrices for the red, green, and blue channels respectively; Step 702, when performing linear weighted summation, not only is the spatial signal energy superposition of neighboring pixels of the same color removed, but also, according to the preset inter-channel signal response overlap coefficient, the signal crosstalk component of neighboring pixels of different colors to the target color channel is removed.

8. The LED display pixel correction method based on image data according to claim 1, characterized in that, The generation process of the correction coefficient matrix also incorporates time-domain evolution characteristics. The method further includes the following steps: Step 801, obtaining multiple historical correction coefficient matrices generated by the LED display at different historical moments; Step 802, calculating the numerical decay rate index of the corresponding pixel coefficients between the currently generated correction coefficient matrix and the historical correction coefficient matrix; Step 803, when the numerical decay rate index exceeds the preset data anomaly threshold, before generating the final correction coefficient matrix, introducing a damping factor based on the data evolution trend to smooth and correct the current calculation results.

9. The LED display pixel correction method based on image data according to claim 1, characterized in that, Step 103 involves constructing local inverse derivation equations, specifically following the least squares criterion, and includes the following steps: Step 901, for each saturated pixel, establish a system of equations containing multiple linear equations, where each linear equation states that the grayscale observation value of the unsaturated associated pixel is equal to the sum of its own intrinsic brightness and the overflow signal energy from the saturated pixel; Step 902, introduce a regularization term to constrain the solution space of the undetermined intrinsic brightness and prevent solution divergence caused by observation noise; Step 903, solve the optimal solution of the overdetermined system of equations, which serves as the over-range theoretical brightness value of the saturated pixel.

10. The LED display pixel correction method based on image data according to claim 1, characterized in that, After generating the correction coefficient matrix in step 106, the following steps are also included: Step 1001, convert the correction coefficient matrix into a fixed-point format lookup table suitable for the LED display control system; Step 1002, send the lookup table to the drive control unit of the LED display screen, so that the drive control unit can call the lookup table in real time to perform reverse pre-distortion processing on the pixel data grayscale value for each frame of image data input during the subsequent display process.

Images