Adaptive image correction method for improving small-pitch LED display effect

Abstract

The present application relates to the technical field of image processing and display control, specifically to an adaptive image correction method for improving small-pitch LED display effect, comprising: acquiring a source image video stream, physical degradation prior parameters of a target output space and degradation image observation data, and performing space-time and pixel coordinate alignment preprocessing; generating an ideal reference image matrix based on a preset ideal lossless mapping model; converting the physical degradation prior parameters into an image degradation operator, and applying the image degradation operator to the ideal reference image matrix to generate a theoretical damaged image matrix; extracting an actual residual space and a theoretical residual space, and calculating the morphological similarity of the two in the spatial topology and time sequence dimensions; when the preset condition is met, extracting an inverse image filtering matrix and superimposing the inverse image filtering matrix with the source image video stream, and outputting an enhanced and restored target digital image stream, so as to realize the suppression of small-pitch LED artifacts and the protection of image original details.

Description

Technical Field

[0001] This invention relates to the field of image processing and display control technology, specifically to an adaptive image correction method for improving the effect of small-pitch LED displays. Background Technology

[0002] Existing small-pitch displays are widely used in scenarios such as command and dispatch, security monitoring, and information display. In related technologies, in order to improve the display effect of small-pitch displays, brightness, color, and uniformity can be compensated at the display end based on fixed correction parameters; or, the screen image can be acquired using an image acquisition device, and display anomalies can be processed through image enhancement, noise reduction, or sharpening.

[0003] However, when the screen exhibits dynamic degradation such as pixel crosstalk, thermal attenuation, and timing trailing, and when ambient light, acquisition noise, and original image textures are mixed together, the former method has low adaptability, while the latter method is prone to misjudging image details as display defects. Therefore, the small-pitch display image correction methods in related technologies are difficult to balance degradation artifact suppression and original detail protection. Summary of the Invention

[0004] The purpose of this invention is to provide an adaptive image correction method to improve the performance of small-pitch LED displays, and to solve the following technical problems:

[0005] To avoid blindly smoothing images or mistaking native image details and non-target noise for defects, targeted and precise compensation is only performed on anomalies that conform to the physical degradation mechanism. This effectively suppresses hardware artifacts in small-pitch LEDs and preserves the original image information completely, ensuring stable restoration effects under different operating stages such as long-term screen aging and thermal fluctuations.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] An adaptive image correction method for improving the performance of small-pitch LED displays, based on an image restoration and synthesis analysis architecture, includes the following steps:

[0008] The system acquires the input source image video stream, the physical degradation prior parameters of the target output space, and the degradation image observation data corresponding to the source image video stream acquired by the image acquisition device, and performs spatiotemporal and pixel coordinate alignment preprocessing on the data.

[0009] Based on a pre-defined ideal lossless mapping model, the source image video stream is converted into an ideal reference image matrix; at the same time, the physical degradation prior parameters are converted into image degradation operators, and the image degradation operators are applied to the ideal reference image matrix to generate a theoretically damaged image matrix in a forward evolution.

[0010] The degraded image observation data and the ideal reference image matrix are differencing each other in the digital pixel domain to extract the actual residual space containing actual environmental noise and hardware degradation; the theoretical damaged image matrix and the ideal reference image matrix are differencing each other to extract the theoretical residual space that only represents the degradation features.

[0011] In terms of spatial topology and temporal dimensions, the morphological similarity between the actual residual space and the theoretical residual space is calculated. Based on the morphological similarity, the original high-frequency texture and non-target noise of the image are stripped. The morphological similarity is compared with a preset similarity threshold. If the morphological similarity is greater than or equal to the preset similarity threshold, the inverse image filtering matrix is ​​extracted based on the theoretical residual space. The inverse image filtering matrix is ​​then superimposed with the source image video stream to output the enhanced and restored target digital image stream. If the morphological similarity is less than the preset similarity threshold, the source image video stream is directly output as the target digital image stream.

[0012] Furthermore, the specific process of preprocessing the data for spatiotemporal and pixel coordinate alignment includes:

[0013] Perform frame parsing and digital color channel separation on the source image video stream;

[0014] The physical degradation prior parameters are converted into a parameter table for digital image convolution and matrix operations. The physical degradation prior parameters are extracted from the pixel topological spacing, circuit layout characteristics and thermal distribution characteristics of the target display end.

[0015] By performing sampling synchronization, digital geometric distortion correction, and background ambient light pixel component stripping on the degraded image observation data, an observation image matrix is ​​obtained that is pixel-level aligned with the source image video stream.

[0016] Furthermore, based on a pre-defined ideal lossless mapping model, the specific process of converting the source image video stream into an ideal reference image matrix includes:

[0017] Assuming that the target output space has an absolutely linear pixel response and no optical or electrical crosstalk between pixels, an ideal lossless mapping model is constructed based on this fundamental physical response mechanism.

[0018] The pixel values ​​in the source image video stream are digitally quantized and reconstructed using an ideal lossless mapping model to generate an ideal reference image matrix that represents the absolutely lossless form that the image should have.

[0019] Furthermore, the specific process of transforming the physical degradation prior parameters into image degradation operators and generating the theoretically damaged image matrix includes:

[0020] The pixel topological spacing parameter is transformed into a two-dimensional point spread function operator that simulates photon overflow;

[0021] The thermal distribution characteristic parameters are transformed into a nonlinear pixel grayscale attenuation matrix operator;

[0022] The circuit layout characteristic parameters are converted into a timing inter-frame brightness and darkness trailing filter operator;

[0023] The two-dimensional point spread function operator, the nonlinear pixel grayscale attenuation matrix operator, and the temporal inter-frame brightness and darkness trailing filter operator are sequentially superimposed or convolved and applied to each pixel of the ideal reference image matrix to generate the theoretically damaged image matrix.

[0024] Furthermore, the actual residual space is obtained by subtracting the degraded image observation data from the ideal reference image matrix pixel by pixel according to the corresponding frame timestamp, two-dimensional spatial coordinates and color channels. It includes physical degradation features, ambient light noise features and image acquisition noise features.

[0025] The theoretical residual space is obtained by subtracting the theoretical damaged image matrix from the ideal reference image matrix pixel by pixel according to the corresponding frame timestamp, two-dimensional spatial coordinates and color channels. It only contains the pure digital state degradation feature fingerprint derived from the parameters.

[0026] Furthermore, the process of calculating the morphological similarity between the actual residual space and the theoretical residual space includes:

[0027] Two-dimensional structural features of the actual residual space and theoretical residual space in the current frame, as well as temporal series features between adjacent frames, are extracted respectively.

[0028] Calculate the two-dimensional cross-correlation function value or structural similarity index value between the two features, or calculate the comprehensive similarity parameter which is dynamically fused from the two-dimensional cross-correlation function value and the structural similarity index value according to the preset weight, and use the comprehensive similarity parameter as the morphological similarity; the preset weight is dynamically allocated by the motion intensity coefficient of the current source image video stream.

[0029] Furthermore, the process of stripping away the original high-frequency texture and non-target noise from the image based on morphological similarity, and outputting the enhanced and restored target digital image stream includes:

[0030] When the morphological similarity is greater than or equal to the preset similarity threshold, the observed image variation features are determined to belong to the known physical degradation of the target. Based on the theoretical residual space, the inverse image filtering matrix is ​​extracted to digitally cancel the source image video stream.

[0031] When the morphological similarity is less than a preset threshold, the observed image variation features are determined to be high-frequency detail textures inherent in the image or random interference from external ambient light. The system determines this as a misidentification and refuses to perform digital cancellation operations or output a zero compensation matrix to protect the original image detail data.

[0032] Furthermore, the inverse image filtering matrix contains inverse components corresponding to the image degradation operator. The inverse components include at least: the spatial domain deconvolution integral corresponding to the two-dimensional point spread function operator, the gray magnitude compensation component corresponding to the nonlinear pixel gray-level attenuation matrix operator, and the inter-frame temporal compensation component corresponding to the temporal inter-frame brightness and darkness trailing filter operator.

[0033] The inverse image filtering matrix re-encodes the grayscale and color channel values ​​of the target pixels affected by degradation in the source image video stream by using a pixel-to-pixel mapping correction method.

[0034] Furthermore, the method also includes an adaptive update step for the closed-loop parameters:

[0035] After outputting the enhanced and restored target digital image stream, the system continuously acquires the degraded image observation data for the next time period and extracts the current morphological similarity and the residual error of the inverse image filtering matrix.

[0036] By utilizing residual errors and employing gradient descent or adaptive filtering algorithms, the weight parameters in the two-dimensional point spread function operator, the nonlinear pixel grayscale attenuation matrix operator, and the temporal inter-frame brightness and darkness trailing filter operator are updated in reverse iteration to generate the image degradation operator used in the next correction cycle.

[0037] The beneficial effects of this invention are:

[0038] 1. This invention extracts the actual residual space and the purely degraded theoretical residual space, and calculates the morphological similarity in the spatiotemporal dimension, which can accurately remove the original high-frequency texture and non-target noise of the image; this mechanism overcomes the problem that traditional methods easily mistake the original details of the image for display defects, and achieves a balance between degradation artifact suppression and original image detail preservation;

[0039] 2. This invention transforms physical prior parameters such as pixel topology, thermal distribution, and circuit layout into image degradation operators such as point diffusion, grayscale attenuation, and inter-frame trailing. This method transforms complex hardware mechanisms into computable digital expressions, enabling accurate forward simulation of dynamic hardware defects such as optical crosstalk, thermal drift, and timing trailing, and providing a reliable basis for directional compensation.

[0040] 3. This invention introduces a closed-loop parameter adaptive update mechanism to continuously extract residual errors and perform reverse iterative updates on the weight parameters of the degradation operator. This mechanism completely breaks the limitation of poor adaptability of traditional fixed correction parameters, enabling the system to automatically evolve with screen aging, thermal fluctuations and environmental changes, ensuring a stable restoration effect during long-term operation.

[0041] 4. This invention re-encodes target pixels affected by degradation based on the inverse image filtering matrix; by combining spatial domain deconvolution integral, grayscale amplitude compensation component and inter-frame temporal compensation component, the system replaces the traditional global indiscriminate filtering process and realizes coordinated directional correction of spatial diffusion, brightness loss and cross-frame ghosting.

[0042] 5. Before data differentiation, this invention performs strict spatiotemporal and pixel coordinate alignment preprocessing, including frame parsing, sampling synchronization, distortion correction, and ambient light stripping. This step effectively eliminates interference from shooting timing misalignment, lens perspective distortion, and complex lighting reflections in the duty room, ensuring that all subsequent comparisons and corrections are based on a unified and clean data benchmark. Attached Figure Description

[0043] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0044] Figure 1 This is a flowchart illustrating the adaptive image correction method for improving the effect of small-pitch LED displays provided in an embodiment of this application. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Please see Figure 1 An adaptive image correction method for improving the effect of small-pitch LED displays is proposed. The method is based on an image restoration and synthesis analysis architecture and includes the following steps: acquiring the input source image video stream, the physical degradation prior parameters of the target output space, and the degradation image observation data of the corresponding source image video stream acquired by the image acquisition device, and performing spatiotemporal and pixel coordinate alignment preprocessing on the data.

[0047] Based on a pre-defined ideal lossless mapping model, the source image video stream is converted into an ideal reference image matrix; at the same time, the physical degradation prior parameters are converted into image degradation operators, and the image degradation operators are applied to the ideal reference image matrix to generate a theoretically damaged image matrix in a forward evolution.

[0048] The degraded image observation data and the ideal reference image matrix are differencing each other in the digital pixel domain to extract the actual residual space containing actual environmental noise and hardware degradation; the theoretical damaged image matrix and the ideal reference image matrix are differencing each other to extract the theoretical residual space that only represents the degradation features.

[0049] In terms of spatial topology and temporal dimensions, the morphological similarity between the actual residual space and the theoretical residual space is calculated. Based on the morphological similarity, the original high-frequency texture and non-target noise of the image are stripped. The morphological similarity is compared with a preset similarity threshold. If the morphological similarity is greater than or equal to the preset similarity threshold, the inverse image filtering matrix is ​​extracted based on the theoretical residual space. The inverse image filtering matrix is ​​then superimposed with the source image video stream to output the enhanced and restored target digital image stream. If the morphological similarity is less than the preset similarity threshold, the source image video stream is directly output as the target digital image stream.

[0050] This embodiment provides an adaptive image correction mechanism to improve the effect of small-pitch LED displays. Specifically, the solution is set to be deployed on an ultra-large small-pitch LED splicing screen in a municipal emergency command center. The splicing screen continuously displays drone fire images, geographic information base maps, thermal imaging overlays, and multiple monitoring windows.

[0051] This scenario is characterized by prolonged high-brightness operation, dense image details, and significant changes in ambient light in the duty room. It is prone to the simultaneous occurrence of mixed image texture details and hardware artifacts, making it suitable as a continuous application scenario for this solution.

[0052] Specifically, the system receives three types of input information: one is the source image and video stream to be displayed, which serves as the raw digital content output by the business system; another is the physical degradation prior parameters of the target display end, reflecting the pixel pitch, driving circuit layout, and thermal state of the splicing screen; and the third is the actual display image acquired by an image acquisition device that can be a calibration camera or photoelectric sensor.

[0053] The system first performs alignment preprocessing to bring the digital source, physical prior, and observed image into the same comparison coordinate system. After completing this basic preparation, the system does not directly perform simple filtering on the observed image. Instead, it first constructs an ideal reference image matrix to express the theoretical luminescence result that the current video content should present under the conditions of completely independent pixels, completely linear response, no thermal drift, and no trailing.

[0054] After obtaining the ideal reference image matrix, the system constructs an image degradation operator based on the prior physical parameters of the target display end; this degradation operator is not an abstract noise model, but a digital mapping of the physical defects of small-pitch LEDs;

[0055] For example, too small a pixel pitch can induce light energy diffusion between adjacent pixels, uneven drive arrangement can cause timing trailing in a specific direction, and high temperature areas can lead to a decrease in local luminous efficiency.

[0056] The system transforms these mechanisms into superimposed degradation operators and applies them to an ideal reference image matrix to obtain a theoretically damaged image matrix. The theoretically damaged result formed in this way is essentially a positive simulation result of what kind of traces known hardware defects will leave on the current image.

[0057] The system extracts two error paths respectively; the first path is to subtract the actual acquired degraded image observation data from the ideal reference image matrix to obtain the actual residual space;

[0058] The actual residual includes real optical crosstalk, thermal drift, electrical trailing, as well as ambient lighting reflection, lens noise, and stray light around the seam.

[0059] The second approach is to subtract the theoretical damaged image matrix from the ideal reference image matrix to obtain the theoretical residual space. This theoretical residual retains only the pure degradation feature fingerprint derived from the physical degradation mechanism, without being mixed with on-site environmental factors.

[0060] The difference between the two is that the former includes all the deviations of the actual observations but contains a high level of environmental and random noise from the acquisition, while the latter removes random interference factors and retains only deterministic degradation simulation features.

[0061] The system compares the morphological similarity of the two types of residuals in two dimensions: spatial topology and temporal continuity. The morphology here is not limited to whether the brightness and darkness values ​​of a single pixel are similar, but focuses on structural features that better reflect the hardware mechanism, such as the direction of bright spot diffusion, the distribution position of edge spikes, the direction of trailing continuation, and the rhythm of cross-frame changes.

[0062] When the actual residuals and theoretical residuals are in high agreement on these characteristics, it indicates that the currently observed anomalies are more likely to originate from known hardware degradation rather than the video content itself.

[0063] For example, although the high-frequency boundaries of river textures on the map also form bright-dark transitions, their spatial organization is not the same as the blocky dark spots formed by thermal drift. The system can distinguish between the two through similarity discrimination.

[0064] After determining that the similarity meets the preset conditions, the system generates an inverse image filtering matrix based on the theoretical residual space, and superimposes the inverse image filtering matrix onto the source image video stream, pre-compensates from the source encoding side, and then outputs it to the display end.

[0065] The inverse image filtering matrix can be understood as a preemptive cancellation of expected defects: if a certain type of area will darken after display, the driving value of the corresponding area is pre-increased appropriately on the source image side;

[0066] If a highlight edge tends to trail to the right, the corresponding direction is pre-constrained digitally; this way, the final light field after physical display is closer to the ideal reference image.

[0067] If frame loss, timestamp jump, or significant viewpoint shift is detected in the image acquisition device during the alignment process, the system will pause the compensation generation of the current cycle, maintain only the correction parameters of the previous stable cycle, or return to zero compensation output to avoid treating normal images as defects due to errors in the comparison benchmark.

[0068] For example, when the on-site duty room temporarily turns on the strong supplementary lighting equipment, resulting in a large area of ​​specular reflection in the observed image, the actual residual will deviate from the theoretical residual by more than the set threshold. The system will regard it as non-target noise and will not trigger the cancellation correction.

[0069] During a nighttime consultation at the municipal emergency command center, the main screen displays an infrared image of a fire line in a mountainous area in the center, a visible light image of a drone on the left, and a terrain contour line and wind field layer on the right.

[0070] Due to continuous high brightness operation of the main screen for several hours, slight uneven brightness dark spots began to appear in some hot areas, and the edges of the high-brightness fire line were accompanied by a narrow trail.

[0071] The system receives the original image stream, module thermal distribution information, and the main screen image captured by the front image acquisition device. After alignment, it generates an ideal reference image and then obtains the theoretically damaged image according to the known small-pitch diffusion and thermal attenuation laws.

[0072] If the comparison reveals that the actual residual and the theoretical residual are highly consistent in the direction of fire line diffusion and the distribution pattern of dark spots, then the system outputs the corresponding reverse compensation.

[0073] If a comparison reveals that a certain area exhibits high-frequency flickering but does not conform to the theoretical defect morphology, and this flickering occurs simultaneously in the external glass reflection area, then that area will not be corrected.

[0074] The purpose of this step is to establish a closed-loop relationship between the ideal state, the degraded simulation state, and the field observation state, so that the system does not blindly smooth the image, but only performs targeted compensation for anomalies that conform to the physical degradation mechanism, thereby suppressing small-pitch LED artifacts and protecting the original details of the image.

[0075] In a preferred embodiment of the present invention, the specific process of performing spatiotemporal and pixel coordinate alignment preprocessing on the data includes: performing frame parsing and digital color channel separation on the source image video stream; converting physical degradation prior parameters into a parameter table for digital image convolution and matrix operations, wherein the physical degradation prior parameters are extracted from the pixel topology spacing, circuit layout characteristics and thermal distribution characteristics of the target display end; performing sampling synchronization, digital geometric distortion correction and background ambient light pixel component stripping on the degraded image observation data to obtain an observation image matrix that is pixel-level aligned with the source image video stream.

[0076] This embodiment provides a data alignment preprocessing mechanism; specifically, in the aforementioned emergency command center scenario, if the source image, physical parameters and observation screen are directly used for subsequent comparison, a typical defect will occur: the same anomaly may be caused by inconsistent shooting time, lens perspective distortion or environmental reflection, rather than degradation of the display screen itself.

[0077] Therefore, this embodiment introduces a pre-alignment process of spatiotemporal and pixel coordinates to establish a consistent data foundation for subsequent residual extraction;

[0078] Specifically, the system first performs frame parsing on the source image video stream and then separates it according to the color channels used by the display driver. The reason for this processing is that the red, green, and blue sub-pixels of small-pitch LEDs may not be completely consistent in terms of packaging structure, luminous efficiency, and thermal sensitivity. If the comparison is still based on mixed brightness, the anomaly of one color channel may be masked by other channels.

[0079] For ease of explanation, a simplified example can be used: Suppose that the current frame contains only a 3×3 pixel area, where the source image value of the central pixel is represented by high, medium, and low levels in the red, green, and blue channels, respectively; the system does not directly model the overall brightness, but retains the independent positional relationship of the three channels, so as to identify the case where only the red channel is thermally attenuated in the later stage.

[0080] Meanwhile, the system converts the physical degradation prior parameters into a parameter table in a unified format; the unified parameter table is a multi-dimensional array matrix containing two-dimensional spatial coordinates, pixel pitch scalars, thermal attenuation coefficients, and temporal delay periods.

[0081] This parameter table does not simply record the original hardware number, but transforms engineering attributes such as pixel pitch being lower than the preset standard, common cable impedance of drive columns exceeding the standard, and regional temperature rise exceeding the set threshold into a structured description that can be called by digital simulation.

[0082] For example, for the upper left area of ​​the splicing screen, the parameter table can mark that the coupling between its adjacent pixels is stronger than that in the central area; for the lower edge near the power module, it can mark that its thermal attenuation risk is higher; for certain column drive links, it can mark that its inter-frame trailing is more likely to propagate along the scanning direction; in this way, the subsequent simulation engine can inject the physical phenomena into the image model item by item.

[0083] Furthermore, the system performs sampling synchronization processing on the degraded image observation data; here, synchronization not only ensures that the camera captures the same frame, but also ensures that the acquisition rhythm corresponds as closely as possible to the display refresh cycle, so as to avoid mistaking the bright state residue of the previous frame as a display anomaly of the current frame;

[0084] Perform digital geometric distortion correction to eliminate spatial misalignment of the correction camera caused by installation angle, lens distortion, or the curvature of the splicing screen;

[0085] Afterwards, the system performs background ambient light pixel component stripping to remove interference from the duty room lighting, control panel reflections, and stray light from the glass curtain wall on the observed image;

[0086] To illustrate its logic, a simplified example can be used: if a bright band with a fixed direction appears continuously in the upper right corner of the observed image, and the position of the bright band changes with the indoor light switch but not with the displayed content, the system will identify it as an ambient light component and remove it, instead of sending it to the hardware defect comparison process.

[0087] If the acquisition device experiences occlusion, defocusing, or overexposure during a certain period, resulting in the inability to form a stable pixel mapping after geometric correction, the system will mark the current observation frame as a low-confidence frame and will not participate in the actual residual generation.

[0088] If the thermal sensor loses connection for a short period of time, the system can call up the thermal distribution data of the most recent stable period as a temporary replacement, but at the same time reduce the compensation intensity of that period to prevent overcorrection based on the expired thermal state.

[0089] If the usable effective pixel area after ambient light stripping is lower than the preset ratio, the system will directly revert to the conservative mode that only relies on the source image and historical stable parameters.

[0090] In the daytime scene of the aforementioned emergency command center, the glass curtain wall directly in front of the main screen introduces oblique sunlight reflection, and the correction camera, because it is installed above the control console, forms a slight overhead view.

[0091] The system analyzes the drone video frames according to the refresh rate and separates the three color channels; it converts the module pixel spacing, drive column arrangement and temperature rise monitoring results into a parameter table; and then performs synchronization, perspective correction and background reflection stripping on the camera images.

[0092] After processing, the map boundary, which originally appeared as a trapezoid in the camera image, was restored to a rectangular coordinate consistent with the digital source. The bright bands caused by sunlight reflection were removed, resulting in an observation image matrix that can be directly used in the difference operation.

[0093] The purpose of this step is to ensure that all subsequent difference, similarity comparison and compensation generation are based on the same time reference, the same spatial coordinates and the same color expression, so as to achieve effective identification of real hardware degradation and avoid misjudgment caused by acquisition errors.

[0094] In a preferred embodiment of the present invention, the specific process of converting a source image video stream into an ideal reference image matrix based on a preset ideal lossless mapping model includes: assuming that the target output space has an absolutely linear pixel response and that there is no optical and electrical crosstalk between pixels, an ideal lossless mapping model is constructed based on this basic physical response mechanism.

[0095] The pixel values ​​in the source image video stream are digitally quantized and reconstructed using an ideal lossless mapping model to generate an ideal reference image matrix that represents the absolutely lossless form that the image should have.

[0096] This embodiment provides a mechanism for constructing an ideal reference image matrix. Specifically, after alignment is completed in the previous embodiment, although the source image and the observed image have entered a unified coordinate system, there is still a drawback if the source image itself is used as the comparison benchmark directly: the data format of the source image reflects more the encoding and transmission logic and is not directly equivalent to the light field result that the display panel should form under an absolutely ideal state. Therefore, this embodiment introduces an ideal lossless mapping model to construct a truly ideal reference image matrix that can be used as a benchmark.

[0097] Specifically, the ideal lossless mapping model is based on first principles, assuming that the response of each pixel is independent, linear and repeatable, there is no optical crosstalk and electrical coupling between pixels, the panel temperature is spatially uniform and stable, and the driving timing does not introduce historical residues.

[0098] The model built in this way is not intended to describe the real panel, but to deliberately construct a reference world without any hardware defects; in this world, each color value of the input image corresponds only to the luminous intensity that should be present at its own pixel location, and is not affected by the neighborhood, time or thermal environment.

[0099] During the mapping process, the system performs digital quantization reconstruction on the pixel values ​​in the source image video stream, converting them from the business-side encoded values ​​into the ideal luminous expression on the display side; the focus of this reconstruction is on unifying the display benchmark, rather than complex calculations.

[0100] A simplified example can be used to illustrate this: Suppose there is a 3×3 region in the source image, with the central pixel being a bright red warning and the surrounding pixels being a dark black background; Under an ideal lossless model, the reference result generated by the system should show that the central highlight is limited to the center, and the surrounding black area is unaffected, that is, no reddish or ghosting will be produced around it.

[0101] For example, a 1px white contour line should maintain a sharp edge in an ideal reference image matrix and should not be widened by adjacent bright areas;

[0102] Since the ideal reference image matrix is ​​the common baseline of the two subsequent residual paths, its construction accuracy directly affects the system's ability to distinguish between defects and textures.

[0103] Without this step, the system might mistakenly identify normal brightness compression or color allocation, which is part of the display conversion chain, as hardware degradation. By establishing an ideal reference image matrix separately, the residual information obtained by subsequent differential analysis can truly focus on the parts that deviate from the ideal display state.

[0104] If the source image itself has encoding anomalies, frame drops or insertions, or missing color space identifiers, making it impossible to directly form a stable ideal mapping, the system will prioritize performing source-end format correction; if the correction fails, the current frame will be marked as a low-confidence source frame and will not be used to generate a new compensation template.

[0105] If a certain service source is a low bitrate compressed video, and there are already compressed ringings at the edges that exceed the preset pixel range, the system can still construct an ideal reference image matrix, but will add a mark indicating that the source content has been distorted to this type of source content to avoid mistaking encoding artifacts for hardware defects.

[0106] In the aforementioned emergency command center, the main screen needs to simultaneously display infrared hotspots of the fire, vector contour lines, and unit identification text. The system converts these digital contents into an ideal reference image matrix based on an ideal lossless mapping model: the boundaries of the hotspots should have a continuous and smooth transition, the strokes of white text should remain sharp, the width of contour lines should be stable and consistent, and the black background area of ​​the map should remain pure and free of light leakage.

[0107] The reference result formed in this way does not depend on whether the current screen is hot or not, nor is it affected by the actual driving circuit layout. Instead, it serves as a unified benchmark for all subsequent comparisons.

[0108] The purpose of this step is to establish an ideal display reference that is separate from the actual degradation, thereby fixing the baseline for subsequent error extraction and avoiding confusion between source content encoding features and display hardware defects.

[0109] In a preferred embodiment of the present invention, the specific process of converting physical degradation prior parameters into image degradation operators and generating a theoretically damaged image matrix includes: converting pixel topological spacing parameters into a two-dimensional point spread function operator simulating photon overflow; and converting thermal distribution characteristic parameters into a nonlinear pixel grayscale attenuation matrix operator.

[0110] The circuit layout characteristic parameters are transformed into a time-series inter-frame brightness and darkness trailing filter operator; the two-dimensional point spread function operator, the nonlinear pixel grayscale attenuation matrix operator, and the time-series inter-frame brightness and darkness trailing filter operator are sequentially superimposed or convolved and applied to each pixel of the ideal reference image matrix to generate a theoretically damaged image matrix.

[0111] This embodiment provides a theoretical damaged image matrix generation mechanism; specifically, an ideal reference image matrix alone is still insufficient to support accurate correction, because it can only express the expected state, but cannot answer how the current specific screen will deviate from the expected state under known hardware conditions; especially in the continuous operation scenario of the emergency command center, the degradation performance of the same map is different under the conditions of cold machine in the early morning, high temperature in the afternoon and long-term monitoring at night.

[0112] Therefore, in this embodiment, the physical degradation prior parameters are specifically transformed into image degradation operators that can be applied to the ideal reference image matrix;

[0113] The system transforms the pixel topological spacing parameter into a two-dimensional point diffusion function operator that simulates photon overflow; the corresponding physical mechanism is that the small-pitch encapsulation structure causes the energy of a certain pixel to diffuse towards the direction of the adjacent pixels when the pixel emits high brightness, resulting in phenomena such as blurred edges and widened bright spots.

[0114] The two-dimensional point diffusion function operator does not describe the abstract and vague, but rather the engineering laws of how a single point of light is distributed in its surrounding locations;

[0115] A simplified example can be used to illustrate this: if only the central pixel is a bright spot in the ideal reference image, theoretically, after degradation, the bright spot will no longer be limited to the center, but will have a weaker brightness spillover in the four or eight neighboring areas, forming a diffusion shape that conforms to the screen packaging structure.

[0116] The system transforms the thermal distribution characteristic parameters into a nonlinear pixel grayscale attenuation matrix operator. The corresponding mechanism is that LED luminous efficiency is related to temperature. After the local temperature rises, the actual brightness under the same driving conditions may decrease, and different color channels are affected to different degrees.

[0117] When the infrared fire point and bright alarm bar are displayed on the main screen for a long time, local heat accumulation will cause dark spots or color shifts in some areas; therefore, the attenuation matrix operator does not need to be applied uniformly across the entire screen, but rather it can be differentiated and corrected according to the heat distribution characteristics of different areas.

[0118] Furthermore, the system transforms the circuit layout characteristic parameters into a timing frame inter-frame brightness and darkness trailing filter operator; the corresponding mechanism is that the drive link length, parasitic capacitance, and scanning method may cause the brightness of the previous frame to affect the next frame, especially at the edge of the brightness and darkness transition, which manifests as trailing or afterimage.

[0119] For example, when a flashing alarm box on an emergency map changes from bright to dark, if there is residual historical charge in the driving column, its edge will remain visible for one or more subsequent frames; the operator therefore reflects coupling in the temporal direction rather than spatial blurring in a single frame.

[0120] When generating the theoretically damaged image matrix, the system sequentially superimposes or convolves the above-mentioned degradation operators onto the ideal reference image matrix;

[0121] This does not mean that a unique order must be used, but rather that the screen should be injected in an orderly manner according to the dominant degradation link in the engineering implementation, so as to preserve the composite form of spatial diffusion, thermal decay and temporal tail in the simulation results.

[0122] To facilitate understanding, a simplified example can be used: a single-pixel white line in an ideal reference image first spreads through two-dimensional points and then becomes extended by a set number of pixels. After thermal attenuation, it darkens in high-temperature areas. After time-series trailing, it leaves a backward trail when the display is moved. The final theoretical damage result is closer to the abnormal performance of the actual screen.

[0123] If a certain type of prior parameter is temporarily missing, such as a failure of the thermal distribution sensor, the system can still generate a partially theoretically damaged image matrix based on the pixel topology and circuit layout, but the credibility weight of the thermally related region will be reduced in subsequent similarity comparisons.

[0124] If a certain area experiences abnormal temperature rise for an extended period and exceeds the safe operating boundary, the system can mark that area as a hardware maintenance priority zone and limit the automatic compensation range to avoid masking physical faults that already require repair through excessive digital compensation.

[0125] During the afternoon high-load operation phase of the aforementioned emergency command center, the central part of the main screen continuously displayed a bright fire hotspot, while the lower right corner scrolled with unit dispatch information; the system knew from the module structure that the pixel spacing in the middle was closer, the drive column in the lower right corner was longer, and the temperature rise in the lower area near the power supply compartment was higher.

[0126] Therefore, the system first models optical crosstalk as bright edges expanding outwards to neighboring pixels, then models thermal attenuation as a decrease in brightness response in the lower region, and models driving trail as a narrow, dark, and bright afterimage appearing behind scrolling text.

[0127] The theoretically damaged image matrix obtained after these degradation operators can more realistically reflect the current damage morphology of the screen under this service load.

[0128] The purpose of this step is to transform the dispersed physical priors into computable, combinable, and verifiable degenerate representations, thereby enabling forward simulation of known hardware defects and providing a basis for subsequent defect matching and reverse compensation.

[0129] In a preferred embodiment of the present invention, the actual residual space is obtained by subtracting the degraded image observation data from the ideal reference image matrix pixel by pixel according to the corresponding frame timestamp, two-dimensional spatial coordinates and color channels, and it includes physical degradation features, ambient light noise features and image acquisition noise features.

[0130] The theoretical residual space is obtained by subtracting the theoretical damaged image matrix from the ideal reference image matrix pixel by pixel according to the corresponding frame timestamp, two-dimensional spatial coordinates and color channels. It only contains the pure digital state degradation feature fingerprint derived from the parameters.

[0131] This embodiment provides a dual-track residual space extraction mechanism. Specifically, based on the existing ideal reference image matrix and theoretically damaged image matrix, if the system still directly compares the on-site observation image and the theoretically damaged image, there will be a significant drawback: both contain a large amount of subject information related to the current image content, such as fire line outlines, text borders, road networks, etc. This subject information will obscure the degradation features that truly need attention.

[0132] Therefore, this embodiment uses dual-track difference to separate the actual residual space and the theoretical residual space, retaining only the part of information that deviates from the ideal state;

[0133] Specifically, the actual residual space is obtained by subtracting the degraded image observation data from the ideal reference image matrix pixel by pixel; before performing the pixel-by-pixel subtraction, the system must first perform photometric dimension conversion and normalization alignment.

[0134] Since degraded image observation data is essentially the distribution of light radiation energy collected by photoelectric sensors, it is affected by exposure time and sensor photoelectric response curve, while the ideal reference image matrix is ​​the driving code value in the digital domain, and the two have different dimensions.

[0135] Therefore, the system pre-acquires the photoelectric conversion curve of the image acquisition device at the factory, and obtains the inverse photoelectric response mapping function by calculating the inverse matrix of the photoelectric conversion curve at the factory.

[0136] The system pre-emptively uses the inverse photoelectric response mapping function of the acquisition device to unify the observation data to the linear luminous flux space, and maps the ideal reference image to the same linear luminous flux space according to the nominal gamma curve of the display panel. This ensures that the subsequent pixel-by-pixel subtraction is performed under the same physical amplitude reference, avoiding the generation of pseudo residuals due to the misalignment of the basic level.

[0137] The pixel-by-pixel correspondence here is based on a unified photometric unit, the same timestamp, the same spatial coordinates, and the same color channel;

[0138] It reflects all deviations of the actual display results from the ideal benchmark, and therefore includes both real hardware degradation and on-site factors such as environmental reflections, lens noise, and local exposure fluctuations.

[0139] A simplified example can be used to illustrate this: if only the central pixel of a certain 3×3 region in the ideal reference image is bright, while there is an additional column of weaker bright shadows to the right of the central pixel in the observed image, and a uniformly brightened area is superimposed in the upper left corner, then the former is more likely to correspond to drive trailing or optical crosstalk, while the latter is more likely to correspond to ambient lighting reflection.

[0140] Both will enter the actual residual space, but their spatial forms are different; in contrast, the theoretical residual space is obtained by subtracting the theoretical damaged image matrix from the ideal reference image matrix; this residual space only expresses the traces left by the known degradation mechanism in theory, and therefore does not contain environmental noise and acquisition noise.

[0141] It is equivalent to a defect fingerprint template, which is used to illustrate how the ideal image will be distorted if the screen is only affected by pixel crosstalk, thermal attenuation and timing trailing. Since both paths are based on the same ideal reference image matrix, the subsequent system can compare whether the actual deviation and the theoretical deviation belong to the same type of phenomenon at the same zero point.

[0142] The significance of dual-track differential technology lies in its ability to process image content and deviation content separately. For high-frequency content such as vector maps, thermal imaging false colors, and flashing alarm boxes in this scenario, if residual stripping is not performed first, the system can easily mistake complex edges for panel artifacts.

[0143] After the dual-track difference, what is truly retained are those additional structures that should not appear relative to the ideal display, which is more conducive to subsequent morphological similarity judgment.

[0144] If the alignment error exceeds the acceptable range, the pixel-by-pixel difference will treat normal edges as large residuals. In this case, the system should return to the alignment process to retry instead of continuing to enter the similarity calculation.

[0145] If the observed image has local occlusion, such as a staff member briefly passing in front of the camera and causing a shadow, the occluded area can be marked as an invalid mask area and excluded when extracting the actual residual, to prevent occlusion information that does not belong to the display system from contaminating the comparison results.

[0146] In the aforementioned emergency command center, the system corrected the wildfire hotspots and topographic contour lines displayed on the main screen; a section of white contour line in the ideal reference image matrix should be clear and continuous, but a narrow double image appeared on the right side of the contour line in the observation screen, and the upper right area of ​​the screen was generally bright due to the influence of sunlight outside the window;

[0147] By subtracting from the ideal reference image matrix, the actual residual obtained by the system simultaneously includes the ghosting on the right side of the contour lines and the large area of ​​brightening in the upper right corner.

[0148] On the other hand, after subtracting the theoretically damaged image matrix from the ideal reference image matrix, only the narrow ghost structure that matches the driving arrangement and pixel crosstalk is retained, without including the large-area brightening caused by sunlight; in this way, the difference and overlapping areas of the two types of residuals are clearly separated.

[0149] The purpose of this step is to extract all deviations in the real world and the pure deviations predicted by physical mechanisms through dual-track difference under a unified benchmark, so as to achieve the parallel construction of observational evidence and theoretical fingerprints required for subsequent defect comparison.

[0150] In a preferred embodiment of the present invention, the process of calculating the morphological similarity between the actual residual space and the theoretical residual space includes: extracting the two-dimensional structural features of the actual residual space and the theoretical residual space in the current frame and the time series features between adjacent frames, respectively; calculating the two-dimensional cross-correlation function value or structural similarity index value between the two features, or calculating a comprehensive similarity parameter dynamically fused by the two-dimensional cross-correlation function value and the structural similarity index value according to a preset weight, and using the comprehensive similarity parameter as the morphological similarity; the preset weight is dynamically allocated by the motion intensity coefficient of the current source image video stream.

[0151] This embodiment provides a morphological similarity calculation mechanism; specifically, although most of the main content of the image has been stripped away by dual-track residual extraction alone, there is still a shortcoming: some high-frequency textures will also leave structural traces after difference, such as dense terrain lines, tree canopy textures and infrared noise clusters, which may be similar to slight crosstalk in local morphology.

[0152] Therefore, this embodiment further compares the actual residual space with the theoretical residual space from two dimensions: spatial structure and time series, in order to improve the ability to identify real hardware artifacts.

[0153] Specifically, the system extracts two-dimensional structural features from the current frame and temporal series features from multiple adjacent frames. The two-dimensional structural features focus on the spatial topology of the residual distribution, such as whether the anomaly is stretched along a fixed direction, distributed in blocky clusters, or forms a regular overflow around the highlight boundary.

[0154] Time series features focus on whether the anomaly exhibits a stable and recurring evolution trend over several consecutive frames. For example, does the trail always appear in the frame after the bright-dark transition, or does the hot spot gradually expand under continuous high brightness load?

[0155] In contrast, ambient reflections are often related to changes in the lighting in the duty room and do not necessarily spread along the edges of the displayed content; the textures inherent in the image content often change with the movement of the scene and do not have stable temporal characteristics consistent with the panel structure.

[0156] In specific calculations, the system can use two-dimensional cross-correlation function values ​​to measure the degree of agreement between the two types of residual structures in terms of position and direction, or it can use structural similarity index values ​​to measure whether their local bright and dark organization patterns are similar.

[0157] The two can also be fused into a comprehensive similarity parameter according to preset weights; specifically, the two-dimensional cross-correlation function value is set to... The structural similarity index value is Then the comprehensive similarity parameter The calculation logic is as follows:

[0158]

[0159] in, and To meet The dynamic weights are determined by the following rules: the motion intensity coefficient K is obtained by calculating the sum of the absolute values ​​of the pixel grayscale differences between adjacent frames of the current source image video stream, and a threshold is set. Represents the motion intensity coefficient of the business screen. When the value is less than a preset static threshold, spatial structural characteristics are determined to be the dominant factor, and the following settings are made: It is between 0.7 and 0.9, and is determined by... Sure ;when When the time series characteristics are greater than or equal to a preset static threshold, the time series characteristics are determined to be the dominant factor, and the following settings are made: It is between 0.2 and 0.4, and is determined by... Sure ;

[0160] For ease of explanation, a simplified example can be given: Suppose that the theoretical residuals exhibit a trailing pattern with a bright center and a weaker right side within a 3×3 area, and the actual residuals also show a similar layout in the same area, then their cross-correlation and structural similarity are both high; if the actual residuals become a bright center with a ring-shaped pattern around the perimeter, then although there is also a bright center, it is inconsistent with the direction of the theoretical trailing pattern, and the overall similarity will decrease.

[0161] The reason for adopting dynamic fusion is that different degradation types have different sensitivities to spatial and temporal dimensions; for static, unevenly bright dark spots, spatial structure is more important than temporal continuity; for the problem of trails after bright and dark flickering, temporal characteristics are more critical.

[0162] Therefore, the system can dynamically adjust the weight of each similarity component in the comprehensive judgment based on the current dominant degradation category, the motion intensity of the business screen, or the reliability of the sensor.

[0163] If the number of consecutive frames is insufficient, for example, if a temporary switch of the service source causes the interruption of historical frames, the system can degenerate to make similarity judgments based only on the two-dimensional structure of the current frame and mark the result as moderately reliable.

[0164] If the image content itself changes drastically during a certain period of time, such as rapid switching between multiple screens or frequent pop-up windows, the time series characteristics are easily affected by business actions. In this case, the system can reduce the weight of the time dimension to avoid mistaking the business switching rhythm as a hardware timing abnormality.

[0165] If there is a significant conflict between the cross-correlation and structural similarity results, the system can temporarily suspend triggering compensation and continue observation in the next cycle to reduce the risk of misjudgment.

[0166] In the aforementioned emergency command center, the dispatch list is scrolling on the right side of the main screen. Occasionally, a narrow rightward trail appears during the text scrolling. The system extracts from the theoretical residual that the trail should be distributed along the scanning direction and is most obvious in the frame after the text turns from bright to dark.

[0167] The same residual stripes with the same direction and rhythm were also detected in the actual residuals, so the two showed high consistency in both space and time. On the contrary, for the high-frequency edges formed by the fine texture of the tree canopy in the drone image, there are complex residuals even in a single frame, but their distribution does not stably follow the driving direction, nor does it maintain a uniform trailing rhythm in multiple frames, so the overall similarity is low.

[0168] The purpose of this step is to elevate the similarity assessment from a simple brightness difference to a judgment on whether the structure and evolution patterns are consistent, thereby achieving reliable matching of physically degraded feature fingerprints and reducing misprocessing of original image details.

[0169] In a preferred embodiment of the present invention, the process of stripping the original high-frequency texture and non-target noise of an image based on morphological similarity and outputting an enhanced and restored target digital image stream includes: when the morphological similarity is greater than or equal to a preset similarity threshold, determining that the observed image variation features belong to known target physical degradation, and extracting an inverse image filtering matrix based on the theoretical residual space to digitally cancel the source image video stream;

[0170] When the morphological similarity is less than a preset threshold, the observed image variation features are determined to be high-frequency detail textures inherent in the image or random interference from external ambient light. The system determines this as a misidentification and refuses to perform digital cancellation operations or output a zero compensation matrix to protect the original image detail data.

[0171] This embodiment provides a compensation decision mechanism based on morphological similarity. Specifically, after the similarity calculation is completed, if the system still performs reverse compensation for all detected anomalies, it will expose a boundary defect of the previous step: even if the similarity analysis is relatively strict, there may still be local mismatches on site, such as contour areas with extremely dense textures, bright spot areas of glass reflections, or smoke boundary areas in drone images.

[0172] If such areas are also forcibly compensated, it can easily cause details to become dull, edges to be miscolored, or even local flickering; therefore, this embodiment adds a clear decision branch.

[0173] Specifically, when the morphological similarity is greater than or equal to a preset threshold, the system determines that the observed image variation features belong to the physical degradation of a known target; the threshold here is not an isolated numerical threshold, but is used to distinguish the boundary between a fingerprint that is sufficiently similar to a theoretically defective fingerprint and one that only looks somewhat similar.

[0174] Reaching the threshold means that the on-site anomalies have a strong correspondence with the theoretical residuals in terms of structural orientation, spatial distribution, and temporal duration. Therefore, the system can generate an inverse image filtering matrix based on this and perform digital cancellation on the source image.

[0175] A simplified example can be used to illustrate this: if the overall similarity of a region is determined to be higher than the threshold, then the region will output non-zero compensation; if another adjacent region has high-frequency residuals but insufficient similarity, then the region will output zero compensation.

[0176] Conversely, when the morphological similarity is lower than the preset threshold, the system determines that the observed anomaly is more likely to be a high-frequency detail of the image itself or random interference from external ambient light, and does not perform digital cancellation, or directly outputs a zero compensation matrix.

[0177] This processing logic is not a passive abandonment, but a proactive protection of the original image content, avoiding the removal of real map textures, plume details, road boundaries or fine strokes of text as defects.

[0178] Especially in emergency command scenarios, detailed information often directly affects the duty personnel's judgment on the location of the fire line, road accessibility, and team dispatch. Therefore, protecting details is an engineering necessity. In engineering implementation, the system can also adopt a conservative strategy for intermediate areas with similarity close to the threshold.

[0179] For example, for regions slightly above the threshold but with insufficient confidence, apply low-intensity compensation first and continue to observe the feedback in the next cycle; for regions slightly below the threshold but approaching the threshold for multiple consecutive cycles, do not compensate temporarily but increase the sampling frequency so as to determine whether it will develop into a stable degradation later.

[0180] If the similarity of a large area of ​​the entire screen increases abnormally at the same time within a certain period, the system should further check whether common errors on the acquisition side have occurred, such as camera focal length drift, sudden changes in exposure strategy or sudden changes in ambient lighting, and should not immediately compensate for the entire screen.

[0181] If the similarity is below the threshold but visible artifacts are continuously detected in key areas by the business, the system can trigger a manual review prompt or switch to a cautious correction mode. The threshold strategy will be adjusted after confirmation by the operations and maintenance department.

[0182] If the residuals continue to expand rapidly after the zero-compensation matrix is ​​output, the system can re-estimate the degenerate prior by combining the closed-loop update mechanism, instead of maintaining the state of rejection of compensation for a long time.

[0183] In the aforementioned emergency command center, the visible light image of the drone on the left side of the main screen contains a large amount of tree branch and leaf texture, while the edge of the central fire line shows a rightward trailing due to the high brightness display; after comparison, the system found that the residual of the fire line edge is highly consistent with the theoretical trailing fingerprint, so it generates corresponding reverse compensation to pre-cancel the bright edge.

[0184] Although the wood texture region also has complex high-frequency residuals, their direction and timing do not conform to the theoretical fingerprint, so the system outputs zero compensation, preserving the detailed layers of the tree crown and plume;

[0185] For example, although the bright spot in the upper right corner caused by the reflection of the glass curtain wall is obvious, it does not conform to the known hardware defect form and will not be compensated.

[0186] The purpose of this step is to establish a selective compensation strategy that corrects only when conditions are met and protects when conditions are not met, thereby achieving precise suppression of hardware artifacts and complete preservation of native image information.

[0187] In a preferred embodiment of the present invention, the inverse image filtering matrix includes inverse components corresponding to the image degradation operator. The inverse components include at least: a spatial domain deconvolution integral corresponding to the two-dimensional point spread function operator, a grayscale amplitude compensation component corresponding to the nonlinear pixel grayscale attenuation matrix operator, and an inter-frame temporal compensation component corresponding to the temporal inter-frame brightness and darkness trailing filter operator. The inverse image filtering matrix performs re-encoding operations on the grayscale values ​​and color channel values ​​of the target pixels affected by degradation in the source image video stream by means of pixel-to-pixel mapping correction.

[0188] This embodiment provides a mechanism for generating and applying an inverse image filtering matrix. Specifically, after determining which anomalies are worth compensating in the previous step, if a single form of compensation is still used, such as only brightening the overall image or only sharpening the spatial image, new defects will be encountered: the correction directions corresponding to different degradation sources are not the same. Optical crosstalk needs to deal with spatial diffusion, thermal drift needs to deal with brightness amplitude attenuation, and driving trail needs to deal with cross-frame historical effects.

[0189] A single compensation method cannot cover these differences simultaneously, so this embodiment refines the inverse image filtering matrix into multiple functional components;

[0190] Specifically, the inverse image filtering matrix contains inverse components corresponding to the image degradation operator. The inverse components include at least: the spatial domain deconvolution integral corresponding to the two-dimensional point spread function operator, the gray magnitude compensation component corresponding to the nonlinear pixel gray-level attenuation matrix operator, and the inter-frame temporal compensation component corresponding to the temporal inter-frame brightness and darkness trailing filter operator.

[0191] The spatial domain deconvolution integral is used to deal with edge blurring or bright spot widening caused by light energy diffusion between pixels. Its function is to bring the display result that was widened due to overflow back to the proper boundary.

[0192] The grayscale amplitude compensation component corresponds to the brightness loss caused by local thermal attenuation or efficiency reduction. Its function is to perform differential gain or suppression on the target region on the source image encoding side.

[0193] Inter-frame timing compensation components address the trailing problem caused by scan-driven and parasitic coupling by combining the content of previous and subsequent frames to correct the encoding of the current frame, thereby reducing the persistence of the afterimage.

[0194] In actual application, the system adopts a pixel-to-pixel mapping correction method to re-encode the target pixels affected by degradation. Here, point-to-point does not mean ignoring neighborhood relationships, but rather that each output position has a clear input mapping responsibility and compensation result landing point to avoid global indiscriminate filtering.

[0195] A simplified example of pixel-level mapping operation can be used to illustrate this: Suppose that in a certain 3×3 region, the central pixel should be bright and the right neighbor pixel should be dark, but due to crosstalk and trailing, the right neighbor will be brighter in actual display; then the system configures a spatial suppression component for the central pixel and an inverse constraint component for the right neighbor pixel in the inverse filtering matrix, and adds a timing cancellation term for the right neighbor in the current frame when the central pixel is bright in the previous frame.

[0196] If the area below is also subjected to thermal attenuation, then the grayscale amplitude compensation component will be superimposed on the area; when performing the re-encoding operation, the system applies a spatial domain deconvolution integral to the source image pixels of the target pixel and its associated neighborhood to obtain the deblurred basic encoding value.

[0197] Perform a pixel-by-pixel multiplication operation on the base encoding value and the local gain coefficient determined by the grayscale amplitude compensation component; subtract the ghosting effect extracted by the combination of the previous frame image-driven encoding and the inter-frame temporal compensation component;

[0198] To clearly define the cascade calculation rules, the current frame matrix of the region where the target pixel of the source image is located is set as follows: The spatial domain deconvolution integral is a matrix. The local gain coefficient of the grayscale amplitude compensation component is Inter-frame temporal compensation components are used for image retention extraction weights. The image drive encoding value of the previous frame is Then the corresponding pixel value in the target digital image stream The calculation is expressed as:

[0199]

[0200] in, This represents a convolution operation. After this deconvolution-multiplication-subtraction cascade correction, if the final output correction target value exceeds the driving encoding threshold of the display module's underlying layer, i.e., the effective range of 0-255, the system will limit the output through a hard truncation function with upper and lower limit protection. The hard truncation function with upper and lower limit protection is defined as follows: when the correction target value is less than 0, the output is 0; when the correction target value is greater than 255, the output is 255; when the correction target value is between 0 and 255, the output is the correction target value, thereby eliminating the risk of false colors and flickering caused by dynamic range exceeding the limit.

[0201] Since different color channels may have different attenuation and diffusion characteristics, compensation can be applied separately for each color channel during re-encoding; for example, the red channel has a more significant brightness attenuation at high temperatures, so the grayscale compensation component can be more sensitive on the red channel.

[0202] If the blue channel is less affected by a certain type of encapsulation crosstalk, the spatial domain deconvolution intensity can be lower; through this combination of location-based, time-based, and channel-based compensation, the system can more closely approximate the real degradation mechanism.

[0203] If the compensation required for a target pixel exceeds the range allowed by the display driver, the system should perform a limiting process to prevent new saturation or color shift caused by overcompensation.

[0204] If multiple compensation components conflict after being superimposed in the same region, for example, spatial deconvolution requires reducing a certain pixel while thermal compensation requires increasing that pixel, the system can coordinate them according to the priority of the main degradation cause or safety constraints.

[0205] If a certain compensation component is found to be close to the limit value for a long period of time, the system can output a maintenance alarm simultaneously, indicating that the hardware degradation in that area may be close to an irreversible state.

[0206] In the aforementioned emergency command center, edge overflow appeared around the infrared fire point in the center of the main screen, the brightness of the high-temperature area below decreased by a set ratio, and the scrolling dispatch text on the right was accompanied by a trail with a grayscale deviation lower than the preset value.

[0207] Based on this, the system generates an inverse image filtering matrix: applying a spatial domain deconvolution integral to the fire point edges to make the displayed boundaries more convergent;

[0208] A grayscale amplitude compensation component is applied to the high-temperature area on the lower side to compensate for insufficient light emission caused by heat; an inter-frame timing compensation component is applied to the scrolling text to reduce the influence of the bright state of the previous frame on the next frame; after point-to-point re-encoding, the final output target digital image stream, after being displayed on the physical panel, is closer to the clear alarm screen that the duty personnel expect to see.

[0209] The purpose of this step is to further transform the confirmed degradation into corresponding precise compensation, thereby achieving coordinated correction at the spatial, grayscale, and temporal levels and improving the consistency between the output image and the ideal display state.

[0210] In a preferred embodiment of the present invention, the method further includes a closed-loop parameter adaptive update step: after outputting the enhanced and restored target digital image stream, the system continuously acquires the degraded image observation data of the next time period and extracts the current morphological similarity and the residual error of the inverse image filtering matrix.

[0211] By utilizing residual errors and employing gradient descent or adaptive filtering algorithms, the weight parameters in the two-dimensional point spread function operator, the nonlinear pixel grayscale attenuation matrix operator, and the temporal inter-frame brightness and darkness trailing filter operator are updated in reverse iteration to generate the image degradation operator used in the next correction cycle.

[0212] This embodiment provides a closed-loop parameter adaptive update mechanism. Specifically, the aforementioned implementation methods can generate relatively accurate compensation results within the current cycle. However, if a fixed degradation operator is used for a long time, new shortcomings will be exposed: the degradation of small-pitch LED screens is not static and will continue to evolve with ambient temperature, runtime, content load, module aging, and maintenance and replacement.

[0213] Especially in 24 / 7 scenarios such as emergency command centers, the screen status varies significantly during the day, at night, and when the screen is continuously lit up during emergencies.

[0214] Therefore, this embodiment introduces a closed-loop parameter adaptive update step, enabling the degradation model to be continuously corrected according to the running state; specifically, after outputting the enhanced and restored target digital image stream, the system continues to collect the degradation image observation data of the next time period and extracts the residual error after correction in this period.

[0215] The residual error here reflects the deviation that has not been eliminated after compensation and the mismatch newly introduced after compensation; if a certain type of spatial spillover still retains obvious edge diffusion after compensation, it indicates that the description of the point spread function operator is still insufficient.

[0216] If some hot areas remain dark after compensation, it indicates that the grayscale attenuation model underestimates the impact of the current temperature rise; if the scrolling text trail is reduced but not disappeared, it indicates that the historical weights of the time-series trailing operator still need to be adjusted.

[0217] The system utilizes the trend of morphological similarity changes and residual errors to perform reverse iterative updates on the weight parameters of the two-dimensional point spread function operator, the nonlinear pixel grayscale attenuation matrix operator, and the temporal inter-frame brightness and darkness trailing filter operator.

[0218] Gradient descent or adaptive filtering can be used here, but in engineering terms, the goal is to make the model gradually approach the actual degradation intensity of the screen under the current working conditions.

[0219] In the actual execution of the adaptive update algorithm, the logic is to use the residual error as the loss function and combine the degradation specificity of various physical mechanisms to perform dimensionality reduction gradient update, rather than blindly performing global multi-parameter joint optimization.

[0220] For example, when the residual error manifests as the spillover area of ​​the light source still being larger than the ideal theoretical area after the initial compensation, the system locks the spillover direction with respect to the spatial domain and calculates the directional gradient, thereby increasing the spatial broadening coefficient of the two-dimensional point spread function.

[0221] When the residual error manifests as the target high-temperature area still failing to reach the target reference brightness after gain compensation, the system identifies it as an underestimation of thermal attenuation and specifically increases the attenuation coefficient step size of the nonlinear pixel grayscale attenuation matrix at the corresponding coordinates.

[0222] When the bright image from the previous frame still has a slight residual in the new frame, the cross-frame residual energy is converted according to the preset learning rate ratio and superimposed on the inter-frame trailing filter memory weight of the next cycle; this iterative constraint of specific targeted matching based on error mode effectively ensures the accurate convergence and physical consistency of reverse update.

[0223] For ease of explanation, a simplified example of local region update can be used for deduction: if a region exhibits fuzzy features caused by spatial spillover even after compensation for three consecutive cycles, the system will appropriately enhance the weight of the spatial degradation operator in that region according to the direction of the directional gradient in the next cycle.

[0224] If another region exhibits over-sharpening or even darkening of edges after compensation, the inverse estimation for that region will be weakened accordingly in the next cycle; thus, the direction of the update is determined by the residual error, rather than remaining fixed.

[0225] This closed-loop mechanism also allows the model to adapt to differences in the operating phase; for example, in the first few minutes after a cold start, thermal decay is weak; as continuous high-brightness operation continues, the thermal correlation weight should gradually increase; when ambient light decreases at night and the proportion of camera noise increases, the system can appropriately suppress overlearning of weak residuals.

[0226] Through continuous updates, the degradation operator is no longer a one-time setting, but becomes a set of parameters that evolve in sync with the actual state of the screen.

[0227] If the quality of the data collected in a certain period is not up to standard, such as camera obstruction, sensor offline, or frequent switching of business sources, the system can pause parameter updates and retain only the existing stable model to avoid mislearning the degradation operator into abnormal periods.

[0228] If the similarity decreases continuously after the update, or the residual error shows a diverging trend, the system can execute a rollback mechanism to restore to the most recent set of stable parameters.

[0229] If the rate of change of parameters in a certain area is abnormally large, in addition to limiting the automatic update range, the system can also output an operation and maintenance alarm, indicating that the module may have a hardware failure acceleration phenomenon.

[0230] During the three consecutive days of handling wildfires at the aforementioned emergency command center, the main screen displayed a high-brightness infrared image for an extended period, causing the temperature in the middle and lower parts to gradually rise.

[0231] On the first night, the system's thermal attenuation compensation was able to eliminate most of the dark spots; on the afternoon of the second day, due to continuous high-brightness operation, the lower area still showed slight darkening. After the system extracted the residual error from the newly acquired observation data, it automatically increased the grayscale attenuation weight of the area.

[0232] On the third day, after the right scrolling schedule bar was replaced with a new driver board, the original trailing mode changed. The system then readjusted the timing trailing operator based on the compensated residual trail, so that the compensation in the next cycle would be more in line with the new hardware status.

[0233] The purpose of this step is to enable the degradation model and compensation strategy to have the ability to adapt to changes over time, thereby achieving continuous tracking of long-term aging of the screen, thermal fluctuations and local hardware changes, and ensuring that the target digital image stream can maintain a stable restoration effect at different operating stages.

[0234] The foregoing detailed one embodiment of the present invention, but this is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. An adaptive image correction method for improving the effect of small-pitch LED displays, characterized in that, The method is based on an image restoration and synthesis analysis architecture and includes the following steps: The system acquires the input source image video stream, the physical degradation prior parameters of the target output space, and the degradation image observation data corresponding to the source image video stream acquired by the image acquisition device, and performs spatiotemporal and pixel coordinate alignment preprocessing on the data. Based on a preset ideal lossless mapping model, the source image video stream is converted into an ideal reference image matrix; at the same time, the physical degradation prior parameters are converted into image degradation operators, and the image degradation operators are applied to the ideal reference image matrix to generate a theoretically damaged image matrix in a forward evolution. The degraded image observation data and the ideal reference image matrix are differencing each other in the digital pixel domain to extract the actual residual space containing actual environmental noise and hardware degradation; the theoretically damaged image matrix is ​​differencing each other in the ideal reference image matrix to extract the theoretical residual space that only represents degradation features. In terms of spatial topology and temporal dimensions, the morphological similarity between the actual residual space and the theoretical residual space is calculated. Based on the morphological similarity, the original high-frequency texture and non-target noise of the image are stripped. The morphological similarity is compared with a preset similarity threshold. If the morphological similarity is greater than or equal to the preset similarity threshold, an inverse image filtering matrix is ​​extracted based on the theoretical residual space. The inverse image filtering matrix is ​​then superimposed on the source image video stream to output the enhanced and restored target digital image stream. If the morphological similarity is less than the preset similarity threshold, the source image video stream is directly output as the target digital image stream.

2. The adaptive image correction method for improving the display effect of small-pitch LEDs according to claim 1, characterized in that, The specific process of performing spatiotemporal and pixel coordinate alignment preprocessing on the data includes: Frame parsing and digital color channel separation are performed on the source image video stream; The physical degradation prior parameters are converted into a parameter table for digital image convolution and matrix operations, wherein the physical degradation prior parameters are extracted from the pixel topological spacing, circuit layout characteristics and thermal distribution characteristics of the target display end; The degraded image observation data is sampled and synchronized, digital geometric distortion is corrected, and background ambient light pixel components are stripped to obtain an observation image matrix that is pixel-level aligned with the source image video stream.

3. The adaptive image correction method for improving the display effect of small-pitch LEDs according to claim 1, characterized in that, The specific process of converting the source image video stream into an ideal reference image matrix based on a preset ideal lossless mapping model includes: Assuming that the target output space has an absolutely linear pixel response and no optical or electrical crosstalk between pixels, the ideal lossless mapping model is constructed based on this fundamental physical response mechanism. The pixel values ​​in the source image video stream are digitally quantized and reconstructed using the ideal lossless mapping model to generate the ideal reference image matrix that represents the absolutely lossless form that the image should have.

4. The adaptive image correction method for improving the display effect of small-pitch LEDs according to claim 1, characterized in that, The specific process of converting the physical degradation prior parameters into image degradation operators and generating the theoretically damaged image matrix includes: The pixel topological spacing parameter is transformed into a two-dimensional point spread function operator that simulates photon overflow; The thermal distribution characteristic parameters are transformed into a nonlinear pixel grayscale attenuation matrix operator; The circuit layout characteristic parameters are converted into a timing inter-frame brightness and darkness trailing filter operator; The two-dimensional point spread function operator, the nonlinear pixel grayscale attenuation matrix operator, and the temporal inter-frame brightness and darkness trailing filter operator are sequentially superimposed or convolved and applied to each pixel of the ideal reference image matrix to generate the theoretically damaged image matrix.

5. The adaptive image correction method for improving the display effect of small-pitch LEDs according to claim 1, characterized in that, The actual residual space is obtained by subtracting the degraded image observation data from the ideal reference image matrix pixel by pixel according to the corresponding frame timestamp, two-dimensional spatial coordinates and color channels. It includes physical degradation features, ambient light noise features and image acquisition noise features. The theoretical residual space is obtained by subtracting the theoretical damaged image matrix from the ideal reference image matrix pixel by pixel according to the corresponding frame timestamp, two-dimensional spatial coordinates and color channels. It contains only the pure digital state degradation feature fingerprint derived from the parameters.

6. The adaptive image correction method for improving the display effect of small-pitch LEDs according to claim 1, characterized in that, The process of calculating the morphological similarity between the actual residual space and the theoretical residual space includes: Two-dimensional structural features of the actual residual space and the theoretical residual space in the current frame, as well as time series features between adjacent frames, are extracted respectively. Calculate the two-dimensional cross-correlation function value or structural similarity index value between the two features, or calculate the comprehensive similarity parameter dynamically fused by the two-dimensional cross-correlation function value and the structural similarity index value according to a preset weight, and use the comprehensive similarity parameter as the morphological similarity; the preset weight is dynamically allocated by the motion intensity coefficient of the current source image video stream.

7. The adaptive image correction method for improving the display effect of small-pitch LEDs according to claim 1, characterized in that, The process of stripping the original high-frequency texture and non-target noise from the image based on the morphological similarity, and outputting the enhanced and restored target digital image stream includes: When the morphological similarity is greater than or equal to the preset similarity threshold, it is determined that the observed image variation features belong to the known physical degradation of the target, and the inverse image filtering matrix is ​​extracted based on the theoretical residual space to digitally cancel the source image video stream; When the morphological similarity is less than a preset threshold, the observed image variation features are determined to be high-frequency detail textures inherent in the image or random interference from external ambient light. The system determines this as a misidentification and refuses to perform digital cancellation operations or output a zero compensation matrix to protect the original image detail data.

8. The adaptive image correction method for improving the effect of small-pitch LED displays according to claim 1, characterized in that, The inverse image filtering matrix includes inverse components corresponding to the image degradation operator. The inverse components include at least: the spatial domain deconvolution integral corresponding to the two-dimensional point spread function operator, the grayscale amplitude compensation component corresponding to the nonlinear pixel grayscale attenuation matrix operator, and the inter-frame temporal compensation component corresponding to the temporal inter-frame brightness and darkness trailing filter operator. The inverse image filtering matrix performs re-encoding operations on the grayscale and color channel values ​​of the target pixels affected by degradation in the source image video stream through pixel-to-pixel mapping correction.

9. The adaptive image correction method for improving the display effect of small-pitch LEDs according to claim 1, characterized in that, The method also includes a closed-loop parameter adaptive update step: After outputting the enhanced and restored target digital image stream, the system continuously acquires the degraded image observation data for the next time period and extracts the current morphological similarity and the residual error of the inverse image filtering matrix. Using the residual error, the weight parameters in the two-dimensional point spread function operator, the nonlinear pixel grayscale attenuation matrix operator, and the temporal inter-frame brightness and darkness trailing filter operator are updated in reverse iteration through gradient descent or adaptive filtering algorithms to generate the image degradation operator used in the next correction cycle.

Images