A digital printing color calibration method and system based on real-time image processing

Abstract

This invention relates to the fields of digital image processing and digital printing control technology, specifically a digital printing color calibration method and system based on real-time image processing. The method comprises: constructing a coordinate set of feature points on the fabric surface using hybrid feature tracking technology; applying an adaptive time-domain prediction algorithm to solve and predict the target deformation field of the fabric at the printing position; and combining this target deformation field with a real-time analyzed fabric texture feature map to dynamically interpolate the original digital image pixel by pixel to generate the final printed image data. By combining highly robust tracking, high-precision adaptive prediction, and high-fidelity image generation technologies, the method significantly improves the geometric accuracy and visual quality of printed products.

Description

Technical Field

[0001] The present invention relates to the technical field of digital image processing and digital printing control, and in particular to a digital printing color calibration method and system based on real-time image processing. Background Art

[0002] When digital image processing is applied to textile printing, high-quality color reproduction relies on accurate pixel-level spatial mapping between the digital image space and the physical fabric surface.

[0003] Current digital image processing techniques typically use online cameras to monitor fabrics and compensate for stretching or skew by calculating an affine transformation of the entire image. This approach only provides linear correction and cannot accurately describe non-rigid deformations, resulting in significant deviations between the generated deformation field and the actual physical state. Furthermore, the system lacks temporal prediction capabilities, and can only analyze and compensate for historical image frames. It cannot predict how the deformation field will evolve within the delay window between image acquisition and printing.

[0004] Therefore, a digital printing color calibration method and system based on real-time image processing is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a digital printing color calibration method and system based on real-time image processing, which realizes high-precision real-time geometric correction under variable working conditions.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A digital printing color calibration method based on real-time image processing, comprising:

[0008] Analyze the real-time acquired image of the fabric to be printed, identify and locate structured texture features in the image using a pattern recognition algorithm, and / or detect and locate preset beacon features using wavelength filtering technology, and construct a two-dimensional coordinate set consisting of texture features and beacon features;

[0009] Comparing the two-dimensional coordinate set with a reference coordinate system, calculating the displacement vector of each feature point, and solving a non-rigid geometric deformation field describing the current state of the fabric; using the non-rigid geometric deformation field as input, applying an adaptive time-domain prediction algorithm, combining the fabric transmission speed and physical delay parameters, to predict the target deformation field when the fabric reaches the printing position;

[0010] An inverse transformation is calculated based on the target deformation field to generate a pixel-level inverse distortion map; local texture analysis is performed on the fabric image to be printed to generate a texture feature map; and the original digital image to be printed is resampled and dynamically pixel interpolated based on the inverse distortion map and the texture feature map to generate final printed image data.

[0011] Preferably, for the texture features, a pattern recognition algorithm is used to identify and extract key feature points that remain stable when the fabric is rotated and scaled; the beacon feature is a fluorescence that can only be displayed under a fixed non-visible light band, and an active light source of the corresponding band is configured for excitation illumination, and a bandpass filter is set on the image acquisition path to detect the center of mass coordinates of the beacon.

[0012] Preferably, the adaptive time domain prediction algorithm is specifically an adaptive Kalman filter algorithm; the adaptive Kalman filter algorithm integrates an online model parameter adaptation method; the online model parameter adaptation method continuously calculates the new information residual between the predicted target deformation field and the non-rigid geometric deformation field actually measured at the next moment; when the physical properties of the fabric change and cause the new information residual to mismatch, the online model parameter adaptation method dynamically adjusts the process noise covariance matrix Q parameters inside the filter to match the new deformation law.

[0013] Preferably, the step of performing dynamic pixel interpolation based on the inverse distortion map and the texture feature map includes:

[0014] The texture feature map calculates a quantized texture descriptor for each pixel area by applying a gray-level co-occurrence matrix and / or a wavelet energy spectrum analysis algorithm to a real-time fabric image; the texture descriptor includes roughness, contrast, and directionality. When performing pixel interpolation, an interpolation strategy mapping function is established. The interpolation strategy mapping function takes the texture descriptor as input and makes a decision based on preset corresponding rules. From a set of interpolation kernels containing different characteristics, an interpolation kernel that matches the texture characteristics of the pixel point currently to be calculated is selected.

[0015] Preferably, the beacon feature forms a pseudo-random dot matrix on the fabric surface, wherein the dot matrix pattern in any local area is unique.

[0016] Preferably, the step of analyzing the statistical characteristics of the innovation residual in the online model parameter adaptive method is specifically:

[0017] The covariance matrix of the innovation residual sequence is calculated within a sliding time window; when the trace of the covariance matrix exceeds a set threshold continuously within a predetermined number of image frames, a dynamic adjustment process is triggered.

[0018] Preferably, the step of selecting an interpolation kernel that matches the texture characteristics of the pixel point for the current pixel point to be calculated is specifically as follows:

[0019] Weighted fusion is performed on n interpolation kernels in a group of the interpolation kernels; the interpolation strategy mapping function outputs a set of normalized weight coefficients according to the texture descriptor, and the weight coefficients are used to linearly combine the calculation results of the n interpolation kernels to generate a final pixel color value.

[0020] A digital printing color calibration system based on real-time image processing, comprising:

[0021] Feature extraction module: Analyzes the real-time image of the fabric to be printed, identifies and locates the structured texture features in the image using a pattern recognition algorithm, and / or detects and locates preset beacon features using wavelength filtering technology, and constructs a two-dimensional coordinate set consisting of texture features and beacon features;

[0022] Deformation prediction module: compares the two-dimensional coordinate set with the reference coordinate system, calculates the displacement vector of each feature point, and solves the non-rigid geometric deformation field that describes the current state of the fabric. Using the non-rigid geometric deformation field as input, an adaptive time-domain prediction algorithm is applied, combining the fabric transmission speed and physical delay parameters, to predict the target deformation field when the fabric reaches the printing position.

[0023] Image generation module: calculates the inverse transformation according to the target deformation field to generate a pixel-level inverse distortion map; performs local texture analysis on the fabric image to be printed to generate a texture feature map; resamples and dynamically interpolates pixels of the original digital image to be printed based on the inverse distortion map and the texture feature map to generate final printed image data.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1. The hybrid strategy of this invention allows the system to directly utilize the natural characteristics of richly textured fabrics, reducing reliance on pre-processing. Furthermore, when dealing with these challenging fabrics, it can detect invisible beacons with high contrast and high signal-to-noise ratio to ensure continuous and accurate tracking. This intelligent switching and fusion capability greatly expands the applicability of this method, enabling its stable application to nearly all types of textile substrates and significantly enhancing the environmental adaptability and operational reliability of the entire geometric calibration system.

[0026] 2. By predicting the fabric's "target deformation field" at the future printing moment, it fundamentally resolves the compensation lag error caused by the physical delay between image acquisition and printing execution. This prediction mechanism eliminates the need for geometric correction as a passive reaction based on historical conditions, but rather as an active prediction of future conditions. This significantly improves printing accuracy, especially under high-speed production conditions. The "adaptive" nature of this prediction algorithm enables it to dynamically adjust internal model parameters through online learning to match the physical properties of different types and batches of fabric. This eliminates the need for tedious offline modeling and manual parameter calibration for each new fabric, significantly shortening production preparation time, reducing material waste caused by trial production, and improving the overall operational efficiency and flexibility of the production line.

[0027] 3. This invention incorporates a dynamic pixel interpolation method based on the fabric's real-time texture when generating the final print image data. By pre-analyzing the fabric's local texture characteristics and dynamically matching the most appropriate interpolation strategy to each image region, this method achieves refined and intelligent control of the final visual effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of a digital printing color calibration method based on real-time image processing proposed in an embodiment of the present invention;

[0029] Figure 2 A flow chart of a parameter adaptation loop of an adaptive Kalman filter algorithm proposed in an embodiment of the present invention;

[0030] Figure 3 This is a system structure diagram of a digital printing color calibration system based on real-time image processing proposed in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0032] Example 1

[0033] See also Figure 1 The present invention provides a digital printing color calibration method based on real-time image processing, and the technical solution is as follows:

[0034] A digital printing color calibration method based on real-time image processing, comprising:

[0035] Analyze the real-time acquired image of the fabric to be printed, identify and locate structured texture features in the image using a pattern recognition algorithm, and / or detect and locate preset beacon features using wavelength filtering technology, and construct a two-dimensional coordinate set consisting of texture features and beacon features;

[0036] Comparing the two-dimensional coordinate set with a reference coordinate system, calculating the displacement vector of each feature point, and solving a non-rigid geometric deformation field describing the current state of the fabric; using the non-rigid geometric deformation field as input, applying an adaptive time-domain prediction algorithm, combining the fabric transmission speed and physical delay parameters, to predict the target deformation field when the fabric reaches the printing position;

[0037] An inverse transformation is calculated based on the target deformation field to generate a pixel-level inverse distortion map; local texture analysis is performed on the fabric image to be printed to generate a texture feature map; and the original digital image to be printed is resampled and dynamically pixel interpolated based on the inverse distortion map and the texture feature map to generate final printed image data.

[0038] Furthermore, for the texture features, a pattern recognition algorithm is used to identify and extract key feature points that remain stable when the fabric is rotated and scaled; the beacon features use fluorescence that can only be displayed under a fixed non-visible light band, and by configuring an active light source of the corresponding band for excitation illumination, a bandpass filter is set on the image acquisition path to detect the center of mass coordinates of the beacon.

[0039] The pattern recognition algorithm is specifically a detection and matching method based on local feature descriptors. The local feature descriptor represents a feature composed of multiple numerical values ​​that can represent the unique identity of the key point and is used for subsequent precise matching. The method first applies convolution to the fabric grayscale image acquired in real time using Gaussian kernels of different scales to construct a Gaussian difference scale space pyramid, and identifies candidate key points that are insensitive to scale and brightness changes by searching for local extreme points in this Gaussian difference scale space pyramid, while eliminating low-contrast and unstable edge response points. The Gaussian difference scale space pyramid simulates the effect of the human eye observing objects at different distances by blurring and subtracting the image to different degrees, so that features can be detected at different scales.

[0040] To ensure rotational invariance of keypoints, the system calculates a gradient direction histogram of pixels in the local neighborhood of each stabilized keypoint and uses the direction of the main peak as the keypoint's principal direction. Within a rotated coordinate system centered on the keypoint and based on its principal direction, the gradient information of its neighborhood is recalculated and divided into multiple subregions to generate a high-dimensional descriptor vector that uniquely represents the keypoint's features.

[0041] To further improve accuracy, the feature point fusion process in this embodiment is confidence-weighted. A confidence score based on clarity or matching score is calculated for each texture feature point extracted by the pattern recognition algorithm. When subsequently solving the non-rigid geometric deformation field, feature point coordinates with higher confidence scores are given greater weight, effectively suppressing the interference of noise points or mismatched points on the overall deformation field solution.

[0042] The feature extraction method disclosed in this invention, by constructing a scale space and generating high-dimensional descriptors, ensures that texture feature points are extremely robust to the inevitable rotation, scaling, and illumination changes that occur during fabric transport. Its innovative hybrid tracking strategy combines this highly robust texture recognition with high signal-to-noise ratio beacon detection, overcoming the technical bottleneck of single methods for extreme smooth, solid-color fabrics and achieving stable tracking of all fabric types. This dual guarantee mechanism greatly expands the scope of application of the technology and, by generating a dense and precise set of feature points, lays a solid and reliable data foundation for subsequent high-precision deformation field solution and prediction.

[0043] Furthermore, the adaptive time domain prediction algorithm is specifically an adaptive Kalman filter algorithm, and its internal parameter adaptive loop process can be found in Figure 2 ; The adaptive Kalman filter algorithm integrates an online model parameter adaptation method; the online model parameter adaptation method continuously calculates the innovation residual between the predicted target deformation field and the non-rigid geometric deformation field actually measured at the next moment; when the physical properties of the fabric change, resulting in a mismatch in the innovation residual, the online model parameter adaptation method dynamically adjusts the process noise covariance matrix Q parameters inside the filter to match the new deformation law. The innovation residual is the error between the model prediction value and the actual measurement value; the process noise covariance matrix Q parameter is a key parameter used to describe the uncertainty of the prediction model itself. Adjusting the Q value is equivalent to adjusting the system's trust in the physical model.

[0044] The method for determining innovation residual mismatch can be implemented by performing real-time online statistical tests on the innovation residual sequence: determining whether the statistical characteristics of the innovation residual sequence significantly deviate from the theoretical characteristics of zero-mean Gaussian white noise within a preset sliding time window. This test includes one or more of the following criteria: first, calculating the sample mean of the innovation residual within the window. If its absolute value continuously exceeds a preset small threshold, it is determined that the forecast has a systematic bias; second, calculating the autocorrelation function of the innovation residual within the window. If its multiple values ​​at non-zero delays exceed the preset confidence interval, it is determined that the uncorrelated white noise characteristics have been lost; and calculating the sample covariance matrix of the innovation residual within the window and comparing it with the innovation covariance matrix calculated by the filter based on the current model parameters. If the difference between the two exceeds the statistical significance level, it is determined that the current model no longer matches the actual physical process. When any one or more of the above test conditions are met, the system determines that "innovation residual mismatch" exists, thereby triggering dynamic adjustment of the process noise covariance matrix Q.

[0045] The dynamic adjustment process is to make the theoretical covariance of the innovation predicted by the filter model consistent with the covariance of the innovation samples statistically calculated from the actual measurement data by modifying the Q value.

[0046] The specific adjustment algorithm is an iterative update mechanism: the current sample covariance matrix is ​​calculated using the sequence of innovation residuals within a sliding time window. A weighted averaging method with a forgetting factor is then used for smoothing updates. The new Q matrix is ​​a linear combination of the previous Q matrix and a correction term calculated based on the latest innovation residuals. The magnitude of this correction term is proportional to the square of the innovation residuals, ensuring that the adjustment is strengthened when the forecast error is large, thereby guiding Q to converge in a direction that better accounts for the current measured error.

[0047] In addition, the online model parameter adaptation method also integrates the mutation event monitoring function. When the instantaneous value of the new information residual exceeds a preset "safety threshold", the system will determine that a sudden event such as a large wrinkle or tear in the fabric has occurred, and immediately send a pause command to the printing execution unit to avoid the production of large areas of defective products. This function serves as the ultimate safety guarantee and minimizes the economic losses caused by sudden physical failures.

[0048] The adaptive prediction method disclosed in the present invention establishes a set of scientific and objective model mismatch judgment criteria by performing multi-dimensional online statistical tests on the new information residuals. It can accurately distinguish between changes in fabric properties and random noise, avoiding system misjudgments and parameter oscillations. Furthermore, the smooth iteration and weighted average parameter update mechanism it adopts ensures that the prediction model can quickly and stably converge to a state that matches the new material when faced with changes in working conditions. This highly robust and stable fully autonomous adaptive capability enables the system to continuously guarantee the highest prediction accuracy for various types of variable fabrics without any human intervention, significantly reducing production and debugging costs.

[0049] Furthermore, the step of performing dynamic pixel interpolation based on the inverse distortion map and the texture feature map includes:

[0050] The texture feature map calculates a quantitative texture descriptor for each pixel area by applying a grayscale co-occurrence matrix and / or a wavelet energy spectrum analysis algorithm to a real-time image of the fabric. The grayscale co-occurrence matrix is ​​a classic statistical method for quantifying texture by analyzing the spatial relationship between pairs of pixels with different grayscale levels. The wavelet energy spectrum analysis algorithm is a method that can simultaneously analyze the characteristics of an image at different frequencies and spatial positions. The texture descriptors include roughness, contrast, and directionality. When performing pixel interpolation, an interpolation strategy mapping function is established. The interpolation strategy mapping function takes the texture descriptor as input and makes a decision based on preset corresponding rules. From a set of interpolation kernels containing different characteristics, an interpolation kernel that matches the texture characteristics of the pixel point currently to be calculated is selected.

[0051] The interpolation strategy mapping function can be implemented as a multidimensional lookup table or a pre-trained decision tree. A multidimensional input vector is composed of multiple quantized texture descriptors, such as "roughness," "contrast," and "directivity." Each entry or leaf node in the lookup table or decision tree stores one or a set of output instructions, which specify which interpolation kernel to select from a toolkit of interpolation kernels or the specific weight coefficients used for weighted fusion of multiple interpolation kernels. For example, a specific mapping rule can be set as follows: when the input vector represents an area of ​​"high roughness" and "high contrast," the mapping function outputs an instruction to select the sharpening Lanczos interpolation kernel with 100% weight; when the input represents an area of ​​"low roughness," the output instruction selects the smoothing bicubic interpolation kernel with 100% weight. For textures in between, a set of non-zero weight coefficients can be output to linearly weight the calculation results of the two interpolation kernels, thereby achieving a smooth transition in interpolation effect.

[0052] The dynamic pixel interpolation process is coupled with an ink bleeding compensation model based on the printed color. At the intersection of dark and light colors, a stronger "anti-bleeding" pixel correction is applied to ensure sharp color boundaries. This results in a final printed image that triple-compensates for macro-stretching, micro-texture, and the physical properties of the ink. This method pre-compensates for the physical bleeding of the ink, resolves the blurring caused by color bleeding, and significantly improves the sharpness of pattern details and color boundaries.

[0053] By employing scientific texture quantification methods such as gray-level co-occurrence matrices, combined with efficient decision-making mechanisms such as lookup tables or decision trees, this method achieves intelligent and refined image generation. Based on precise analysis of fabric microtexture, this method makes adaptive decisions at the pixel level, matching rendering strategies to specific regions. This overcomes the technical bottleneck of traditional single interpolation algorithms, which often compromise between clarity and smoothness. It sharpens rough textures to maintain clarity of detail, while softening smooth areas to ensure smooth transitions. The resulting product combines the sharpness of high-frequency patterns with the softness of large color blocks.

[0054] Furthermore, the beacon feature forms a pseudo-random dot matrix on the fabric surface, wherein the dot matrix pattern in any local area is unique.

[0055] This pseudo-random dot matrix can be constructed based on the generation principle of a two-dimensional de Bruijn sequence. This two-dimensional de Bruijn sequence can generate a large, non-periodic binary array in which any local binary pattern observed within a sliding window of a preset size appears only once in the entire array. By mapping the "1" and "0" of this binary array to "printing" and "not printing" of fluorescent ink, respectively, a locally unique beacon pattern can be formed on the fabric.

[0056] By adopting a pseudo-random dot matrix based on a two-dimensional de Bruijn sequence, traditional relative displacement tracking is upgraded to fast and robust absolute coordinate positioning. Due to the uniqueness of any local pattern, the system does not rely on continuous tracking and can instantly calculate its global coordinates by analyzing any local image. This feature not only greatly improves the efficiency of system initialization, but also enables instant repositioning after tracking is accidentally lost due to occlusion, giving the system extremely high fault tolerance. It fundamentally eliminates the inevitable cumulative error problem in traditional incremental tracking, ensures geometric positioning accuracy from start to finish in long-distance printing tasks, and significantly enhances system reliability.

[0057] Furthermore, the steps of analyzing the statistical characteristics of the innovation residuals by the online model parameter adaptive method are specifically as follows:

[0058] The covariance matrix of the innovation residual sequence is calculated within a sliding time window; when the trace of the covariance matrix exceeds a set threshold continuously within a predetermined number of image frames, a dynamic adjustment process is triggered.

[0059] During the debugging phase, the values ​​of the covariance matrix trace at the debugging order are continuously collected. By statistically analyzing the large amount of sample data collected, the mathematical expectation and standard deviation are calculated. The threshold can be set based on the principles of statistical process control to a level that can clearly distinguish normal fluctuations from abnormal deviations. The predetermined number of image frames can be set within the range of 10 to 30 frames.

[0060] The parameter calibration method disclosed in this invention transforms the threshold setting problem, which originally relied on experience, into a clear, quantifiable engineering practice. By determining the trigger boundary based on statistical principles, the objectivity and high reliability of mismatch judgment are ensured, and random noise interference can be effectively filtered out to avoid system misjudgments or slow responses. This standardized calibration process not only significantly improves the operational stability of the adaptive system, but also reduces the complexity and human dependency of equipment debugging, and has high engineering practical value.

[0061] Furthermore, the step of selecting an interpolation kernel that matches the texture characteristics of the pixel point for the current pixel point to be calculated is specifically as follows:

[0062] Weighted fusion is performed on n interpolation kernels in a group of the interpolation kernels; the interpolation strategy mapping function outputs a set of normalized weight coefficients according to the texture descriptor, and the weight coefficients are used to linearly combine the calculation results of the n interpolation kernels to generate a final pixel color value.

[0063] The weight coefficients are the fusion ratios corresponding to specific texture vectors, determined experimentally or through optimization algorithms during the system calibration phase. When performing weighted fusion, the system queries the corresponding set of weight coefficients based on the texture vector of the current pixel. The system uses these weight coefficients to linearly combine the results of the n interpolation kernels to generate a final pixel color that takes into account the characteristics of different textures; n is at least 2.

[0064] By weightedly fusing multiple interpolation kernels, this invention achieves a continuous and smooth transition between different interpolation effects, avoiding the visual abrupt changes that can occur when switching algorithms. Because the fusion weights are pre-determined through an optimization algorithm, the trade-off between clarity and smoothness is precisely controlled based on data. This "offline optimization, online query" model makes real-time computation extremely efficient, significantly improving the visual consistency and quality of the final product without sacrificing speed.

[0065] This invention provides a comprehensive, high-precision, real-time calibration method. First, its hybrid feature tracking technology ensures the system's robust applicability across a wide range of fabrics, from smooth to rough. Second, its innovative adaptive prediction algorithm eliminates lag errors in high-speed printing through pre-judgment compensation and automatically adapts to material changes, significantly improving both accuracy and efficiency. Finally, its unique texture-aware dynamic interpolation technology intelligently balances pattern clarity and color smoothness at the pixel level, resulting in printed products with both sharp details and a soft texture, enhancing visual quality.

[0066] Example 2

[0067] This embodiment further describes an operation mode of a digital printing color calibration system based on real-time image processing in an actual application scenario. Figure 3 On a high-speed production line at digital printing equipment manufacturer A, a geometric pattern composed of fine lines, requiring extremely high precision, is accurately printed on a Lycra blend fabric with high elasticity and low surface friction coefficient used in high-end custom swimwear.

[0068] Here's how it works:

[0069] Feature Extraction Module: In this scenario, to address the challenges of swimsuit fabric's high elasticity and the fine or missing texture features in some areas, the system pre-prints a layer of invisible fluorescent pseudo-random dots along both edges of the fabric using a specialized printhead before feeding it into the printer. The pattern is constructed based on the principle of a two-dimensional de Bruijn sequence and is invisible to the naked eye. When the fabric enters the system, the module's image acquisition component actively illuminates the fabric with UV light of the corresponding wavelength (for example, 365nm) and uses a bandpass filter in front of the camera lens to capture a clear fluorescent beacon pattern with a high signal-to-noise ratio. The module's built-in pattern recognition algorithm based on local feature descriptors (such as the SURF algorithm) simultaneously analyzes the fabric's knit texture, extracting thousands of stable structured feature points. The module then intelligently fuses these two sets of data based on confidence weighting. For each texture feature point, the system calculates a confidence score (between 0 and 1) based on its matching score and contrast. When constructing the coordinate set, the fluorescent beacon points are weighted to 1.0, the highest value, while the texture points are weighted according to their respective confidence scores. By solving a weighted least squares problem, the module ultimately constructs a highly dense, precise, and completely unaffected two-dimensional coordinate set composed of tens of thousands of points.

[0070] Deformation Prediction Module: Upon receiving the continuous coordinate data stream from the Feature Extraction Module, the Deformation Prediction Module immediately compares it with the reference coordinate system, calculating in real time the non-rigid geometric deformation field of the swimsuit fabric at the current observation location due to high tension. The module's built-in adaptive Kalman filter algorithm is immediately activated. Based on the current transmission speed and system latency, it predicts the target deformation field when the fabric reaches the print head a few tenths of a second later. By analyzing the innovation residuals between the predictions and the actual observations, it learns the specific elastic coefficients of the swimsuit fabric in real time. As the fabric's stretching behavior changes, the algorithm automatically adjusts its internal prediction model parameters to ensure consistently accurate deformation predictions for this highly elastic fabric. Furthermore, the module incorporates a sudden event detection function. If the instantaneous innovation residual exceeds a safe threshold, it identifies an unexpected event, such as a fabric tear, and pauses printing, preventing excessive waste.

[0071] The image generation module calculates the inverse transformation based on the target deformation field output by the deformation prediction module and generates a pixel-level inverse distortion map, which is used to perform geometric corrections by pre-stretching or pre-compressing the original precise geometric pattern. It also performs local texture analysis on the real-time fabric image to generate a feature map describing the knitted texture of the swimsuit fabric surface. When performing pixel interpolation, this module relies on a multidimensional lookup table (LOT) that serves as an interpolation strategy mapping function. It retrieves the optimal interpolation instruction from the LUT based on the geometric position of the current pixel (derived from the inverse distortion map) and the texture characteristics (derived from the texture feature map). In relatively smooth areas, the instruction uses a softer bicubic interpolation kernel, or even a weighted blend of the two. For example, in the "grooves" of the fabric texture, a sharper interpolation kernel is used to ensure edge clarity of the geometric lines, while a softer interpolation kernel is used in relatively smooth areas. Finally, the module is coupled with an ink penetration compensation model based on printing color to perform "anti-bleeding" corrections on pixels at the junction of dark and light colors, ultimately generating a set of printing data that has been triple-compensated for geometry, texture, and ink physical properties.

[0072] Through its highly robust hybrid feature extraction module, the digital printing color calibration system based on real-time image processing first solved the industry's difficult problem of stable tracking on highly challenging elastic fabrics. On this basis, its adaptive prediction module can learn the deformation patterns of specific fabrics in real time and, through precise advance compensation, completely eliminate geometric distortion under high-speed production. Furthermore, the image generation module combines macro-geometric correction with micro-texture perception to ensure that the edges of fine patterns remain sharp and clear, and color transitions are natural. In summary, the system synergistically solves the three core issues of stability, accuracy, and ultimate visual quality, making high-quality printing production possible under highly challenging and high-speed conditions.

[0073] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A digital printing color calibration method based on real-time image processing, characterized in that: include: Analyze the real-time acquired image of the fabric to be printed, identify and locate structured texture features in the image using a pattern recognition algorithm, and / or detect and locate preset beacon features using wavelength filtering technology, and construct a two-dimensional coordinate set consisting of texture features and beacon features; Comparing the two-dimensional coordinate set with a reference coordinate system, calculating the displacement vector of each feature point, and solving a non-rigid geometric deformation field describing the current state of the fabric; using the non-rigid geometric deformation field as input, applying an adaptive time-domain prediction algorithm, combining the fabric transmission speed and physical delay parameters, to predict the target deformation field when the fabric reaches the printing position; Calculating an inverse transformation based on the target deformation field to generate a pixel-level inverse distortion map; Performing local texture analysis on the fabric image to be printed to generate a texture feature map; resampling and dynamic pixel interpolation of the original digital image to be printed according to the inverse distortion mapping and the texture feature map to generate final printing image data.

2. The digital printing color calibration method based on real-time image processing according to claim 1, characterized in that: Aiming at the texture features, a pattern recognition algorithm is used to identify and extract key feature points that remain stable when the fabric is rotated and scaled; the beacon feature is fluorescence that can only be displayed under a fixed non-visible light band. By configuring an active light source of the corresponding band for excitation illumination, a bandpass filter is set on the image acquisition path to detect the center of mass coordinates of the beacon.

3. The digital printing color calibration method based on real-time image processing according to claim 1, characterized in that: The adaptive time-domain prediction algorithm is specifically an adaptive Kalman filter algorithm; the adaptive Kalman filter algorithm integrates an online model parameter adaptation method; the online model parameter adaptation method continuously calculates the innovation residual between the predicted target deformation field and the non-rigid geometric deformation field actually measured at the next moment; When the physical properties of the fabric change, resulting in a mismatch in the innovation residual, the online model parameter adaptation method dynamically adjusts the process noise covariance matrix Q parameters inside the filter to match the new deformation law.

4. The digital printing color calibration method based on real-time image processing according to claim 1, characterized in that: The step of performing dynamic pixel interpolation according to the inverse distortion map and the texture feature map comprises: The texture feature map calculates a quantized texture descriptor for each pixel area by applying a gray-level co-occurrence matrix and / or a wavelet energy spectrum analysis algorithm to a real-time fabric image; the texture descriptor includes roughness, contrast, and directionality. When performing pixel interpolation, an interpolation strategy mapping function is established. The interpolation strategy mapping function takes the texture descriptor as input and makes a decision based on preset corresponding rules. From a set of interpolation kernels containing different characteristics, an interpolation kernel that matches the texture characteristics of the pixel point currently to be calculated is selected.

5. The digital printing color calibration method based on real-time image processing according to claim 2, characterized in that: The beacon feature forms a pseudo-random dot matrix on the fabric surface, wherein the dot matrix pattern in any local area is unique.

6. The digital printing color calibration method based on real-time image processing according to claim 3, characterized in that: The steps of analyzing the statistical characteristics of the innovation residuals by the online model parameter adaptive method are specifically as follows: The covariance matrix of the innovation residual sequence is calculated within a sliding time window; when the trace of the covariance matrix exceeds a set threshold continuously within a predetermined number of image frames, a dynamic adjustment process is triggered.

7. The digital printing color calibration method based on real-time image processing according to claim 4, characterized in that: The step of selecting an interpolation kernel that matches the texture characteristics of the pixel point to be calculated currently comprises: performing weighted fusion on n interpolation kernels in a group of the interpolation kernels; the interpolation strategy mapping function outputs a set of normalized weight coefficients according to the texture descriptor, and linearly combines the calculation results of the n interpolation kernels using the weight coefficients to generate a final pixel color value.

8. A digital printing color calibration system based on real-time image processing, characterized in that: include: Feature extraction module: Analyzes the real-time image of the fabric to be printed, identifies and locates the structured texture features in the image using a pattern recognition algorithm, and / or detects and locates preset beacon features using wavelength filtering technology, and constructs a two-dimensional coordinate set consisting of texture features and beacon features; Deformation prediction module: compares the two-dimensional coordinate set with the reference coordinate system, calculates the displacement vector of each feature point, and solves the non-rigid geometric deformation field that describes the current state of the fabric. Using the non-rigid geometric deformation field as input, an adaptive time-domain prediction algorithm is applied, combining the fabric transmission speed and physical delay parameters, to predict the target deformation field when the fabric reaches the printing position. Image generation module: calculates the inverse transformation according to the target deformation field to generate a pixel-level inverse distortion map; performs local texture analysis on the fabric image to be printed to generate a texture feature map; resamples and dynamically interpolates pixels of the original digital image to be printed based on the inverse distortion map and the texture feature map to generate final printed image data.

Images