Digital pattern-based flexible deformation compensation for digital printing product production method

Abstract

The present application relates to a digital pattern-based flexible deformation compensation digital printing product production method, belonging to the technical field of digital printing, first obtaining surface micro-topological structure data by scanning cloth, then using conditional generative adversarial network to construct deformation defect prediction simulation engine, using the engine to simulate ink blurring based on the currently used cloth and the original design pattern, predicting the area where the pattern misplacement occurs, using reverse rendering compensation algorithm to reversely modify the local pixel structure or dot shape of the pattern for the area, finally generating a compensation pattern, and completing digital printing according to the compensation pattern. Before printing, the present application realizes the pre-position quality control through the strategy of "virtual trial and error first, physical jet printing later", reduces the rate of defective products, and avoids the waste of fabric.

Description

Technical Field

[0001] This invention relates to the field of digital printing technology, and more specifically to a method for producing digitally printed products based on flexible deformation compensation of digital patterns. Background Technology

[0002] The dyeing and finishing industry is one of my country's earliest developed and internationally competitive traditional industries, and it is the most crucial technological link in enhancing the added value of textiles. Digital printing, with its advantages of low energy consumption and low emissions, has become a technology choice that aligns with environmental trends in recent years.

[0003] In digital printing, the digital printing machine prints a pre-generated pattern onto the fabric using inkjet technology. Because fabric is flexible and prone to deformation, misalignment of the printed pattern can occur during the printing process, resulting in defects and discrepancies between the digital design and the actual printed effect. Furthermore, heterogeneous equipment produces varying results when handling different fabrics, making it difficult to effectively eliminate defects caused by fabric deformation and misalignment during digital printing. This impacts the production efficiency of large-scale customized production.

[0004] In existing technologies, the main method for controlling defects in digital printing patterns is to control the amount of ink jetted. However, the difference in ink absorption rate when heterogeneous equipment processes different fabrics leads to color difference problems. Therefore, it is not possible to use a single ink jetting amount. When using different fabrics, it is necessary to increase the calculation and trial and error costs, which affects the efficiency of mass customized production. At the same time, some patterns can only show the best effect on certain high-end fabrics, which undoubtedly increases production costs. Summary of the Invention

[0005] This invention discloses a digital printing product production method based on digital pattern flexible deformation compensation. Before printing, a strategy of "virtual trial and error first, then physical printing" is adopted to achieve pre-printing quality control.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The digital printing product manufacturing method based on digital pattern flexible deformation compensation includes the following steps: Step 1): Scan the fabric to be printed using a scanning camera to obtain the microscopic topological structure data of the fabric surface, and use the Yule-Nielsen model of light scattering inside the fiber and the point diffusion function of ink penetration to establish the physical basis describing the interaction between ink and fiber. Step 2): Construct a deformation defect prediction model engine using a conditional generative adversarial network. Through this engine, learn the ink diffusion pattern on the fiber. Based on the obtained microscopic topological structure data of the fabric surface and the original printed pattern, simulate the printing effect of the current fabric and determine the location of the generated pattern misalignment defect. Step 3): For the misaligned areas of the pattern, the local pixel structure or dot shape of the pattern is modified in reverse through the reverse rendering compensation algorithm to finally generate a compensation pattern; Step 4): The digital printing machine performs digital printing based on the compensation pattern.

[0007] Furthermore, in step 1), assuming the fabric fiber surface is a Lambertian body, the pixel brightness received by the scanning camera... I Depends on surface albedo ρ Surface normal vector n and the direction of the light source l The relationship between them: , In the above formula, I ( u,v ) represents image coordinates ( u,v Brightness observation value at ( ); ρ ( u,v ) represents the surface albedo, which is the color and reflectivity of the material itself; n ( u,v ) is the unit surface normal vector of the image coordinate point, used to describe the tilt angle of the fiber; l Indicates the direction of a known light source; The scanning camera is kept stationary, and light is shone on the fabric to be scanned from multiple different directions to obtain multiple images. The surface normal vector is then deduced through matrix operations. , m = ρ ( u,v ) n ( u,v ), L T Represents the light source direction matrix L transpose, I This represents the brightness vector observed at the same pixel under illumination from multiple directions. L For the light source matrix; in, ; The normal vector was obtained. n x , n y , n z After that, convert it into a height map Z( x , y Then solve the gradient integral: , In the above formula, p and q They represent the height at...x , y Local slope in direction, dx, dy Indicates along x direction and y The small integral step size in the direction, i.e. the distance between adjacent sampled pixels or sampled points; , Using the height map obtained above, the root mean square roughness is calculated. S q Quantifying the fuzziness of fabric fibers: , In the above formula, S q It is a surface roughness index; A The area of ​​the sampling region; The reference plane height.

[0008] Furthermore, in steps 1) and 2), the Yule-Nielsen model is used to describe the physical and visual differences caused by light scattering inside the fabric, and the point spread function (PSF) is convolved with the original printed pattern; then, the original digital image is input into the deformation defect prediction model engine, and the simulated fabric printing effect image is output.

[0009] Furthermore, the objective function of the conditional generative adversarial network is as follows: , In the above formula, This represents the adversarial loss of a conditional generative adversarial network. G For generator, D For discriminator, x This represents the original pattern and fabric topology features in the conditional input. y This indicates the printing result obtained from actual scanning or calibration. z This represents random disturbances or latent variables. E This represents the mathematical expectation operation; the first term on the right side of the above formula is the constraint discriminator identifying real samples, and the second term is the constraint generator generating a simulation result that closely approximates the real printing effect; To ensure that the generated pattern structure is consistent with the original Figure 1 To, Join L 1 Distance constraints: .

[0010] Furthermore, in step 3), when modifying the local pixel structure or dot shape of the pattern in reverse, the solution is minimized using the following formula: , In the above formula, Iinput This represents the compensation input pattern finally obtained by the system; argmin represents the scheme that minimizes the objective function among all candidate input patterns. F ( I (fabric_topology) represents the physical printing and ink diffusion process under given fabric topology conditions; I target This represents the target design diagram; λ is the regularization weight. Ω(I) Regularization terms to suppress excessive modification and noise.

[0011] This invention achieves proactive quality control through a strategy of "virtual trial and error first, followed by physical printing." When compensating for printing defects, it does not make general adjustments to the entire fabric, but dynamically generates compensation schemes based on the microscopic physical differences of each inch of fabric, resulting in extremely high precision. Through targeted algorithmic compensation, the same digital pattern presents a visually consistent effect on different fabric media with varying physical properties. By shifting quality control from post-production inspection to pre-production simulation, potential defects are eliminated before the ink is sprayed, greatly reducing the defect rate and the waste of expensive fabrics. Attached Figure Description

[0012] Figure 1 This is a schematic diagram illustrating the principle of flexible deformation compensation for digital patterns in this embodiment. Detailed Implementation

[0013] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0014] This embodiment discloses a digital printing product production method based on digital pattern flexible deformation compensation, combined with... Figure 1 The production process is described below: Step 1) After the original digital design pattern is input into the corresponding production equipment, the digital pattern flexible deformation compensation system first uses the scanning camera of the digital printing machine to scan the fabric to be printed, obtaining the microscopic topological structure information of the fabric surface. During the scanning process, the Lambertian reflection model is first used: assuming the fabric fiber surface is a Lambertian body, the pixel brightness received by the scanning camera... I Depends on surface albedo ρ Surface normal vector n and the direction of the light source l The relationship between them.

[0015] , In the above formula, I ( u,v ) represents image coordinates ( u,v Brightness observation value at ( ); ρ( u,v ) represents the surface albedo, which is the color and reflectivity of the material itself; n ( u,v ) is the unit surface normal vector of the image coordinate point, used to describe the tilt angle of the fiber; l This indicates the direction of the known light source. The above Lambert reflection formula uses... I ( u,v As an observation, it is necessary to solve for the coordinates of each pixel. u,v The mixing vector at ) m , m Includes geometric shape and material information. m ( u,v )= ρ ( u,v ) n ( u,v Then, the subsequent formula will... m The surface albedo and unit normal vector are obtained by decomposition.

[0016] To obtain the 3D structure, the system keeps the scanning camera stationary while illuminating it from multiple different directions, resulting in multiple images. Matrix operations are then used to inversely solve for the surface's normal vectors. , The above formula is used for m ( u,v )= ρ ( u,v ) n ( u,v Perform least squares estimation, where, L T Represents the light source direction matrix L transpose, I This represents the brightness vector observed at the same pixel under illumination from multiple directions. L This is the light source matrix.

[0017] Among them, the recovered normal vector n and albedo ρ for: , Step 2), obtained the normal vector ( n x , n y , n z After that, it needs to be converted into a specific height map Z( x , y Since the fabric's microstructure is uneven, the gradient integral is then solved: , The above integral formula is used to recover the height map from the normal vector gradient. p and q They represent the height at... x , y Local slope in direction, dx, dy Indicates along x direction and y The small integral step size in the direction, i.e. the distance between adjacent sampled pixels or sampled points.

[0018] , The system uses the height map obtained above to quantify fiber hairiness, typically using root mean square roughness. S q express: , In the above formula, S q This is a surface roughness index; the higher the value, the more severe the fuzziness, and the easier it is for ink to bleed. A The area of ​​the sampling region; The reference plane height.

[0019] The reason for scanning the microscopic topology is that traditional printing only considers macroscopic flatness, but fabric is a porous and non-uniform medium. The fuzziness and weave density of fibers directly affect the diffusion path of ink. Without obtaining this microscopic physical data, it is impossible to accurately predict the final landing point and degree of ink bleeding.

[0020] Step 3) Using the aforementioned information, simulate the ink penetration effect on fabric in a GAN (Conditional Generative Adversarial Network) simulation engine. The Yule-Nielsen model is the most commonly used formula in the printing industry for predicting dot gain, used to describe the physical and visual differences caused by light scattering within paper or fabric. , In the above formula, R This is the final predicted reflectance (the color perceived by the human eye); R i The reflectivity of the ink; R s The reflectivity of the substrate (fabric); a The proportion of physical area covered by ink dots is not a fixed constant, but a local ink dot coverage proportion predicted based on fiber roughness, porosity, local texture direction, ink droplet volume and original pattern dot density. n It is the Yule-Nielsen factor, which is related to the light scattering ability of the medium. nInitial empirical values ​​can be set according to the fabric type first, and then calibrated by scanning samples or printing calibration data. Therefore, it can be used as a calibration parameter in the same batch of fabrics, and adjusted according to the scattering ability of the medium in cross-material scenarios.

[0021] Meanwhile, to simulate microscopic penetration, a point spread function (PSF) is typically used to convolve with the original pattern: , In the above formula, I predicted The printed pattern is predicted after convolution; I original Represents the original input digital pattern; * indicates a convolution operation; PSF fabric This represents a two-dimensional point diffusion function determined by the fabric height, roughness, porosity, and fiber orientation. h, w This indicates the sampling location of the diffusion nucleus in the longitudinal and transverse directions.

[0022] Subsequently, a GAN is used to generate predicted defects. The original digital image is input, and a simulated fabric printing effect image is output. This process uses a conditional generative adversarial network, whose objective function is as follows: , In the above formula, This represents the adversarial loss of a conditional generative adversarial network. G For generator, D For discriminator, x This represents the original pattern and fabric topology features in the conditional input. y This indicates the printing result obtained from actual scanning or calibration. z This represents random disturbances or latent variables. E This represents the mathematical expectation operation; the first term on the right side of the above formula is the constraint discriminator identifying real samples, and the second term is the constraint generator generating a simulation result that closely approximates the real printing effect.

[0023] To ensure that the generated pattern structure is consistent with the original Figure 1 To, usually added L 1 Distance constraints: , Step 4): If fabric deformation in a certain area is found to cause misalignment defects in the printed pattern, the system will use a reverse rendering compensation algorithm to modify the local pixel structure or dot shape of the pattern to offset the misalignment defects. In this process, the model calls the reverse rendering compensation algorithm to modify the pixel structure of the defective area. This is a minimization problem; the system's goal is to find a new input pattern. I inputThis ensures that the result after the physical inkjet printing process F is closest to the target design drawing. I target .

[0024] , In the above formula, I input This represents the compensation input pattern finally obtained by the system; argmin represents the scheme that minimizes the objective function among all candidate input patterns. F ( I (fabric_topology) represents the physical printing and ink diffusion process under given fabric topology conditions; I target This represents the target design diagram; λ is the regularization weight. Ω(I) Regularization terms to suppress excessive modification and noise.

[0025] The reason for using reverse modification of the pixel structure is that the system found that a certain fiber would cause excessive ink diffusion and blur the pattern. The algorithm changes the shape of the dots in the original pattern (for example, by pre-shrinking the edge of the pattern or changing the dot distribution). In this way, when the ink inevitably diffuses on the physical fabric, the final effect will just offset the deformation and restore the original design.

[0026] This process ultimately achieved the following effects: 1) Consistent visual effects across materials: Whether on silk, cotton, linen, or synthetic fibers, targeted algorithmic compensation ensures that the same digital pattern presents a consistent visual effect to the naked eye on fabrics with different physical properties. 2) Quality control is moved from post-production inspection to pre-production simulation. Potential visual defects are eliminated before the printhead actually sprays ink, greatly reducing the defect rate and waste of expensive fabrics. 3) Pixel-level precise control: Instead of uniformly adjusting the entire fabric, the printing scheme for the corresponding area is dynamically adjusted based on the microscopic differences of each inch of fabric.

[0027] After compensation is completed, the system will first generate an optimized printing pattern scheme before issuing a printing permit to the digital printing all-in-one machine. Then, it will be handed over to the printhead for execution. The final product should achieve consistent visual effects across materials. This scanning-simulation-modification process will be completed before the printhead starts printing.

[0028] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for producing digital printing products based on flexible deformation compensation of digital patterns, characterized in that, Includes the following steps: Step 1): Scan the fabric to be printed using a scanning camera to obtain the microscopic topological structure data of the fabric surface, and use the Yule-Nielsen model of light scattering inside the fiber and the point diffusion function of ink penetration to establish the physical basis describing the interaction between ink and fiber. Step 2): Construct a deformation defect prediction model engine using a conditional generative adversarial network. Through this engine, learn the ink diffusion pattern on the fiber. Based on the obtained microscopic topological structure data of the fabric surface and the original printed pattern, simulate the printing effect of the current fabric and determine the location of the generated pattern misalignment defect. Step 3): For the misaligned areas of the pattern, the local pixel structure or dot shape of the pattern is modified in reverse through the reverse rendering compensation algorithm to finally generate the compensation pattern; Step 4): The digital printing machine performs digital printing based on the compensation pattern.

2. The method for producing digital printing products based on flexible deformation compensation of digital patterns according to claim 1, characterized in that: In step 1), assuming the fabric fiber surface is a Lambertian body, the pixel brightness received by the scanning camera... I Depends on surface albedo ρ Surface normal vector n and the direction of the light source l The relationship between them: ， In the above formula, I ( u,v ) represents image coordinates ( u,v Brightness observation value at ( ); ρ ( u,v ) represents the surface albedo, which is the color and reflectivity of the material itself; n ( u,v ) is the unit surface normal vector of the image coordinate point, used to describe the tilt angle of the fiber; l Indicates the direction of a known light source; The scanning camera is kept stationary, and light is shone onto the fabric to be scanned from multiple different directions to obtain multiple images. The surface normal vector is then deduced through matrix operations. ， m = ρ ( u,v ) n ( u,v ), L T Represents the light source direction matrix L transpose, I This represents the brightness vector observed at the same pixel under illumination from multiple directions. L For the light source matrix; in, ; The normal vector was obtained. n x , n y , n z After that, convert it into a height map Z( x , y Then solve the gradient integral: ， In the above formula, p and q They represent the height at... x , y Local slope in direction, dx, dy Indicates along x direction and y The small integral step size in the direction, i.e. the distance between adjacent sampled pixels or sampled points; ， Using the height map obtained above, the root mean square roughness is calculated. S q Quantifying the fuzziness of fabric fibers: ， In the above formula, S q It is a surface roughness index; A The area of ​​the sampling region; The reference plane height.

3. The method for producing digital printing products based on flexible deformation compensation of digital patterns according to claim 1, characterized in that: In steps 1) and 2), the Yule-Nielsen model is used to describe the physical and visual differences caused by light scattering inside the fabric, and the point spread function (PSF) is convolved with the original printed pattern. Subsequently, the original digital image is input into the deformation defect prediction model engine, and the simulated fabric printing effect image is output.

4. The method for producing digital printing products based on flexible deformation compensation of digital patterns according to claim 3, characterized in that: The objective function of a conditional generative adversarial network is as follows: ， In the above formula, This represents the adversarial loss of a conditional generative adversarial network. G For generator, D For discriminator, x This represents the original pattern and fabric topology features in the conditional input. y This indicates the printing result obtained from actual scanning or calibration. z This represents random disturbances or latent variables. E This represents the mathematical expectation operation; the first term on the right side of the above formula is the constraint discriminator identifying real samples, and the second term is the constraint generator generating a simulation result that closely approximates the real printing effect; To ensure that the generated pattern structure is consistent with the original image, add... L 1 Distance constraints: 。 5. The method for producing digital printing products based on flexible deformation compensation of digital patterns according to claim 1, characterized in that: In step 3), when modifying the local pixel structure or dot shape of the pattern in reverse, the following formula is used for minimization: ， In the above formula, I input This represents the compensation input pattern finally obtained by the system; argmin represents the scheme that minimizes the objective function among all candidate input patterns. F ( I (fabric_topology) represents the physical printing and ink diffusion process under given fabric topology conditions; I target This represents the target design diagram; λ is the regularization weight. Ω(I) Regularization terms to suppress excessive modification and noise.

Images