A method, apparatus and storage medium for eliminating direct jet feathering by printing color prediction

Abstract

The application discloses a method and device for eliminating direct jet feathering channel by printing color prediction and a storage medium, relates to the technical field of printing, and comprises the following steps: first, a sample set containing different image features and historical printing data with / without feathering channel are acquired; image features are extracted based on the two, a neural network model containing an encoder-decoder is constructed and trained, and the neural network model is caused to learn the corresponding relationship between the image features and a feathering channel elimination color adjustment strategy; a to-be-printed image is input into the trained model, the model divides a high-risk area, matches optimal color adjustment parameters, and processes the image; and the processed image is sent to a printing device to perform printing, so that the direct jet feathering channel is eliminated; in addition, the model can be optimized in combination with medium characteristic parameters, the model is incrementally trained through actual printing effect, and the adaptability and accuracy of eliminating the feathering channel are improved.

Description

Technical Field

[0001] This invention relates to the field of printing technology, and more specifically, to a method, apparatus, and storage medium for eliminating direct injection feathering through printing color prediction. Background Technology

[0002] In the field of direct-to-garment printing, feathering (pass marks) is a core technical bottleneck restricting print quality. Essentially, it refers to periodic stripes formed by uneven ink droplet superposition at the junctions of different strokes during printhead scanning. It is widely present in batch printing scenarios such as textile printing, advertising inkjet printing, and industrial labeling. Current technologies for addressing feathering primarily focus on hardware calibration and manual parameter adjustments, such as adjusting printhead stepping accuracy through mechanical structures and manually optimizing ink volume distribution. However, these solutions have significant limitations. From a technical perspective, the generation of feathering is not only related to the hardware status of the printing equipment but also deeply coupled with the physical characteristics of the printing medium. The medium's absorbency determines the ink penetration rate, surface roughness affects the ink droplet diffusion morphology, and color reflectivity is related to the uniformity of color after printing. These three factors collectively cause feathering to exhibit differentiated characteristics for different media and image content. For example, high-absorbency cotton fabrics are prone to feathering due to insufficient ink volume, while low-roughness plastic films are more prone to feathering with blurred edges due to ink droplet accumulation. Current solutions struggle to overcome the limitations of static calibration in adapting to dynamic scenarios: on the one hand, manual adjustments rely on the experience of technicians, requiring adjustments to parameters such as brightness and contrast for different images, such as grayscale transition images, solid color blocks, and fine lines. Each adjustment is time-consuming and it is difficult to guarantee the consistency of printing results across different batches. On the other hand, traditional machine vision inspection can only identify feathering that has already occurred and cannot optimize it in advance through color prediction. This leads to the need for reprinting after problems occur, resulting in a waste of materials and time. In addition, existing models mostly use fixed algorithms to process all image types and do not perform differentiated optimization for high-risk areas of feathering, such as line edges and color block junctions, further reducing the accuracy of feathering elimination.

[0003] In summary, existing technologies suffer from problems such as low efficiency of manual debugging, difficulty in adapting to multiple media, and lack of predictive optimization capabilities. Summary of the Invention

[0004] In order to overcome the problems of low efficiency of manual debugging, difficulty in adapting to multiple media, and lack of predictive optimization capabilities in existing technologies, this invention discloses a method, device and storage medium for eliminating direct injection feathering through printing color prediction, which can effectively solve the above-mentioned technical problems.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] A method for eliminating direct injection feathering by predicting print color includes the following steps:

[0007] A set of image samples and historical printing data are acquired. The image sample set includes images with different grayscale transitions, solid color blocks, and line combinations. The historical printing data includes printed sample images with feathering and corresponding original images without feathering. Based on the image sample set and historical printing data, an image processing model is trained. The image processing model is used to learn feathering elimination color adjustment strategies corresponding to different image features. An image to be printed is acquired and input into the trained image processing model. The image processing model matches the optimal color adjustment strategy according to the features of the image to be printed and processes the image to be printed based on the strategy. The processed image to be printed is sent to the printing device, and the printing device is controlled to perform the printing operation to eliminate direct injection feathering.

[0008] Preferably, the step of training the image processing model based on the image sample set and historical printing data includes: extracting features from the images in the image sample set and historical printing data to obtain grayscale distribution features, color channel features, and line structure features; constructing a neural network model, which includes an encoder and a decoder, wherein the encoder is used to extract image features and the decoder is used to output color adjustment parameters; taking printed sample images with feathering lines from the historical printing data as input data and the corresponding original images without feathering lines as target output data, inputting them into the neural network model, optimizing the model parameters through a backpropagation algorithm, enabling the model to learn pixel-level dot-dodge and burn-down processing strategies, and obtaining the trained image processing model.

[0009] Preferably, the neural network model is at least one of convolutional neural network, generative adversarial network or Transformer network, and the loss function of the neural network model is a combination function of mean squared error loss and structural similarity index loss.

[0010] Preferably, the image processing model matches the optimal color adjustment strategy based on the features of the image to be printed, including: the image processing model divides the image to be printed into regions and identifies high-risk regions in the image that may produce feathering; for different regions after division, the color characteristics and texture features of the regions are extracted respectively, and the corresponding color adjustment parameters are matched from a preset color adjustment strategy library based on the extracted features, wherein the color adjustment parameters include brightness adjustment value, contrast adjustment coefficient and color gamut compression ratio.

[0011] Preferably, the method further includes an image processing model optimization step: collecting the processed image output by the printing device, and acquiring the actual printing effect image through an image acquisition device; comparing the actual printing effect image with a preset target effect image and calculating the deviation value between the two; determining whether the deviation value exceeds a preset deviation threshold; when the deviation value exceeds the preset deviation threshold, using the actual printing effect image and the corresponding image to be printed as new training data to incrementally train the image processing model and update the color adjustment strategy in the model.

[0012] Preferably, the historical printing data also includes characteristic parameters of different printing media, including media absorptivity, surface roughness, and color reflectivity; during the training of the image processing model, the media characteristic parameters are correlated with the corresponding color adjustment strategies, so that the trained image processing model can automatically adapt to the corresponding color adjustment strategy according to the characteristic parameters of the medium to be printed.

[0013] Preferably, the processing of the image to be printed based on this strategy includes: for continuous tone regions in the image to be printed, using a gradient smoothing algorithm to adjust the transition effect of pixel values ​​within the region; for edge line regions in the image to be printed, using a contour enhancement algorithm to optimize the continuity and clarity of the lines; and for solid color block regions in the image to be printed, using a uniformity calibration algorithm to eliminate the difference in brightness within the color block.

[0014] Preferably, an electronic device includes:

[0015] The data acquisition module is used to acquire image sample sets, historical printing data and images to be printed. The image sample sets include images with different grayscale transitions, solid color blocks and line combinations. The historical printing data includes printed sample images with feathering, corresponding original images without feathering, and characteristic parameters of different printing media.

[0016] The model training module is used to train an image processing model based on the image sample set and historical printing data. The image processing model is used to learn feathering elimination color adjustment strategies corresponding to different image features and different printing media characteristics.

[0017] The image processing module is used to input the image to be printed into the trained image processing model, control the image processing model to match the optimal color adjustment strategy according to the features of the image to be printed and the characteristic parameters of the printing medium, and process the image to be printed based on the strategy.

[0018] The print control module is used to send the processed image to be printed to the printing device and control the printing device to perform the printing operation, thereby eliminating the direct injection feathering path.

[0019] Preferably, it further includes a model optimization module and an image acquisition module; the image acquisition module is used to collect the processed image output by the printing device to obtain the actual printing effect image; the model optimization module is used to compare the actual printing effect image with the preset target effect image, calculate the deviation value, and when the deviation value exceeds the preset deviation threshold, use the actual printing effect image and the corresponding image to be printed as new training data to incrementally train the image processing model and update the color adjustment strategy in the model.

[0020] Preferably, a computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method for eliminating direct injection feathering by predicting print color as described above.

[0021] Compared with existing technologies, the beneficial effects of this invention are as follows: This technical solution addresses the pain points of existing technologies through technological innovation, resulting in significant technological gains: Regarding the problem of low efficiency in manual debugging, the solution first constructs an image sample set covering multiple gray levels, color blocks, and lines, along with historical printing data with / without feathering. This allows the image processing model to autonomously learn color adjustment strategies corresponding to different image features. When printing, only the image needs to be input, and the model can automatically identify high-risk areas, match optimal parameters (brightness, contrast, color gamut compression ratio), and complete the processing, eliminating the need for manual image-by-image debugging and shortening the single-image processing cycle, making it particularly suitable for batch printing scenarios. Regarding the difficulty in adapting to multiple media, the model incorporates characteristic parameters (absorption rate, roughness, reflectivity) of different printing media during training, and learns through association. By establishing a mapping relationship between media characteristics and adjustment strategies, the model can automatically adapt parameters when dealing with different media such as cotton, plastic, and paper. It also provides differentiated processing (gradient smoothing, contour enhancement, and uniformity calibration) for areas such as continuous tones, edge lines, and solid color blocks, overcoming the limitations of manual adjustment in handling diverse media and image types. Addressing the lack of pre-prediction optimization capabilities, the model learns the feathering pattern based on historical printing data. During the pre-printing stage, it predicts high-risk areas in advance through feature matching and applies optimization strategies to avoid reprinting after feathering occurs. Furthermore, through monitoring deviations in actual printing results and incremental training, the model's strategy library is continuously updated to ensure a steady improvement in prediction and optimization capabilities over long-term use. Ultimately, this achieves simultaneous breakthroughs in direct-to-garment feathering elimination efficiency, media adaptability, and predictive optimization capabilities. Attached Figure Description

[0022] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other embodiments can be derived from the provided drawings without creative effort.

[0023] Figure 1 This is a diagram illustrating the steps of the method of the present invention;

[0024] Figure 2 This is a structural diagram of the electronic device of the present invention. Detailed Implementation

[0025] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent.

[0026] To better illustrate this embodiment, some parts in the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions;

[0027] It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings.

[0028] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0029] Example 1

[0030] Please see Figure 1 A method for eliminating direct injection feathering by predicting print color includes the following steps:

[0031] A set of image samples and historical printing data are acquired. The image sample set includes images with different grayscale transitions, solid color blocks, and line combinations. The historical printing data includes printed sample images with feathering and corresponding original images without feathering. Based on the image sample set and historical printing data, an image processing model is trained. The image processing model is used to learn feathering elimination color adjustment strategies corresponding to different image features. An image to be printed is acquired and input into the trained image processing model. The image processing model matches the optimal color adjustment strategy according to the features of the image to be printed and processes the image to be printed based on the strategy. The processed image to be printed is sent to the printing device, and the printing device is controlled to perform the printing operation to eliminate direct injection feathering.

[0032] The step of training an image processing model based on the image sample set and historical printing data includes: extracting features from the images in the image sample set and historical printing data to obtain grayscale distribution features, color channel features, and line structure features; constructing a neural network model, which includes an encoder and a decoder, wherein the encoder is used to extract image features and the decoder is used to output color adjustment parameters; taking printed sample images with feathering lines from the historical printing data as input data and the corresponding original images without feathering lines as target output data, inputting them into the neural network model, optimizing the model parameters through a backpropagation algorithm, enabling the model to learn pixel-level dot-dodge and burn-down processing strategies, and obtaining the trained image processing model.

[0033] The neural network model is at least one of convolutional neural network, generative adversarial network or Transformer network, and the loss function of the neural network model is a combination function of mean squared error loss and structural similarity index loss.

[0034] The image processing model matches the optimal color adjustment strategy based on the features of the image to be printed, including: dividing the image to be printed into regions and identifying high-risk regions in the image that may produce feathering; extracting the color characteristics and texture features of different regions after division, and matching the corresponding color adjustment parameters from a preset color adjustment strategy library based on the extracted features, wherein the color adjustment parameters include brightness adjustment value, contrast adjustment coefficient and color gamut compression ratio.

[0035] It also includes an image processing model optimization step: collecting the processed image output by the printing device, and acquiring the actual printing effect image through an image acquisition device; comparing the actual printing effect image with the preset target effect image and calculating the deviation value between the two; determining whether the deviation value exceeds the preset deviation threshold; when the deviation value exceeds the preset deviation threshold, using the actual printing effect image and the corresponding image to be printed as new training data to incrementally train the image processing model and update the color adjustment strategy in the model.

[0036] The historical printing data also includes characteristic parameters of different printing media, including media absorptivity, surface roughness, and color reflectivity. During the training of the image processing model, the media characteristic parameters are correlated with the corresponding color adjustment strategies, so that the trained image processing model can automatically adapt to the corresponding color adjustment strategy according to the characteristic parameters of the printing media.

[0037] The processing of the image to be printed based on this strategy includes: for continuous tone regions in the image to be printed, using a gradient smoothing algorithm to adjust the transition effect of pixel values ​​within the region; for edge line regions in the image to be printed, using a contour enhancement algorithm to optimize the continuity and clarity of the lines; and for solid color block regions in the image to be printed, using a uniformity calibration algorithm to eliminate the difference in brightness within the color block.

[0038] During direct inkjet printing, feathering often occurs due to uneven ink diffusion on paper or other media, inkjet accuracy issues of the print head, and improper processing of image edge transitions. Feathering manifests as blurred edges, unnatural color transitions, or streaks or marks on the printed image, severely affecting print quality. Traditional manual adjustment of printing parameters or simple image preprocessing methods are difficult to effectively solve this problem.

[0039] Collect image samples containing different grayscale transitions, solid color blocks, and line combinations. These image samples should cover various common print image types, such as landscape photos (rich grayscale transitions), text documents (mainly solid color blocks), and line patterns (line combinations). You can select from public image databases, or take or design your own images that meet the requirements. For example, select natural landscape images with grayscale transitions from the ImageNet database, extract solid color block text pages from office document templates, and draw line combination patterns of different thicknesses and directions from design software.

[0040] Image samples are preprocessed, including cropping and scaling to a uniform resolution, such as 300dpi × 300dpi, and normalization is performed to adjust the pixel value range to the [0, 1] interval so that the data format is consistent during subsequent model training, thereby improving training efficiency and model stability.

[0041] Collect printed sample images with feathering lines and their corresponding original images without feathering lines. This can be done through actual printing experiments. Use direct-to-garment printers of different brands and models to print image samples on various printing media, such as plain paper, photo paper, and specialty paper. Observe and screen out printed samples with obvious feathering lines, and record the corresponding original image files to ensure that the original images and printed sample images are completely consistent in content, except that the printed sample images have feathering lines due to the printing process.

[0042] Simultaneously, the characteristic parameters of different printing media are recorded, including media absorptivity (determined by experimentally measuring the penetration depth and diffusion range of ink in the media, such as observing the size of the diffusion area formed by ink droplets on the media surface using an optical microscope), surface roughness (measured using a surface roughness measuring instrument to obtain the microscopic profile height deviation value of the media surface), and color reflectivity (measured using a spectrophotometer to measure the reflection intensity of light of different wavelengths on the media surface to obtain the color reflectivity curve). These parameters are associated and stored with the corresponding printed sample images and original images to form a complete dataset.

[0043] Feature extraction is performed on images from the image sample set and historical printing data. Computer vision algorithms, such as grayscale histogram statistics, are used to obtain the grayscale distribution characteristics of the images, analyze the distribution of the number of pixels with different grayscale values, and understand the brightness and contrast of the images. Color space conversion, such as conversion from RGB to HSV or LAB color space, and channel separation operations are used to obtain color channel features, extracting information such as brightness, hue, and saturation of the image in different color channels so that subsequent models can be adjusted for different color characteristics. Edge detection algorithms, such as the Canny algorithm, and texture analysis methods, such as grayscale co-occurrence matrix, are used to obtain line structure features, identify the position, direction, and thickness of edge lines in the image, as well as the periodicity and directionality of textures, providing a basis for subsequent special processing of edge and texture regions.

[0044] Convolutional Neural Networks (CNNs) are chosen as the basic architecture for building neural network models. CNNs have excellent image feature extraction capabilities and can automatically learn local spatial correlations in images. The model consists of two parts: an encoder and a decoder. The encoder is composed of multiple convolutional layers and pooling layers. The convolutional layers use convolutional kernels of different sizes, such as 3×3 and 5×5, to perform convolution operations on the input image, extracting low-level features such as edges and textures, and high-level semantic features such as object shapes and region distributions. The pooling layers downsample the output of the convolutional layers, reducing the spatial dimension of the feature map, reducing computation, and preserving important feature information. For example, in the first convolutional layer of the encoder, 32 3×3 convolutional kernels are used to convolve the input image, and the activation function is ReLU. Then, a 2×2 max pooling layer is used for downsampling. Subsequent convolutional layers gradually increase the number of convolutional kernels, such as 64 and 128, to extract more abstract features.

[0045] The decoder consists of multiple deconvolutional layers and upsampling layers. The deconvolutional layers are used to gradually restore the feature maps extracted by the encoder to the same spatial size as the input image. The upsampling layers further improve the resolution of the feature maps through interpolation and other methods. The last deconvolutional layer of the decoder outputs color adjustment parameters, including brightness adjustment values, contrast adjustment coefficients, and color gamut compression ratios, which are used to guide subsequent image color adjustment operations. For example, the last deconvolutional layer outputs a parameter map with the same size as the input image and 3 channels (corresponding to brightness, contrast, and color gamut compression parameters, respectively).

[0046] The system uses printed sample images with feathering patterns from historical printing data as input data and the corresponding original images without feathering patterns as target output data. These are input into a pre-built neural network model, and the backpropagation algorithm is used to optimize the model parameters. During training, the loss value between the predicted image and the target output image is calculated. The loss function is a combination of mean squared error loss (MSE) and structural similarity index loss (SSIM). MSE loss measures pixel-level error, enabling the model to learn how to adjust the color value of each pixel to approximate the target image. SSIM loss focuses more on the structural information of the image, including brightness, contrast, and texture structure, which helps the model maintain the overall structure and visual effect of the image while eliminating feathering patterns. For example, the loss function can be expressed as: Loss = α × MSE + (1 - α) × (1 - SSIM), where α is a weight coefficient between 0 and 1, which is adjusted according to actual training needs, such as taking α = 0.5.

[0047] The model parameters are iteratively updated using optimization algorithms such as stochastic gradient descent (SGD) or Adam. An appropriate learning rate is set, such as an initial learning rate of 0.001. During training, the learning rate is dynamically adjusted according to the changes in the loss value. For example, if the loss value decreases slowly after several consecutive iterations, the learning rate is halved. After multiple iterations, such as more than 10,000 iterations, the loss value of the model is gradually reduced, and finally the trained image processing model is obtained.

[0048] The image to be printed is obtained and input into the trained image processing model. The image processing model first divides the image into regions and identifies high-risk areas in the image that may produce feathering. For example, by analyzing the grayscale gradient changes in the image, regions with abrupt grayscale transitions, such as image edges and color block boundaries, are classified as high-risk areas. At the same time, the line structure feature recognition algorithm learned during the previous training process is used to identify line regions in the image, which are also prone to feathering.

[0049] For the different regions after segmentation, the color characteristics of each region are extracted, such as average brightness, color distribution histogram, and texture features, such as texture directionality and texture density. Based on the extracted features, the corresponding color adjustment parameters are matched from a preset color adjustment strategy library. The color adjustment strategy library is learned during model training based on image samples and historical printing data, and stores the mapping relationship between different image features and color adjustment parameters. For example, for an edge region with a sharp grayscale transition and strong texture directionality, the model matches color adjustment parameters from the strategy library with a brightness adjustment value of -0.1 (reduce brightness to enhance contrast), a contrast adjustment coefficient of 1.2 (increase contrast), and a color gamut compression ratio of 0.9 (appropriately compress the color gamut to avoid color overflow).

[0050] Based on the matched color adjustment parameters, the colors of the images to be printed in different regions are adjusted. For continuous tone regions, a gradient smoothing algorithm is used to adjust the transition effect of pixel values ​​within the region. The gradient smoothing algorithm calculates the gray level difference between adjacent pixels within the region and smooths pixels with large gray level differences to make the color transition more natural. For example, for a gray level transition region, if the gray level values ​​of adjacent pixels A and B are 100 and 150 respectively, the gradient smoothing algorithm will adjust their gray level values ​​according to a certain smoothing coefficient, such as 0.5, so that the gray level value of A increases to 125 and the gray level value of B decreases to 125, thereby reducing the abrupt change in gray level transition.

[0051] For edge line regions, a contour enhancement algorithm is used to optimize the continuity and clarity of the lines. The contour enhancement algorithm enhances the contrast of the line edges, making the lines more prominent. For example, the Sobel operator is used to perform edge detection on the edge line regions to obtain an edge intensity map. Then, the original image is weighted and adjusted according to the edge intensity map to enhance the pixel values ​​of the line edges, making the lines clearer. For solid color block regions, a uniformity calibration algorithm is used to eliminate the difference in brightness within the color block. The uniformity calibration algorithm calculates the average brightness and color value of the pixels in the color block region and corrects pixels that deviate from the average value, making the color within the color block more uniform. For example, for a red color block region, its average red value is calculated to be 200. If the red value of a certain pixel in the region is 180, the uniformity calibration algorithm will adjust its red value to be close to 200, thereby eliminating the difference in brightness within the color block.

[0052] The processed image to be printed is sent to the printing device, which is then controlled to perform the printing operation. On the printing device side, the processed image data is received through the printing driver, and the ink jetting action of the print head is controlled according to the image data. The ink is sprayed onto the printing medium according to the adjusted color value and position information, thereby eliminating direct jet feathering during the printing process and obtaining a high-quality printed image.

[0053] Collect the processed images output from the printing equipment and acquire the actual printed image using an image acquisition device, such as a high-resolution scanner or a professional camera. Ensure that the resolution and color reproduction capability of the image acquisition device are high enough to accurately reflect the details and color information of the printed image. For example, use a scanner with a resolution of at least 600 dpi to scan the printed image in RGB color mode to obtain a digital file of the actual printed image.

[0054] The actual printed image is compared with the preset target image. The target image can be the original image to be printed after ideal color adjustment, or a standard image designed according to actual printing requirements. During the comparison, an image similarity evaluation algorithm, such as the Structural Similarity Index (SSIM) algorithm, is used to calculate the similarity value between the two images. At the same time, the deviation value between the two images is also calculated. The deviation value can be obtained by averaging the absolute values ​​of the differences in color values ​​between corresponding pixels in the two images. For example, for the red channel of the two images, the deviation value can be expressed as: Deviation value (R) = (1 / N) ×Σ|AR(i,j) - BR(i,j)|, where N is the total number of pixels in the image, and AR(i,j) and BR(i,j) are the red channel pixel values ​​of the actual printed image and the target image at position (i,j), respectively.

[0055] The system determines whether the deviation value exceeds a preset deviation threshold. If the deviation value exceeds the preset deviation threshold, it indicates that the current image processing model has a large error when processing the image, and the model needs to be optimized. The actual printed image and the corresponding image to be printed are used as new training data to incrementally train the image processing model. During incremental training, the new data is input into the model along with the original training data, and the model parameters are readjusted so that the model can better adapt to new image features and printing conditions. For example, if it is found that the model has a large deviation when processing images with a certain special texture during incremental training, the parameters of the convolutional layers related to texture feature extraction in the model are adjusted to optimize the texture feature extraction and color adjustment strategies, thereby improving the model's performance when processing similar images. After multiple incremental trainings, the color adjustment strategy in the model is continuously updated so that the image processing model can be continuously optimized to better eliminate direct injection feathering.

[0056] Example 2

[0057] Please see Figure 2 An electronic device, comprising:

[0058] The data acquisition module is used to acquire image sample sets, historical printing data and images to be printed. The image sample sets include images with different grayscale transitions, solid color blocks and line combinations. The historical printing data includes printed sample images with feathering, corresponding original images without feathering, and characteristic parameters of different printing media.

[0059] The model training module is used to train an image processing model based on the image sample set and historical printing data. The image processing model is used to learn feathering elimination color adjustment strategies corresponding to different image features and different printing media characteristics.

[0060] The image processing module is used to input the image to be printed into the trained image processing model, control the image processing model to match the optimal color adjustment strategy according to the features of the image to be printed and the characteristic parameters of the printing medium, and process the image to be printed based on the strategy.

[0061] The print control module is used to send the processed image to be printed to the printing device and control the printing device to perform the printing operation, thereby eliminating the direct injection feathering path.

[0062] It also includes a model optimization module and an image acquisition module; the image acquisition module is used to collect the processed image output by the printing device to obtain the actual printing effect image; the model optimization module is used to compare the actual printing effect image with the preset target effect image, calculate the deviation value, and when the deviation value exceeds the preset deviation threshold, use the actual printing effect image and the corresponding image to be printed as new training data to incrementally train the image processing model and update the color adjustment strategy in the model.

[0063] The data acquisition module includes a high-speed network interface for quickly acquiring image sample sets, historical printing data, and images to be printed from a network server or local storage device. The network interface supports multiple network protocols, such as TCP / IP and FTP, and can perform stable data transmission with different data sources.

[0064] Equipped with a large-capacity solid-state drive (SSD) for local storage of acquired data, the SSD has fast read and write speeds, which can meet the high-efficiency requirements of data storage and retrieval during data processing. For example, a 1TB SSD can store a large number of image samples and print data, ensuring that the device will not be interrupted due to insufficient data storage during operation.

[0065] It has data preprocessing capabilities, enabling automatic cropping, scaling, and normalization of acquired image sample sets and historical print data. The preprocessing software employs multi-threading technology, allowing it to process multiple images simultaneously, thus improving data preparation efficiency. For example, for a batch of image samples downloaded from the network, the preprocessing software can complete cropping and normalization of all images to a uniform resolution within minutes, providing consistent data for subsequent model training.

[0066] It can communicate with printing devices to obtain information such as the model of the printing device and the type of printing media. Based on this information, it can filter out data similar to the current printing conditions from the historical printing data stored locally, and use it as reference data for model training or optimization. For example, when the device is connected to a specific model of inkjet printer and the printing media is photo paper, the data acquisition module will automatically filter out image data printed on that model of printer and photo paper from the historical printing data so that the model can better adapt to the current printing environment.

[0067] The model training module is equipped with a high-performance graphics processing unit (GPU), such as the NVIDIA Tesla V100. GPUs have powerful parallel computing capabilities, which can accelerate the training process of neural network models. When training complex models such as convolutional neural networks, GPUs can handle a large number of convolution operations and matrix multiplication operations simultaneously, reducing training time from several days to several hours or even less. For example, when training an image processing model that contains multiple convolutional and decoding layers, the NVIDIA Tesla V100 GPU can complete the model training in 24 hours, while using a regular CPU may take several weeks.

[0068] It includes a high-capacity memory module for storing a large amount of intermediate data generated during training, such as feature maps and gradient values. Sufficient memory capacity can avoid frequent data exchanges during training and improve training efficiency. For example, during training, the feature maps and gradient values ​​generated by each layer of the model need to be temporarily stored in memory for subsequent backpropagation calculations. 64GB of memory can meet the high memory requirements of the model during training.

[0069] Deep learning frameworks, such as TensorFlow or PyTorch, are used to build and train neural network models. These frameworks provide neural network layer components and optimization algorithm libraries, making it easy to quickly build and train complex neural network models. For example, in TensorFlow, the Keras API can be used to quickly define model components such as convolutional layers, pooling layers, and deconvolutional layers, and the model can be trained by calling built-in optimization algorithms.

[0070] It has model training monitoring capabilities, which can display the change curves of indicators such as loss value and accuracy in real time during the training process. The monitoring software displays these indicators through a graphical interface, allowing users to intuitively understand the progress of model training. For example, during the training process, users can see through the monitoring interface that the loss value gradually decreases from the initial high value and the accuracy gradually increases, thereby judging whether the model is learning effectively and whether the training parameters need to be adjusted.

[0071] It supports model saving and loading functions, which can save the trained image processing model to the local storage device as a file, and can also load the saved model from the local storage device, so that users can use the model on different devices or in different scenarios. For example, after a user completes the model training on one device, he / she can save the model to the hard drive, and then load the model on another device to perform image processing operations, so as to realize model sharing and reuse.

[0072] The image processing module shares GPU and memory resources with the model training module. It utilizes the computing power of the GPU to perform image processing operations. During image processing, a large number of pixel-level calculations and feature extraction operations are required. The GPU can process these computing tasks in parallel, improving image processing efficiency. For example, when adjusting the color of a high-resolution image, the GPU can simultaneously perform brightness and contrast adjustment calculations on multiple pixels in the image, shortening the image processing time.

[0073] It can receive the image to be printed and input it into the trained image processing model. It converts the image data from the storage format to the input format acceptable to the model and applies the color adjustment parameters output by the model to the image to be printed. For example, it converts the image to be printed from JPEG format to the tensor format required by the model input. Then, it applies the brightness adjustment value, contrast adjustment coefficient and other parameters output by the model to the image pixel by pixel to obtain the processed image.

[0074] It has image region segmentation and feature extraction functions. It can segment the image to be printed according to the model’s instructions, identify high-risk areas, and extract the color characteristics and texture features of each area. For example, it can use image segmentation algorithms to divide the image into continuous tone areas, edge line areas and solid color block areas, and then calculate the color mean, standard deviation, texture directionality and other features of each area, so as to provide a basis for the model to match the optimal color adjustment strategy.

[0075] It supports a variety of image processing algorithms, such as gradient smoothing, contour enhancement, and uniformity calibration. These algorithms are integrated in a modular form. The appropriate algorithm is selected to process the image based on the color adjustment strategy output by the model. For example, when the model indicates that continuous tone regions should be processed, the software calls the gradient smoothing algorithm module to smooth the pixel values ​​in the region, making the color transition more natural.

[0076] The print control module includes a Universal Serial Bus (USB) interface and a parallel interface for connecting to different types of printing devices. The USB interface supports high-speed data transmission, meeting the data transmission speed requirements of modern printing devices; the parallel interface is compatible with some older printing devices, ensuring device versatility. For example, connecting a high-speed inkjet printer via the USB interface allows for quick transmission of processed image data to the printer for printing.

[0077] It can communicate with printing devices, send processed image data to be printed, and control the printing devices to perform printing operations. The software generates corresponding printing instructions based on the model of the printing device and the type of printing media. The image data and printing instructions are sent to the printing device through the printing driver. For example, for a specific model of inkjet printer, the software will generate printing instructions adapted to the printer based on parameters such as the printer's printhead resolution and ink type, ensuring that the printing device can correctly receive and execute the printing task.

[0078] It has a printing status monitoring function, which can obtain printing status information of the printing device in real time, such as printing progress, ink remaining, printhead temperature, etc. The monitoring software receives status information through the communication interface with the printing device and displays it to the user on the device interface. For example, the user can see the printing progress bar gradually filling up through the device interface to understand the completion status of the printing task. At the same time, the user can also check the ink remaining and replace the ink cartridge in time to avoid printing interruption.

[0079] The model optimization module shares hardware resources with the data acquisition module and the model training module. It uses a high-speed network interface to obtain images of actual printed results and uses GPU and memory for model optimization calculations.

[0080] It has image acquisition and control functions, which can control image acquisition devices, such as scanners or cameras, to capture images of the actual printed effect. It can set the acquisition parameters of the image acquisition device according to the size and resolution of the printed image to ensure that the acquired image can accurately reflect the printing effect. For example, for an A4-sized printed image, the software will set the scanner's scanning resolution to 600dpi and the scanning mode to RGB color mode to obtain a high-quality image of the actual printed effect.

[0081] The software can compare the actual printed image with the preset target image and calculate the deviation value. It uses an image similarity evaluation algorithm and deviation calculation formula to automatically calculate the similarity and deviation value between the two images and display the results to the user. For example, the software will display the SSIM similarity value and deviation value between the actual printed image and the target image on the interface, so that the user can intuitively understand the difference between the printed effect and the target effect.

[0082] When the deviation value exceeds the preset deviation threshold, the actual printed image and the corresponding image to be printed are automatically used as new training data to incrementally train the image processing model. The software will automatically adjust the model training parameters, such as the learning rate and the number of training iterations, and optimize the model according to the characteristics of the new data. For example, if the new data contains images with special textures, the software will appropriately increase the learning rate of the convolutional layers related to texture feature extraction, so that the model can better learn the color adjustment strategies of these special textures, thereby improving the model's performance when processing similar images.

[0083] The image acquisition module includes a high-resolution scanner and a camera, such as an SLR camera or an industrial camera. The scanner is used to acquire images printed on flat media such as paper, providing high-resolution, high-color-fidelity image data. The camera can be used to acquire images printed on special shapes or three-dimensional media, with flexible shooting angle and lighting control functions. For example, for images printed on photo paper, a scanner can obtain clear and detailed image acquisition results; while for images printed on curved objects, a camera can be used to capture images from different angles to obtain complete image information.

[0084] It can control the acquisition parameters of the scanner and camera, such as scanning resolution, scanning mode, camera exposure time, and aperture size. The software provides a user interface, through which users can manually set acquisition parameters, or let the software automatically select appropriate parameters according to the preset acquisition task. For example, when the acquisition task is to acquire high-contrast text images, the software will automatically set the scanner's scanning mode to black and white binary mode to improve the clarity of the text; when shooting color patterns, the software will automatically adjust the camera's white balance and exposure parameters according to the color range of the pattern to ensure that the acquired image has accurate colors.

[0085] It has an automated image acquisition function, which can automatically complete the image acquisition process according to the preset acquisition task. For example, the user can set an acquisition task to require the device to automatically acquire a printed image at a certain time interval, or automatically start the acquisition program after detecting that the printing device has completed the printing task. The software will automatically control the scanner or camera to acquire images according to the set task and save the acquired images to the specified storage location for convenient subsequent model optimization operations.

[0086] Example 3

[0087] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described above for eliminating direct injection feathering by predicting print color.

[0088] The computer-readable storage medium is a storage device for storing a computer program that implements a method for eliminating direct-injection feathering by predicting print color. It can be an optical disc, such as a CD-ROM or DVD-ROM, a hard disk, such as a mechanical hard disk or a solid-state hard disk, a flash memory card, such as an SD card or a USB flash drive, or other computer storage media. The computer program stored on the storage medium contains instruction code that can be executed by a computer processor to implement the various steps of the method for eliminating direct-injection feathering by predicting print color.

[0089] The same or similar labels correspond to the same or similar parts;

[0090] The terms used to describe positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent.

[0091] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all implementation methods here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A method for eliminating direct injection feathering by predicting printing color, characterized in that, Includes the following steps: Acquire image sample sets and historical printing data. The image sample sets include images with different grayscale transitions, solid color blocks, and line combinations. The historical printing data includes printed sample images with feathering and corresponding original images without feathering. Based on the image sample set and historical printing data, an image processing model is trained. The image processing model is used to learn feathering elimination color adjustment strategies corresponding to different image features. Obtain the image to be printed, input the image to be printed into the trained image processing model, the image processing model matches the optimal color adjustment strategy according to the features of the image to be printed, and processes the image to be printed based on the strategy; The processed image to be printed is sent to the printing device, which is then controlled to perform the printing operation to eliminate direct injection feathering. The image processing model matches the optimal color adjustment strategy based on the features of the image to be printed, including: The image processing model divides the image to be printed into regions and identifies high-risk areas in the image that may produce feathering paths. For the different regions after division, the color characteristics and texture features of the regions are extracted respectively. Based on the extracted features, the corresponding color adjustment parameters are matched from the preset color adjustment strategy library. The color adjustment parameters include brightness adjustment value, contrast adjustment coefficient and color gamut compression ratio. The historical printing data also includes characteristic parameters of different printing media, including media absorptivity, surface roughness, and color reflectivity. During the training of the image processing model, the medium characteristic parameters are associated with the corresponding color adjustment strategy, so that the trained image processing model can automatically adapt to the corresponding color adjustment strategy according to the characteristic parameters of the medium to be printed. The processing of the image to be printed based on this strategy includes: For continuous tone regions in the image to be printed, a gradient smoothing algorithm is used to adjust the transition effect of pixel values ​​within the region; For edge line areas in the image to be printed, a contour enhancement algorithm is used to optimize the continuity and clarity of the lines; For solid color areas in the image to be printed, a uniformity calibration algorithm is used to eliminate the differences in brightness within the color blocks.

2. The method as described in claim 1, characterized in that, Based on the image sample set and historical printing data, an image processing model is trained, including: Feature extraction is performed on the images in the image sample set and historical printing data to obtain the grayscale distribution features, color channel features and line structure features of the images; Construct a neural network model, which includes an encoder and a decoder. The encoder is used to extract image features, and the decoder is used to output color adjustment parameters. The image sample with feathering from the historical printing data is used as input data, and the corresponding original image without feathering is used as the target output data. These are input into the neural network model, and the model parameters are optimized through the backpropagation algorithm so that the model learns pixel-level dot-dodge and burn-down processing strategies, resulting in a trained image processing model.

3. The method as described in claim 2, characterized in that, The neural network model is at least one of convolutional neural network, generative adversarial network or Transformer network, and the loss function of the neural network model is a combination function of mean squared error loss and structural similarity index loss.

4. The method as described in claim 1, characterized in that, It also includes image processing model optimization steps: Collect the processed images output by the printing equipment and acquire the actual printing effect image through the image acquisition device; The actual printed image is compared with the preset target image, and the deviation value between the two is calculated. Determine whether the deviation value exceeds a preset deviation threshold. If the deviation value exceeds the preset deviation threshold, use the actual printed image and the corresponding image to be printed as new training data to incrementally train the image processing model and update the color adjustment strategy in the model.

5. An electronic device for implementing the method according to any one of claims 1-4, characterized in that, include: The data acquisition module is used to acquire image sample sets, historical printing data and images to be printed. The image sample sets include images with different grayscale transitions, solid color blocks and line combinations. The historical printing data includes printed sample images with feathering, corresponding original images without feathering, and characteristic parameters of different printing media. The model training module is used to train an image processing model based on the image sample set and historical printing data. The image processing model is used to learn feathering elimination color adjustment strategies corresponding to different image features and different printing media characteristics. The image processing module is used to input the image to be printed into the trained image processing model, control the image processing model to match the optimal color adjustment strategy according to the features of the image to be printed and the characteristic parameters of the printing medium, and process the image to be printed based on the strategy. The print control module is used to send the processed image to be printed to the printing device and control the printing device to perform the printing operation, thereby eliminating the direct injection feathering path.

6. The electronic device as claimed in claim 5, characterized in that, It also includes a model optimization module and an image acquisition module; The image acquisition module is used to collect the processed images output by the printing device and obtain the actual printing effect image; The model optimization module is used to compare the actual printed image with the preset target image, calculate the deviation value, and when the deviation value exceeds the preset deviation threshold, use the actual printed image and the corresponding image to be printed as new training data to incrementally train the image processing model and update the color adjustment strategy in the model.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method for eliminating direct injection feathering by predicting print color as described in any one of claims 1-4.

Images