A multi-feature fusion traditional painting paper production process decision method

Abstract

This invention relates to the fields of intelligent manufacturing and digital image processing technology, specifically a multi-feature fusion method for decision-making in the production process of traditional calligraphy and painting paper. The method includes the following steps: constructing a traditional calligraphy and painting paper production database, where each data record in the database is associated with at least raw material formula data, process parameter data, and corresponding finished product image data; obtaining a reference image of the target traditional calligraphy and painting paper; extracting multi-dimensional visual features from the reference image to characterize its visual properties; and by constructing a database covering raw materials, processes, finished product images, and performance, transforming the tacit knowledge of craftsmen into storable and computable structured data, effectively solving the risk of loss of traditional skills due to reliance on oral transmission. This method, through image feature extraction and multimodal similarity matching, transforms subjective visual judgments into objective quantitative indicators, significantly improving the accuracy and repeatability of traditional colored paper restoration.

Description

Technical Field

[0001] This invention relates to the fields of intelligent manufacturing and digital image processing technology, specifically a decision-making method for the production process of traditional calligraphy and painting paper that integrates multiple features. Background Technology

[0002] Traditional calligraphy and painting papers, such as Xue Tao paper and hard yellow paper from the Tang Dynasty, and porcelain blue paper and burnished paper from the Song Dynasty, are cultural treasures representing the deep integration of Chinese papermaking techniques and natural dyeing methods. Made from handmade paper as the base material, they utilize natural dyes from plants and minerals, supplemented by special processes such as mordant dyeing, calendering, and gold sprinkling. They possess not only unique visual beauty and a warm texture but also carry profound historical and cultural value, and are widely used in ancient bookbinding, calligraphy and painting creation, classical writing, and religious rituals. However, the inheritance, restoration, and innovative production of this precious intangible cultural heritage face severe technological challenges in the modern era. The core dilemma lies in the fact that the production process decisions of traditional calligraphy and painting papers rely heavily on the individual experience and tacit knowledge of craftsmen, lacking the support of quantifiable, standardized, and reproducible modern technologies. This leads to numerous problems such as difficulties in the inheritance of skills, insufficient restoration accuracy, low production efficiency, and unstable product quality.

[0003] Currently, existing technologies related to traditional calligraphy and painting paper production process decisions can be mainly divided into three categories: artisan craftsmanship based on purely traditional experience, standardized production models based on modern chemical dyeing, and surface pattern generation technologies based on computer-aided design. Although these technologies solve some problems at certain levels, they all have inherent technical drawbacks closely related to the core problem that this invention aims to solve.

[0004] In conclusion, there is an urgent need in this field for an innovative technical solution that can make implicit craftsmanship experience explicit and structured, quantify vague process perception into calculable parameters, and build an intelligent decision-making pathway from visual requirements to process implementation, so as to completely solve the fundamental technical bottlenecks faced by traditional crafts in the process of inheritance and innovation. Summary of the Invention

[0005] The purpose of this invention is to provide a method for detecting and warning of abnormal electricity market prices based on deep learning, so as to overcome the shortcomings of the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multi-feature fusion method for decision-making in the production process of traditional calligraphy and painting paper, comprising the following steps: S1. Database construction steps: Construct a traditional calligraphy and painting paper production database. Each data record in the database is associated with at least raw material formula data, process parameter data, and corresponding finished product image data. S2. Target feature extraction step: Obtain a reference image of the target traditional calligraphy and painting paper, and extract multi-dimensional visual features from the reference image to characterize its visual properties; S3. Similarity matching step: Calculate the similarity between the target multi-dimensional visual features extracted in step S2 and the multi-dimensional visual features corresponding to each finished product image data stored in the database, and select a candidate sample set based on the calculation results. S4. Process Decision Step: Based on the raw material formulation data and process parameter data associated with the candidate sample set obtained in step S3, the decision model outputs a recommended combination of production process parameters.

[0007] As a preferred embodiment of this scheme, in step S1, the raw material formulation data includes dye type, dye pretreatment parameters, and material-to-water ratio; the process parameter data includes extraction temperature, extraction time, mordant type and concentration, mordating time, drying temperature and humidity; each data record in the database is also associated with and stored with finished product performance index data, including color difference value ΔE, folding endurance, and lightfastness rating.

[0008] As a preferred embodiment of this solution, in step S2, the multi-dimensional visual features include at least color features and texture features; wherein... The color features are extracted using either the color histogram method or the dominant color extraction method. The color histogram method involves converting the image to the HSV color space and calculating the histograms of the H, S, and V channels. The dominant color extraction method involves using the K-means clustering algorithm to cluster the RGB values ​​of the image pixels to extract representative dominant colors. The texture features are extracted using the gray-level co-occurrence matrix method, which includes calculating the gray-level co-occurrence matrix and extracting at least one statistic from it, including contrast, correlation, energy, and homogeneity.

[0009] As a preferred embodiment of this scheme, in step S3, the similarity calculation adopts the cosine similarity method; the fusion method of the multi-dimensional visual features is: after assigning preset weights to features of different dimensions, a weighted calculation is performed to obtain the comprehensive similarity; wherein, the weight assigned to the color feature is greater than the weight assigned to the texture feature.

[0010] As a preferred embodiment of this scheme, in step S4, the decision model is a rule base model; the rule base model has at least one preset decision rule, including: when the similarity of a single sample in the candidate sample set is higher than a first preset threshold, the combination of production process parameters associated with that single sample is directly output as a recommended scheme.

[0011] In a preferred embodiment of this scheme, in step S4, the decision model is an optimization algorithm model. The optimization algorithm model uses the process parameters associated with the candidate sample set as the initial solution, takes minimizing the difference between the finished product image features predicted based on the recommended process parameters and the target multi-dimensional visual features as the objective function, and performs iterative search through the optimization algorithm to output an optimized combination of production process parameters.

[0012] As a preferred embodiment of this scheme, the optimization algorithm is a genetic algorithm; the operation of the genetic algorithm includes: Encoding: Encoding production process parameters into chromosomes using real numbers; Initial population construction: The process parameters of the candidate sample set are used as part of the initial individuals, and combined with randomly generated individuals to form the initial population; Fitness assessment: The fitness function is used to evaluate the quality of an individual, and the fitness function is negatively correlated with the objective function; Iterative evolution: A new generation of population is generated through selection, crossover, and mutation operations until the termination condition is met.

[0013] In a preferred embodiment of this scheme, when the reference image contains gradient colors or specific texture regions, the target feature extraction step in step S2 further includes: Image region segmentation: A texture segmentation algorithm is used to separate specific texture regions from the background regions in an image; Partition feature extraction: The background color area is divided into multiple sub-regions, and the color features of each sub-region are extracted to form a feature sequence representing color gradation; Texture feature enhancement: Extract and enhance texture features for the specific texture region; In step S3, the similarity matching step adopts a multi-stage matching strategy, first performing preliminary screening based on background color features, and then performing final screening based on texture features.

[0014] As a preferred embodiment of this scheme, step S5, the feedback optimization step, is also included: adding the finished product image data of the calligraphy and painting paper actually produced according to the recommended production process parameter combination, the corresponding production process parameters, and the measured performance indicators as a new data record to the traditional calligraphy and painting paper production database.

[0015] The present invention also discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the steps of the above-mentioned multi-feature fusion traditional calligraphy and painting paper production process decision-making method.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: By constructing a database encompassing raw materials, processes, finished product images, and performance, the tacit knowledge of artisans is transformed into storable and computable structured data, effectively addressing the risk of loss of traditional skills due to their reliance on oral transmission. This method, through image feature extraction and multimodal similarity matching, transforms subjective visual judgments into objective quantitative indicators, significantly improving the accuracy and repeatability of traditional colored paper restoration.

[0017] This method can quickly match historical data with target images or generate new solutions through optimization algorithms, greatly reducing the time and resource costs required by traditional "trial and error" methods and significantly improving production efficiency. Its decision model has both rule-based judgment and intelligent optimization capabilities, which can not only meet the needs of rapid restoration with high confidence, but also handle parameter optimization in complex scenarios, combining high efficiency and accuracy.

[0018] The system possesses incremental learning capabilities, enabling it to evolve by continuously absorbing new production data, thus constantly enriching and optimizing its process knowledge base. This standardized and replicable decision-making method is not only applicable to high-precision fields such as cultural relic restoration, but also lays a solid technical foundation for the large-scale production, personalized customization, and cultural and creative product development of traditional calligraphy and painting paper, effectively promoting the living inheritance and modern transformation of intangible cultural heritage skills. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: Rapid restoration decision based on traditional colored paper using high similarity matching Please see Figure 1 As shown in the figure, this embodiment elaborates in detail the complete application process of the traditional calligraphy and painting paper production process decision-making method with multi-feature fusion in restoring the specific target of "Xue Tao paper" in the Tang Dynasty.

[0022] S101: Database Construction and Data Preprocessing By creating a rich and well-structured database of traditional calligraphy and painting paper production, we can provide data support for subsequent feature matching and intelligent decision-making.

[0023] Data Collection Scope and Standard Formulation: First, the scope of the database is defined. This database aims to collect typical traditional colored papers from the Han Dynasty to the Qing Dynasty, such as Xue Tao paper and hard yellow paper from the Tang Dynasty, and porcelain blue paper and embossed paper from the Song Dynasty. Data sources include: (1) Colored paper samples successfully restored by the laboratory based on ancient documents (such as "Qimin Yaoshu" and "Wenfang Sipu"); (2) Images and corresponding replica sample data obtained by high-precision digital scanning of ancient colored paper artifacts in cooperation with museums; (3) Data on fine colored papers made by experienced craftsmen according to traditional techniques. To ensure data quality, a unified data collection standard is established: all finished product images must be scanned under a standard light source box using a high-resolution scanner and embedded with color profiles. Performance indicators must be measured according to the prescribed standards using a calibrated colorimeter, folding endurance meter, and lightfastness meter.

[0024] Database structure design and data entry: The database is constructed using a relational database, and the core data tables include: Raw material formula table: Records the unique identifier, dye type, dye pretreatment parameters, and material-to-water ratio for each data entry.

[0025] Process parameter table: The sample_ID is linked to the raw material formula table to record the extraction temperature (Extraction_Temp, unit °C), extraction time (Extraction_Time, unit min), mordant type (Mordant_Type, such as alum, ferrous sulfate), mordant concentration (Mordant_Concentration, unit %), mordant time (Mordant_Time, unit min), drying temperature (Drying_Temp, unit °C), and drying humidity (Drying_Humidity, unit %).

[0026] Finished Product Data Table: Linked to the above two tables via Sample_ID, it stores the storage path (Image_Path), color difference value (Delta_E), folding endurance (Folding_Endurance, per cycle), and lightfastness grade (Lightfastness_Grade) of the finished product image.

[0027] Data preprocessing and enhancement: Perform necessary preprocessing before and after data entry.

[0028] (a) For image data, normalize the size (scale to 2000x2000 pixels) and crop the background to ensure that only the paper area is retained.

[0029] (b) For numerical data in process parameters, normalization is performed to scale them to the [0,1] range to eliminate the influence of dimensions and facilitate subsequent algorithm processing.

[0030] (c) Data augmentation techniques are employed to generate new data samples from existing image data through random rotation (±5°) and brightness / contrast fine-tuning (±10%), thereby expanding the dataset and improving the robustness of the model. Ultimately, the initial database constructed in this embodiment contains more than 500 valid data records, covering historical samples of various color systems such as red, yellow, blue, green, and purple.

[0031] S102: Target Feature Extraction and Vectorization Representation It transforms user-provided, unstructured target images into structured feature vectors that computers can understand and process, representing their visual essence.

[0032] Target Image Input and Preprocessing: The user provides a digital reference image of the target Tang Dynasty "Xue Tao Paper" to be restored (e.g., a high-resolution photograph from a museum collection). After receiving the image, the system first performs preprocessing: (1) Color Correction: Ensure that the color space of the image is consistent with the image in the database (sRGB). (2) Region Extraction: If the background of the target image is complex, an image segmentation algorithm (such as GrabCut or a semantic segmentation model based on U-Net) is used to accurately extract the colored paper area and eliminate background interference. (3) Image Normalization: The extracted colored paper area image is scaled to the same size as the image in the database (2000x2000 pixels).

[0033] Multi-dimensional visual feature extraction: Extract color and texture features in parallel from the preprocessed target image.

[0034] Color feature extraction (using color histogram method and dominant hue extraction method): Color histogram: The image is converted from the RGB color space to the HSV color space, which better matches human color perception (hue H, saturation S, lightness V). Histograms are calculated for the H, S, and V channels separately. To maintain discriminative power while controlling feature dimensionality, the H channel is divided into 36 bins (each bin is 10° apart), and the S and V channels are each divided into 16 bins. The histograms of the three channels are then concatenated into a one-dimensional vector, which serves as the color histogram feature vector F_color_hist, with dimensions 36 + 16 + 16 = 68.

[0035] Dominant color extraction: The K-means clustering algorithm is used to cluster the RGB values ​​of all pixels in the image. In this embodiment, the number of clusters K=5, thereby extracting the 5 most representative main colors in the image. The RGB values ​​of these 5 colors (a total of 15 values) are sorted in descending order of their pixel proportion in the image, and then concatenated into a one-dimensional vector, which serves as the dominant color feature vector F_color_dominant, with a dimension of 15.

[0036] Texture feature extraction (using the gray-level co-occurrence matrix method): First, the color image is converted to a grayscale image. Then, the Gray-Level Co-occurrence Matrix (GLCM) is calculated, which describes the probability that a pair of pixels have gray levels i and j under specific spatial relationships (distance and orientation). In this embodiment, four orientations are selected (0°, 45°, 90°, 135°), with a pixel distance d=1. For the GLCM generated in each orientation, four commonly used texture statistics are calculated: Contrast, Correlation, Energy, and Homogeneity. Finally, the mean and standard deviation of each statistic in the four orientations are calculated to obtain an 8-dimensional texture feature vector F_texture_glcm.

[0037] Feature vector fusion and normalization: The extracted feature vectors are concatenated to form the final comprehensive feature vector F_target used to represent the target image.

[0038] Right now F_target=Concat(F_color_hist,F_color_dominant,F_texture_glcm), Its total dimensions are 68 + 15 + 8 = 91. Finally, F_target is L2 normalized to make its modulus 1, in order to improve the stability and accuracy of subsequent similarity calculations.

[0039] S103: Multimodal Similarity Matching and Candidate Sample Selection The core of this step is to compare the feature vector of the target image with the feature vectors of a massive number of samples in the database, find the most visually similar historical samples, and form a candidate set.

[0040] Batch calculation of database sample features: During system initialization or database update, the feature vectors of all finished images in the database are calculated offline in batches and stored in the feature field of the database. The calculation method is exactly the same as S102, ensuring the consistency of the feature space.

[0041] Similarity measurement and weighted fusion: Cosine similarity is used as the measurement standard to calculate the similarity Sim_i between the target feature vector F_target and the feature vector F_db_i of the i-th sample in the database.

[0042] The formula for calculating cosine similarity is: Sim_i=(F_target·F_db_i) / (||F_target||*||F_db_i||).

[0043] Since the feature vectors have been L2 normalized, the calculation can be simplified to Sim_i = F_target • F_db_i. To more accurately reflect the differences in the contribution of different feature dimensions to the final visual effect, this embodiment introduces a weighting mechanism. According to expert experience, color features have the highest visual importance for the restoration of calligraphy and painting paper, followed by texture features. Therefore, a weight of 0.5 is assigned to the color histogram feature F_color_hist, a weight of 0.2 is assigned to the dominant color feature F_color_dominant, and a weight of 0.3 is assigned to the texture feature F_texture_glcm.

[0044] The formula for calculating the weighted combined similarity Sim_combined_i is: Sim_combined_i=0.5*Sim_color_hist_i+0.2*Sim_color_dominant_i+0.3*Sim_texture_glcm_i.

[0045] Threshold screening and candidate set generation: After calculating the comprehensive similarity between all samples and the target, they are sorted from highest to lowest similarity value. A similarity threshold θ = 0.85 is set. All samples with Sim_combined_i >= 0.85 are selected to form a candidate sample set. In this embodiment, three samples meet the condition, with similarities of 0.91 (sample A), 0.88 (sample B), and 0.86 (sample C). The Sample_ID, process parameters, and finished product performance corresponding to these three samples will be used for the next step of decision-making.

[0046] S104: Production Process Decisions and Parameter Output This step generates the final recommended combination of production process parameters based on the candidate sample set and through an intelligent decision-making model.

[0047] Decision model selection and execution: Since the similarity of the candidate samples in this embodiment is high (>0.85) but there are differences, and no single sample has a similarity far exceeding that of others (e.g., >0.95), an optimization algorithm model is used to make decisions in order to obtain a better solution that may be superior to the existing candidates.

[0048] Objective function definition: The objective function aims to minimize the difference between the image features of the predicted output paper and the target image features. It is defined as follows: Loss=1-Sim_combined(F_predicted,F_target) F_predicted is a feature vector of the finished product image predicted based on a set of process parameters (which can be obtained through a trained neural network model or a physical simulation model; for simplicity, this embodiment assumes that it can be approximated by interpolation of candidate sample features). Our goal is to find a set of process parameters P* that minimizes Loss(P*).

[0049] Optimization algorithm (genetic algorithm) settings: Encoding: Each production process parameter (such as material-to-water ratio, extraction temperature, mordant concentration, etc.) is encoded as a chromosome (individual). Each parameter is treated as a gene and encoded using real numbers.

[0050] Initial population: Using the process parameters of candidate samples A, B, and C as 3 high-quality initial individuals, 17 individuals are randomly generated (within a reasonable range of parameters) to form an initial population of size 20.

[0051] Fitness function: Fitness = 1 / (Loss + ε), where ε is a minimum value to prevent division by zero errors. The higher the fitness, the better the individual.

[0052] Genetic operations: The crossover probability is set to 0.8, using simulated binary crossover (SBX); the mutation probability is set to 0.1, using polynomial mutation. The selection mechanism is roulette wheel selection.

[0053] Iterative optimization: Run the genetic algorithm for 50 generations. In each generation, for each individual (i.e., a set of process parameters), estimate its corresponding predicted feature vector F_predicted by interpolating from its nearest neighbors in the database, and then calculate its fitness. After 50 generations of evolution, the population converges, and the process parameters represented by the individual with the highest fitness are the recommended solutions.

[0054] Decision Output: The optimization algorithm in this embodiment ultimately recommends the following set of process parameters: using sappanwood as dye (particle size 2-5mm), a material-to-water ratio of 1:1.8, an extraction temperature of 98℃, an extraction time of 28 minutes, using a 2.5% alum solution as mordant, a mordating time of 10 minutes, and finally air-drying at 26℃ and 70% humidity. The system outputs this parameter combination to the user in a structured form (such as JSON or XML format) or directly to the production control system.

[0055] To further explain, the current production process decisions for traditional calligraphy and painting paper heavily rely on the personal experience and intuitive judgment of craftsmen. This tacit knowledge, which is "more felt than explained," is difficult to quantify, standardize, and pass on, resulting in high restoration costs, long cycles, and unstable results. This invention constructs a digital database covering the entire chain of "raw materials-process-image-performance," transforming scattered, unstructured experiential knowledge into systematic, computable structured data. By introducing image processing technology, multi-dimensional visual features such as color histograms, dominant hues, and GLCM textures are extracted from target images, transforming the subjective and vague "likeness" problem into an objective and precise feature vector similarity calculation problem. Finally, through a similarity-based intelligent decision-making model (such as a rule base or optimization algorithm), quantified and repeatable optimal process parameters are output, thereby transforming the traditional process from "experience-driven" to "data-driven" and "model-driven." This method not only significantly improves the accuracy and efficiency of traditional colored paper restoration, providing reliable technical support for cultural relic restoration and ancient book reproduction, but more importantly, it opens up a brand-new technical path for the digital archiving, scientific inheritance and innovative development of traditional intangible cultural heritage skills.

[0056] Example 2: Fast Decision-Making and Incremental Learning Mechanism Based on Rule Base Based on Example 1, this example focuses on the fast decision rule adopted when the confidence of the matching result is extremely high, and introduces an incremental learning mechanism to achieve the self-optimization of the system.

[0057] S201: Rule base decision-making in high-confidence matching scenarios Suppose that in the similarity matching in step S103, there exists a sample D in the database, whose comprehensive similarity with the target "Xue Tao Jian" image reaches 0.96, which is much higher than other candidate samples (such as sample E, whose similarity is 0.87) and the preset high confidence threshold (θ_high=0.93).

[0058] Rule Trigger: At this point, the system will not start the computationally complex genetic algorithm, but will instead trigger a pre-set decision rule in the rule base: "If there is a single sample in the candidate sample set whose similarity is higher than the high confidence threshold θ_high, then the complete production process parameters of that sample will be directly adopted as the final recommended solution." Decision Output: The system directly outputs all process parameters corresponding to sample D: sappanwood, material-to-water ratio 1:1.5, extraction at 100℃ for 30 min, mordant treatment with 2% alum solution for 8 min, and drying at 26℃ / 70% humidity. This process eliminates the iterative calculations of the optimization algorithm, resulting in extremely fast decision-making speed. It is suitable for scenarios with high timeliness requirements and ideal matching accuracy.

[0059] S202: Incremental Learning and Dynamic Database Updates In order to enable the system to continuously evolve with the accumulation of production practice, this embodiment introduces an incremental learning mechanism.

[0060] New Data Entry: Regardless of the decision model used by the system (the optimization algorithm in Example 1 or the rule base in this example), after the user completes actual production according to the process parameters recommended by the system, a new colored paper sample X is generated. The user needs to enter the data of this new sample into the system.

[0061] Data quality verification: Before data entry, the system automatically verifies the data of new samples. For example, it checks whether the format and resolution of the finished image meet the standards, and whether performance indicators (such as ΔE) are within a reasonable range. If the data is abnormal, the user will be prompted to remeasure or confirm.

[0062] Feature Calculation and Database Entry: After verification, the system automatically calculates the feature vector of the finished product image of the new sample (method is the same as S102), and then adds the complete record (raw material formula, process parameters, finished product image path, feature vector, performance index) as a new data entry to the traditional calligraphy and painting paper production database.

[0063] Model optimization (optional): If the system uses some data-based machine learning models (e.g., regression models for predicting finished product characteristics under given process parameters), these models can be retrained periodically with an updated and larger database to improve their prediction accuracy, thereby indirectly improving the quality of future decisions.

[0064] Furthermore, the decision-making module of this invention possesses high flexibility and adaptability. It does not rigidly employ a single algorithm, but intelligently selects the most suitable decision-making strategy based on the confidence level of the matching results. In high-confidence scenarios, a rule base is used for rapid response, greatly improving decision-making efficiency; when matching results are relatively close or scattered, an optimization algorithm is initiated for deep search to explore the optimal solution that may surpass historical experience. This hybrid decision-making mechanism cleverly balances the two key indicators of "efficiency" and "accuracy." In addition, by introducing an incremental learning mechanism, the system is no longer a static, closed knowledge base, but an "intelligent agent" capable of learning and continuously evolving from each production practice. Each new successful production case enriches the system's knowledge reserves, continuously expanding the database, making feature matching more accurate, and the decision model increasingly intelligent. This not only effectively solves the dilemma of traditional skills being lost due to the loss of artisans, but also enables traditional craft knowledge to be continuously optimized and upgraded through "version iteration," like modern software, injecting strong vitality into the living transmission of intangible cultural heritage.

[0065] Example 3: Complex Process Decisions for Gradient Colors and Special Textures This embodiment will demonstrate the ability of the method of the present invention to handle more challenging restoration targets, such as traditional colored paper that requires the production of gradient color effects or specific surface textures (such as burnishing).

[0066] S301: Decomposition and Representation of Complex Target Features The user's target is an image of a Song Dynasty "embossed paper," characterized by a pale yellow background, raised and recessed printed textures on the surface, and a subtle color gradient from dark yellow to light yellow from the edge to the center.

[0067] Image region segmentation and feature extraction: The system first needs to recognize and handle this complexity.

[0068] Texture region segmentation: Using texture segmentation algorithms (such as segmentation based on Gabor filters or local binary mode LBP), the "brushed" regions with obvious bumps and textures in the image are separated from the smooth "base" regions.

[0069] Gradient color analysis: For the "base color" region, instead of extracting global color features, it is divided into multiple concentric rings or grid regions. Color features (such as the average dominant hue of the region) are extracted independently for each sub-region, thus discretizing the continuous color gradient into a color feature sequence F_gradient=[C1,C2,C3,...], where Ci represents the representative color of the i-th region.

[0070] Texture feature enhancement: For the "etched" area, in addition to using GLCM, more refined texture features, such as Local Binary Pattern (LBP) features, are extracted to better capture its microscopic concave and convex structure, forming a texture feature vector F_texture_lbp.

[0071] Composite feature vector construction: Construct a composite feature vector for this complex target, which includes background color gradient sequence information, burnishing area texture information, etc.

[0072] S302: Process Decision Based on Multi-Stage Matching For such complex targets, a single global matching approach may not be effective. This embodiment employs a multi-stage matching strategy.

[0073] Background color matching stage: The system first ignores textures and focuses on matching background colors and gradient effects. A preliminary search is performed in the database to select a set of candidate samples (Candidate_color) that are relatively similar in color and gradient effect.

[0074] Texture matching stage: Based on Candidate_color, the similarity of the embossed texture is further compared. The final candidate sample set Candidate_final is selected based on the similarity in both color and texture.

[0075] Process reasoning and combination: The decision model needs to handle two process steps: dyeing and burnishing.

[0076] Staining process decision: Based on the staining process parameters of the samples in Candidate_final, an optimization algorithm similar to that in Example 1 is used to recommend a staining scheme to achieve the target gradient background color (which may involve multiple local brushings of dye solutions of different concentrations).

[0077] Burnishing process decision: Burnishing process parameters (such as the pattern ID of the carved woodblock and pressure of 0.3-0.5MPa) are extracted from the database. The system can recommend the woodblock pattern and corresponding pressure parameters that are most similar to the texture of the target image.

[0078] Furthermore, the method of this invention is not only applicable to the restoration of a single uniform color, but its powerful scalability also enables it to handle more complex artistic effects on traditional calligraphy and painting papers. By decomposing complex targets into multiple visual elements such as color, texture, and gradation, and employing a phased, multi-level matching and decision-making strategy, the system can analyze and reproduce the superb composite skills of ancient craftsmen. For special surface treatment processes such as burnishing and gold sprinkling, this invention manages and matches them as independent process modules, enabling the system to handle complex processes combining "dyeing" and "post-processing." This demonstrates the profound understanding and powerful expressive ability of the method of this invention towards traditional crafts, covering the production decision-making needs of various traditional calligraphy and painting papers from simple to complex, showcasing its enormous application potential in the development of high-end cultural and creative products and the customization of advanced calligraphy and painting creation materials.

[0079] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-feature fusion method for decision-making in the production process of traditional calligraphy and painting paper, characterized in that, Includes the following steps: S1. Database construction steps: Construct a traditional calligraphy and painting paper production database. Each data record in the database is associated with at least raw material formula data, process parameter data, and corresponding finished product image data. S2. Target feature extraction step: Obtain a reference image of the target traditional calligraphy and painting paper, and extract multi-dimensional visual features from the reference image to characterize its visual properties; S3. Similarity matching step: Calculate the similarity between the target multi-dimensional visual features extracted in step S2 and the multi-dimensional visual features corresponding to each finished product image data stored in the database, and select a candidate sample set based on the calculation results. S4. Process Decision Step: Based on the raw material formulation data and process parameter data associated with the candidate sample set obtained in step S3, the decision model outputs a recommended combination of production process parameters.

2. The method for decision-making in the production process of traditional calligraphy and painting paper according to claim 1, characterized in that, In step S1, the raw material formulation data includes dye type, dye pretreatment parameters and material-to-water ratio; the process parameter data includes extraction temperature, extraction time, mordant type and concentration, mordating time, drying temperature and humidity; each data record in the database is also associated with and stored finished product performance index data, including color difference value ΔE, folding endurance and lightfastness rating.

3. The method for decision-making in the production process of traditional calligraphy and painting paper based on multi-feature fusion according to claim 1, characterized in that, In step S2, the multi-dimensional visual features include at least color features and texture features; wherein, The color features are extracted using either the color histogram method or the dominant color extraction method. The color histogram method involves converting the image to the HSV color space and calculating the histograms of the H, S, and V channels. The dominant color extraction method involves using the K-means clustering algorithm to cluster the RGB values ​​of the image pixels to extract representative dominant colors. The texture features are extracted using the gray-level co-occurrence matrix method, which includes calculating the gray-level co-occurrence matrix and extracting at least one statistic from it, including contrast, correlation, energy, and homogeneity.

4. The method for decision-making in the production process of traditional calligraphy and painting paper based on multi-feature fusion according to claim 3, characterized in that, In step S3, the similarity calculation adopts the cosine similarity method; the fusion method of the multi-dimensional visual features is: after assigning preset weights to features of different dimensions, a weighted calculation is performed to obtain the comprehensive similarity; wherein, the weight assigned to color features is greater than the weight assigned to texture features.

5. The method for decision-making in the production process of traditional calligraphy and painting paper according to claim 1, characterized in that, In step S4, the decision model is a rule base model; the rule base model has at least one preset decision rule, including: when the similarity of a single sample in the candidate sample set is higher than a first preset threshold, the combination of production process parameters associated with that single sample is directly output as a recommended solution.

6. The method for decision-making in the production process of traditional calligraphy and painting paper according to claim 1, characterized in that, In step S4, the decision model is an optimization algorithm model. The optimization algorithm model uses the process parameters associated with the candidate sample set as the initial solution, and takes minimizing the difference between the finished product image features predicted based on the recommended process parameters and the target multi-dimensional visual features as the objective function. It performs iterative search through the optimization algorithm to output an optimized combination of production process parameters.

7. The method for decision-making in the production process of traditional calligraphy and painting paper based on multi-feature fusion according to claim 6, characterized in that, The optimization algorithm is a genetic algorithm; the operation of the genetic algorithm includes: Encoding: Encoding production process parameters into chromosomes using real numbers; Initial population construction: The process parameters of the candidate sample set are used as part of the initial individuals, and combined with randomly generated individuals to form the initial population; Fitness assessment: The fitness function is used to evaluate the quality of an individual, and the fitness function is negatively correlated with the objective function; Iterative evolution: A new generation of population is generated through selection, crossover, and mutation operations until the termination condition is met.

8. The method for decision-making in the production process of traditional calligraphy and painting paper according to claim 1, characterized in that, In step S2, when the reference image contains gradient colors or specific texture regions, the target feature extraction step further includes: Image region segmentation: A texture segmentation algorithm is used to separate specific texture regions from the background regions in an image; Partition feature extraction: The background color area is divided into multiple sub-regions, and the color features of each sub-region are extracted to form a feature sequence representing color gradation; Texture feature enhancement: Extract and enhance texture features for the specific texture region; In step S3, the similarity matching step adopts a multi-stage matching strategy, first performing preliminary screening based on background color features, and then performing final screening based on texture features.

9. The method for decision-making in the production process of traditional calligraphy and painting paper according to claim 1, characterized in that, It also includes step S5, a feedback optimization step: adding the finished product image data of the calligraphy and painting paper actually produced according to the recommended production process parameter combination, the corresponding production process parameters, and the measured performance indicators as a new data record to the traditional calligraphy and painting paper production database.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the traditional calligraphy and painting paper production process decision-making method with multi-feature fusion as described in any one of claims 1 to 9.

Images