Method for constructing wood decoration texture generation system based on design characteristics

Abstract

The invention discloses a method for constructing a wood decoration texture generation system based on design features. The method comprises the following steps: acquiring images of wood patterns, establishing a database, and collecting visual preference data; performing ICC analysis on intra-group correlation coefficients, and screening out images; calculating gray GLCM feature values, and extracting visual features; pCA (principal component analysis) is carried out; a wood decoration texture data set conforming to the design features is constructed; training the data set by using the model, and learning design features of the wood texture image; generating a wood decoration texture image with specific design characteristics by using the trained model; according to the method, the generated wood decoration texture image is evaluated, the design features and the design generation method are integrated, the wood decoration texture generation system based on the design features is obtained, and the technical characteristics that the wood pattern image can be efficiently generated, effective fusion of the pattern features is achieved, development of a target product is promoted, and the diversified requirements of wood patterns are met are achieved.

Description

Technical Field

[0001] The present invention relates to a method for a wood decoration texture generation system, and more specifically, to a method for constructing a wood decoration texture generation system based on design features, and belongs to the field of artificial intelligence visual technology. Background Art

[0002] Wood decorative texture design plays a vital role in the wood and decoration industry. It not only enhances the aesthetics of wood products, but also directly affects the market competitiveness of products and consumers' purchasing decisions. Faced with the growing demand for personalization and customization, wood decorative texture design has become a key driving force for industry innovation and development.

[0003] The current design of wood decorative textures mainly relies on the personal experience and aesthetic intuition of designers, lacking systematic and scientific guidance, resulting in unstable design quality, low innovation efficiency, overly subjective design process, and waste of design resources. Existing design tools and methods also fail to meet the special needs of wood decorative texture design, lacking effective tools to capture and express the characteristics of wood texture. When processing large-scale wood texture images, there are still problems such as long calculation time, high requirements for input data quality, and complex processing procedures. In addition, existing image development tools are also difficult to meet the needs of wood texture design in terms of professionalism and uniqueness, and it is difficult to accurately interpret the natural characteristics of wood.

[0004] The above problems limit the development potential and innovation speed of wood decorative texture design. Therefore, it is urgent to solve these challenges through scientific research and technological innovation to achieve the stability of design quality and the improvement of innovation efficiency, and meet the market and consumer demand for high-quality, personalized wood decorative texture products. Summary of the invention

[0005] In order to solve the above-mentioned problems of the prior art, the present invention provides a method for constructing a wood decorative texture generation system based on design features, which has the technical characteristics of being able to efficiently generate wood pattern images, and realizing effective fusion of pattern features, promoting the development of target products, and meeting the diversified needs of wood patterns.

[0006] In order to achieve the above object, the present invention is implemented by the following technical solutions:

[0007] The present invention provides a method for constructing a wood decoration texture generation system based on design features, the method comprising the following steps:

[0008] Step 1): Collect images of wood patterns, establish a sample image database of wood patterns, conduct a subjective visual preference experiment, and collect users' visual preference data on wood texture;

[0009] Step 2): Perform intraclass correlation coefficient (ICC) analysis on the collected data to screen out images with features that are strongly correlated with visual preference;

[0010] Step 3): Calculate the eigenvalues ​​of the gray level co-occurrence matrix GLCM of the screened image and extract visual features, which include energy, entropy, correlation and contrast;

[0011] Step 4): PCA principal component analysis is performed on the eigenvalues ​​of the gray level co-occurrence matrix GLCM to reduce the dimension and extract the design features to illustrate how the design features affect the visual preference of wood texture;

[0012] Step 5): Construct a wood decoration texture dataset that meets the design characteristics to provide effective training materials for subsequent image generation algorithms;

[0013] Step 6): Use the CycleGAN model to train the classified data in the dataset and learn the design features of wood texture images through deep learning methods;

[0014] Step 7): Reuse the trained CycleGAN model to generate wood decoration texture images with specific design features;

[0015] Step 8): Evaluate the generated wood decoration texture images to ensure that they meet the design requirements and have the expected visual effects;

[0016] Step 9): Integrate design features and design generation methods to obtain a wood decorative texture generation system based on design features for designers to use and other technical features.

[0017] Preferably, the intraclass correlation coefficient ICC analysis formula for the collected data in step 2) is as follows:

[0018] ICC n =[(X 1 -X avg )(X 2 -X avg )…(X 31 -X avg )] / [(X 1 -X avg ) 2 (X 2 -X avg ) 2 …(X 31 -X avg ) 2 ]

[0019] Among them, ICCn is the intra-group correlation coefficient value corresponding to each picture, X1 to X31 are the visual preference selection scores of 31 wood textures respectively, and Xavg is the average of the sum of X1 to X31. When the ICC value is high, it means that the consistency of visual preference judgment between experimental observers is high, which helps to accurately screen out wood textures with consistent visual preference characteristics; when the ICC value is low, it means that there are large differences in visual preference judgments between observers.

[0020] Preferably, the step of calculating the eigenvalue of the gray level co-occurrence matrix GLCM in step 3) comprises: using a common image analysis tool MATLAB, inputting a corresponding GLCM eigenvalue calculation code, and inputting a wood texture image for calculation to obtain an output numerical result;

[0021] Among them, the eigenvalues ​​selected for calculating GLCM are: E-mean, E-variance, EN-mean, EN-variance, CON-mean, CON-variance, COR-mean, COR-variance, D-mean and D-variance, a total of 10 eigenvalues;

[0022] Where E stands for energy, EN stands for entropy, CON stands for contrast, COR stands for correlation, and D stands for inverse moment;

[0023] The grayscale surface of the wood texture image is taken as a surface in three-dimensional space, and the calculation formula is as follows:

[0024]

[0025]

[0026]

[0027]

[0028]

[0029] In which oxy is used as the coordinate plane of the image pixels, and the total number of pixels in the horizontal direction x is Nx, then the horizontal spatial domain of the pixels is Lx = {1, 2, ..., Nx}; the total number of pixels in the vertical direction is Ny, then the vertical spatial domain of the pixels is Ly = {1, 2, ..., Ny}; z is used as the grayscale coordinate axis of the pixels, and the highest level of grayscale quantization of each pixel on the z axis is Ng, then the grayscale quantization set of the pixels is G = {0, 1, 2, ..., Ng}; the image function f is represented as a Lx × Ly image pixel set, and each pixel of this pixel set has a value G in the Ng grayscale layers, that is, it is represented as f: Lx × Ly → G; the interval distance parameter of the pixel pair is d, and the direction angle parameter of the pixel pair is The number is θ, when θ is parallel to the x-axis, the value is zero, and the counterclockwise rotation is the positive direction. The probability P(i,j,d,θ) that a pair of pixels separated by a distance of d pixels in the θ direction in the texture image have grayscale layers i and j respectively is counted, and the grayscale co-occurrence matrix C(d,θ) is generated, which is recorded as C(d,θ)=[P(i,j,d,θ)]; by calculating through the input function, the specific values ​​of the wood texture image in the four directions are obtained, and a representative texture feature value is obtained using the values ​​of the four directions. After obtaining the calculation results of all images, the calculation results are fitted and analyzed with the visual preference selection scores of the corresponding images. After visualizing the data according to the analysis results, an intuitive distribution law diagram of GLCM feature quantities under different visual preference effects is obtained.

[0030] Preferably, in step 4), the traditional data analysis software SPSS is selected to perform principal component analysis PCA, by inputting the 10 GLCM feature data of the image into SPSS, and then using its PCA analysis function to select and obtain the main analysis results, the main analysis results include correlation matrix, total variance explanation, component matrix, KMO and Bartlett test.

[0031] Preferably, the construction of the wood decoration texture image dataset in step 5) includes: data collection, data screening and classification, data preprocessing, data enhancement, and data set division;

[0032] In the data collection phase: collect images of various wood decorative textures, including actual photographed wood samples, existing design sample libraries, and open source image libraries on the Internet;

[0033] In the data screening and classification stage: the collected images need to be screened and classified to remove low-quality, blurred or repeated images to ensure the accuracy of the data set; classification is carried out according to the wood species and texture characteristics so that the characteristics of different textures can be accurately captured in the subsequent training and generation process;

[0034] In the data preprocessing stage: the screened images need to be preprocessed, including size adjustment, format conversion, and normalization;

[0035] Data enhancement stage: Generate more texture images through rotation, scaling, flipping, and cropping operations to increase the diversity and generalization ability of the data set;

[0036] Dataset division stage: The preprocessed and enhanced dataset is divided into training set, validation set and test set. The training set is used to train the CycleGAN model, the validation set is used to adjust the model parameters and monitor the training process, and the test set is used to evaluate the generation effect of the model.

[0037] Preferably, the specific operation process of step 6) is to input the classified wood pattern image dataset into the generative adversarial network CycleGAN; the training environment is set as: python3.8.9, tensorflow2.2.0, tensorflow_examples, PyCharm Community Edition 2021, win11;

[0038] Hardware requirements include: processor 11th Gen Intel (R) Core (TM) i7-11800H, RAM 16G, graphics card RTX3060;

[0039] In the specific implementation, the two training data sets, which are wood pattern picture sets of two styles A and B, are divided into two folders respectively; style one is divided into: trainA, testA; style two is divided into: trainB, testB.

[0040] Preferably, in step 7), 15 wood textures of different types can be designed, and the wood texture scores of the original visual preference results between 1 and 3 partitions are made to reach the range of 3 to 5 points by designing the feature model and the design generation of CycleGAN. The test image is further input into CycleGAN training and the result is output.

[0041] Preferably, in step 8), the output image results are divided into each output image using the WDF model according to the visual preference quantification method, all images with results between 3 and 5 points are screened out, and the generated wood decoration texture image is evaluated again.

[0042] Preferably, step 9) integrates the wood decoration texture image generation method integrating design features into the system to obtain a wood decoration texture design application platform, wherein the generation design platform can be implemented in an iPad system interface, a computer system interface, and a mobile phone system interface.

[0043] Beneficial effects: To achieve the stabilization of the quality of wood decorative texture design and the improvement of innovation efficiency, to meet the market and consumer demand for high-quality, personalized wood decorative texture products; to improve the texture and overall visual effect of generated images, to provide technical support for promoting innovation and development in the field of wood decorative texture design. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a WDF model diagram of the wood decoration texture design feature model of the present invention.

[0045] Figure 2 It is one of the flow diagrams of the method of the present invention.

[0046] Figure 3 This is the second schematic diagram of the process flow of the method of the present invention.

[0047] Figure 4 It is a schematic diagram of the process flow of the method of the present invention.

[0048] Figure 5 30 representative wood pattern types are used as experimental materials in the embodiments of the present invention.

[0049] Figure 6 It is a grayscale image of the experimental material of the embodiment of the present invention.

[0050] Figure 7 This is a wood texture image classification diagram based on ICC values ​​of the present invention.

[0051] Figure 8 This is a pixel layer data display diagram of the gray level co-occurrence matrix GLCM calculation of the present invention.

[0052] Fig. 9 This is one of the data display diagrams of the relationship layer of the wood decoration texture design feature analysis according to the embodiment of the present invention.

[0053] Fig.10 This is the second data display diagram of the wood decoration texture design feature analysis relationship layer according to the embodiment of the present invention.

[0054] Fig.11 This is the third data display diagram of the relationship layer of the wood decoration texture design feature analysis according to the embodiment of the present invention.

[0055] Fig.12 It is a comparison diagram of the visual preference results of the input image and the output image of the present invention.

[0056] in: Figure 1 For wood decorative texture, we first need to establish the relationship between subjective visual perception and objective physical quantity representation, and propose the wood decorative texture image design feature model - WDF model.

[0057] Figure 7When the ICC value is ≥0.6, this type of image has a higher visual preference consistency and is more likely to have more significant design features. DETAILED DESCRIPTION

[0058] The following is in conjunction with the instruction manual Figure 1-12 , the present invention is further illustrated, but the present invention is not limited to the following embodiments.

[0059] The present invention proposes to construct a feature model hypothesis relationship, realize the quantitative characterization of the subjective visual effect of wood texture by using objective visual physical quantities, and analyze and clarify the key design features. It enriches the research on the theory of wood texture design and provides a new solution for the design and development of wood decorative texture. The image generation technology is integrated to replace the traditional manual design behavior to improve the design efficiency. Combined with the design feature theory, the design generation of wood decorative texture under the intervention of design features is realized, which effectively improves the design efficiency and design quality of wood decorative texture images.

[0060] Technical solution / principle of this application:

[0061] 1. Image Evaluation: Visual Preference Experiment

[0062] (Application layer: obtaining subjective visual preference data)

[0063] 2. Image classification: Intraclass correlation coefficient (ICC) analysis

[0064] (Application layer: preliminary screening of images with design features)

[0065] ICC n =[(X 1 -X avg )(X 2 -X avg )…(X 31 -X avg )] / [(X 1 -X avg ) 2 (X 2 -X avg ) 2 …(X 31 -X avg ) 2 ]

[0066] 3. Quantification of wood texture visual features: Gray-level co-occurrence matrix GLCM calculation (pixel layer: obtaining objective physical quantity data)

[0067]

[0068]

[0069] 4. Design feature screening: principal component analysis (PCA) (feature layer: obtain design features and their relationships)

[0070] Wood Decorative Texture Design Features:

[0071] Wood grain visual preference = 0.65 wood grain primitive information + 0.18 wood grain pattern distribution + 0.08 wood grain direction;

[0072] Wood Grain Primitive

[0073] 0.948EN-mean-0.922D-mean-0.864COR-mean+0.857CON-mean-0.851E-mean+0.834CON-variance;

[0074] Wood grain layout = 0.636E-variance + 0.602EN-variance + 0.532D-variance;

[0075] Wood grain order = 0.779COR-variance.

[0076] Wood grain primitives: reflect the basic unit or basic element characteristics of wood texture. The complexity, contrast intensity and uniformity of the primitives greatly affect the visual effect of the primitives.

[0077] Wood grain layout: reflects the regularity and uniformity of the distribution of basic texture units in the picture.

[0078] Wood grain order: reflects the unity and chaos of the composition of the primitives in the texture picture, as well as the isotropy or anisotropy, etc.

[0079] like Figure 1-11 The figure shows a specific embodiment of a method for constructing a wood decoration texture generation system based on design features. The embodiment is a method for constructing a wood decoration texture generation system based on design features. The method comprises the following steps:

[0080] Step 1): Collect images of wood patterns, establish a sample image database of wood patterns, conduct a subjective visual preference experiment, and collect users' visual preference data on wood textures; provide data support for screening wood texture images with significant visual preference characteristics through intraclass correlation coefficient ICC analysis;

[0081] A specific experiment: 3 wood texture images will appear in the middle of each experimental page, that is, the subjects will see 3 images for each question at the same time, and need to drag and sort them according to the visual preference of the image in a short time. After the sorting is completed, you can click the next question. There are 30 questions in total. The test images appear randomly. The number and probability of each image appearing will be as equal as possible. All images will appear equally in the first 10 questions. For the last 20 questions, the order of appearance of subsequent question images will be adjusted according to the subject's previous choices. The selection time for each question is controlled at about 5 seconds, and the subjects are required to sort as quickly as possible when observing, rather than choosing after thinking. Short breaks are allowed when not observing, and the experimental time for each subject does not exceed 5 minutes.

[0082] The experiment invited 31 subjects, corresponding to identities X1 to X31, including 16 males and 15 females, from different regions, aged between 22 and 45 years old. More than half of them have design experience, a few have some knowledge of wood decorative texture products, and most do not have professional knowledge related to wood decorative texture. The subjects were asked to observe the wood texture images multiple times and select and rank the visual preferences of each image. The questionnaire and images are presented by electronic devices; this application can also use multiple visual recognition scoring devices to replace people for scoring;

[0083] Step 2): Perform intra-group correlation coefficient (ICC) analysis on the collected data to screen out images with features that are strongly correlated with visual preference; provide representative samples for subsequent objective visual physical quantity calculation and design feature extraction;

[0084] The intraclass correlation coefficient ICC analysis formula for the collected data is as follows:

[0085] ICC n =[(X 1 -X avg )(X 2 -X avg )…(X 31 -X avg )] / [(X 1 -X avg ) 2 (X 2 -X avg ) 2 …(X 31 -X avg ) 2 ]

[0086] Among them, ICCn is the intraclass correlation coefficient value corresponding to each picture, X1 to X31 are the visual preference selection scores of 31 wood textures, and Xavg is the average of the sum of X1 to X31. When the ICC value is high, it means that the consistency of visual preference judgment between experimental observers is high, which helps to accurately screen out wood textures with consistent visual preference characteristics; when the ICC value is low, it means that there are large differences in visual preference judgments between observers;

[0087] Step 3): Calculate the eigenvalues ​​of the gray-level co-occurrence matrix GLCM of the screened image, extract visual features such as energy, entropy, correlation and contrast; achieve the purpose of objective visual physical quantity calculation of the wood texture image, and provide original physical quantity data for extracting design features through principal component analysis PCA dimensionality reduction;

[0088] The calculation steps of the eigenvalue of the gray-level co-occurrence matrix GLCM include: using the commonly used image analysis tool MATLAB, inputting the corresponding GLCM eigenvalue calculation code, and inputting the wood texture image for calculation to obtain the output numerical result;

[0089] Among them, the eigenvalues ​​selected for calculating GLCM are: E-mean, E-variance, EN-mean, EN-variance, CON-mean, CON-variance, COR-mean, COR-variance, D-mean and D-variance, a total of 10 eigenvalues;

[0090] Where E stands for energy, EN stands for entropy, CON stands for contrast, COR stands for correlation, and D stands for inverse moment;

[0091] The calculation needs to determine the position space of the image, that is, the direction and distance between pixels. Because different position space settings will generate different GLCM results, thus reflecting the texture characteristics of the image at different scales. The technical solution of this application takes values ​​from four different directions of 0°, 45°, 90°, and 135° for each calculated image, and takes the average of the eigenvalues ​​in the four directions to represent the corresponding eigenvalue results of the image. Mean is the mean of the calculation results in the four directions, and variance is the standard deviation of the calculation results in the four directions.

[0092] The grayscale surface of the wood texture image is taken as a surface in three-dimensional space, and the calculation formula is as follows:

[0093]

[0094]

[0095]

[0096]

[0097]

[0098] In which oxy is used as the coordinate plane of the image pixels, and the total number of pixels in the horizontal direction x is Nx, then the horizontal spatial domain of the pixels is Lx = {1, 2, ..., Nx}; the total number of pixels in the vertical direction is Ny, then the vertical spatial domain of the pixels is Ly = {1, 2, ..., Ny}; z is used as the grayscale coordinate axis of the pixels, and the highest level of grayscale quantization of each pixel on the z axis is Ng, then the grayscale quantization set of the pixels is G = {0, 1, 2, ..., Ng}; the image function f is represented as a Lx × Ly image pixel set, and each pixel of this pixel set has a value G in the Ng grayscale layers, that is, it is represented as f: Lx × Ly → G; the interval distance parameter of the pixel pair is d, and the direction angle parameter of the pixel pair is The number is θ, when θ is parallel to the x-axis, the value is zero, and the counterclockwise rotation is the positive direction. The probability P(i,j,d,θ) that a pair of pixels separated by a distance of d pixels in the θ direction in the texture image have grayscale layers i and j respectively is counted, and the grayscale co-occurrence matrix C(d,θ) is generated, which is recorded as C(d,θ)=[P(i,j,d,θ)]; by calculating through the input function, the specific values ​​of the wood texture image in the four directions are obtained, and a representative texture feature value is obtained using the values ​​of the four directions. After obtaining the calculation results of all images, the calculation results are fitted and analyzed with the visual preference selection scores of the corresponding images. After visualizing the data according to the analysis results, an intuitive distribution law diagram of GLCM feature quantities under different visual preference effects is obtained.

[0099] Step 4): PCA principal component analysis is performed on the eigenvalues ​​of the gray level co-occurrence matrix GLCM to reduce the dimension and extract the design features to illustrate that the design features affect the visual preference of wood texture; by performing principal component analysis PCA, the purpose of data dimensionality reduction and design feature extraction is achieved, providing key design feature basis for the development of WOODAI, a design platform for wood decoration texture image generation;

[0100] The traditional data analysis software SPSS was selected for principal component analysis (PCA). The 10 GLCM feature data of the image were input into SPSS, and then its PCA analysis function was used to obtain the main analysis results, including the correlation matrix, total variance explanation, component matrix, KMO and Bartlett test.

[0101] Constructing the design feature model theory: Based on the PCA analysis results, constructing the design feature model theory, and establishing the quantitative relationship between design features and visual preference of wood texture;

[0102] Step 5): Construct a wood decoration texture dataset that meets the design characteristics to provide effective training materials for subsequent image generation algorithms;

[0103] The construction of wood decoration texture image dataset includes: data collection, data screening and classification, data preprocessing, data enhancement and data set division;

[0104] In the data collection phase: collect images of various wood decorative textures, including actual photographed wood samples, existing design sample libraries, and open source image libraries on the Internet;

[0105] It includes the online website of The wood database https: / / www.wood-database.com / , Wood Pattern Atlas and Wood Pattern Atlas (2024-LY-045). During the collection process, special attention should be paid to the clarity and resolution of the image to ensure the accuracy of subsequent processing.

[0106] In the data screening and classification stage: the collected images need to be screened and classified to remove low-quality, blurred or repeated images to ensure the accuracy of the data set; classification is carried out according to the wood species and texture characteristics so that the characteristics of different textures can be accurately captured in the subsequent training and generation process;

[0107] In the data preprocessing stage: the filtered images need to be preprocessed, including resizing, format conversion, and normalization. These operations help to unify the standards of the data set, improve training efficiency and generation quality. At the same time, for some special texture features, enhancement processing may be required to highlight their characteristics.

[0108] Data enhancement stage: Generate more texture images through rotation, scaling, flipping, and cropping operations to increase the diversity and generalization ability of the data set;

[0109] Data augmentation significance Since the number of wood decoration texture samples is small, data augmentation technology can effectively expand the dataset. Through operations such as rotation, scaling, flipping, and cropping, more texture images can be generated to increase the diversity and generalization ability of the dataset. The dataset processing code is shown in Appendix A.

[0110] Dataset division stage: The preprocessed and enhanced dataset is divided into training set, validation set and test set. The training set is used to train the CycleGAN model, the validation set is used to adjust the model parameters and monitor the training process, and the test set is used to evaluate the generation effect of the model.

[0111] Step 6): Use the CycleGAN model to train the classified data in the dataset and learn the design features of wood texture images through deep learning methods;

[0112] The specific operation process is to input the classified wood pattern image dataset into the generative adversarial network CycleGAN; the training environment is set as: python3.8.9, tensorflow2.2.0, tensorflow_examples, PyCharmCommunity Edition 2021, win11;

[0113] Hardware requirements include: processor 11th Gen Intel (R) Core (TM) i7-11800H, RAM 16G, graphics card RTX3060;

[0114] In the specific implementation, the two training data sets, which are wood pattern picture sets of two styles A and B, are divided into two folders respectively; style one is divided into: trainA, testA; style two is divided into: trainB, testB.

[0115] There is no strict requirement for the number of files, but it is preferred that the training set is larger than the test set (train files are larger than test files), and the loss function is defined, and then the corresponding part of the code is modified to make the code run smoothly without errors. The code loops to train the adversarial generative model, and the selection of features changes with the number of loops, and the image quality also improves with the iteration of the number of loops.

[0116] Step 7): Reuse the trained CycleGAN model to generate wood decoration texture images with specific design features; the images generated by the design feature model training based on the Cycle Generative Adversarial Network (CycleGAN) have clearer textures, stronger layers, more harmonious and natural light and dark combinations, and significantly improved overall visual preference effects, while maintaining the features of the original images. The generated images maintain a high feature correlation with the original images. The algorithm can better integrate the style of the input image, retain the texture features, and has the ability to design and generate wood decoration texture images with original style characteristics;

[0117] Specifically, 15 different types of wood textures were designed. Through the design of feature models and CycleGAN design generation, the wood texture scores of the original visual preference results between 1 and 3 partitions were increased to 3 to 5 points. The test images were further input into CycleGAN training and the results were output. Cycle Generative Adversarial Network (CycleGAN): It can capture small sample features and generate texture design images. At the same time, it can iterate infinitely based on a feature to simulate the designer's behavior. This algorithm makes up for the shortcoming of the small number of wood decorative texture samples.

[0118] Visual preference experiment and ICC intraclass correlation coefficient analysis:

[0119] When the ICC value is ≥0.6, this type of image has a higher visual preference consistency and is more likely to have more significant design features.

[0120] In this experiment, there are 15 images with significant texture visual features.

[0121] Step 8): Evaluate the generated wood decoration texture images to ensure that they meet the design requirements and have the expected visual effects;

[0122] The output image results are divided by the WDF model and according to the visual preference quantification method, and all images with results between 3-5 points are screened out, and the generated wood decoration texture images are evaluated again. Specifically, the same method is used to generate the visual preference selection scores of 31 wood textures in the step: Specifically, 31 subjects are evaluated again on the generated wood decoration texture images, and the subjects are divided into two groups. The first group of subjects can see the first 7 of the 15 images before generation and the last 8 after generation of the wood decoration texture images, and the second group of subjects can see the first 7 of the 15 images after generation and the last 8 before generation of the wood decoration texture images. At the same time, 5 images that did not participate in this experiment were added to balance the visual preference of the entire group of images. As before, all subjects only scored the visual preference item. Finally, when studying the results, the two groups of data are observed comprehensively. The wood grain images after generation are used as the experimental group, and the wood grain images before generation are used as the control group. If the visual preference selection results of the generated images can be improved and reach the range of 3 to 5 points, it can be proved that CycleGAN can be successfully used with the WDF theoretical model to generate wood decoration texture design images that meet the visual preference standards.

[0123] Step 9): Integrate the design features and the design generation method to obtain a wood decoration texture generation system (wood decoration texture design application platform) based on the design features for use by designers.

[0124] The wood decoration texture image generation method that integrates design features is integrated into the system to obtain a wood decoration texture design application platform, thereby forming a complete set of design feature-driven wood decoration texture design processes and operating specifications. It is implemented as a concrete operation interface, allowing designers to easily design and generate textures. Among them, the generation design platform can be implemented in the iPad system interface, computer system interface, and mobile phone system interface. The platform information architecture includes: "Home", "Home", "Help" and "My". Designers register and log in to the platform system through the homepage, understand the overall functions, control elements and operation processes of the platform on the help page, and implement the design of wood decoration textures under design feature control and visual preference control on the homepage.

[0125] Finally, it should be noted that the present invention is not limited to the above embodiments, and there are many variations. All variations that can be directly derived or associated with the content disclosed by ordinary technicians in this field should be considered as the protection scope of the present invention.

Claims

1. A method for constructing a wood decoration texture generation system based on design features, characterized in that: include Step 1): Collect images of wood patterns, establish a sample image database of wood patterns, conduct a subjective visual preference experiment, and collect users' visual preference data on wood texture; Step 2): Perform intraclass correlation coefficient (ICC) analysis on the collected data to screen out images with features that are strongly correlated with visual preference; Step 3): Calculate the eigenvalues ​​of the gray level co-occurrence matrix GLCM of the screened image and extract visual features, which include energy, entropy, correlation and contrast; Step 4): PCA principal component analysis is performed on the eigenvalues ​​of the gray level co-occurrence matrix GLCM to reduce the dimension and extract the design features to illustrate how the design features affect the visual preference of wood texture; Step 5): Construct a wood decoration texture dataset that meets the design characteristics to provide effective training materials for subsequent image generation algorithms; Step 6): Use the CycleGAN model to train the classified data in the dataset and learn the design features of wood texture images through deep learning methods; Step 7): Reuse the trained CycleGAN model to generate wood decoration texture images with specific design features; Step 8): Evaluate the generated wood decoration texture images to ensure that they meet the design requirements and have the expected visual effects; Step 9): Integrate the design features and the design generation method to obtain a wood decoration texture generation system based on the design features for use by designers.

2. The method for constructing a wood decoration texture generation system based on design features according to claim 1, characterized in that: The intraclass correlation coefficient ICC analysis formula for the collected data in step 2) is as follows: ICC n [(X1-X avg )(X2-X avg )…(X 31 -X avg )] / [(X1-X avg ) 2 (X2–X avg ) 2 …(X 31 -X avg ) 2 ] Among them, ICCn is the intra-group correlation coefficient value corresponding to each picture, X1 to X31 are the visual preference selection scores of 31 wood textures respectively, and Xavg is the average of the sum of X1 to X31. When the ICC value is high, it means that the consistency of visual preference judgment between experimental observers is high, which helps to accurately screen out wood textures with consistent visual preference characteristics; when the ICC value is low, it means that there are large differences in visual preference judgments between observers.

3. A method for constructing a wood decoration texture generation system based on design features according to claim 1 or 2, characterized in that: The step of calculating the eigenvalue of the gray-level co-occurrence matrix GLCM in step 3) includes: using a common image analysis tool MATLAB, inputting a corresponding GLCM eigenvalue calculation code, and inputting a wood texture image for calculation to obtain an output numerical result; Among them, the eigenvalues ​​selected for calculating GLCM are: E-mean, E-variance, EN-mean, EN-variance, CON-mean, CON-variance, COR-mean, COR-variance, D-mean and D-variance, a total of 10 eigenvalues; Where E stands for energy, EN stands for entropy, CON stands for contrast, COR stands for correlation, and D stands for inverse moment; The grayscale surface of the wood texture image is taken as a surface in three-dimensional space, and the calculation formula is as follows: In which oxy is used as the coordinate plane of the image pixels, and the total number of pixels in the horizontal direction x is Nx, then the horizontal spatial domain of the pixels is Lx = {1, 2, ..., Nx}; the total number of pixels in the vertical direction is Ny, then the vertical spatial domain of the pixels is Ly = {1, 2, ..., Ny}; z is used as the grayscale coordinate axis of the pixels, and the highest level of grayscale quantization of each pixel on the z axis is Ng, then the grayscale quantization set of the pixels is G = {0, 1, 2, ..., Ng}; the image function f is represented as a Lx × Ly image pixel set, and each pixel of this pixel set has a value G in the Ng grayscale layers, that is, it is represented as f: Lx × Ly → G; the interval distance parameter of the pixel pair is d, and the direction angle parameter of the pixel pair is The number is θ, when θ is parallel to the x-axis, the value is zero, and the counterclockwise rotation is the positive direction. The probability P(i,j,d,θ) that a pair of pixels separated by a distance of d pixels in the θ direction in the texture image have grayscale layers i and j respectively is counted, and the grayscale co-occurrence matrix C(d,θ) is generated, which is recorded as C(d,θ)=[P(i,j,d,θ)]; by calculating through the input function, the specific values ​​of the wood texture image in the four directions are obtained, and a representative texture feature value is obtained using the values ​​of the four directions. After obtaining the calculation results of all images, the calculation results are fitted and analyzed with the visual preference selection scores of the corresponding images. After visualizing the data according to the analysis results, an intuitive distribution law diagram of GLCM feature quantities under different visual preference effects is obtained.

4. The method for constructing a wood decoration texture generation system based on design features according to claim 2, characterized in that: In step 4), the traditional data analysis software SPSS was selected to perform principal component analysis (PCA). The 10 GLCM feature data of the image were input into SPSS, and then the PCA analysis function was used to obtain the main analysis results, including the correlation matrix, total variance explanation, component matrix, KMO and Bartlett test.

5. The method for constructing a wood decoration texture generation system based on design features according to claim 4, characterized in that: In step 5), the construction of the wood decoration texture image dataset includes: data collection, data screening and classification, data preprocessing, data enhancement, and data set division; In the data collection phase: collect images of various wood decorative textures, including actual photographed wood samples, existing design sample libraries, and open source image libraries on the Internet; In the data screening and classification stage: the collected images need to be screened and classified, and the images with poor quality, blurry or duplicate images need to be removed to ensure the accuracy of the data set; Classification is performed based on wood species and texture characteristics so that the characteristics of different textures can be accurately captured in the subsequent training and generation process; In the data preprocessing stage: the screened images need to be preprocessed, including size adjustment, format conversion, and normalization; Data enhancement stage: Generate more texture images through rotation, scaling, flipping, and cropping operations to increase the diversity and generalization ability of the data set; Dataset division stage: The preprocessed and enhanced dataset is divided into training set, validation set and test set. The training set is used to train the CycleGAN model, the validation set is used to adjust the model parameters and monitor the training process, and the test set is used to evaluate the generation effect of the model.

6. The method for constructing a wood decoration texture generation system based on design features according to claim 5, characterized in that: Step 6) The specific operation process is to input the classified wood pattern image dataset into the generative adversarial network CycleGAN; the training environment is set as: python3.8.9, tensorflow2.2.0, tensorflow_examples, PyCharm Community Edition 2021, win11; Hardware requirements include: processor 11th Gen Intel (R) Core (TM) i7-11800H, RAM 16G, graphics card RTX3060; In the specific implementation, the two training data sets, which are wood pattern picture sets of two styles A and B, are divided into two folders respectively; style one is divided into: trainA, testA; style two is divided into: trainB, testB.

7. The method for constructing a wood decoration texture generation system based on design features according to claim 6, characterized in that: In step 7), 15 different types of wood textures can be designed. By designing the feature model and the design generation of CycleGAN, the wood texture score of the original visual preference result between 1 and 3 partitions is increased to 3 to 5 points. The test image is further input into CycleGAN training and the result is output.

8. The method for constructing a wood decoration texture generation system based on design features according to claim 2 or 7, characterized in that: Step 8) The output image results are divided into each image according to the visual preference quantification method using the WDF model, and all images with results between 3 and 5 points are screened out, and the generated wood decoration texture images are evaluated again.

9. The method for constructing a wood decoration texture generation system based on design features according to claim 8, characterized in that: Step 9) Integrate the wood decoration texture image generation method integrating design features into the system to obtain a wood decoration texture design application platform, wherein the generation design platform can be implemented in an iPad system interface, a computer system interface, and a mobile phone system interface.

Images