Traditional pattern image style migration system and method based on multi-modal feature decoupling and dynamic propagation

Abstract

The invention provides a traditional pattern image style migration system and method based on multi-modal feature decoupling and dynamic propagation, and relates to the technical field of computer vision and digital media, and the method comprises the steps: collecting color features, texture features and vectorization composition features of a Tujia brocade image, and selecting a standard image; based on a low-rank adaptive technology, training a low-rank adaptive model by using the characteristics acquired by the Tujia brocade image; splitting an image to be migrated into migrated images according to frames, selecting key frame images, and obtaining key frame migrated images through the trained low-rank adaptive model; establishing a Tujia brocade image feature difference evaluation system, and determining a Tujia brocade image style range based on the system; and calculating the style offset condition of the key frame migration image and the standard Tujia brocade image, selecting a final fusion image from the key frame migration image according to the style range of the Tujia brocade image, and processing the final fusion image through a Poisson fusion technology to obtain a style migration image.

Description

Technical Field

[0001] The present invention relates to the field of computer vision and digital media technology, and in particular to a traditional pattern image style migration system and method based on multimodal feature decoupling and dynamic propagation. Background Art

[0002] The digital protection and inheritance of traditional patterns face the challenge of insufficient adaptability of style transfer technology. Existing image style transfer methods are mostly based on general data set training, which makes it difficult to capture the multimodal characteristics (color, texture, composition) unique to intangible cultural heritage (such as Tujia brocade), resulting in cultural semantic distortion and style deviation in the transfer results. Especially for dynamic images, existing technologies lack a differentiated processing mechanism for key frames and cannot balance style consistency and computational efficiency. In addition, traditional methods have a vague definition of the scope of pattern style and often rely on manual experience, making it difficult to quantitatively evaluate the degree of fit between the transfer effect and the native culture.

[0003] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0004] The purpose of the present invention is to provide a traditional pattern image style transfer system and system based on multimodal feature decoupling and dynamic propagation to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation, specifically including:

[0007] The data acquisition module is used to obtain Tujia brocade images from the local intangible cultural heritage public database and select a Tujia brocade image as the standard Tujia brocade image. The K-means algorithm is used to extract the color features of the Tujia brocade image. The gray-level co-occurrence matrix is used to extract the texture features of the Tujia brocade image. The Canny hard edge detection algorithm is used to extract the vectorized composition features of the Tujia brocade image.

[0008] The model analysis module is used to train the parameters of the diffusion model based on low-rank adaptive technology, using the color features, texture features, and vectorized composition features of each Tujia brocade image to generate a low-rank adaptive model adapted to the traditional pattern. The image to be migrated is split into multiple migration images according to the frame, and the key frame images are selected by the Frobenius norm of adjacent frames. The key frame images are input into the trained low-rank adaptive model to obtain the key frame migration images.

[0009] A weight analysis module is used to establish a Tujia brocade image feature difference evaluation system and determine the style range of Tujia brocade images based on the Tujia brocade image feature difference evaluation system;

[0010] The style deviation module is used to calculate the style deviation between the key frame migration image and the standard Tujia brocade image, select the final fusion image from the key frame migration image according to the style range of the Tujia brocade image, and process the final fusion image using the Poisson fusion technology to obtain the style migration image.

[0011] Furthermore, the color features include the color difference, main color type and main color type ratio of the Tujia brocade image; the texture features of the Tujia brocade image include the contrast, correlation, energy and homogeneity of the grayscale co-occurrence matrix of the Tujia brocade image at 0°, 45°, 90° and 135°; the vectorized composition features of the Tujia brocade image include the edge distribution map of the Tujia brocade image.

[0012] Furthermore, the specific logic for calculating the Frobenius norm of adjacent frames is as follows: construct an image migration matrix based on the image to be migrated, each pixel in the image corresponds to an element of the migration matrix, and the pixel value is the element value of the migration matrix. The Frobenius norm of adjacent frames is calculated. If the Frobenius norm is greater than a preset threshold, the migration image with the later time label in the two frames is selected as the key frame image. The specific formula for calculating the Frobenius norm is:

[0013] ΔF e =||E e+1 -E e || F

[0014] Where ΔF e is the Frobenius norm of adjacent frames, E e+1 is the image migration matrix of the next frame in the adjacent frames, E e is the image migration matrix of the previous frame in the adjacent frame.

[0015] Furthermore, the Tujia brocade image feature difference evaluation system specifically includes color cultural differences, texture structure differences and composition semantic differences;

[0016] The specific formula for calculating color cultural differences is:

[0017]

[0018] Among them, C For color cultural differences, u i is the three-dimensional vector of the i-th main color, u iOis the standard three-dimensional vector of the i-th main color, ||||2 is the norm operator, and I is the number of main color types;

[0019] The specific formula for calculating texture structure differences is:

[0020]

[0021] Among them, t is the texture structure difference, GLCM j is the feature vector of the jth texture feature, GLCM jO is the standard feature vector of the jth texture feature, the feature vector of the texture feature is a 4-dimensional vector, expressed as GLCM j =(glcm j0 ,glcm j45 ,glcm j90 ,glcm 135 ), where glcm j0 Represents the eigenvalue of the j-th texture feature of the gray-level co-occurrence matrix in the 0° direction, glcm j45 is the eigenvalue of the jth texture feature of the gray-level co-occurrence matrix at 45°, glcm j90 is the eigenvalue of the j-th texture feature of the gray-level co-occurrence matrix in the 90° direction, glcm 135 is the eigenvalue of the jth texture feature of the gray-level co-occurrence matrix in the 135° direction, SSIM (GLCM j -GLCM jO ) is GLCM j With GLCM jO The similarity between

[0022] The specific formula for calculating the semantic difference of composition is:

[0023] ζ l =Iou(M k ,M kO )

[0024] Among them, lk For the semantic difference of composition, M k is the binary matrix of the marginal distribution graph, M kO is a binary matrix of the standard edge distribution graph, which is obtained by analyzing the edge distribution graph. Specifically, the edge part of the edge distribution graph is marked as 1, and the other parts are marked as 0. k ,M kO ) is M k and M kO The intersection and union ratio of

[0025] The standard three-dimensional vector of the main color, the standard feature vector of the texture feature and the binary matrix of the standard edge distribution map are obtained by analyzing the selected standard Tujia brocade image.

[0026] Furthermore, the color culture differences, composition semantic differences, and texture structure differences between all Tujia brocade images are calculated, and the average color culture difference, average composition semantic difference, and average texture structure difference are calculated. The style range of Tujia brocade images is determined based on the average color culture difference, average composition semantic difference, and average texture structure difference. The style range of Tujia brocade images is specifically as follows:

[0027]

[0028] in, To average color cultural differences, is the average texture structure difference, is the average composition semantic difference.

[0029] Furthermore, the specific logic for selecting the final fusion image is as follows: the color culture differences, composition semantic differences, and texture structure differences between the key frame migration image and the standard Tujia brocade image are calculated respectively. If the calculated color culture differences, composition semantic differences, and texture structure differences are all within the Tujia brocade image style range, then the corresponding key frame migration image is determined to be the final fusion image.

[0030] The present invention further provides a method for traditional pattern image style transfer based on multimodal feature decoupling and dynamic propagation, characterized in that the method is executed to implement the entire content of the traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation, specifically including:

[0031] Step 1: Obtain Tujia brocade images from the local intangible cultural heritage public database and select one Tujia brocade image as the standard Tujia brocade image. Use the K-means algorithm to extract the color features of the Tujia brocade image, use the gray-level co-occurrence matrix to extract the texture features of the Tujia brocade image, and use the Canny hard edge detection algorithm to extract the vectorized composition features of the Tujia brocade image.

[0032] Step 2: Based on low-rank adaptive technology, the color features, texture features, and vectorized composition features of each Tujia brocade image are used to train the parameters of the diffusion model to generate a low-rank adaptive model adapted to the traditional pattern. The image to be migrated is split into multiple migration images according to the frame. The key frame images are selected by the Frobenius norm of adjacent frames. The key frame images are input into the trained low-rank adaptive model to obtain the key frame migration images.

[0033] Step 3: Establish a Tujia brocade image feature difference evaluation system, and determine the style range of Tujia brocade images based on the Tujia brocade image feature difference evaluation system;

[0034] Step 4: The style deviation between the key frame migration image and the standard Tujia brocade image is analyzed. The final fusion image is selected from the key frame migration image according to the style range of the Tujia brocade image. The final fusion image is processed using the Poisson fusion technique to obtain the style migration image.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] This solution significantly improves the cultural fidelity and automation level of traditional pattern style transfer through multimodal feature decoupling and dynamic propagation mechanisms. First, based on the low-rank adaptive model, the color, texture and vectorized composition features are integrated to solve the style distortion problem caused by feature coupling in traditional methods, making the transfer results more in line with the artistic characteristics of Tujia brocade. Secondly, the Frobenius norm is used to dynamically filter key frames and quantify style offsets, ensuring temporal consistency while reducing the computational load. Finally, the constructed difference assessment system works synergistically with Poisson fusion technology to achieve the objective definition and natural fusion of style ranges, effectively avoiding the subjective bias of manual intervention, and providing quantifiable technical support for the digitization of intangible cultural heritage. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Schematic diagram of the overall system structure of the present invention.

[0038] Figure 2 Schematic diagram of the overall method of the present invention. DETAILED DESCRIPTION

[0039] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0040] It should be noted that, unless otherwise defined, the technical or scientific terms used in the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative position relationships. When the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0041] Example:

[0042] See also Figure 1 , the present invention provides a technical solution:

[0043] A traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation, specifically including:

[0044] The data acquisition module is used to obtain Tujia brocade images from the local intangible cultural heritage public database and select a Tujia brocade image as the standard Tujia brocade image. The K-means algorithm is used to extract the color features of the Tujia brocade image. The gray-level co-occurrence matrix is used to extract the texture features of the Tujia brocade image. The Canny hard edge detection algorithm is used to extract the vectorized composition features of the Tujia brocade image.

[0045] The color features include the color difference, main colors and proportion of each main color of the Tujia brocade image; the texture features of the Tujia brocade image include the contrast, correlation, energy and homogeneity of the grayscale co-occurrence matrix of the Tujia brocade image at 0°, 45°, 90° and 135°; the vectorized composition features of the Tujia brocade image include the edge distribution map of the Tujia brocade image.

[0046] The model analysis module is used to train the parameters of the diffusion model based on low-rank adaptive technology, using the color features, texture features, and vectorized composition features of each Tujia brocade image to generate a low-rank adaptive model adapted to the traditional pattern. The image to be migrated is split into multiple migration images according to the frame, and the key frame images are selected by the Frobenius norm of adjacent frames. The key frame images are input into the trained low-rank adaptive model to obtain the key frame migration images.

[0047] The parameters of the diffusion model are trained based on the adaptive model using the color features, texture features, and vectorized composition features of each Tujia brocade image. Existing technologies can be used. Specifically, trainable parameters are inserted into the CrossAttention layer of Stable Diffusion v1.5:

[0048] w′=w+αBA,(B∈R 768*64 ,A∈R 64*768 )

[0049] Where w′ is the updated trainable parameter, w is the trainable parameter before the update, and α is the scaling factor. Usually, α = 0.8

[0050] R is the image dimension space for training, and the total image dimension is R 768*768 , B is the downlink matrix, which is the result of the downlink 64-dimensional mapping of the trained image dimension space, and A is the uplink matrix, which is the result of the uplink 64-dimensional mapping of the total trained image dimension space;

[0051] Training parameter configuration: Optimizer: AdamW-8bit (β1=0.9, β2=0.99); Learning rate: Cosine annealing schedule (initial 4e-4→final 1e-6); Batch size: 4 (512×512 resolution); Regularization: L2 coefficient λ=0.01.

[0052] The specific logic for calculating the Frobenius norm of adjacent frames is as follows: construct an image migration matrix based on the image to be migrated. Each pixel in the image corresponds to an element of the migration matrix, and the pixel value is the element value of the migration matrix. The Frobenius norm of adjacent frames is calculated. If the Frobenius norm is greater than the preset threshold, the migration image with the later time label in the two frames is selected as the key frame image. The specific formula for calculating the Frobenius norm is as follows:

[0053] ΔF e =||E e+1 -E e || F

[0054] Where ΔF e is the Frobenius norm of adjacent frames, E e+1 is the image migration matrix of the next frame in the adjacent frames, E e is the image migration matrix of the previous frame in the adjacent frame.

[0055] A weight analysis module is used to establish a Tujia brocade image feature difference evaluation system and determine the style range of Tujia brocade images based on the Tujia brocade image feature difference evaluation system;

[0056] The Tujia brocade image feature difference evaluation system specifically includes color cultural differences, texture structure differences and composition semantic differences;

[0057] The specific formula for calculating color cultural differences is:

[0058]

[0059] Among them, C For color cultural differences, u i is the three-dimensional vector of the i-th main color, u iO is the standard three-dimensional vector of the i-th main color, ||||2 is the norm operator, and I is the number of main color types;

[0060] Color cultural differences quantify the cultural consistency of the color distribution between the image and the standard Tujia brocade, focusing on protecting the symbolic meaning of ethnic colors and the accuracy of visual perception. i and u iO is the CIE-Lab three-dimensional coordinate of the i primary colors. The Lab space is consistent with human eye perception. i -u iO ||2 Euclidean distance, which calculates the color space distance between the main color and the standard main color. The larger the Euclidean distance, the greater the color space distance between the main color and the standard main color, and the worse the color cultural style similarity between the two images. This indicates that the color deviates further from the standard color and the more serious the cultural distortion.

[0061] The specific formula for calculating texture structure differences is:

[0062]

[0063] Among them, t is the texture structure difference, GLCM j is the feature vector of the jth texture feature, GLCM jO is the standard feature vector of the jth texture feature, the feature vector of the texture feature is a 4-dimensional vector, expressed as GLCM j =(glcm j0 ,glcm j45 ,glcm j90 ,glcm 135 ), where glcm j0 Represents the eigenvalue of the j-th texture feature of the gray-level co-occurrence matrix in the 0° direction, glcm j45 is the eigenvalue of the jth texture feature of the gray-level co-occurrence matrix at 45°, glcm j90 is the eigenvalue of the j-th texture feature of the gray-level co-occurrence matrix in the 90° direction, glcm 135is the eigenvalue of the jth texture feature of the gray-level co-occurrence matrix in the 135° direction, SSIM (GLCM j -GLCM jO ) is GLCM j With GLCM jO The similarity between

[0064]

[0065] Among them, μ jx is the mean of the eigenvalues of the jth texture feature at 0°, 45°, 90° and 135°, μ y is the mean of the standard values of the eigenvalues of the j-th texture feature at 0°, 45°, 90° and 135°, σ jx is the variance of the eigenvalues of the jth texture feature at 0°, 45°, 90° and 135°, σ jy is the variance of the standard value of the eigenvalue of the jth texture feature at 0°, 45°, 90° and 135°, σ jxy The covariance between the eigenvalue of the jth texture feature at 0°, 45°, 90° and 135° and the standard value of the eigenvalue.

[0066] The texture structure difference reflects the comprehensive texture structure difference between the selected image and the standard image. It can directly select the structural difference between the image and the standard image at the structural context level. This difference is evaluated by the similarity index SSIM. SSIM is the similarity index between the texture structures of the two images. The larger the value, the more similar the texture features are; the lower the value, the greater the difference. The SSIM method is more consistent with the human eye's perception of texture structure (taking into account brightness, contrast, and structural information).

[0067] In order to avoid a single texture feature dominating the texture structure difference, the feature values of the texture features are all normalized texture feature values;

[0068] The specific formula for calculating the semantic difference of composition is:

[0069] ζ l =Iou(M k ,M kO )

[0070] Among them, lk For the semantic difference of composition, M k is the binary matrix of the marginal distribution graph, M kO is a binary matrix of the standard edge distribution graph, which is obtained by analyzing the edge distribution graph. Specifically, the edge part of the edge distribution graph is marked as 1, and the other parts are marked as 0. k ,M kO ) is M k and MkO The intersection-and-union ratio of compositional semantic differences reflects the geometric structure and topological relationship of the pattern totem. Larger compositional semantic differences can ensure the semantic integrity of cultural symbols (such as the symmetry of the diamond skeleton).

[0071] The standard three-dimensional vector of the main color, the standard feature vector of the texture feature and the binary matrix of the standard edge distribution map are obtained by analyzing the selected standard Tujia brocade image.

[0072] The style deviation module is used to calculate the style deviation between the key frame migration image and the standard Tujia brocade image, select the final fusion image from the key frame migration image according to the style range of the Tujia brocade image, and process the final fusion image using the Poisson fusion technology to obtain the style migration image.

[0073] The color culture differences, composition semantic differences, and texture structure differences between all Tujia brocade images were calculated, and the average color culture differences, average composition semantic differences, and average texture structure differences were calculated. The style range of Tujia brocade images was determined based on the average color culture differences, average composition semantic differences, and average texture structure differences. The style range of Tujia brocade images is specifically as follows:

[0074]

[0075] in, To average color cultural differences, is the average texture structure difference, is the average composition semantic difference.

[0076] The style range of Tujia brocade images can constrain the style and ensure that the selected image is within the predetermined style range.

[0077] The specific logic for selecting the final fusion image is as follows: the color cultural differences, composition semantic differences, and texture structure differences between the key frame migration image and the standard Tujia brocade image are calculated respectively. If the calculated color cultural differences, composition semantic differences, and texture structure differences are all within the style range of the Tujia brocade image, the corresponding key frame migration image is judged to be the final fusion image.

[0078] Using Poisson fusion to synthesize a style-transferred image from style-transferred keyframes can be accomplished using existing techniques. Specifically, this involves calculating the gradients of the overlapping regions between adjacent stylized keyframes and fusing this gradient information into the target frame using the Poisson equation to ensure a natural transition between color and texture. Finally, the fused keyframes are time-aligned with the non-keyframes of the original video using optical flow or interpolation algorithms to produce a final, style-consistent video. This is a conventional technique used by those skilled in the art and will not be elaborated upon here.

[0079] The present invention further provides a method for traditional pattern image style transfer based on multimodal feature decoupling and dynamic propagation, characterized in that the method is executed to implement the entire content of the traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation, specifically including:

[0080] Step 1: Obtain Tujia brocade images from the local intangible cultural heritage public database and select one Tujia brocade image as the standard Tujia brocade image. Use the K-means algorithm to extract the color features of the Tujia brocade image, use the gray-level co-occurrence matrix to extract the texture features of the Tujia brocade image, and use the Canny hard edge detection algorithm to extract the vectorized composition features of the Tujia brocade image.

[0081] Step 2: Based on low-rank adaptive technology, the color features, texture features, and vectorized composition features of each Tujia brocade image are used to train the parameters of the diffusion model to generate a low-rank adaptive model adapted to the traditional pattern. The image to be migrated is split into multiple migration images according to the frame. The key frame images are selected by the Frobenius norm of adjacent frames. The key frame images are input into the trained low-rank adaptive model to obtain the key frame migration images.

[0082] Step 3: Establish a Tujia brocade image feature difference evaluation system, and determine the style range of Tujia brocade images based on the Tujia brocade image feature difference evaluation system;

[0083] Step 4: The style deviation between the key frame migration image and the standard Tujia brocade image is analyzed. The final fusion image is selected from the key frame migration image according to the style range of the Tujia brocade image. The final fusion image is processed using the Poisson fusion technique to obtain the style migration image.

[0084] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0085] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software systems depends on the specific application and design constraints of the technical solution.

[0086] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0087] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation, characterized by: Specifically include: The data acquisition module is used to obtain Tujia brocade images from the local intangible cultural heritage public database and select a Tujia brocade image as the standard Tujia brocade image. The K-means algorithm is used to extract the color features of the Tujia brocade image. The gray-level co-occurrence matrix is used to extract the texture features of the Tujia brocade image. The Canny hard edge detection algorithm is used to extract the vectorized composition features of the Tujia brocade image. The model analysis module is used to train the parameters of the diffusion model based on low-rank adaptive technology, using the color features, texture features, and vectorized composition features of each Tujia brocade image to generate a low-rank adaptive model adapted to the traditional pattern. The image to be migrated is split into multiple migration images according to the frame, and the key frame images are selected by the Frobenius norm of adjacent frames. The key frame images are input into the trained low-rank adaptive model to obtain the key frame migration images. A weight analysis module is used to establish a Tujia brocade image feature difference evaluation system and determine the style range of Tujia brocade images based on the Tujia brocade image feature difference evaluation system; The style deviation module is used to calculate the style deviation between the key frame migration image and the standard Tujia brocade image, select the final fusion image from the key frame migration image according to the style range of the Tujia brocade image, and process the final fusion image using the Poisson fusion technology to obtain the style migration image.

2. The traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation according to claim 1, characterized in that: The color features include the color difference, main color type and main color type ratio of the Tujia brocade image; the texture features of the Tujia brocade image include the contrast, correlation, energy and homogeneity of the grayscale co-occurrence matrix of the Tujia brocade image at 0°, 45°, 90° and 135°; the vectorized composition features of the Tujia brocade image include the edge distribution map of the Tujia brocade image.

3. The traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation according to claim 1, characterized in that: The specific logic for calculating the Frobenius norm of adjacent frames is as follows: construct an image migration matrix based on the image to be migrated. Each pixel in the image corresponds to an element of the migration matrix, and the pixel value is the element value of the migration matrix. The Frobenius norm of adjacent frames is calculated. If the Frobenius norm is greater than the preset threshold, the migration image with the later time label in the two frames is selected as the key frame image. The specific formula for calculating the Frobenius norm is as follows: ΔF e =||And e+1 -AND e || F Where ΔF e is the Frobenius norm of adjacent frames, E e+1 is the image migration matrix of the next frame in the adjacent frames, E e is the image migration matrix of the previous frame in the adjacent frame.

4. The traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation according to claim 2, characterized in that: The Tujia brocade image feature difference evaluation system specifically includes color cultural differences, texture structure differences and composition semantic differences; The specific formula for calculating color cultural differences is: Among them, C For color cultural differences, u i is the three-dimensional vector of the i-th main color, u iO is the standard three-dimensional vector of the i-th main color, ||||2 is the norm operator, and I is the number of main color types; The specific formula for calculating texture structure differences is: Among them, t is the texture structure difference, GLCM j is the feature vector of the jth texture feature, GLCM jO is the standard feature vector of the jth texture feature, the feature vector of the texture feature is a 4-dimensional vector, expressed as GLCM j =(glcm j0 ,glcm j45 ,glcm j90 ,glcm 135 ), where glcm j0 Represents the eigenvalue of the j-th texture feature of the gray-level co-occurrence matrix in the 0° direction, glcm j45 is the eigenvalue of the jth texture feature of the gray-level co-occurrence matrix at 45°, glcm j90 is the eigenvalue of the j-th texture feature of the gray-level co-occurrence matrix in the 90° direction, glcm 135 is the eigenvalue of the jth texture feature of the gray-level co-occurrence matrix in the 135° direction, SSIM (GLCM j -GLCM jO ) is GLCM j With GLCM jO The similarity between The specific formula for calculating the semantic difference of composition is: ζ l =Iou(M k ,M kO ) Among them, lk For the semantic difference of composition, M k is the binary matrix of the marginal distribution graph, M kO is a binary matrix of the standard edge distribution graph, which is obtained by analyzing the edge distribution graph. Specifically, the edge part of the edge distribution graph is marked as 1, and the other parts are marked as 0. k ,M kO ) is M k and M kO The intersection and union ratio of The standard three-dimensional vector of the main color, the standard feature vector of the texture feature and the binary matrix of the standard edge distribution map are obtained by analyzing the selected standard Tujia brocade image.

5. The traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation according to claim 4, characterized in that: The color culture differences, composition semantic differences, and texture structure differences between all Tujia brocade images were calculated, and the average color culture differences, average composition semantic differences, and average texture structure differences were calculated. The style range of Tujia brocade images was determined based on the average color culture differences, average composition semantic differences, and average texture structure differences. The style range of Tujia brocade images is specifically as follows: in, To average color cultural differences, is the average texture structure difference, is the average composition semantic difference.

6. The traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation according to claim 5, characterized in that: The specific logic for selecting the final fusion image is as follows: the color cultural differences, composition semantic differences, and texture structure differences between the key frame migration image and the standard Tujia brocade image are calculated respectively. If the calculated color cultural differences, composition semantic differences, and texture structure differences are all within the style range of the Tujia brocade image, the corresponding key frame migration image is judged to be the final fusion image.

7. A traditional pattern image style transfer method based on multimodal feature decoupling and dynamic propagation, characterized by: The method is executed to implement the entire content of the traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation according to any one of claims 1 to 6, specifically including: Step 1: Obtain Tujia brocade images from the local intangible cultural heritage public database and select one Tujia brocade image as the standard Tujia brocade image. Use the K-means algorithm to extract the color features of the Tujia brocade image, use the gray-level co-occurrence matrix to extract the texture features of the Tujia brocade image, and use the Canny hard edge detection algorithm to extract the vectorized composition features of the Tujia brocade image. Step 2: Based on low-rank adaptive technology, the color features, texture features, and vectorized composition features of each Tujia brocade image are used to train the parameters of the diffusion model to generate a low-rank adaptive model adapted to the traditional pattern. The image to be migrated is split into multiple migration images according to the frame. The key frame images are selected by the Frobenius norm of adjacent frames. The key frame images are input into the trained low-rank adaptive model to obtain the key frame migration images. Step 3: Establish a Tujia brocade image feature difference evaluation system, and determine the style range of Tujia brocade images based on the Tujia brocade image feature difference evaluation system; Step 4: The style deviation between the key frame migration image and the standard Tujia brocade image is analyzed. The final fusion image is selected from the key frame migration image according to the style range of the Tujia brocade image. The final fusion image is processed using the Poisson fusion technique to obtain the style migration image.

Images