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

Abstract

The application provides a traditional pattern image style transfer system and method based on multi-modal feature decoupling and dynamic propagation, and relates to the field of computer vision and digital media technology. Color features, texture features and vectorized composition features of Tujia brocade images are collected, and a standard image is selected; based on low-rank adaptive technology, a low-rank adaptive model is trained using the features collected from the Tujia brocade images; the image to be transferred is split into a transfer image and a selected key frame image according to frames, and the key frame transfer image is obtained through the trained low-rank adaptive model; a Tujia brocade image feature difference evaluation system is established, and the Tujia brocade image style range is determined based on this; the style deviation of the key frame transfer image and the standard Tujia brocade image is calculated, the final fusion image is selected from the key frame transfer image according to the Tujia brocade image style range, and the final fusion image is processed through Poisson fusion technology to obtain a style transfer image.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of computer vision and digital media, in particular to a traditional pattern image style transfer system and method based on multi-modal feature decoupling and dynamic propagation. BACKGROUND

[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 sets for training, which are difficult to capture the unique multi-modal features (color, texture, composition) of intangible cultural heritage (such as Tujia brocade), resulting in problems such as cultural semantic distortion and style deviation in the transfer results. Especially for dynamic images, existing technologies lack differentiated processing mechanisms for key frames, and cannot balance style consistency and computational efficiency. In addition, traditional methods have a fuzzy range of defining pattern style, often relying on manual experience, making it difficult to quantitatively evaluate the fit of the transfer results and the original culture.

[0003] The above information disclosed in the background section is only intended to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0004] The purpose of the present application is to provide a traditional pattern image style transfer system and method based on multi-modal feature decoupling and dynamic propagation to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0006] A traditional pattern image style transfer system based on multi-modal feature decoupling and dynamic propagation, specifically comprising:

[0007] A data acquisition module for obtaining Tujia brocade images from a local intangible cultural heritage public database and selecting one Tujia brocade image as a standard Tujia brocade image, using K-means algorithm to extract color features of the Tujia brocade image, extracting texture features of the Tujia brocade image through a gray level co-occurrence matrix, and using a Canny hard edge detection algorithm to extract vectorized composition features of the Tujia brocade image;

[0008] A model analysis module for training parameters of a diffusion model using color features, texture features, and vectorized composition features of each Tujia brocade image based on low-rank adaptive technology to generate a low-rank adaptive model adapted to traditional patterns; splitting the image to be transferred into multiple transfer images according to frames, selecting key frame images through adjacent frame Frobenius norm, and inputting the key frame images into the trained low-rank adaptive model to obtain key frame transfer images;

[0009] The weight analysis module is configured to establish a Tujia brocade image feature difference evaluation system and determine a Tujia brocade image style range based on the Tujia brocade image feature difference evaluation system.

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

[0011] Further, the color feature includes a color difference, a main color type, and a main color type proportion of the Tujia brocade image; the texture feature of the Tujia brocade image includes a contrast, a correlation, an energy, and a homogeneity of a gray level co-occurrence matrix of the Tujia brocade image at 0°, 45°, 90°, and 135°; and the vectorized composition feature of the Tujia brocade image includes an edge distribution map of the Tujia brocade image.

[0012] Further, a specific logic for calculating the Frobenius norm of the adjacent frames is as follows: an image transition matrix is constructed according to the image to be transitioned, each pixel point in the image corresponds to an element of the transition matrix, and the pixel value is the element value of the transition matrix. The Frobenius norm of the adjacent frames is calculated. If the Frobenius norm is greater than a preset threshold, the transition image with a later time label is selected as the key frame image. The specific formula for calculating the Frobenius norm is as follows:

[0013]

[0014] wherein, is the Frobenius norm of the adjacent frames, is the image transition matrix of the later frame in the adjacent frames, is the image transition matrix of the earlier frame in the adjacent frames.

[0015] Further, the Tujia brocade image feature difference evaluation system specifically includes a color culture difference, a texture structure difference, and a composition semantic difference.

[0016] The specific formula for calculating the color culture difference is as follows:

[0017]

[0018] wherein, is the color culture difference, is a three-dimensional vector of the i-th main color, is a standard three-dimensional vector of the i-th main color, is a three-dimensional vector of the i-th main color, is a three-dimensional vector of the i-th main color, is a norm operator, is a number of main color types.

[0019] The specific formula used to calculate texture structure differences is as follows:

[0020]

[0021] in, Due to differences in texture structure, For the first Feature vectors of texture features For the first A standard feature vector for each texture feature, wherein the feature vector of each texture feature is a 4-dimensional vector, represented as follows: ,in, The gray-level co-occurrence matrix represents the first position in the 0° direction. Feature values ​​of each texture feature To represent the gray-level co-occurrence matrix in the 45° direction... Feature values ​​of each texture feature To represent the gray-level co-occurrence matrix in the 90° direction... Feature values ​​of each texture feature To represent the gray-level co-occurrence matrix in the 135° direction... Feature values ​​of each texture feature for and Similarity between them

[0022] The specific formula used to calculate semantic differences in composition is as follows:

[0023]

[0024] in, For semantic differences in composition, This is a binary matrix representing the edge distribution map. This is a binary matrix of a standard edge distribution map, obtained through edge distribution map analysis. Specifically, the edge portions of the edge distribution map are marked as 1, and other portions are marked as 0. for and The intersection and union ratio;

[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] Further, the color culture difference, the composition semantic difference and the texture structure difference of all Tujia brocade images and the Tujia brocade images are calculated, the average color culture difference, the average composition semantic difference and the average texture structure difference are calculated, and the Tujia brocade image style range is determined according to the average color culture difference, the average composition semantic difference and the average texture structure difference, and the Tujia brocade image style range is specifically:

[0027]

[0028] wherein, the average color culture difference is, the average texture structure difference is, the average composition semantic difference is.

[0029] Further, the specific logic for selecting the final fusion image is that the color culture difference, the composition semantic difference and the texture structure difference of the key frame transition image and the standard Tujia brocade image are calculated respectively, and if the calculated color culture difference, composition semantic difference and texture structure difference are within the Tujia brocade image style range, the corresponding key frame transition image is determined as the final fusion image.

[0030] The application further provides a traditional pattern image style transfer method based on multi-modal feature decoupling and dynamic propagation, which is executed to realize the whole content of the traditional pattern image style transfer system based on multi-modal feature decoupling and dynamic propagation, and specifically includes:

[0031] Step 1: obtaining Tujia brocade images from a local intangible cultural heritage public database and selecting one Tujia brocade image as a standard Tujia brocade image, extracting color features of the Tujia brocade images using a K-means algorithm, extracting texture features of the Tujia brocade images through a gray level co-occurrence matrix, and extracting vectorized composition features of the Tujia brocade images using a Canny hard edge detection algorithm;

[0032] Step 2: based on a low-rank adaptive technology, training parameters of a diffusion model using the color features, the texture features and the vectorized composition features of each Tujia brocade image to generate a low-rank adaptive model suitable for traditional patterns; splitting a to-be-transferred image into multiple transition images according to frames, selecting a key frame image through adjacent frame Frobenius norm, inputting the key frame image into the trained low-rank adaptive model to obtain a key frame transition image;

[0033] Step 3: establishing a Tujia brocade image feature difference evaluation system and determining a Tujia brocade image style range based on the Tujia brocade image feature difference evaluation system;

[0034] Step 4: The style migration of the key frame transition image and the standard Tujia brocade image is determined, the final fusion image is selected according to the style range of the Tujia brocade image, and the final fusion image is processed by the Poisson fusion technology to obtain the style migration image.

[0035] Compared with the prior art, the beneficial effects of the present application are:

[0036] The present scheme significantly improves the cultural fidelity and automation level of traditional pattern style migration through multi-modal feature decoupling and dynamic propagation mechanism. Firstly, based on the low-rank adaptive model, color, texture and vectorized composition features are fused, which solves the style distortion problem caused by feature coupling in traditional methods, so that the migration result is more consistent with the artistic characteristics of Tujia brocade. Secondly, the key frame is dynamically selected and the style deviation is quantified by Frobenius norm, which reduces the computational load while ensuring temporal consistency. Finally, the difference evaluation system and Poisson fusion technology work together to realize the objective definition and natural fusion of the style range, effectively avoiding the subjective bias of manual intervention, and providing quantifiable technical support for digital non-heritage. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 It is the overall system structure diagram of the present application.

[0038] Figure 2 It is the overall method flow diagram of the present application. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below in combination with specific embodiments.

[0040] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present application should be understood as the usual meaning understood by those skilled in the art to which the present application belongs. The "first", "second" and similar words used in the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Include" or "contain" and similar words mean that the elements or objects before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connected" or "connected" and similar words are not limited to physical or mechanical connection, but can include electrical connection, whether direct or indirect. "Up", "down", "left", "right" and the like only represent relative positional relationship, when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0041] Embodiment:

[0042] Please refer to Figure 1The application provides a technical solution:

[0043] A traditional pattern image style transfer system based on multi-modal feature decoupling and dynamic propagation, specifically comprising:

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

[0045] The color features include color differences, main colors, and proportions of the main colors of the Tujia brocade image; the texture features of the Tujia brocade image include contrasts, correlations, energies, and homogeneities of the gray level co-occurrence matrix of the Tujia brocade image at 0°, 45°, 90°, and 135°; and the vectorized composition features of the Tujia brocade image include an edge distribution map of the Tujia brocade image.

[0046] A model analysis module is configured to train parameters of a diffusion model based on a low-rank adaptive technique using the color features, the texture features, and the vectorized composition features of each Tujia brocade image to generate a low-rank adaptive model adapted to a traditional pattern; split a to-be-transferred image into a plurality of transfer images according to frames, select a key frame image through a Frobenius norm of adjacent frames, input the key frame image into the trained low-rank adaptive model, and obtain a key frame transfer image.

[0047] Training the parameters of the diffusion model based on the adaptive model using the color features, the texture features, and the vectorized composition features of each Tujia brocade image can adopt an existing technique, specifically by inserting trainable parameters in a CrossAttention layer of Stable Diffusion v1.5:

[0048] wherein, is an updated trainable parameter, is a pre-updated trainable parameter, is a scaling factor, usually, 0.8

[0049] is a trained image dimension space, and the total image dimension is , is a downlink matrix, and the result of downlink 64-dimensional mapping of the trained image dimension space is is an uplink matrix, and the result of uplink 64-dimensional mapping of the total trained image dimension space is

[0050] Training parameter configuration: optimizer: Learning rate: cosine annealing schedule (initial 4e-4 to final 1e-6); batch size: 4 (512x512 resolution); regularization: L2 coefficient lambda = 0.01.

[0051] The specific logic for calculating the Frobenius norm of adjacent frames is as follows: an image migration matrix is constructed according to the image to be migrated, each pixel point 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 the adjacent frames is calculated, and if the Frobenius norm is greater than a predetermined 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:

[0052]

[0053] wherein, is the Frobenius norm of the adjacent frames, is the image migration matrix of the later frame in the adjacent frames, is the image migration matrix of the earlier frame in the adjacent frames.

[0054] The weight analysis module is configured to establish a Tujia brocade image feature difference evaluation system and determine a Tujia brocade image style range based on the Tujia brocade image feature difference evaluation system.

[0055] The Tujia brocade image feature difference evaluation system specifically includes color culture difference, texture structure difference, and composition semantic difference.

[0056] The specific formula for calculating the color culture difference is:

[0057]

[0058] wherein, is the color culture difference, is the three-dimensional vector of the i-th dominant color, is the standard three-dimensional vector of the i-th dominant color, is the norm operator, is the number of dominant color categories. The color culture difference quantifies the cultural consistency of the image and the standard Tujia brocade in color distribution, and focuses on protecting the symbolic meaning and visual perception accuracy of national colors. and

[0059] are the CIE-Lab three-dimensional coordinates of the i-th dominant color, and the Lab space is consistent with the perception of the human eye, ​​​Euclidean distance is used to calculate the color space distance between the primary color and the standard primary color. The larger the Euclidean distance, the greater the color space distance between the primary color and the standard primary color, and the worse the cultural similarity between the two images. This indicates that the colors deviate further from the standard colors and the more severe the cultural distortion.

[0060] The specific formula used to calculate texture structure differences is as follows:

[0061]

[0062] in, Due to differences in texture structure, For the first Feature vectors of texture features For the first A standard feature vector for each texture feature, wherein the feature vector of each texture feature is a 4-dimensional vector, represented as follows: ,in, The gray-level co-occurrence matrix represents the first position in the 0° direction. Feature values ​​of each texture feature To represent the gray-level co-occurrence matrix in the 45° direction... Feature values ​​of each texture feature To represent the gray-level co-occurrence matrix in the 90° direction... Feature values ​​of each texture feature To represent the gray-level co-occurrence matrix in the 135° direction... Feature values ​​of each texture feature for and Similarity between them

[0063]

[0064] in, For the first The mean of the feature values ​​of each texture feature at 0°, 45°, 90° and 135°. For the first The mean of the standard values ​​of the texture features at 0°, 45°, 90° and 135°. For the first The variance of the feature values ​​of each texture feature at 0°, 45°, 90° and 135° For the first The variance of the standard values ​​of the texture features at 0°, 45°, 90°, and 135° No. The covariance of texture features at 0°, 45°, 90° and 135° with respect to the standard values ​​of the features.

[0065] The texture structure difference reflects the comprehensive texture structure difference between the selected image and the standard image, and can directly select the structural difference between the selected image and the standard image from the structural context level. The difference is reflected by the similarity index to be evaluated, is the similarity index between the texture structures of the two images. The greater the value, the more similar the texture features; the lower the value, the greater the difference, The method is more in line with the perception of the human eye to the texture structure (taking into account brightness, contrast, and structural information).

[0066] In order to avoid the dominance of a single texture feature in the texture structure difference, the feature values of the texture features are normalized.

[0067] The specific formula for calculating the composition semantic difference is:

[0068]

[0069] wherein, is the composition semantic difference, is a binary matrix of the edge distribution map, is a binary matrix of the standard edge distribution map, which is obtained by edge distribution map analysis, specifically by marking the edge part in the edge distribution map as 1 and the other parts as 0, is and is the intersection-over-union ratio; the composition semantic difference reflects the geometric structure and topological relationship of the pattern totem, and a larger composition semantic difference can ensure the semantic integrity of the cultural symbol (such as the symmetry of the diamond skeleton).

[0070] 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.

[0071] The style deviation module is configured to calculate the style deviation of 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 Tujia brocade image style range, and process the final fusion image by Poisson fusion technology to obtain a style migration image.

[0072] The color cultural difference, the composition semantic difference, and the texture structure difference between all Tujia brocade images and the Tujia brocade image are calculated, and the average color cultural difference, the average composition semantic difference, and the average texture structure difference are calculated. The Tujia brocade image style range is determined according to the average color cultural difference, the average composition semantic difference, and the average texture structure difference. The Tujia brocade image style range is specifically:

[0073]

[0074] wherein, is the average color culture difference, is the average texture structure difference, is the average composition semantic difference.

[0075] The Tujia brocade image style range can constrain the style to ensure that the selected image is within the predetermined style range.

[0076] The specific logic for selecting the final fusion image is: the color culture difference, the composition semantic difference and the texture structure difference of the key frame transition image and the standard Tujia brocade image are calculated respectively, if the calculated color culture difference, composition semantic difference and texture structure difference are within the Tujia brocade image style range, it is judged that the corresponding key frame transition image is the final fusion image.

[0077] The existing technology can be used to synthesize the style transition key frame image into a style transition video using the Poisson fusion technology: specifically including then calculating the gradient of the overlapping area between adjacent stylized key frames, and using the Poisson equation to fuse the gradient information into the target frame to ensure natural color and texture transition; finally, the fused key frame and the non-key frame of the original video are time-aligned through optical flow or interpolation algorithm to generate the final style-consistent video. This is a routine technical means for those skilled in the art, and will not be described here.

[0078] The application further provides a traditional pattern image style transition method based on multi-modal feature decoupling and dynamic propagation, which is executed to realize the entire content of the traditional pattern image style transition system based on multi-modal feature decoupling and dynamic propagation, and specifically includes:

[0079] Step 1: obtain Tujia brocade images from a local intangible cultural heritage public database and select a Tujia brocade image as a standard Tujia brocade image, use K-means algorithm to extract color features of the Tujia brocade image, extract texture features of the Tujia brocade image through a gray level co-occurrence matrix, and use a Canny hard edge detection algorithm to extract vectorized composition features of the Tujia brocade image;

[0080] Step 2: based on a low-rank adaptive technology, use the color features, texture features and vectorized composition features of each Tujia brocade image to train parameters of a diffusion model to generate a low-rank adaptive model suitable for traditional patterns; split the image to be transitioned into multiple transition images according to frames, select a key frame image through adjacent frame Frobenius norm, input the key frame image into the trained low-rank adaptive model, and obtain a key frame transition image;

[0081] Step 3: establish a Tujia brocade image feature difference evaluation system, and determine a Tujia brocade image style range based on the Tujia brocade image feature difference evaluation system;

[0082] Step 4: The style migration of the key frame transition image and the standard Tujia brocade image is determined, the final fusion image is selected according to the style range of the Tujia brocade image, and the style transition image is obtained by processing the final fusion image through the Poisson fusion technology.

[0083] The above formulas are dimensionless values calculated, the formula is obtained by software simulation of a large number of collected data to obtain the nearest real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0084] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. Those skilled in the art can realize that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are executed by hardware or software systems depends on the specific application and design constraints of the technical solutions.

[0085] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, which can be located in one place or distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0086] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation, characterized in that, Specifically, it includes: 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 a 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, and 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 the color features, texture features and vectorized composition features of each Tujia brocade image using low-rank adaptive technology, and generate a low-rank adaptive model that adapts to traditional patterns; the image to be transferred is split into multiple transfer images according to frames, and 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 key frame transfer images. The weighting analysis module is used to establish an evaluation system for the differences in the features of Tujia brocade images, and to determine the range of Tujia brocade image styles based on the evaluation system for the differences in the features of Tujia brocade images. The style deviation module is used to calculate the style shift between the keyframe transfer image and the standard Tujia brocade image. Based on the style range of the Tujia brocade image, the final fused image is selected from the keyframe transfer image, and the style transfer image is obtained by processing the final fused image through Poisson fusion technology. The Tujia brocade image feature difference evaluation system specifically includes color culture differences, texture structure differences, and compositional semantic differences; The specific formula used to calculate cultural differences in color is as follows: in, Due to differences in color culture, For the first A three-dimensional vector of a primary color. For the first A standard three-dimensional vector of a primary color, Norm operators, Number of main color types; The specific formula used to calculate texture structure differences is as follows: in, Due to differences in texture structure, For the first Feature vectors of texture features For the first A standard feature vector for each texture feature, wherein the feature vector of each texture feature is a 4-dimensional vector, represented as follows: ,in, The gray-level co-occurrence matrix represents the first position in the 0° direction. Feature values ​​of each texture feature To represent the gray-level co-occurrence matrix in the 45° direction... Feature values ​​of each texture feature To represent the gray-level co-occurrence matrix in the 90° direction... Feature values ​​of each texture feature To represent the gray-level co-occurrence matrix in the 135° direction... Feature values ​​of each texture feature for and Similarity between them The specific formula used to calculate semantic differences in composition is as follows: in, For semantic differences in composition, This is a binary matrix representing the edge distribution map. This is a binary matrix of a standard edge distribution map, obtained through edge distribution map analysis. Specifically, the edge portions of the edge distribution map are marked as 1, and other portions are marked as 0. for and The intersection and union ratio; 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.

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 types, and proportion of main color types in the Tujia brocade image; the texture features of the Tujia brocade image include the contrast, correlation, energy, and homogeneity of the gray-level 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: An image transfer matrix is ​​constructed based on the image to be transferred. Each pixel in the image corresponds to an element of the transfer matrix, and the pixel value is the element value of the transfer matrix. The Frobenius norm of adjacent frames is calculated. If the Frobenius norm is greater than a preset threshold, the transferred image with the later time label in the two frames is selected as the keyframe image. The specific formula for calculating the Frobenius norm is as follows: in, The Frobenius norm of adjacent frames, This is the image migration matrix of the next frame in an adjacent frame. This is the image migration matrix of the previous frame in adjacent frames.

4. The traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation according to claim 3, characterized in that: The differences in color culture, compositional semantics, and texture structure among all Tujia brocade images are calculated. The average differences in color culture, compositional semantics, and texture structure are then calculated. Based on these average differences, the style range of Tujia brocade images is determined. Specifically, the style range of Tujia brocade images is as follows: in, To average out cultural differences in color, For average texture structure differences, This represents the average semantic difference in composition.

5. A traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation according to claim 4, characterized in that: The specific logic for selecting the final fused image is as follows: calculate the differences in color culture, compositional semantics, and texture structure between the keyframe transfer image and the standard Tujia brocade image respectively. If the calculated differences in color culture, compositional semantics, and texture structure are all within the style range of the Tujia brocade image, then the corresponding keyframe transfer image is determined to be the final fused image.

6. A traditional pattern image style transfer method based on multimodal feature decoupling and dynamic propagation, characterized in that: The method is executed to achieve all the contents of the traditional pattern image style transfer system based on multimodal feature decoupling and dynamic propagation as described in any one of claims 1-5, specifically including: Step 1: Obtain Tujia brocade images from the local intangible cultural heritage open 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, extract the texture features of the Tujia brocade image through the gray-level co-occurrence matrix, 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 parameters of the diffusion model are trained using the color features, texture features and vectorized composition features of each Tujia brocade image to generate a low-rank adaptive model that adapts to traditional patterns; the image to be transferred is split into multiple transfer images according to frames, 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 key frame transfer images. Step 3: Establish an evaluation system for the differences in the features of Tujia brocade images, and determine the range of Tujia brocade image styles based on the evaluation system. Step 4: Determine the style shift between the keyframe-transferred image and the standard Tujia brocade image. Based on the style range of the Tujia brocade image, select the final fused image from the keyframe-transferred image and process the final fused image using Poisson fusion technology to obtain the style-transferred image.

Images