Cultural Inheritance in the Digital Age: Intangible Cultural Heritage in Virtual Costume Design
Pubblicato online: 22 set 2025
Ricevuto: 13 gen 2025
Accettato: 19 apr 2025
DOI: https://doi.org/10.2478/amns-2025-0959
Parole chiave
© 2025 Li’na Zhao et al., published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.
With the rapid development of digital technology, cultural content has been more widely disseminated and more diversified forms of expression with the support of various digital platforms such as the Internet and mobile Internet. Cultural inheritance and innovation in the digital era not only maintain the basic characteristics of traditional culture, but also have a richer way of expression. Cultural inheritance and cultural development in the digital era is not only a new way of cultural development, but also a new way of cultural thinking and cultural innovation. In this context, the application of intangible cultural heritage in clothing design is an important form of cultural inheritance and innovation [1–4].
China has accumulated a rich intangible cultural heritage in its long history, and in recent years, some fashion designers at home and abroad have been looking for inspiration from Chinese intangible cultural heritage, digging inspiration, and designing a large number of distinctive clothing works, which has injected new vitality into the inheritance and protection of Chinese intangible cultural heritage. Along with the continuous transformation of the aesthetic concept of the public, people have put forward diversified and personalized demands for clothing performance, which requires the innovative application of intangible cultural heritage in clothing design, and should focus on meeting the various needs of consumers, and designers should have a deep understanding of the cultural connotation of intangible cultural heritage, so as to give the clothing works with the essence of intangible cultural heritage in the clothing design. At the same time, it is necessary to consider the aesthetic standard of modern clothing, promote the organic integration of the relevant elements of intangible cultural heritage and modern clothing design, and then promote the protection and inheritance of China’s intangible cultural heritage [5–8].
The inheritance and protection of traditional culture is a long-term and arduous project. Digital technology provides new methods and means for traditional culture protection. Through digital technology, it can realize the digital restoration of traditional culture, so that its traditional culture can be continued, but also can realize the digital preservation of cultural heritage, avoiding the loss of cultural heritage caused by the erosion of time and environment. Literature [9] suggests building digital platforms, interactive systems and competitions for intangible cultural heritage to break the limitations and practical dilemmas of traditional communication methods, so as to realize scientific protection, public recognition and innovative inheritance of cultural heritage. Literature [10] explored the role of modern information technology in the protection of ICH as well as the deficiencies and characteristics of digital protection of ICH, and outlined the potential application of computer digital technology in the protection of ICH in order to realize the digital protection of ICH. Literature [11] examined the role of digital technology in the preservation and transmission of NRH. It was pointed out that digital technology promotes the inheritance, transformation and application of NRH, but it also faces challenges in terms of management mechanism and technology. Literature [12] discusses the importance of non-legacy resources and the development status of digital protection of non-legacy in various countries, and based on the Internet thinking and user thinking, it discusses big data, artificial intelligence and other technologies, as well as the inheritance of non-legacy and the innovative development of the ways and means, with a view to providing reference for the development of non-legacy resources in the new era. Literature [13] puts forward coping strategies for the problems existing in digital virtual technology in the protection of non-heritage, by virtue of digital virtual technology to change new concepts, develop new channels, and use more ways and innovative digital virtual technology can realize the inheritance and development of non-heritage. Literature [14] analyzes the connotation of non-heritage according to its own attributes, and discusses the use of human-computer interaction, networks and other digital technologies to improve the interactivity and audience experience of non-heritage communication, aiming to provide innovative paths for the communication of non-heritage. Literature [15] examines the development of cultural heritage types and institutional protection in China, reveals the deficiencies therein, and puts forward practical suggestions based on the application of network and digital technologies for NRH protection worldwide.
Intangible cultural heritage as the crystallization of human creativity and wisdom, has a rich historical value and unique cultural charm, the application of clothing product design is innovative but also an important way of protection, while through the in-depth excavation of intangible cultural heritage and the use of it, but also for the product to inject unique cultural connotations, to enhance its cultural taste and artistic value. Literature [16] through the CLO three-dimensional virtual simulation technology, Wenzhou blue anti-dyeing and non-heritage digital empowerment combined to achieve the three-dimensional human body modeling to establish the research and development process, so as to innovate the form of Wenzhou blue anti-dyeing customized clothing design display, in order to improve the efficiency of the design of apparel products and reduce the cost, and at the same time also for the Wenzhou blue anti-dyeing of the non-heritage culture of inheritance and development of providing a new path. Literature [17] analyzes Manchu embroidery culture and specific cases based on Manchu embroidery patterns, emphasizing that combining modern clothing design with embroidery helps to promote Manchu culture and display the characteristics of clothing. Literature [18] examined the application of Hanbok in modern clothing design, emphasizing that Hanbok is the intangible cultural heritage of Fengxian County in Jiangxi Province and an important symbol of traditional Chinese culture, which is an important reference value for the creative design of modern clothing. Literature [19] based on the feasibility of combining Shuijia Horsetail Embroidery and clothing design, elucidates the specific application of non-heritage in clothing design, which not only puts forward new ideas for the inheritance, protection and development of non-heritage, but also has a reference value for the development and design of non-heritage derivatives. Literature [20] describes the application of embroidery in Chinese intangible cultural heritage in contemporary clothing. It is pointed out that embroidery is a folk craft commonly inherited in China, which possesses very high artistic value and has significant reference value in contemporary clothing design. Literature [21] outlines the application of traditional opera costume elements in modern costume design, and on this basis, it explores the creation of opera elements in modern costume design, aiming to provide reference for the development of modern costume design.
In this paper, through the intangible cultural heritage characteristic elements extraction system, the K-means algorithm and morphological methods are used to extract patterns, colors and contours from the images of intangible cultural heritage, and to construct the image processing model of intangible cultural heritage virtual clothing design. The extracted characteristic elements of intangible cultural heritage are integrated into the virtual costume design, and after 3D costume modeling, the virtual costume design based on the characteristic elements of intangible cultural heritage is carried out, so that the elements of intangible cultural heritage fit the theme expression and mood performance of the costume. The performance of the model in this paper is examined by evaluating the SSIM and PSNR values of the non-heritage images extracted and reconstructed by the model. Finally, the shadow play elements in the costume design are taken as an example, and their hue, saturation and brightness are analyzed in depth.
China’s intangible cultural heritage (ICH) is a valuable cultural treasure of the Chinese nation, carrying the genes of Chinese civilization and manifesting the aesthetic pursuit and spirituality of the nation. However, in the process of modernization, many NLCs are facing the dilemma of inheritance and are in urgent need of innovation and development. Incorporating non-legacy elements into modern clothing design is a vivid practice of implementing the spirit of this important instruction. This design concept and practice can not only let the non-heritage new vitality, but also for clothing design to inject cultural connotation, creating a unique aesthetic value.
As an important part of the fashion industry, clothing plays an important role in cultural heritage and innovation. Clothing design is a comprehensive discipline that integrates multiple attributes such as art, culture and business, and its creation process involves many fields such as aesthetics, anthropology, psychology and sociology. Introducing non-heritage elements into clothing design is an innovative practice driven by designers’ cultural consciousness and mission, which is not only the inheritance and development of national culture, but also the expansion and innovation of clothing design language.
Image extraction techniques have been widely used in medical, biological, industrial automation, military security, textile, clothing and other fields [22]. In recent years, the textile and clothing fields have also gradually used image extraction technology for clothing retrieval, textile detection and processing of various types of clothing and fabric images, etc., while the application in the images of intangible cultural heritage is less, this section uses the K-means clustering algorithm and mathematical morphology algorithm, which are commonly used in the image extraction technology, to carry out the feature extraction of the patterns, lines, and color elements in the images of intangible cultural heritage with a view to provide new ideas for the digital protection of intangible cultural heritage and cross-disciplinary application and inheritance.
Matlab GUI is a user interface that uses graphical display and is a tool for human-computer information interaction. It can not only be embedded into existing simulation programs, but also can directly display the graphical results obtained through human-computer interaction. As long as the user is familiar with the mastery of the operation process can easily and skillfully operate the interface, which provides great convenience for research workers.
There are usually two methods for user interface design: one is to use the method of writing Matlab files directly to complete the GUI design. The other is to generate the related program files through the development environment GUIDE. When GUIDE is used to build the GUI, the designed GUI interface can be stored as a FIG file, which automatically forms the corresponding Matlab file containing the initialization code of the GUI and the control code used to create the interface structure.
We adopt the method of writing Matlab files directly to realize the design of the color extraction interface, and the method of constructing the GUI with GUIDE to complete the design of the pattern and line extraction interface. The interface of the extraction system is required to present the characteristic elements of Tianjin regional folk art in real time, and the schematic diagram of the system extraction interface design is shown in Figure 1. The extraction application system is divided into four layers, namely, input layer, algorithm layer, database and application layer, the algorithm layer includes line feature extraction, pattern feature extraction and color feature extraction, while the database corresponds to the extracted elements, including line library, pattern library and color library, and finally the application layer.
System extraction interface design schematic
Different patterns may have similar shapes, and the same pattern will be deformed due to different shooting angles, the size of the original image will be changed due to the difference in shooting distance, and even the noise intensity in the segmentation process will be changed due to the clarity of the pattern, all of which will have an impact on the pattern recognition and extraction results. Therefore, researchers need to further improve and upgrade the algorithm to reduce the time-consuming feature recognition and extraction, and improve efficiency and accuracy. Some of the algorithms that have been used in recent years in pattern segmentation of apparel images are clustering algorithm, JSEG, Grab Cut, SLIC, HOG feature algorithm, LBP, CNN, etc.
Comparison found that, in addition to trying to apply the algorithms used in other fields to the field of clothing, many scholars also use a combination of commonly used pattern extraction algorithms, and different algorithms have different application areas due to different functions. Among them, clustering algorithm and convolutional neural network are more widely used in pattern extraction, and both have achieved better segmentation effect in the feature recognition and extraction of images.K-means algorithm is simple in principle, fast convergence speed, strong robustness, and is most widely used in the industry, and for the dataset of unknown characteristics, it can be used to do the preliminary operation first, and for the weak computer skills of the apparel industry, the algorithm is easy to understand and easy to use. Personnel, the algorithm is easy to understand and easy to learn, so this paper selects the K-means algorithm commonly used in clustering algorithms.
Color space and conversion
The color space commonly used in the image processing process are RGB, CMYK, CIE-Lab and HSV color space, and different color spaces can be converted by formula.
RGB color space is based on the principle of additive color mixing, by the red, green and blue primary colors of light constitutes the color space, for the three colors of the division of the range of 0-255. CMYK color space is based on the principle of subtractive color mixing, through the reflection of light to show the color of the color space, which is widely used in the industrial printing industry. HSV color space is a model based on the characteristics of human observation of color, which uses H (hue), S (saturation), and V (lightness) to describe the color characteristics, and can directly reflect the relationship between colors, with higher flexibility. Therefore, everyday designers prefer to use HSV space when expressing colors.
In order to allow users to more intuitively perceive the visual effect of color and facilitate the calculation of color, the subject will be converted from the RGB color space to the HSV color space for subsequent color value analysis.
Generally images are stored in RGB form in the computer, image processing with the help of computer c language involves opencv computer vision library, in which the image format defaults to BGR mode, so first the image BGR mode is converted to RGB mode, then RGB is used as a medium and finally converted to HSV mode. In this, the formulas involved in RGB to HSV are shown below. Let
Selection of color extraction methods
In the field of image color extraction, in order to use quantitative data information to reflect the color, scholars at home and abroad have introduced computer image processing techniques based on clustering algorithms in their research. K-means is widely used in clustering algorithms due to its simple principle and operability. Some studies have also shown that Gaussian mixture model also has a better segmentation effect on trees, drawings and other types of images. However, both algorithms have the defects that the initial clustering center is difficult to choose and the number of class clusters k value is difficult to determine, so they are optimized first.
K-means algorithm belongs to the grouping method based on the principle of distance between data, and its basic principle is: randomly select k points from the data set as the initial clustering center, calculate the Euclidean distance from each point in the data set to the initial clustering center and use it as a similarity criterion, and assign the sample points to the class clusters represented by the clustering center with the largest similarity [23–24]. Based on the similarity between the dataset and the clustering centers, the positions of the clustering centers are continuously updated until the clustering centers no longer change.
Gaussian mixture modeling is a method of clustering with statistical mixture models [25]. It is based on the principle that the probability density function of the Gaussian mixture density model, assuming that the pixel Y of the image is a mixture of multiple Gaussian distributions:
Initialize parameter
Repeat the following two steps continuously until the iteration is terminated.
Step E: Observe the data with the current solution
M Step:
Optimization of clustering algorithm
Optimization of selecting initial clustering centers
Traditional clustering algorithms for color clustering of images have two forms for the selection of the initial clustering center: hue mode and grayscale mode. Hue mode refers to the initial clustering center selected by hue, along the hue
The color matching style of non-heritage art will be different in different era background and regional environment. The optimized way of selecting initial cluster centers can better select different initial cluster centers according to the pixel distribution of different images and improve the accuracy of color extraction.
Determination of the number of class clusters k value
Traditional clustering algorithms generally need to rely on personal experience to input the estimated value of the number of clusters k, compare the segmentation effect of the image under different number of clusters and then adjust the value of k, which is only applicable to experiments with a small sample size. In order to obtain the optimal number of clusters for experimental objects, reduce manual input and improve the efficiency of experiments, the elbow method is introduced to estimate the number of clusters. The elbow method uses a formula to calculate the sum of squared errors (SSE) from the sample points of each cluster to their respective centers of mass and uses it as a performance measure, with smaller values indicating greater convergence of the clusters. In this process, a maximum number of possible class clusters is randomly specified
Mathematical Morphology is an algorithm proposed by J. Serra and Marceron in France in 1964, which is now widely used in computer vision, pattern recognition, image analysis and processing [26]. Analyzing and processing images in many fields, for example, buildings on the ground from high resolution satellite images, biological cells, bacteria, viruses, red blood cells of lizards. In medicine, such as MRI images, electrocardiograms, and even the human retina, morphology has become one of the common and necessary algorithms in the field of image processing, and the study of its special technical principles and applications is still under constant exploration and optimization.
There are two basic operations of morphology: expansion, erosion, open operation, and closed operation, and these basic operations can also be used to derive and combine into a variety of practical algorithms for morphology, such as: top hat and bottom hat transformations, hit and miss transformations, watershed transformations, morphology gradient, particle analysis, etc. Structural elements are an important concept in morphology, and the application of morphological theory to process images is also realized based on structural elements. Morphology was essentially designed for binary images and later it was continued to be used for grayscale activities and images.
Erosion is an operation that eliminates pixel points at the image boundary and shrinks the image boundary inward, it can also be used to eliminate targets in the original image that are smaller than the structural elements, and has the effect of removing noise. The operator of erosion is Θ,
The purpose of processing the image using the erosion operation is to shrink the image to be processed by one week, and the exact size of the inward shrinkage by one week is affected by the structural elements, and the simulation of the erosion operation is shown in Fig. 2.
Erosion operation simulation diagram
The principle and role of expansion is just the opposite of corrosion, expansion is an operation to expand the target image boundary. Generally used to connect the image is not continuous, there are breakpoints in the boundary, complete and smooth image boundary breakpoints will affect the image of the complete region of the subsequent calculation of statistical operations.
The operator for expansion is ⊕.
Dilation operation simulation diagram
Erosion shrinks the target area, essentially shrinking the image boundary inward. Expansion, on the other hand, expands the target area, causing the image boundary to expand outward. Since the changes brought about by erosion and expansion occur only at the edges of the image, the edges of the objects in the image can be extracted at this point by subtracting the two images. We are able to get the non-legacy pattern line contour elements after noise removal, expansion, erosion, and image binarization followed by inversion.
3D apparel modeling technology refers to the process of using computer software to transform apparel design into a three-dimensional digital model. In the growing development of digital technology today, the apparel industry is increasingly concerned about the application of 3D apparel modeling technology. 3D apparel modeling technology includes human body modeling, apparel modeling and virtual display, etc. 3D apparel modeling technology can improve the efficiency of apparel production, and can be apparel design and production process needs to be solved in a way to intuitively show the problem.
With the development of science and technology, there are now many powerful 3D design, scene creation, animation production and other modeling software on the market. In the field of apparel design, there are two main modeling methods used to create 3D apparel models: one is professional 3D apparel modeling by importing apparel prints, and common software includes Style3D, CLO3D, VStitcher, Marvelous Designer, etc. The other is to directly model apparel in 3D. The other is direct 3D modeling of garments, and such software has Blender, CINEMA 4D, Rhino, etc.
Through analysis and comparison, this thesis chooses software in the field of professional clothing design such as Style3D and CLO 3D for the practice of 3D modeling of clothing, and software in the field of 3D modeling such as Blender, CINEMA 4D and Rhino for the study of digital presentation of traditional bamboo weaving techniques.
Digital non-heritage pattern elements and clothing theme interact with each other, digital non-heritage pattern elements design strategy shown in Figure 4, from the figure can be seen, the pattern design elements because of closely follow the clothing theme for the natural environment, the social environment, the main spirit of the elements of the expression, and around this core level to start a series of design behavior, making the entire visual identity system between the elements more harmoniously The elements of the whole visual identification system are harmonized to form a whole, and the main idea is conveyed powerfully and effectively.
Digital pattern element design strategy
Digital non-heritage patterns inherently contain a new type of “text” with digital aesthetic characteristics, whether it is data or verbal text, and the process of designers using computers to create digital patterns, i.e., “coding” at the computer level, is the process of creating symbolic meanings and transforming symbols at the technological level. The process of designers using computers to create digital patterns, i.e. “coding” at the computer level, is the creation of symbolic meanings and the transformation of symbols at the technical level. The designer derives the elements and visual representations from the meaning of the theme, while the public thinks back to the meaning of the creation from the visual representation of the garment and analyzes the theme from the surface to the inside according to their own feelings. In this process, the viewers have different experiences of the meaning of the clothes according to their subjective thinking. Under the opportunity of the development of digital technology, the aesthetics of digital technology into the context of clothing aesthetics, in order to realize the significance of digital patterns applied to clothing, thus creating a new realm of clothing digital aesthetics. Designers should pay attention to the symbolic significance of the “text” in the process of digital pattern creation, touching the hearts of consumers and arousing emotional resonance, so as to make it fit the expression of clothing mood. Complete the viewer’s “decoding” process of digital patterns.
The expression path of digitized non-heritage patterns in the clothing context is shown in Figure 5. Figure 5 explains how the digitized pattern fits the pattern language with the clothing context during the creation process. It can be understood from two dimensions, one is the aesthetic dimension of “visual” and “aesthetic”, and the other is the functional dimension of “digital” and “technology”. The first is the aesthetic dimension of “vision” and “aesthetics”, and the second is the functional dimension of “digital” and “technology”, which embodies an aesthetic way of thinking characterized by digital technology, and which is interlinked, influenced and interacted with each other in the design process.
The expression path of digital design in the context of clothing
In the experiments in this chapter, the coding size is set to 1024, the input image size is 256 × 256, the batch size is 8, the discriminator enabled rounds are 10, the number of iterations is set to 500, the parameters are updated by the Adam optimizer, the initial learning rate is set to 0.0003, and the hyperparameter
In the experiments in this section, three main evaluation metrics are used to assess the quality of the reconstructed images, namely MSE, SSIM, and PSNR.
MSE: Mean Square Error, which indicates the expected value of the sum of the squares of the differences between the pixel values of the processed image and the original pixel values. For reconstructed images, a larger value indicates a poorer quality of the reconstructed image.
SSIM: Structural similarity, a full-reference image quality evaluation metric, which takes into account the three aspects of brightness, contrast, and structure, and is used to measure the similarity between images. The value of SSIM is in the range of [0,1], and the closer the value is to 1, it means that the higher the similarity between the images is, i.e., the lower the distortion of the images is.
PSNR: PSNR measures the image quality based on the mean square error between image pixels and calculates the signal-to-noise ratio between images, i.e., the ratio of the signal portion to the noise portion of an image, which is used to evaluate the performance of image compression or recovery algorithms.
In this paper, three existing state-of-the-art models are selected for experiments, namely VQ-VAE-2, VQGAN, and PeCo.VQVAE-2 is an improved version of the VQ-VAE model.PeCo is a perceptual contrast learning based model for learning data representation. It uses the perceptual contrast loss function to learn the feature representation of the data, and optimizes the quality of the data representation by maximizing the similarity within the same sample and minimizing the similarity between different samples. In this experiment, 20,000 pieces of data are selected from the dataset, of which the training set is 12,000 and the test set is 8,000, to train the four models and evaluate the quality of reconstructed NRI images.
The results of the quality assessment of reconstructed non-heritage images are shown in Table 1. The experimental results show that the method in this paper exhibits higher reconstruction accuracy and better visual quality in the garment image reconstruction task, and is able to more accurately reconstruct the textures and folds on the garment images. Compared with other methods, it performs better in preserving garment details and overall structure.
In terms of the quality assessment metrics of intangible cultural genetic images, the reconstructed images by this paper’s method achieved the best reconstruction SSIM and PSNR in both the training and test sets, which were 0.887 and 24.68 in the training set, and 0.815 and 23.95 in the test set, respectively.
The evaluation results of different methods
| Method | SSIM(train) | SSIM(val) | PSNR(train) | PSNR(val) |
|---|---|---|---|---|
| VQ-VAE | 0.795 | 0.689 | 21.56 | 20.42 |
| PeCo | 0.825 | 0.742 | 20.87 | 20.06 |
| VQVAE-2 | 0.859 | 0.793 | 22.75 | 22.38 |
| Ours | 0.887 | 0.815 | 24.68 | 23.95 |
In order to verify the effectiveness of feature matching in the discriminator, the experiments used the same configuration, while using different types of loss functions, and recorded the loss during the training process, and in order to more conveniently show the changes in the quality of the reconstructed image with the number of iterations, and also recorded the Mean Squared Error (MSE) of the original image of the immaterial cultural heritage and the reconstructed image during the training process. The mean square error and and loss of different methods during the training process are shown in Figure 6. The initial mean square error is relatively low, about 0.35, mainly because the costume image contains many white background regions. The network structure is unbalanced due to the mutual game between the GAN generator and the discriminator, resulting in large fluctuations in the model loss. However, in comparison, the overall MSE of the model in this paper is smaller. At the 2500th round, the reconstruction of this paper’s model is already comparable to that of another model at the 5000th round. By the 5000th round, the MSE of the reconstructed image of this paper’s method is reduced to as low as 0.0019, and the reconstructed image can clearly reproduce the texture and color information of the ICH.
Comparison of results for different losses
This subsection takes shadow play as an example to analyze the color composition of shadow elements in virtual costume design. According to the role classification of shadow people, five typical character role images, namely, raw role, dan role (small dan, old dan), pure role, and ugly role, are selected as the samples for this experimental test.
According to the image color extraction algorithm proposed in the previous section, the single shadow images are clustered for the first time to get the main color of each image. On this basis, the main colors obtained from the image clustering of each character type are then clustered twice using the density peak clustering algorithm to obtain the main color of each character.
In order to quantitatively analyze the color composition of the head stubble of different characters of the shadow, the distribution of the main color of each character is visualized in the HSV color space,and the deviation of the main color of each character in the three color channels of H, S and V is quantified.
The different color accounts for the five roles are shown in Figure 7, and the hue values of the colors used for the different role images are shown in Table 2. Among them, Lao Dan, Sheng, Jing and Chou all have 9 role types, and Xiao Dan has 8 role types.
Proportion of different colors in 5 roles
Hue distribution of 5 roles
| Role | Different color hue (°) | |||||||
|---|---|---|---|---|---|---|---|---|
| Red | Orange | Yellow | Yellow-green | Green | Blue-green | Cyan | Indigo | |
| Lao Dan | 345 | 50, 52, 64, 65, 70 | 154 | 191 | ||||
| Xiao Dan | 10, 21 | 36, 46 | 48, 63 | 92 | 113 | |||
| Sheng | 1, 6, 358 | 42 | 55, 56 | 131 | 165 | 213 | ||
| Jing | 1, 359 | 32, 39 | 48, 62 | 110 | 152 | 213 | ||
| Chou | 12 | 38 | 49, 53, 63, 71, 74 | 143 | 174 | |||
Combined with Figure 7 and Table 2, it is found that the main colors used in the five types of roles all contain the colors red and yellow in common, and there are differences in the types of colors used and the proportion of each color used in the five types of images due to the needs of the roles portrayed in each of the five types of images.
Among them, the image of the Old Dan has the least number of colors used, containing only four colors: red, yellow, green and cyan. As can be seen from Figure 7, in terms of the proportion of colors used, yellow accounts for the most, about 67.75%, mainly used in the hair part of the shadow figure, and this characteristic of the colors used is closely related to the actual shadow craft. In the production process, in order to highlight the aging characteristics of the character of the old Dan, a lighter color is chosen as the character’s hair color. As the shadow is carved on the treated donkey skin, the donkey skin after the oil treatment shows a light yellow color, which in turn creates the yellow character’s hair color. The other three colors, red, lime green and green, are mainly used as decorative accents for the head stubble. This bright use of color with large areas of yellow plus red, green and green accents highlights the dignified and simple character of the old Dan as well as the aging character.
The main color tones of Xiaodan are concentrated in H ∈ [10°, 113°], with a total of five colors, with yellow and green as the base tones and red, orange and yellow-green as the secondary colors. As can be seen from Figure 7, compared with the old Dan and the Sheng, Jingsheng, and Jiuzhuang roles: the five colors used by the young Dan have the most even distribution, with the proportion of each color basically ranging from 11.27% to 30.05%, and most of them are used for the head decorative colors. This colorful headdress color, and Yang carved hollow white face form a strong contrast, more highlights the exquisite beauty of Xiaodan characters.
In the clown role, yellow, green and green are the main colors, and red, orange and green are the secondary colors. Its characters use color variations, intended to portray the strange and funny character traits. The shades used by the raw and pure characters are similar, and the range of colors used is the widest among the five characters, H ∈ [1°, 213°]. Among them, red, blue and cyan colors account for more, and the overall tone is on the cool side, which highlights the majesty and seriousness of the male characters portrayed by Shengjiao and Jingshijiao.
The color saturation distribution of the main colors of the 5 characters is shown in Fig. 8, and the percentage of saturation is shown in Table 3.The selected colors of the 5 characters are dominated by medium and high saturation, and S ∈ [34, 100] accounts for 76.13% of the overall proportion. In comparison, the five characters differ in the range of saturation distribution and the proportion of medium and high saturation.
Color saturation distribution of 5 roles
Saturation deviation of main colors
| Role | Max | Min | Mean | SD | Low saturation proportion (%) | Medium saturation proportion (%) | High saturation proportion (%) |
|---|---|---|---|---|---|---|---|
| 0~33 | 34~66 | 67~100 | |||||
| Lao Dan | 85 | 0 | 52.16 | 25.18 | 12.04 | 43.98 | 43.98 |
| Xiao Dan | 93 | 0 | 67.45 | 34.06 | 24.75 | 15.45 | 59.80 |
| Sheng | 78 | 1 | 57.86 | 24.52 | 12.04 | 64.23 | 23.73 |
| Jing | 86 | 1 | 51.66 | 26.74 | 35.25 | 42.45 | 22.30 |
| Chou | 90 | 0 | 55.84 | 32.42 | 35.25 | 12.04 | 52.71 |
As can be seen in Table 3, the saturation used in the main color of Xiaodan is the highest, with a high saturation percentage of 59.8%, followed by Clown, Old Dan, Sheng and Jing. The high saturation of the small Dan color mainly appears in the character’s headdress, Dan corner headdress with very colorful headdress for the character portrayal, headdress with high saturation of color and the character of the white face to form a contrast, more show the character of the handsome and delicate. The headdresses of Shengjiao headdresses are similar to those of Danjiao, with the only difference being whether or not there are earrings or bangs on the face, so they also have high color saturation, with medium and high color saturation accounting for 87.96% of the headdresses.
Among the five types of roles, the range of color saturation chosen for the head stubble of the ugly and pure roles is the largest. In order to portray the comical and humorous characters, the clowns are not only weird and exaggerated in their image treatment, but also extremely bold in their use of color,so the color saturation involved is relatively broad. In the jingjiao characters are mostly big flower face, both red, green, black color yin carving real face, but also the white face of the evil look of the yang carving, the character of the character changes in the facial color to show. This kind of color and carving method makes the image characteristics of jingjiao characters very distinctive, so the saturation of color involves a wide range.
The proportion of brightness of different characters is shown in Figure 9, and the brightness analysis is shown in Table 4.The range of color brightness chosen for the five characters is basically the same, and it is involved in the range of 3~100 brightness, which is a large span. In comparison, there are differences in the saturation distribution range and the proportion of medium and high brightness among the five roles. The color brightness chosen for Lao Dan’s head stubble is more even than that of the other 4 roles, with less deviation. This is consistent with the hue analysis above, in which the color used for the Lao Dan category of characters is more simple, with a larger proportion of yellow. This decorative effect makes the characters have a sense of simplicity and generosity, and fits the character.
Value distribution of 5 roles
Value deviation of main colors
| Role | Max | Min | Mean | SD | Low brightness proportion (%) | Medium brightness proportion (%) | High brightness proportion (%) |
|---|---|---|---|---|---|---|---|
| 0~33 | 34~66 | 67~100 | |||||
| Lao Dan | 99 | 5 | 50.23 | 24.65 | 32.26 | 45.17 | 22.57 |
| Xiao Dan | 99 | 6 | 51.67 | 25.94 | 36.48 | 24.56 | 38.96 |
| Sheng | 99 | 6 | 47.49 | 28.75 | 45.17 | 32.26 | 22.57 |
| Jing | 99 | 8 | 49.52 | 29.04 | 32.26 | 45.17 | 22.57 |
| Chou | 99 | 3 | 56.83 | 28.68 | 22.57 | 45.17 | 32.26 |
As can be seen from Table 4, the high brightness of the colors used for the small Dan accounted for the highest percentage, which is because the character’s head stubble is mainly carved using the yang process, in addition to the facial contour lines and facial features depicted in black lines, the other parts of the skeleton white processing, so increasing the overall brightness of the colors. The overall brightness of the clown character is high and the deviation is large, and this irregular use of color helps to uglify the character’s appearance and the quirky, nonsensical character traits presented.
The article designs and develops the extraction system of genetic elements of intangible culture, uses K-means algorithm and morphological algorithm and so on to extract patterns, colors and contours of intangible cultural heredity, and applies them to the design of clothing to construct the image processing model of virtual clothing design of intangible heritage. The performance of the model and the non-heritage colors in the virtual clothing are analyzed.
The SSIM and PSNR values of the reconstructed images by this paper’s method are significantly better than those of other models, with SSIM values of 0.887 and 0.815 on the training set and test set, respectively, and PSNR values of 24.68 and 23.95, respectively.In terms of feature matching of the images, the model of this paper has a smaller overall MSE, which can more accurately reflect the texture and color of the NRH.
Taking the shadow in the virtual costume design as an example, the yellow color of the old Dan element is 67.75%, which is the highest percentage of the color. The main color tones of the young Dan are concentrated in H ∈ [10°, 113°], with yellow and green as the main colors. In the clown role, yellow and lime green are the dominant colors. The color range of the Shengjiao and Jingjiao is H ∈ [1°, 213°], in which red and blue-green are the dominant colors. The colors of the five characters are all dominated by medium-high saturation, with an overall percentage of 76.13%, and the main color of Xiaodan is the most saturated, with a high saturation percentage of 59.8%.The brightness range of the colors of the five characters is basically the same, with the brightness being in the range between [3, 100].