Research and Practice of Reconstructing Traditional Non-legacy Clothing Patterns Based on Computer Graphics Technology

Intangible cultural heritage is an important way for China to publicize and promote the deep historical and cultural heritage of the nation. The number of intangible cultural heritage in China is countless, and a number of them have been listed in the world-class intangible cultural heritage, which greatly reflects the cultural self-confidence belonging to China [1–3]. With the emergence of digital technology, the way of information dissemination has undergone a great transformation, the development of the cultural industry is also more closely linked with digital technology, network technology, etc., mankind has formally stepped into the digital era, the whole society is constantly moving towards the digital transformation, and how to apply the products of these new times to the protection of intangible Chinese cultural heritage and research work has become an important issue in the academic community [4–6].

The costumes of Chinese nationalities are the vivid carriers of their cultures, carrying their histories, cultures, regional characteristics and livelihoods, and the costumes of each nationality are a flowing epic woven through “fingertip skills” [7–8]. Intangible cultural heritage related to textile and clothing is the unique aesthetic and exquisite skills of all Chinese ethnic groups in the local specific living environment, lifestyle and unique living customs of each ethnic group. Spinning, dyeing, weaving and embroidery are the front-end crafts of clothing, and clothing is the end product. From the principle of “holistic protection” of intangible cultural heritage, the traditional crafts and folk cultures embedded in clothing must be regarded as a whole, and then combined with the whole of the cultural elements of the clothing for research [9–11]. Therefore, the living inheritance of textile and clothing NRH can help promote Chinese cultural identity and enhance cultural self-confidence and pride [12–13].

In the context of the digital era, the intangible cultural heritage of ethnic minorities in order to achieve their own continuation and development should follow the current wave of digital information technology, take advantage of the creative cultural expressions of digital technology, fully develop and utilize the value of intangible cultural heritage, and realize the digital transformation of intangible cultural heritage to achieve the real meaning of inheritance, protection and development [14–16]. Therefore, it is necessary to explore the traditional ornamental design elements of non-heritage clothing, explore its artistic value and cultural essence and its modern translation, to realize the wide dissemination of non-heritage clothing national culture in the information media adapted to the modern social communication, so as to achieve the goal of better protection and inheritance of non-heritage clothing intangible cultural heritage [17–19].

In this paper, the steps for redesigning non-heritage decorative patterns include the extraction of figurative graphics, the refinement of traditional images, and the deconstruction and re-creation of three aspects. At the same time, computer graphics technology is used to help people obtain better-fitting clothing styles and enhance the consumer experience. Simple 2D texture synthesis of apparel is formed by using square texture blocks and texture synthesis of block collage samples. Block sampling is used to ensure the randomness of the synthesised apparel texture for apparel texture synthesis. The grayscale correlation matrix is used to find the correlation value of the non-heritage costume texture and calculate the optimal size of the texture block. A Graphcut algorithm improved by A* algorithm is proposed to connect the pixel points of the region of the synthesised texture block. Use 3D reconstruction technology to combine the extracted texture blocks to reconstruct the texture of the non-heritage costume. Simulation experiments are conducted to analyze the aesthetic evaluation of reconstructed non-heritage clothing textures using computer image technology. Also combined with empirical research to analyze consumer attitudes towards reconstructed non-heritage clothing textures.

2

Application of computer graphics technology in non-heritage costume and clothing design

2.1

Refinement of typical patterns, innovative design

The redesign of the pattern modeling of non-heritage decorations can be carried out by extracting figurative graphics, distilling and deforming traditional patterns, and deconstructing and recreating three aspects.

The figurative graphics of non-heritage decorations, such as the classic image of mother butterfly, dragon and fish, and fish playing with lotus pattern, can be presented in an intuitive way to accurately grasp the patterns, shapes and colours of non-heritage decorations, and combined with the modern aesthetics, the non-heritage decorations can be presented to the public as prototypes. The techniques of collage and reproduction are usually used to combine cultural creative elements with products. Of course, such collage and reproduction should firstly be in line with the design concept of the product itself, reflecting the simple and childish beauty of the non-heritage decorations, so as to be accepted by the public.

One of the most widely used methods of expression is the distillation and deformation of non-heritage decorations. This kind of graphics is based on the traditional graphics of non-heritage, combined with the design techniques of symmetry, distance and proximity, reality and falsehood of graphic design, and the graphic re-creation of non-heritage patterns.

The design method of deconstruction and re-creation mainly involves dismantling a complete graphic, turning it into several basic elements, and then combining and reconstructing it. The bipartite continuous and quadripartite continuous patterns in non-heritage patterns are classic graphics. They are made up of multiple groups of lines or dots that are repeated, and multiple groups of geometric patterns that are similar. Select the representative cultural elements of non-heritage ornament, use the way of dislocation, cutting, continuity, arrangement, etc., and reconstruct the new graphics in line with modern aesthetics on the principle of formal aesthetics of non-heritage ornament.

2.2

The application of computer graphics technology in the design of clothing ornamentation

With the continuous development of the social economy, people’s economic expenditures on clothing have further increased. At present, with the continuous innovation of China’s computer technology, with the help of graphic image technology, it can assist the majority of consumers in completing the selection of clothes online for decoration work. Consumers can upload their favourite tattoo pictures to the relevant fitting software, and then with the help of the software, they can match the clothes they choose with their own image, and feel the effect of tattoos intuitively. At the same time, people can use image processing technology to choose clothing products that are more suitable for their own needs, so as to improve their consumer experience.

3

Extraction and reconstruction of non-heritage costumes based on computer imaging techniques

3.1

2D Texture Synthesis

3.1.1

Basic concepts

In block collage-based texture synthesis of sample images, square texture blocks are usually used. The texture block boundary is shown in Fig. 1, (the grey area in Fig. 1 is shown, and square texture blocks are used throughout this paper) [20]. The boundary of the texture block refers to the region with a certain width at the edge of the texture block, and since only the texture block overlapping with the synthesised region at the boundary in the synthesis process has a value for discussion, the boundary of the texture block usually refers to the part of the texture block overlapping with the synthesised region. As in FIG. 1, the texture is synthesised block by block from left to right and from top to bottom in scan line order (FIG. 1(a)(b)(c) demonstrates the synthesis order), where the grey block is the current texture block to be synthesised, the other square blocks are the synthesised texture blocks, and the orange contour line region is not only the boundary of the current texture block to be synthesised, but also the region overlapping the synthesised portion.

Neighbourhood Matching of Pixel Points and Boundary Matching of Texture Blocks the neighbourhood error of two pixel points is defined as the sum of the errors of the RGB values of the corresponding pixels in their neighbourhood. Specifically, the error of a neighbourhood N₁, N₂ of the same shape determined by two pixel points is defined as the L₂ distance between them, i.e.: 1 $d (N_{1}, N_{2}) = \sum_{p \in N_{1}, q \in N_{2}} s q r t {{(R (p) - R (q))}^{2} + {(G (p) - G (q))}^{2} + {(B (p) - B (q))}^{2}}$ where the function R(pixel), G(pixel), B(pixel) represents the red, green and blue primary colours of the texture image respectively.

3.1.2

Block synthesis algorithm

1)

ChaosMosaic-based texture synthesis algorithm

This paper uses a recursive system in the field of deterministic chaos. By deterministic chaos, we mean a state of disorder and irregularity generated by a nonlinear dynamic system whose history as it occurs uniquely determines future behaviour. It maps the image T (obtained by repeated mapping) to itself such that the point (xⁱ, yⁱ) in it is mapped to the point (xⁱ⁺¹, y^l+1) by means of equation (2), where l refers to the number of recursions and m refers to the length and width of the synthesised texture: 2 $\begin{array}{l} x^{i + 1} = (x^{j} + y^{j}) \mod m \\ y^{i + 1} = (x^{i} + 2 y^{i}) \mod m \end{array}$

2)

ImageQuilting algorithm

Because the synthesis method of one point at a time has a large limitation, in the synthesis process, in fact, a considerable part of the subsequent point selection has been decided by the synthesised point. Compared with the previous algorithms, this algorithm has greatly improved the time of texture synthesis, the visual effect of the synthesised texture, and avoided the problems of blurring and serious misalignment of texture elements that the previous algorithms are prone to cause.

The path with the smallest error is calculated by the following method: let B1, B2 overlap along the vertical edges, the overlap region is $B_{1}^{o v}$ and $B_{2}^{o v}$ , and the error surface is defined as $e = (B_{1}^{o v} - B_{2}^{o v})$ . The error of each point in the last row of the overlap region is obtained by the formula (3): 3 $E_{i, j} = e_{i, j} + \min (E_{i - 1, j - 1}, E_{i, j - 1}, E_{i + 1, j - 1})$

The point with the smallest error is taken and the optimal segmentation path is obtained by backtracking. For the overlap in the horizontal direction, a similar method can be used to obtain it. When both horizontal and vertical directions overlap, the two paths will meet in the middle, and the path with the smallest error is selected as the segmentation boundary. From the above algorithm, we can know that the algorithm is very simple, but the synthesis effect is very good. The problem of this algorithm is that sometimes the texture appears too much duplication and some boundaries do not match.

3)

Texture synthesis algorithm based on block sampling

To ensure the randomness of the synthesised texture I_out, a set Ψ_B is used to contain all texture blocks in the input image I_in whose boundaries and $E_{o u t}^{k}$ match. If B_{(x, y)} is used to mark the texture block in I_in whose lower left corner point is (x, y), Ψ_B can be expressed as follows: 4 $Ψ_{B} = {B_{(x, y)} | d (E_{B (x, y)}, E_{o u t}^{k}) < d_{\max}, B \in I_{i n}}$

Here d_max refers to the maximum error of the boundary zone. A texture block in Ψ_B is randomly selected as the current texture block to be synthesised B_k. If Ψ_B is empty, the texture block in I_in with the minimum boundary zone error is selected as B_k.

3.2

Texture Block Size Adaptive Algorithm in Block Texture Synthesis

3.2.1

Extracting texture correlation eigenvalues

Let a N × M full-width image have L + 1 grey levels, and G(k, l) and G(m, n) are two texture points of the image with grey values i and j respectively. obviously 0 ≤ i, j ≤ L, and 1 ≤ k, m ≤ N, 1 ≤ l, n ≤ M. So there exists a second-order joint probability density function P(i, j | d, θ). Where d is the distance between these two texture points, and θ is the angle between the line between the two texture points and the horizontal axis. Under the 3 × 3-window condition, respectively, 0°, 45°, 90°, 135°,…, 315°, the texture information on the image can be obtained by taking the grey scale correlation matrix (L × L dimensions) from a certain region or in the full image. The corresponding greyscale spatial correlation matrix can be calculated using equation (5): 5 $C (d, θ) = [P (i, j | d, θ)]$ where angles θ are spaced at 45° intervals, and the symbiotic rate P(i, j | d, θ) can be calculated as follows: 6 $\begin{array}{l} P (i, j | d, 0 °) = N O . {[(k, l), (m, n)] \\ \in (N \times M) | m - k = 0, n - l = d, G (k, l) = i, G (m, n = j)} \end{array}$ 7 $\begin{array}{l} P (i, j | d, 45 °) = N O . {[(k, l), (m, n)] \\ \in (N \times M) | m - k = d, n - l = d, G (k, l) = i, G (m, n = j)} \end{array}$ 8 $\begin{array}{l} P (i, j | d, 90 °) = N O . {[(k, l), (m, n)] \\ \in (N \times M) | m - k = d, n - l = 0, G (k, l) = i, G (m, n = j)} \end{array}$ 9 $\begin{array}{l} P (i, j | d, 135 °) = N O . {[(k, l), (m, n)] \\ \in (N \times M) | m - k = d, l - n = d, G (k, l) = i, G (m, n = j)} \end{array}$ where NO. denotes the number of elements in the set.

Clearly, C(d,θ) is a symmetric matrix and the following simple relation exists when the θ corners are 0°,45°,90°,135°,180°,225°,270°,315°, respectively: 10 $\begin{array}{l} C (d, 0 °) = C (d, 180 °), C (d, 45 °) = C (d, 225 °) \\ C (d, 90 °) - C (d, 270 °) C (d, 135 °) - C (d, 315 °) \end{array}$

In this way, the eigenvalue of texture correlation can be derived from the gray scale correlation matrix [P(i, j | d, θ)] using equation (11) [21]: 11 $F = [\sum_{i = 1}^{L} \sum_{j = 1}^{L} (i, j) P (i, j) - u_{x} u_{y}] / σ_{x} σ_{y}$ where u_x, u_y is the mean of P_x(i), P_y(j), σ_x, σ_y is the standard deviation of P_x(i), P_y(j), and $P_{x} (i) = \sum_{j = 1}^{L} P (i, j)$ , $P_{y} (j) = \sum_{i = 1}^{L} P (i, j)$ .

3.2.2

Optimal Size of Texture Blocks

Therefore, the optimal size of the texture block is calculated based on the texture correlation eigenvalues using equation (12) [22]: 12 $K = 9 e^{1 / F}$

3.3

An improved texture synthesis algorithm

3.3.1

Graphcut Algorithm

The Graphcut algorithm takes the pixel points in the overlap region of synthesized blocks A and selected synthesized blocks B. Each pixel point in the overlapping region corresponds to each vertex in the graph, and the corresponding edges are established with a four-adjacent relationship between the pixel points, while the non-overlapping portions of the two pixel blocks of A and B are abstracted as a source and a sink point, respectively, in the graph, and are connected to each pixel point in the edges of the overlapping region [23]. The weights of the edges between each neighboring vertex in the graph are calculated using equation (13) to obtain the error values of the neighboring pixel points: 13 $M (s, t, A, B) = ▯ A (s) - B (s) ▯ + ▯ A (t) - B (t) ▯$ where M(s,t, A, B) represents the weight of the edge between adjacent pixel points s, t, A(s) represents the color value of the pixel points of the overlapping region in the composited block, and B(s) represents the color value of the pixel points of the overlapping region in the matched block.

3.3.2

Improved Graphcut Algorithm

In order to make the selected optimal path does not contain such high weighted edges, this paper improves the Graphcut algorithm and proposes a new algorithm to find the optimal path to avoid the existence of high weighted edges on the path, so that the discontinuity of the inter-seam between blocks is not easy to be noticed by the human eye.

The idea of the algorithm is as follows: using the idea that the Graphcut algorithm views the pixel points in the overlapping region as a weighted graph, the weights of the edges of the neighboring pixel points are computed using Eq. (13) in order to obtain the value of the error between the neighboring pixel points. The maximum weighted edges and the sum of weights of these shortest paths are compared, and a path with no high weights on the path and a smaller sum of weights is selected as the optimal path.

Compared with other algorithms for finding the shortest path, the time complexity and space complexity of the A* algorithm is only O(n) (where n is the number of nodes extended by the A* algorithm), so it is clear that the A* algorithm has a significant advantage. Here we introduce the A* algorithm.

The A* algorithm is an efficient algorithm applied to the optimal path search between points. A* algorithm is a kind of heuristic search algorithm. By heuristic search algorithm, we mean an algorithm that preferentially searches along nodes with enlightening and specific information because these nodes may be on the optimal path to reach the goal based on an evaluation function of the nodes. The idea of A* algorithm is the following three. 1)

Assume that there is an evaluation function f that can help determine the next optimal node to expand. A convention is used so that a small value of f indicates that a good node has been found.

2)

The next node to be extended is the node with the smallest value of f.

3)

The process terminates when the next node to be extended is the target node.

3.4

Algorithm for reconstructing 3D model based on color texture of dress view

3.4.1

Three-dimensional model reconstruction

The problem to be solved by this algorithm is to assign color values to the facets partitioned in planes so that 1 set of input images maximally satisfies the dress view completeness, i.e., the image drawn at the same viewpoint is as close as possible to the original image. Here the 3D view space is partitioned with 3 sets of parallel planes distributed in 3 coordinate planes and the surface of the view is a Lambertian surface, so that the radiance of each point on the view surface is isotropic in character and it is possible to describe the radiance of the point in terms of color. For this reason, the reconstruction yields a new projection of the 3D model that accurately maintains the color, texture, and pixel resolution of the original image.

Definition 1

Suppose a finite set of opaque, colored facets view Θ the surface to the set Ω of all facets in view space Φ obtained by dividing, under facets and pixels infinitesimally small, the facets that are in the same plane as a facet layer Φ_θ, and set Θ as a plane parallel to the facet layer as a reference plane, with the reference plane located outside of the reconstruction space and the normal vectors of facets pointing to the reference plane, and the distance between Φ and Θ is ||Φ||_Θ, then: 14 $Φ_{Θ}^{d} = {Φ | | | Φ | |_{Θ} = d}$ where Θ denotes a three-dimensional view, such as a BMP, JPEG image, or MPEG video, Φ denotes the view space, Ω denotes the set of all pixels on an image, and d is the distance of Φ from the reference plane Θ.

Definition 2

Suppose that the set Φ is formed from all the facets in the reconstruction space, i.e., $Φ = \cup_{i = 1}^{'} Φ_{Θ}^{d} i$ , d₁ < d₂… < d_r then all the facets that make up the surface of the view space are contained in Φ.

Definition 3

Suppose that view Θ and an image Ω with viewpoint at v, there exists a face sheet Φ ∈ Θ, Φ visible to V, projecting Φ onto Ω yields pixels p, Φ = Θ(p), and for each image Ω_n in a set of images Ω₁,Ω₂,…,Ω_n and for each pixel p on Ω_n, there exists a face sheet Φ ∈ Θ such that Φ = Θ(p), the view Θ is complete for the set of images, where the color of pixel p ∈ Ω is color(p, Ω), and the color of face sheet Φ is color(Φ, Θ).

Definition 4

Suppose that for a complete view Θ, there is there is color(p,Ω) = color(Θ(p),Θ), ∀_p ∈ Ω, ∀Ω ∈ {Ω₁, Ω₂,…,Ω_n}. If view Θ coincides with image Ω₁, Ω₂,…, Ω_n, then for this set of input images Ω₁, Ω₂,…, Ω_n, the resulting set of facets represents the 3D view model as: 15 $Θ = {Φ_{p} |_{p} \in Ω_{i}, 1 \leq i \leq n}$ where Φ_p is the facet that is consistent with the input image.

To ensure that the reconstruction process is progressively consistent, i.e., each step of the reconstruction is consistent with the input image, a simulation consistency definition of the incomplete set of facets is introduced5.

Definition 5

A facet set is said to emulate the input image consistently by assuming that its projection is identical to the overlap of each input image. That is, for every facet Φ ∈ Θ′ in a facet set Θ′, then, for pixel point p ∈ Ω_Ω, q ∈ Ω_j. on input image Ω₁, Ω₂,…,Ω_n, facet Φ = Θ(p) = Θ(q), then: 16 $c o l o r (\begin{matrix} p, Ω_{Ω}) = c o l o r (q, Ω_{j}) = c o l o r (Φ, Θ) \end{matrix}$

If, Θ′ is simulation consistent for image Ω₁, Ω₂,…,Ω_n and for the set of facets: 17 $Θ_{Θ}^{d} = \cup_{i = 1}^{r} {Φ | Θ \in Φ_{Θ}^{d} i I Θ, d_{i} < d}$

According to the above 3D view model coherence formula (15) with facet set simulation consistency (17), there are the following 2 theorems, Theorem 1 If Θ is a coherent view, then for each Φ ∈ Θ, ${Φ} \cup Θ_{Θ}^{d}$ is a simulation consistent facet set. Theorem 2 if view Θ is complete and for each Φ ∈ Θ, $Φ | \cup Θ_{Θ}^{d}$ is a simulation consistent face sheet set, then Θ is a coherent view.

3.4.2

Reconstruction of non-heritage clothing ornamentation

1)

Data structure of the view model

Known, defined: 18 $\begin{array}{l} {\vec{Θ}}_{i} Ω = {Φ | Φ \in Θ, | | Φ | |_{Θ} \leq d_{i}}, {\vec{Θ}}_{i}^{'} Ω = {Φ | Φ \in Θ, | | Φ | |_{Θ} = d Ω}, \\ Φ^{″} Ω = {Φ | Φ \in Φ Ω, | | Φ | |_{Θ} \leq d_{i}}, Φ_{i}^{''} = Φ_{Θ}^{d} k - Φ_{i}^{'} Φ_{1}^{''}, 1 \leq k \leq r \end{array}$ $$\begin{array}{l}{{\vec \Phi }_i}\Omega = \left\{ {\Phi |\Phi \in \Theta ,||\Phi |{|_\Theta } \le {d_i}} \right\},{\vec \Theta}_{i}^{\prime}\Omega = \left\{ {\Phi |\Phi \in \Theta ,||\Phi |{|_\Theta } = d\Omega } \right\},\\{\rm{ }}{\Phi ^{\prime\prime}}\Omega = \left\{ {\Phi |\Phi \in \Phi \Omega ,||\Phi |{|_\Theta } \le {d_i}} \right\},\Phi _i^{\prime\prime} = \Phi_{\Theta}^{d}k - \Phi_{i}^{\prime}\Phi _1^{\prime\prime},1 \le k \le r\end{array}$$

Then there are ${\vec{Θ}}_{1} \in Φ_{1}^{''}$ ${\overrightarrow \Theta _1} \in \Phi _1^{''}$ , and $Φ_{1}^{''}$ $\Phi _1^{''}$ simulations are consistent.

PROOF: Assuming that ${\vec{Θ}}_{i} - 1 \in Φ^{''}$ ${\overrightarrow \Theta _i} - 1 \in \Phi ''$ , $Φ_{i - 1}^{''}$ $\Phi _{i - 1}^{''}$ simulate consistently, by Consistency Theorem 1, there is $Φ_{i}^{''} = Φ_{i - 1}^{''} \cup Φ_{i}^{'''}$ $\Phi _i^{''} = \Phi _{i - 1}^{''} \cup \Phi _i^{''''}$ simulate consistently.

Analyze and expand according to Definition 2, i.e: 19 $Φ_{0} = Φ = \cup_{i = 1}^{r} Φ_{Θ^{d}}^{d}, d_{1} < d_{2} < ... < d_{r}$

The projection of Φ in the projection of $Φ_{k}^{'} = {Φ | Φ \in Φ_{Θ^{k}}^{'}, | Φ} \cup Φ_{k - 1}$ does not coincide with the overlapping portion of the input image 1 ≤ k ≤ r: 20 $Φ_{k} = Φ_{k - 1} - Φ^{'}, 1 \leq k \leq r$

From Eqs. (18) and (19), we can see that $Θ^{'} \subseteq Φ_{i}^{''}$ $\Theta ' \subseteq \Phi _i^{''}$ , and therefore have $Θ_{i} = Θ_{i - 1} \cup \oplus_{i}^{'} \subseteq Φ_{i}^{''}$ ${\Theta _i} = {\Theta _{i - 1}} \cup \oplus _i^\prime \subseteq \Phi _i^{''}$ By induction, step-by-step calculation, we have: $Φ_{r}^{''}$ $$\Phi _r^{''}$$ simulation consistent, $Θ = Θ_{r} \subseteq Φ_{r}^{''} = Φ_{r}$ $$\Theta = {\Theta _r} \subseteq \Phi _r^ = {\Phi _r}$$ . Since Θ⋂ |Φ| Φ ∈ Φ_r, Φ are inconsistent with the input image = ∅, we have Θ ∈ Φresult, and Φresult is the set of simulation consistent facets. It is clear that Φresult is complete since the reconstruction process removes only the facets in Φ that are inconsistent with the input image. By the coherence theorem 2, Φresult is a coherent view, i.e., Φ_result =|Φ| Φ ∈ Φ_r, Φ_r of the projections of Φ are consistent with the overlapping portion of the input image, then Φresult = Θ, and hence, Φresult = ⊕ is the model of the desired view.

2)

Image Consistency Processing

The consistent processing and dynamic distribution of weights for the combined image features need to solve the two kinds of features, color and texture, and their combination. For color features, this paper adopts Gaussian consistency, so that a small number of oversized or oversmall element values have little effect on the distribution of element values after consistency, and when calculating the similarity distance, the components can be made to have the same weight.

Definition 6

Assume that for any view Θ_i, its N -dimensional feature vectors F_i = [f_i,1, f_i,2,…,f_i,n], λ(P) are a set of cumulative histogram expressions with different ranges of values, which are computed according to the algorithm for cumulative histograms using the Gaussian consistent treatment formula, i.e.: 21 $λ (P) = (λ_{x 1}, λ_{x 2}, λ_{x 3}, \dots, λ_{x i}, \dots, λ_{x n})$ 22 $f_{i, j}^{(N)} = (f_{i, j} - m_{j}) / σ_{j}$ where $f_{i, j}^{(N)}$ denotes the individual f_i,j transformations into consistent eigenvalues with a N(0,1) distribution, and m_j and σ_j denote the mean and standard deviation of the distance values, respectively.

If 3σ_j is used for consistency, the probability that the value of $f_{i, j}^{(N)}$ falls in the interval [-1, 1] can be up to 99%. In practice, the values of f_i,j outside the [-1, 1] interval are set to one 1 or 1 to ensure that all values of f_i,j fall in the [-1, 1] interval. In this way, the feature vectors of the colors, i.e., the consistent cumulative histogram for each image, are obtained.

For texture features, according to the texture algorithm of the co-production matrix, the 4 parameters of texture consistency, texture contrast, texture entropy, and texture correlation Q1 ~ Q4 can be obtained from a 1 grayscale co-production matrix. Again, Gaussian Consistency is applied to these parameters to obtain the feature vectors of the texture.

The external shape combination, using Gaussian consistentization, is to make each eigenvector of the combined features have the same position in the similarity distance calculation, and is done by linear transformation so that 99% of the $D_{I, Θ}^{(N)}$ values fall in the interval [0, 1]. Its main calculation formula is: 23 $D_{I, J} = dis(F_{I}, F_{J})$ 24 $D_{I, Θ}^{(N)} = [(D_{I Θ} - M_{Θ}) / (3 σ_{Θ}) + 1] / 2,$ where D_I,J denotes the similarity distance between the feature vectors F_I and F_J corresponding to the 2 images I and J, I, J = 1,2,…,MI ≠, M are the number of images in this image library, and $D_{I, Θ}^{(N)}$ denotes the similarity distance of the view Θ.

4

Research and analysis of non-heritage costume ornamentation

4.1

Aesthetic evaluation of non-heritage clothing ornamentation reconstructed by computer image technology

4.1.1

Image database

The reconstruction of non-heritage clothing ornamentation using computer graphics techniques generates a total of 3654 pictures in the Datta image library, and each ornamentation picture will be scored by more than 10 users, with scores ranging from a minimum of 1 to a maximum of 7, with higher scores representing the more users consider them to be good-looking.

When using picture evaluation as a classification problem, the difference between low and high beauty scores should be as large as possible, considering the reduction of noise interference from anomalous samples and the avoidance of excessive classification errors. Therefore, with reference to Datta’s study, pictures with scores higher than 5.8 are considered to be high aesthetic (beautiful), pictures with scores lower than 4.2 are considered to be low aesthetic (unattractive), and pictures with intermediate scores are not considered. Thus, there were 632 “beautiful” images and 823 “unattractive” images, for a total of 1,455 images.

4.1.2

Algorithm performance

Fig. 2 shows the comparison of error rate and ROC performance of different methods, Fig. (a) shows the ROC performance and Fig. (b) shows the recognition error rate, which compares the error rate of pattern recognition of the existing non-legacy clothing and apparel reconstruction methods and the method of this paper, and it can be seen that the error rate of this paper’s method is the lowest is 22.587% from the figure. Additionally, the ROC graph is used to compare classification performance, and it is a commonly used method for comparing classification performance. When the threshold value of clothing ornamentation changes, different true positive and false positive rates can be obtained, and the graph is made according to the relationship between the two rates. Where the area under the curve is larger, the better the performance of classification is represented. As can be seen from Fig. (a), the area of the classification model in this paper reaches a maximum of 84.899%, so the performance should also be optimal.

As for the clothing pattern synthesis model, the Graphcut algorithm was applied for synthesis as described in the previous section. As for the image samples, 3654 images from the Datta image library are used. In this paper, the A* method is used for multi-parameter optimization, and the optimal parameters γ =0.0658, C=10, ε =0.5285. The performance of the model is also evaluated through 5-fold validation. The range of the final predicted scores is consistent with the range of the actual scores which are from 1 to 7. Figure 3 shows the regression performance comparison between this paper’s method and other algorithms, Figure 3 shows some commonly used indicators for evaluating regression performance and compares this paper’s method and other algorithms for comparison. Among them, the correlation of this paper’s method for the reconstruction of non-heritage clothing ornamentation is 0.748, while the correlation predicted by the Datta method is only 0.675, and for the various error indicators, this paper’s method also has a smaller error compared to the Datta method.

4.1.3

Comparison of ratings

The scores of non-heritage clothing ornamentation and the scores of clothing ornamentation reconstructed by computer imaging technology are compared in Figure 4. The vertical axis of the figure is the score after the reconstruction of the algorithm, and the horizontal axis is the original score of non-heritage clothing ornamentation, and it can be seen that both of them are in a certain linear relationship, in addition to the sporadic scattering of a few points in the figure, the other points in the figure, with the rise of the horizontal axis, and the vertical axis also increase linearly within the range of 4-6 points.

4.2

Consumer Attitudes toward Reconstructed NRM Clothing Patterns

4.2.1

Consumer Preferences for Pattern Arrangement

According to the traditional cultural relics of intangible cultural heritage clothing patterns in various periods, it is summarized that the intangible cultural heritage decoration patterns of transmissionclothing can be divided into: “horizontal composition”, “linear composition”, “oblique composition” and “group pattern composition”, which are changed and combined under the size, distance and decoration posture of intangible cultural heritage clothing decoration elements. Figure 5 shows the score of consumers’ preference for the arrangement of intangible cultural heritage clothing ornaments, and it can be concluded that the consumer groups in the 21st century are more inclined to group pattern and oblique composition, with average scores of 7.025 and 6.896, respectively.

4.2.2

Feedback on Consumer Identification

Through the SPSS26 software, the consumer group identity feedback analysis, Table 1 for the non-heritage clothing tattoo satisfaction evaluation score mean value, as shown in Table 1, the table are measured using a 7-level scale, the score from 1 to 7 expression, 4 for the center of the numerical axis, expressed in the middle of the attitude, the 1-3 interval on behalf of the no, agree which 1 represents “very much disagree “The value of 5-7 represents agreement, where 7 represents “strongly agree”. The mean value of consumers’ interactive sense, aesthetic sense and overall evaluation of non-heritage clothing pattern design is above 6 points, and consumers are very satisfied with the evaluation of these three items, while the evaluation of the fashion sense and the price lower than 300 yuan interval is above 5 points, indicating that consumers are more satisfied with these two items.

Table 1.

The mean of satisfaction of the garment

/	N	Minimum value	Maximum value	Mean	SD
1.The design of the garment is fashionable	185	1	7	5.562	1.264
2.The design of the garment is interactive	185	3	7	6.065	1.038
3.The design of the unlicted clothing is aesthetic	185	2	7	6.065	0.963
4.General evaluation of the design of non-reprinted clothing	185	2	7	6.065	0.936
5.The price is located within 300 yuan	185	1	7	5.978	1.147
6.The price is located within 700 yuan	185	1	7	4.359	1.758
Valid case number	185

4.2.3

Description of Consumers’ Attitude towards Non-legacy Clothing Patterns

Using the Likert five-level questionnaire set from the user’s appearance experience, wearing experience, emotional experience and quality experience of the four dimensions of the question to understand the consumer’s willingness to buy as shown in Table 2, showing some data fluctuations, the average of the four experience data were 3.345, 3.248, 3.485 and 3.321, the vast majority of respondents have expressed their fondness for the non-heritage clothing ornament and think that that it brings positive emotions.

Table 2.

The user is descriptive statistics of non-relicted clothing

/	N	Minimum value	Maximum value	Mean value	Standard deviation
Appearance experience	185	1.358	5	3.345	1.026
Dress experience	185	1.356	5	3.248	0.975
Emotional experience	185	1.352	5	3.485	0.975
Quality experience	185	1.356	5	3.321	1.065

5

Conclusion

In this paper, based on the synthesis algorithm of block, combined with the shortest path optimization A* algorithm, an improved texture synthesis Graphcut algorithm is constructed to extract the texture features of the non-heritage clothing and realize the reconstruction of the non-heritage clothing ornamentation through the three-dimensional model. Combining simulation experiments and empirical investigations, we analyze the rendering effect of reconstructing non-heritage clothing ornamentation. 1)

In the algorithm performance comparison, the ROC curve area of this paper’s model reaches a maximum value of 84.899%, and at the same time the error rate is the lowest, only 22.587%, so the performance is optimal among the four algorithms.

2)

Consumers preferred the group and diagonal compositions for the patterns generated by the computer graphics technology, with the average scores of 7.025 and 6.896, respectively, and the mean values of the interactive sense, aesthetic sense, and the overall evaluation of the reconstructed NRM clothing pattern were above 6 from the feedback of the consumers’ recognition level, indicating that the consumers were very satisfied with the three items.

3)

In the survey of purchase intention from the four dimensions of appearance, wearing experience, emotional experience and quality experience of the non-heritage clothing tattoos, the mean values of the four experience data are 3.345, 3.248, 3.485 and 3.321 respectively, which indicates that most consumers love the non-heritage clothing tattoos and believe that they can bring positive emotions.

Language:: English

Publication timeframe:: 1 times per year
Journal Subjects:: Life Sciences, Life Sciences, other, Mathematics, Applied Mathematics, General Mathematics, Physics, Physics, other

Journal RSS Feed

Research and Practice of Reconstructing Traditional Non-legacy Clothing Patterns Based on Computer Graphics Technology

Xiaoxuan Nie

Li’na Zhao

Published Online: Mar 19, 2025

Received: Oct 24, 2024

Accepted: Feb 18, 2025

DOI: https://doi.org/10.2478/amns-2025-0406

KeywordsBlock sampling texture synthesis, Grayscale correlation matrix, Computer graphics, Graphcut algorithm, Non-heritage clothing ornamentation

© 2025 Xiaoxuan Nie et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Keywords
Block sampling texture synthesis, Grayscale correlation matrix, Computer graphics, Graphcut algorithm, Non-heritage clothing ornamentation