Geometric modeling and computer-aided creation methods in traditional cultural costume design

The basic process of apparel design is shown in Figure 1, which mainly includes six tasks: measuring the body data of the customized person, determining the apparel design style, drawing the apparel design plate, sample production, customized person’s sample fitting, pattern correction, and production of ready-to-wear clothes. For customers, the core of clothing design focuses on the fit of the clothing design, the selection of clothing styles, and the time period of customized clothing. For businessmen, the core of the customized business lies in the five major aspects of clothing style design, human body data measurement, pattern making, sample correction, and garment production.

Figure 1.

Basic process of clothing design

Therefore, the digital apparel design process needs to meet the needs of both the customizer and the merchant. The virtual garment design should be able to provide a clear and realistic appearance of the garment. The virtual clothing file should be able to export the clothing design pattern file that meets the individual figure of the customizer. The virtual clothing model should be able to facilitate the merchant in modifying the details of the clothing design according to the customizer’s requirements. The virtual clothing simulation effect should accurately reflect the customization of real clothing designs and the dress effects of the four major functions. All these requirements need to be met with the corresponding CLO3D tools. Thus, the main CLO3D tool sets involved are summarized and sorted out based on the functional characteristics of each tool. The study is based on CLO3D version 5.2. The main tool sets used are Virtual Model Editor, Virtual Model Tape, Virtual Model 3D Pen, 2D Sheet Edit, Fabric Attribute Editor, Sewing and Fabric Simulation Tools.

The traditional pattern making methods are mainly three-dimensional cutting through the mannequin and flat pattern making through plane geometry. In the design and production of customized garments, the three-dimensional cutting method requires the customization of the mannequin to match the body shape of the customized person, which consumes more fabric. Flat pattern making is not intuitive enough and some styles cannot be created using it. And both methods need to sample sewing and correct it. However, this link is also time-consuming and labor-intensive. Therefore, based on the analysis of the customized clothing customization process and the understanding of the function of the CLO3D tool, a new customized clothing design and production method has been designed. The process is as follows: 1)

Determine the clothing style and mark the appropriate virtual model tape on the 3D window virtual model according to the clothing style.

In terms of the composition of the three-dimensional virtual fitting technology, it mainly includes three-dimensional body measurement technology, three-dimensional body modeling technology and three-dimensional clothing modeling technology, which interact with each other to determine the final simulation effect. Among them, three-dimensional body measurement technology for three-dimensional human modeling provides an important data base, only in the three-dimensional body measurement link to achieve a certain degree of accuracy, in order to generate a sufficiently accurate three-dimensional human body model, and then carry out a series of three-dimensional virtual fitting activities. The three-dimensional human body modeling technology is used to display three-dimensional clothing on a digital virtual table, allowing for the three-dimensional clothing to be worn on the virtual table for display.

Noise can be categorized in various ways. From the statistical point of view, any statistical feature that does not change with time is called smooth noise, and the statistical feature that changes with time is called non-smooth noise. From the statistical point of view of noise amplitude distribution, its density function has Gaussian and Rayleigh type, which are called Gaussian noise and Rayleigh noise respectively, and the noise with uniform spectral distribution is called white noise.

The following four types of noise are generally common in image filtering:

Let f(x, y) be the image signal, n(x, y) be the noise, and the output after the noise affects the signal is g(x, y). 1)

Additive noise: (1)g(x,y)=f(x,y)+n(x,y)

Its output waveform is a superposition of noise and signal, characterized by n(x, y) and signal-independent. Such as the general electronic linear amplifier noise, regardless of the size of the input signal, its output is always superimposed with the noise. 2)

Multiplicative noise: (2)g(x,y)=f(x,y)[1+n(x,y)]=f(x,y)+f(x,y)n(x,y)

Its output is a superposition of two parts, the second noise term signal is affected by f(x, y). The larger f(x, y) is, the larger the second term is, i.e. the noise term is limited by the signal. For example, optical quantum noise and negative grain noise all increase with signal. Multiplicative noise model and its analytical calculations are more complex, usually the signal changes are very small, the second term is approximately unchanged, at this time, the model can be used to deal with additive noise. Usually, it is always assumed that the signal and noise are independent of each other. 3)

Quantization noise

Quantization noise is the main source of noise in digital images, its size shows the difference between the digital image and the original image. The best way to reduce this noise is to use the optimal quantization measure of selecting the quantization level according to the gray level probability density function.

For each image point (m,n) in a given image [f(i,j)]_N×N, the average gray scale value of all M pixel points in its neighborhood S is: (3)f¯(m,n)=1M∑(i,j)∈sf(i,j)

Considered as the grayscale of the image pixel (m,n) after the neighborhood averaging technique, i.e., replacing the original grayscale of a pixel with the average of the grayscale of all pixels in the neighborhood of that pixel. The shape of the neighborhood S determines in what kind of geometric space to average, and its size determines for how many values to average, the shape and size of the neighborhood S should be determined according to the image characteristics. The shapes generally taken are square, circle, cross, line, circle, etc. The shape and size of S can be kept constant during the whole image processing, or can be changed according to the local statistical characteristics of the image: point (m,n) is generally located in the center of S, for the noise smoothing of the boundary points, (m,n) can be not necessary to reside in the center of S.

If S is a neighborhood of 3 × 3 and point (m,n) is located in the center, then: (4)f¯(m,n)=19∑i=−11∑j=−11f(m+i,n+j)

Let noise n be additive and spatially uncorrelated with expectation zero and variance σ², g be an undisturbed image, and noise-containing image f be at point (i, j): (5)f(i,j)=g(i,j)+n(i,j)

The image processed by neighborhood averaging is: (6)f¯(m,n)=1M∑(i,j)∈sf(i,j)=1M∑(i,j)∈sg(i,j)+1M∑(i,j)∈sn(i,j)

Because of the randomness of the noise and the fact that neighborhood averaging can reduce the variance of the noise, a threshold T can be appropriately chosen so that when the absolute value of the difference between the gray level of a pixel point and the average value of the gray level of the neighborhood is greater than this threshold T, so that the gray level of this point is equal to the average value of the neighboring gray level, or else it is still the original gray level, that is: (7)f(m,n)={ 1M∑(i,j)∈sf(i,j),|f(m,n)−1M∑(i,j)∈sf(i,j)|≥T f(m,n), Other

This allows larger noise to be filtered out with little effect on the signal. Threshold T can be determined according to the general or local characteristics of the image, e.g. it is chosen to be equal to a number of times the mean square deviation of the image gray level, or it is chosen to be a fraction of a gray level.

The purpose of image enhancement is to process an image in order to improve its visual appearance or to make it easier for a person or computer to analyze and process. The process of image enhancement highlights a part of the information while it may suppress another part.

Currently used enhancement techniques can be categorized into image domain based methods and transform domain based methods depending on the space in which their processing takes place.

In the first category, the processing is performed directly in the space where the image is located, i.e., the pixels are manipulated directly in the space consisting of pixels;

In the second category, the image is processed indirectly in the transform domain of the image.

Also, within the first class of methods there are two groups:

One is pixel-point based, which means that each processing of the image is performed on each pixel, and the enhancement process processes each pixel independently of the other pixels;

The second is template-based, that is, each processing of the image is performed on small sub-images (templates). The null domain enhancement method can be expressed as equation (8): (8)g(x,y)=EH[f(x,y)]

where f(x, y) and g(x, y) are the images before and after enhancement, respectively, and EH represents the enhancement operation. If EH is defined on each (x, y), EH is a point operation; if EH is defined on some neighbor of (x, y), (x, y) becomes a template operation. If the template is smallest, it can be a pixel point, at which point it becomes a point operation, so point operations can be seen as a special case of template operations.

The null domain transformation enhancement can be expressed as follows:

When the value of g(x, y) depends on the value of f(x, y) at (x, y), and EH is a grayscale transformation. If s and denote the grayscale values of f(x, y) and g(x, y) at position (x, y) respectively, then at this point equation (8) can be written as: (9)t=EH(s)

A few common types of enhancement methods based on point operations, also called grayscale transformations, are: 1)

Direct transformation of each pixel in f by EH operation to obtain g.

Canny edge detection is a relatively new edge detection operator with good edge detection performance, which has been more and more widely used in image processing. It designs the Canny edge detection algorithm based on the optimal criterion for image edge detection: 1)

Convolution is first performed with a 2D Gaussian filter template to eliminate noise.

In order to give a simple evaluation of the final simulation design work, a questionnaire was used. The questionnaire contains two aspects, on the one hand, it is the survey of the degree of traditional cultural wind of each type of simulation design, and the available items are divided into six levels from 0 to 5, in which, 0 means no degree of traditional cultural wind, 1 means weak degree of traditional cultural wind, 2 means weak degree of traditional cultural wind, 3 means medium degree of traditional cultural wind, 4 means strong degree of traditional cultural wind, and 5 means strong degree of traditional cultural wind; On the other hand, it is the evaluation of the simulation design effect of each type, and the available items are divided into four levels from 0 to 3, where 0 indicates bad design effect, 1 indicates average design effect, 2 indicates good design effect, and 3 indicates very good design effect. After the questionnaire was designed, it was distributed by e-mail to 22 interested persons, including 2 professors of the School of Fashion, 17 students of fashion, 2 designers and 1 other person. A total of 22 copies were distributed and 22 copies were recovered, with 22 valid questionnaires, a recovery rate of 100% and a validity rate of 100%. After the questionnaires were recovered, the questionnaires were subjected to simple mathematical processing, and the data obtained (which were retained in bits of decimals) are shown in Table 5.

As can be seen from the table, the evaluation of the degree of traditional cultural flavor of the simulation design of the 11 types is mainly concentrated between 3-4, which belongs to the category of traditional styles from medium to strong, and the order of the strength and weakness of the traditional cultural flavor of the simulation design of the various types is basically the same as the order of the strength and weakness of the original types, and the evaluation of the design effect is mainly concentrated between 2-3, which belongs to the category of good, and the best is the design of the 2 and 3 types. That is, according to the 11 types of design methods summarized in this thesis and the general application of the techniques for each design element to express the traditional cultural winds, the simulation design completed does convey the corresponding degree of traditional cultural flavors and the effect is good.