Abstract: The curved shape of female breasts is complex, and the breast shape of women wearing the same cup bra is also different. The existing size parameters such as linearity, circumference, volume and two-dimensional angle are difficult to effectively represent the three-dimensional shape of the breast, thus affecting the fitness of the bra. The purpose of this paper is to propose a subdivision method based on spatial vector angle representation of breast shape, so as to realize the classification and discrimination of three-dimensional breast shape of young women.
In this paper, three-dimensional point cloud data of 209 young women aged 18-25 were scanned and obtained. With the help of auxiliary points, lines and planes, six breast spatial vector angles were constructed as clustering indicators. The optimal clustering number was determined by elbow method and K-means was used for clustering. Learning vector quantization (LVQ) was used to construct a breast shape discrimination model, and the fitness of the subdivided bra was evaluated through digital clothing pressure and real fitting experiments.
In this paper, the breast type 70B, which occupies the largest proportion under the Chinese standard in the sample, was taken as the research object, and the spatial vector angle was used to represent the stereoscopic shape of the breast. Six spatial vector angles that could represent the stereoscopic shape of the breast were constructed, including four local spatial vector angle parameters that represented the upper left, upper right, lower left and lower right shape of the breast divided by BP point as the center. Two global spatial vector angle parameters were used to characterize breast stiffness and sagging. By using k-means clustering, the breast shape can be divided into moderate and introverted (50.00%), flat and low-cut (16.07%), and full and expanded (33.93%) one. Digital clothing pressure and real fitting experiments show that the fit of the subdivided bra is better than the benchmark bra. The LVQ neural network model was established to identify breast shape of young women with an accuracy of 93.33%.
In this paper, the three-dimensional shape of female breasts is represented by the spatial vector angle to realize the morphological subdivision and discrimination of female breasts. The constructed model of breast shape LVQ neural network can provide theoretical reference for bra customization. Digital clothing pressure and real fitting experiments show that the subdivided bra is better than the benchmark bra, which provides a reference for different types of breast bra fitting structure design.

Key words: space vector angle, 3D anthropometry, breast shape, LVQ neural network, fitness

摘要： 采用空间向量角来表征乳房形态，提出了一种青年女性乳房三维形态分类方法。首先获取209名18~25岁青年女性的乳房三维点云数据，并构建了6个乳房形态空间向量角作为聚类指标，采用k-means聚类，以手肘法确定最佳聚类数，运用学习向量量化神经网络（Learning vector quantization，LVQ）构建乳房形态判别模型，实现女性乳房形态的细分与判别。对细分后的乳房类型制作相应的文胸样板，通过虚拟试衣和实物试穿进行合体性评估，结果表明：青年女性乳房占比最多的为70B文胸号型，乳房细分为适中内敛型、平坦低胸位型、丰满外扩型。构建的LVQ神经网络青年女性乳房形态判别模型，准确率达93.33%。数字化服装压力和真实试穿实验表明，细分后的文胸合体性得到了有效提高，为不同类型乳房的文胸合体性结构设计提供参考。

关键词: 空间向量角, 三维人体测量, 乳房形态, LVQ神经网络, 合体性