The purpose of this research is to analyze Japanese women's breast shape based on body surface data described by a three-dimensional (3-D) human body shape model with a bi-cubic B-spline structure and to classify them. The data used for analysis were forty-nine 3-D control points selected from the right breast area on the model surface for each of 556 Japanese women aged 19 through 63 years. We examined the covariance matrix of the data using the principal component analysis method after normalization of their 3-D coordinates with the bust width for reducing the size factor. As a result, we obtained four principal components, which described 77% of breast shape. Then Japanese women's breast shape was classified into five classes in the principal component space using the first, second, third and forth-principal component scores. They could cover 92% of Japanese women's breasts. Therefore, we tried to analyze breast shape by clustering in order to classify all the breasts. For the cluster analysis we prepared two kinds of data; (1) principal component scores and (2) the normalized scores (μ=0, σ=1) of (1). With the clustering (1) and (2) we obtained four classes and five classes, respectively. Properties and advantages of the three kinds of classifications were also discussed. The classification of the principal component space is based on standard deviations of principal component scores, and therefore the resultant classes do not have clear boundaries. The classification according to the cluster analysis (1) can reflect the actual distribution of breast shape. In contrast the clustering (2) gives classification reflecting more principal components and tending to generate more classes than the clustering (1).
Read full abstract