基于椭圆傅里叶算法的青年女性躯干形态分类

The torso morphology classification of young females based on ellipse Fourier

为探究人体躯干形态间的差异,本文于212份人体三维点云数据中随机选择120个样本进行过胸凸矢状轮廓曲线的提取,结合椭圆傅里叶分析对轮廓曲线进行研究。计算并分析不同最大谐波次数下的拟合误差,确定实验用最大谐波次数为15;运用15次谐波对所有样本的轮廓曲线进行椭圆傅里叶变换,获取能客观描述曲线形态的规格化描述子。通过对规格化后的描述子进行主成分分析,最终提取10个主成分。利用混合F统计量确定最佳分类数,以成分得分为依据,运用K-means聚类方法将样本形态分为5类。研究得出:各分类之间于肩部、胸部、腹部、臀部、后背弯曲程度及侧面厚度等方面均存在较为明显的差异,依据椭圆傅里叶描述子可以对躯干形态进行合理有效的分类。

Clothing plays the role of protection,decoration and beautification in human body.At the same time,its shape and structure are also affected by human body shape to a large extent.Therefore,the analysis and classification of human body shape is the basis of improving clothing structure and fitness.According to this,many mathematicians at home and abroad have carried out analysis and research on human form,completed the classification of human form by using different classification basis,and achieved some research results.At present,the classification methods of human body shape are mainly divided into two categories:qualitative classification and quantitative classification.The qualitative classification method is based on observation and scientific analysis,and uses language,numbers or letters to describe the characteristics of human body,while the quantitative classification method is based on a certain data basis and uses it as an index to classify the types of human body.Quantitative classification is more commonly used in the study of body shape classification,but the classification basis selected in most relevant studies is the local characteristic size of the human body,which lacks a complete description of the shape of the human body curve and has certain limitations.In order to explore the differences between human trunk shapes through the complete description of human body contour curve,120 samples were randomly selected from 212 human body three-dimensional point cloud data to extract the sagittal contour curve of the thoracic convexity,and the extracted sagittal contour curve of the thoracic convexity was described and studied combined with elliptic Fourier analysis.The fitting errors under different maximum harmonic numbers were calculated and analyzed,and the maximum harmonic number used in the experiment was determined to be 15;the contour curves of all samples were transformed by the 15^(th)harmonic to obtain the normalized descriptor which could objectively describe the shape of the curve.The normalized descriptors were analyzed by principal component analysis,and finally 10 principal components were extracted.The best classification number was determined by mixed F statistics.Based on the component score,the sample morphology was divided into five categories by K-means clustering method.Based on the application of elliptic Fourier algorithm to describe the contour curve,the ellipse Fourier descriptors were combined with principal component analysis and cluster analysis to finish the analysis and classification of human body shape.The results show that there are obvious differences in shoulder,chest,abdomen,buttocks,back bending degree and lateral thickness among the classifications.The trunk shape can be classified reasonably and effectively according to the elliptical Fourier descriptor.Based on the complete curve shape of human body,the classification of human trunk shape is completed,which provides a new idea and research basis for the classification of human body shape.At the same time,it is proposed and verified that elliptic Fourier transform can be applied to the analysis of human body curve shape.On this basis,the follow-up research can expand the sample size and further explore it in combination with human body size data.