深層学習を用いた日本人の額形状の男らしさ／女らしさの特徴分析

抄録

In facial reconstruction by plastic surgery, masculine and feminine are one of the factors that affect the degree of perfection. It is useful to analyze the characteristics for creating a natural shape as a male or female. In this study, 3D models were created by measuring the facial shapes of Japanese men and women (75 men, 75 women). 100 cross-sectional images of the forehead were made for each person from the 3D model. These images and the gender were used as teaching signals, and the CNN was trained by them. Next, we visualized the characteristics found by the learned CNN using Grad-CAM, and performed curvature analysis on the characteristic parts. As a result, the male characteristic was the swell of superciliary arch and the female characteristic was the roundness of the frontal tuber.