Minh Trieu Nguyen, Truong Thinh Nguyen
College of Technology and Design, University of Economics Ho Chi Minh City-UEH, Ho Chi Minh City 72516, Vietnam.
Diagnostics (Basel). 2023 Feb 26;13(5):891. doi: 10.3390/diagnostics13050891.
Measuring and labeling human face landmarks are time-consuming jobs that are conducted by experts. Currently, the applications of the Convolutional Neural Network (CNN) for image segmentation and classification have made great progress. The nose is arguably one of the most attractive parts of the human face. Rhinoplasty surgery is increasingly performed in females and also in males since surgery can help to enhance patient satisfaction with the resulting perceived beautiful ratio following the neoclassical proportions. In this study, the CNN model is introduced to extract facial landmarks based on medical theories: it learns the landmarks and recognizes them based on feature extraction during training. The comparison between experiments has proved that the CNN model can detect landmarks depending on desired requirements. Anthropometric measurements are carried out by automatic measurement divided into three images with frontal, lateral, and mental views. Measurements are performed including 12 linear distances and 10 angles. The results of the study were evaluated as satisfactory with a normalized mean error (NME) of 1.05, an average error for linear measurements of 0.508 mm, and 0.498° for angle measurements. Through its results, this study proposed a low-cost automatic anthropometric measurement system with high accuracy and stability.
测量和标注人脸地标是一项由专家完成的耗时工作。目前，卷积神经网络（CNN）在图像分割和分类方面的应用取得了巨大进展。鼻子可以说是人脸最具吸引力的部位之一。隆鼻手术在女性中越来越普遍，在男性中也逐渐增多，因为手术有助于提高患者对术后按照新古典比例呈现的美观比例的满意度。在本研究中，引入了基于医学理论的CNN模型来提取面部地标：它在训练过程中通过特征提取学习地标并识别它们。实验之间的比较证明，CNN模型可以根据所需要求检测地标。人体测量通过自动测量进行，分为正面、侧面和颏部视图的三张图像。测量包括12个线性距离和10个角度。研究结果被评估为令人满意，归一化平均误差（NME）为1.05，线性测量的平均误差为0.508毫米，角度测量的平均误差为0.498°。通过研究结果，本研究提出了一种低成本、高精度且稳定的自动人体测量系统。
