Background: Age estimation technology has a wide range of applications and has become a popular research area in computer vision. Wrinkles are a sign of aging skin and make a person visually appear to be older. Many people are opting for new aesthetic medicine treatments to make themselves look younger. To increase the efficacy of aesthetic medicine treatments, it is important to provide physicians with an objective assessment of the difference between pre- and post-surgery age prediction by face imaging. Thus, this study aimed to use deep learning techniques to build an age estimation model using the wrinkle features of human faces.
Objective: This study used deep learning techniques to build an age estimation model using facial wrinkle features. Using the FaceScape public dataset, the model was trained with Asian face photos and to predict the actual age based on face photos. The model was then used to determine the difference between the patient's age before and after surgery, and to objectively evaluate the efficacy of aesthetic medicine treatment. The results are expected to provide reference data for corporate R&D, product R&D or technology R&D in the future.
Methods: This observational study used samples from the FaceScape public database for model development and testing. The database records included color photos of age, gender, different viewing angles, and different expressions. The study subjects were 20 years old and above, and those under 20 were not included. In addition, pre- and post-surgery photographs of Liftera V were collected from a clinic in Taipei City, Taiwan, for further model validation and analysis. Three models, DeepCNN, VGG16, and ResNet50, were trained based on MAE and MSE losses, to find out the age area attributes (wrinkles) and to predict the age according to the literature.
Results: A total of 300 people of different age groups and 1692 face images were selected from the FaceScape database of Multi-view data and neutral images, excluding images with facial occlusions (e.g., wearing glasses). to the models were trained using the images, and the actual age was predicted based on the face images. The DeepCNN model achieved the best results (4.133) in age prediction with the MAE loss function, and was chosen as the model. The statistical analysis of the age prediction for the pre- and post-surgery photos showed that the actual ages were similar to the predicted ages on the pre-surgery photos, and the predicted ages on the pre-surgery photos were lower than that of the post-surgery photos.
Conclusion: Our experimental results showed that the proposed AI deep learning approach was effective in predicting facial skin age in real-world clinical settings, based on our small testing dataset.
目 錄
中文摘要 i
Abstract iii
目 錄 v
圖目錄 vii
表目錄 ix
第一章 緒論 1
第一節 研究背景 1
第二節 研究動機與目的 3
第二章 文獻探討 4
第一節 人臉年齡估計 4
第二節 人臉皮膚老化和皺紋的成因與表現 7
第三節 皺紋治療方法 10
第四節 文獻探討啟示 11
第五節 文獻限制 11
第三章 研究設計與方法 12
第一節 資料來源 12
第二節 研究對象 13
第三節 研究流程與架構 14
第四節 研究工具與分析方法 19
第四章 研究結果 20
第一節 FaceScape Multi-view data樣本人口與百分比 20
第二節 DeepCNN、VGG16與ResNet50模型訓練分析 21
第三節 年齡預測分析 26
第五章 討論 37
第一節 研究成果 37
第二節 研究優勢與限制 39
第三節 結論與建議 40
參考文獻 41
附件 45
參考文獻
一、中文部分
王政坤（2012）。醫學美容與皮膚保養。藝群國際。
譚軍（2014）。鐳射皮膚再生美容。湖南科技學技術出版社。
Bazin, R., & Flament F.（2010）。皮膚老化圖譜 第二冊 亞洲人［裘惠霞、應穎、龍曉慧、甄雅賢譯］。MED'COM。(原著出版於2010)
宋正宇（2018年07月08日）。「術前＆術後照」可信嗎？。Doctor Beauty。取自: https://www.dr-beauty.net/post-view.php?ID=1940
安心醫美（無日期）。醫美行為大調查。取自: https://beconfidentyou.com/survey
Huang, T.（2018年09月27日）。機器/深度學習:基礎介紹-損失函數(loss function)。取自: https://chih-sheng-huang821.medium.com/機器-深度學習-基礎介紹-損失函數-loss-function-2dcac5ebb6cb

二、英文部分
Abirami, B., Subashini, T. S., & Mahavaishnavi, V. (2020). Gender and age prediction from real time facial images using CNN. Materials Today: Proceedings, 33, 4708–4712. https://doi.org/10.1016/j.matpr.2020.08.350
Asterasys (n.d.) Liftera V. Retrieved from: https://asterasys.com/products/liftera/liftera-
vision/
Bobrov, E., Georgievskaya, A., Kiselev, K., Sevastopolsky, A., Zhavoronkov, A., Gurov, S., Rudakov, K., del Pilar Bonilla Tobar, M., Jaspers, S., & Clemann, S. (2018). PhotoAgeClock: Deep learning algorithms for development of non-invasive visual biomarkers of aging. Aging, 10(11), 3249–3259. https://doi.org/10.18632/aging.101629
Chandrakala, M., & Durga Devi, P. (2021). Two-stage classifier for face recognition using HOG features. Materials Today: Proceedings, 47, 5771–5775. https://doi.org/10.1016/j.matpr.2021.04.114
Huynh, H. T., & Nguyen, H. (2020). Joint age estimation and gender classification of Asian faces using wide ResNet. SN Computer Science, 1(5), 284. https://doi.org/10.1007/s42979-020-00294-w
Gordon, R.S. & Brieva, J.C. (2012) Unilateral dermatoheliosis. The New England Journal of Medicine; 366:e25. DOI: 10.1056/NEJMicm1104059.

Mayo Clinic (n.d.) Wrinkles. Retrieved from: https://www.mayoclinic.org/diseases-
conditions/wrinkles/symptoms-causes/syc-20354927
Nouveau-Richard, S., Yang, Z., Mac-Mary, S., Li, L., Bastien, P., Tardy, I., Bouillon, C., Humbert, P., & de Lacharrière, O. (2005). Skin ageing: A comparison between Chinese and European populations. Journal of Dermatological Science, 40(3), 187–193. https://doi.org/10.1016/j.jdermsci.2005.06.006
Othmani, A., Taleb, A. R., Abdelkawy, H., & Hadid, A. (2020). Age estimation from faces using deep learning: A comparative analysis. Computer Vision and Image Understanding, 196, 102961. https://doi.org/10.1016/j.cviu.2020.102961
Silvia, S., & Supangkat, S. H. (2020). A literature review on facial age estimation researches. 2020 International Conference on ICT for Smart Society (ICISS), 1–6. https://doi.org/10.1109/ICISS50791.2020.9307606
Wang, M., & Deng, W. (2021). Deep face recognition: A survey. Neurocomputing, 429, 215–244. https://doi.org/10.1016/j.neucom.2020.10.081
Yang, H., Zhu, H., Wang, Y., Huang, M., Shen, Q., Yang, R., & Cao, X. (2020). FaceScape: A large-scale high quality 3D Face dataset and detailed riggable 3d face prediction. ArXiv:2003.13989 [Cs]. http://arxiv.org/abs/2003.13989
Yang, J., Liu, P., Jiang, Y., & Li, S. (2017). Age Estimation of Asian Face Based on Feature Map of Texture Difference Model. In F. Xhafa, S. Patnaik, & Z. Yu (Eds.), Recent Developments in Intelligent Systems and Interactive Applications (Vol. 541, pp. 290–295). Springer International Publishing. https://doi.org/10.1007/978-3-319-49568-2_41
Yap, M. H., Batool, N., Ng, C.-C., Rogers, M., & Walker, K. (2021). A Survey on Facial Wrinkles Detection and Inpainting: Datasets, Methods, and Challenges. IEEE Transactions on Emerging Topics in Computational Intelligence, 5(4), 505–519. https://doi.org/10.1109/TETCI.2021.3075723
Zhang, B. H., Chen, K., Lu, S. M., Nakfoor, B., Cheng, R., Gibstein, A., Tanna N., Thorne C.H., & Bradley, J. P. (2021). Turning back the clock: Artificial intelligence recognition of age reduction after face-lift surgery correlates with patient satisfaction. Plastic and Reconstructive Surgery, 148(1), 45-54.