人臉美感的自動評估與分析

Analysis and Automatic Evaluation of Human Facial Attractiveness

胡竣凱(Chun-Kai Hu)

指導教授 ： 劉震昌

國立暨南國際大學/科技學院/資訊工程學系/碩士(2017年)

摘要

為什麼人們在看到一張臉的時候只需要花到幾秒的時間就能判別出那張臉美不美?好不好看?甚至還能幫那張臉給出一個分數?那是因為人們從小到大看過了無數張的面孔，所以在心裡上會學習產生一個對美感的尺度，所以可以在幾秒鐘內評斷一張臉的面孔美不美。但是一張看上去美麗的臉孔是由什麼所組成的呢?通常是擁有漂亮立挺的五官、白嫩的肌膚等，這些要素我們可以稱為構成一張漂亮的臉的特徵。所以本論文在探討讓電腦機器可否跟人類一樣看過了大量的照片影像後可以擁有一個與一般人類相仿的美感尺度，可以自動去辨別一張面孔的美感，並給予出一個美感分數。本論文會透過網路爬蟲技術收集大量影像照片來當實驗資料，共抓取40,532張照片影像，擷取影像中人臉幾何特徵後再利用機器學習的中的Linear Regression 與 Support Vector Machine兩種方法進行實驗，得到1~5之間的美感分數，最後機器預測出來的Root Mean Squared Error 達到0.5833。

關鍵字

機器學習 ； 人臉美感 ； 網路爬蟲

並列摘要

Why can we distinguish whether a face is beauty or not by taking a short glance to it, and even score it? The reason is that we have seen many faces since we were born and we would learn the sense of beauty in the brain. As a result, depending on our learned standard, we can tell anyone if a face looks great in a few seconds. However, what composes a pretty facial appearance? It may include features like a clearly outlined face, a crystal-clear skin and so on. Sometimes, with those features, we would say he or she has a beauty face.
This thesis discusses whether a computer can automatically pick up a “beauty face” among numerous faces and even give it “a score of beauty” for us like a human does, after learning from a large amount of face pictures. Through a web crawler, we collected totally 40,532 face pictures as the experimental dataset. Geometrical facial features were measured from the facial landmarks in these pictures and then these features were trained by two machine learning approaches: Linear Regression and Support Vector Machine. The facial beauty score starts from 1 to 5. The Root Mean Squared Error of the facial beauty estimation is about 0.5833.

並列關鍵字

Machine Learning ； Facial Beauty ； Web Crawler

參考文獻

[1] Nancy L. Etcoff. (1999). Survival of the Prettiest：The Science of Beauty. America: Anchor.

[2] Wiki：Golden Ratio，https://en.wikipedia.org/wiki/Golden_ratio.

[3] 三庭五眼-台灣Word，http://www.twword.com/wiki/%E4%B8%89%E5%BA%AD%E4%BA%94%E7%9C%BC.

[4] D. Sutic, I. Breskovic, R. Huic, and I. Jukic, “Automatic Evaluation of Facial Attractiveness,” IEEE Proc. Intl. Conf. on MIPRO, 33rd, pp. 1339–1342, 2010.

[5] D. Xie, L. Liang, L. Jin, J. Xu, M. Li, “SCUT-FBP: A Benchmark Dataset for Facial Beauty Perception,” IEEE Proc. Intl. Conf. on Systems, Man and Cybernetics, pp. 1821–1826, 2015.