人臉美感知覺：影像統計數和對稱感的影響

The Perception of Facial Beauty: The Contribution of Image Statistics and Perceived Symmetry

影響人臉好看程度的因素很多，本研究在探討評估人臉美感如何受到影像統計數（如：對稱係數、頻譜分布的斜率）以及對稱感的影響。我們從人臉資料庫中挑出228張人臉，並進行影像處理，產生完全對稱、不完全對稱以及正常人臉三類實驗刺激。每張人臉皆調整成相同的明暗對比，並計算該人臉的對稱係數以及頻譜分布（斜率）。實驗參與者需要針對每張人臉進行美感分數、對稱分數、情緒正負向特質以及情緒激動程度做判斷。實驗一採用完全對稱以及不完全對稱的人臉刺激探討相同人臉的對稱變化對各項評分間的影響。研究結果顯示完全對稱的人臉刺激對稱係數越接近1，且斜率值較大，實驗者亦評為較美、較對稱、較具正向情緒而情緒激動分數較低。美感分數的變化可以用對稱感變化分數與情緒正負向特質變化來估計，而其他向度皆無預測力。在實驗二中則是邀請另外一批受試者對正常人臉實驗刺激作相同的判斷，實驗結果顯示除了情緒激動程度外，對稱係數、斜率、情緒正負向以及主觀對稱分數皆可利用線性迴歸的方式估計美感原始分數的高低。此外，隨機選取實驗二一半的人臉刺激所得的迴歸係數亦可去預測另外一半人臉刺激的美感分數。最後，我們採用路徑分析探討影像統計數等物理量如何影響或是調節對稱感、情緒正負向特質評估進而影響美感判斷，為後續人臉美感研究提供一個架構。

The judgement of facial beauty was influenced by many factors. In current study, we investigated how the image statistics (including slope of amplitude spectrum and symmetry index) and perceived symmetry influenced facial attractiveness. 228 faces were normalized to the same rms contrast and modified to produce its symmetrical face and asymmetrical face by morphing technique. The task of the observers was to rate each face for its attractiveness, perceived symmetry, valence and arousal in the emotion dimension. In Experiment 1, we used symmetrical and asymmetrical faces as stimuli. Our results showed that the symmetrical face has higher values of symmetry index and slope, and was rated as more attractive, symmetrical and with higher positive emotion. Furthermore, the facial attractiveness could be predicted by symmetry index, slope, perceived symmetric and emotional valence by linear regression. In Experiment 2, we used original faces and asked other participants to do the same task. The results were similar to Experiment 1. The attractiveness rating for randomly chosen faces could be also predicted by the rest of the faces by using the same parameter weights in linear regression. We also adapted path analysis to investigate how the physical properties (symmetry index and slope) influences participants' attractiveness rating via perceived symmetry and emotion judgement. Our results suggested the beauty of the faces was contributed by the image statistics, perceived symmetry and valence of the emotion.