學習角度對3-D臉孔再認的影響: 臉孔鏡像對稱優勢之檢驗

How Does Learning Different Viewpoints Affect Recognition of 3-D Faces? A Test of the Mirror-Symmetry Advantage

葉娟妤(Chuan-Yu Yeh)

指導教授 ： 襲充文

國立中正大學/社會科學院/心理學系/碩士(2012年)

摘要

兩種不同的理論嘗試著解釋跨角度的臉孔辨識如何達成：不受角度影響的單一表徵、或是由多張不同角度的臉孔所組成的角度依賴表徵，Freiwald及Tsao (2010)發現頂葉神經細胞階層性的對臉孔呈現角度激發，其中前端細胞群特化為對鏡像對稱的兩個不同角度激發，此一發現被認為是達成角度不變性的主要機制。本研究使用電腦轉換之三維臉孔模型做為實驗刺激討論學習臉孔的不同角度如何影響臉孔的再認，並且檢驗臉孔鏡像對稱假說。在實驗一中，我們操弄四個水準的學習和再認臉孔的角度差異，顯示角度差異越大，再認的表現越差。在實驗二中，操弄學習角度為單側臉孔或雙側臉孔，結果顯示學習30度的臉孔再認的表現高於學習60度的臉孔，然而，鏡像對稱優勢僅在後者出現。實驗三在再認時比較對稱角度和學習角度，發現對稱角度的再認會一致的劣於學習角度。為了檢驗鏡像對稱假說，實驗四比較內插角度、外插角度以及外插-對稱角度，結果顯示鏡像對稱優勢的確補償了外插角度的劣勢。整體而言，本研究的結果不僅支持Freiwald & Tsao (2010)的發現，並且發現鏡像對稱優勢在不同條件下對臉孔再認的影響。

關鍵字

鏡像對稱性 ； 角度不變性 ； 角度依賴 ； 臉孔辨識 ； 三維臉孔模型

並列摘要

Recognizing a face from a specific viewpoint may be achieved by either a single view-independent representation or a set of view-dependent representations depicting the appearance from different perspectives. More specifically, based on neuronal recording in monkeys, Freiwald & Tsao (2010) have recently proposed that tuning to identity of mirror-symmetrical views may be the underlying mechanisms for achieving view invariance in face recognition. Here in four experiments we used computer-generated 3-D face models via post-production to investigate how the learning of different viewpoints may affect face recognition, and to test mirror-symmetry hypothesis. In Experiments 1A and 1B, four levels of angular disparity between learning and testing views were manipulated, and the results of revealed a tendency of decrement in recognition accuracy with increment in the angular disparity between learning and test views. In Experiment 2, the effect of mirror symmetry on face recognition was examined by presentating faces either within a single visual field or across both visual fields. The results showed a clear advantage of learning faces with the viewpoint of 30˚ than with the viewpoint of 60˚. However, the advantage of mirror symmetry was evident only with the latter than the former view. In Experiment 3, the mirror-symmetrical views were tested against the learned views, and results showed consistent inferior performances with mirror-symmetrical views. Finally, in Experiment 4, the mirror-symmetry hypothesis was examined by comparing recognition performances between interpolated views, extrapolated views and extrapolated-symmetrical views. The results suggested that the mirror-symmetrical view may help overcome the disadvantage of the extrapolated view, although the overall performance indicated face recognition is mainly view-dependent. Take together, our findings not only lend (partial) support to Freiwald & Tsao (2010)’s conjecture, but also demonstrate the constraints on the effects that mirror-symmetrical views may have in face recognition.

並列關鍵字

view-invariance ； viewpoint-dependence ； face recognition ； mirror-symmetry ； 3-D face models

參考文獻

Shyi, G. C.-W., & He, H.-M. (2011). Effects of familiarityand expression variation on face recognition and generalization. Chinese Journal of Psychology, 53(4), 437-470.

Aaronson, D., & Watts, B. (1987). Extensions of Grier’s computational formulas for A’ and B’ to below-chance performance. Psychological Bulletin, 102, 439–442.

Benton, C. P., Jennings, S. J., & Chatting, D. J. (2006). Viewpoint dependence in adaptation to facial identity. Vision Research, 46, 3313–3325.

Biederman, I. (1987). Recognition-by–components: A theory of human image understanding. Psychological Review, 94, 115–147.

Biederman, I. (2000). Recognizing depth-rotated objects: A review of recent research and theory. Spatial Vision, 13, 241–253.