上下半部臉孔特徵對表情分類作業的貢獻

Abstract

本研究的目的乃是以複合臉(composite face)及視覺分類技術來研究視覺系統中臉部表情辨識因子的統計特性，並在刺激空間中量化快樂及悲傷兩種表情的分類界線。易言之，我們量測臉部表情辨識作業如何隨著上半部、下半部臉部表情強度的變化而有所不同。然後，我們使用一個多變項訊號偵測理論的模型，推估表情辨識因子的內在表徵的統計特性。本研究有三個實驗，第一，在視野中央呈現且上下臉密合；第二，視野中央呈現但上下臉錯開44分視角；第三，在視野邊緣呈現且上下臉密合。在每個實驗中，觀察者需判斷所呈現的刺激是快樂臉或悲傷臉。結果顯示，上下臉是否錯開對於表情辨識並無影響，這顯示了表情辨識是由各個特徵分別決定的歷程。另外，較之在中央視野，受試者容易把一張在視野邊緣呈現的影像看成是悲傷的臉。我們的結果可以由多變項訊號偵測理論的模型來加以解釋，數學模型的結果也顯示，上下部分的臉對於臉部表情的分類的貢獻是互相獨立的。

We used a composite face paradigm and a visual classification task to investigate the property of facial expression processors in the visual system and the decision boundary for happy/sad classification in a stimulus space. We measured how the facial expression classification depends on the intensity of a particular expression in either the upper or the lower faces. We then applied a model based on the multidimensional signal detection theory to estimate the statistical properties of the internal representations, or the tuning properties of the expression processors. There were three test conditions: two foveal viewing conditions in which the upper and lower parts of the test image are either aligned or had a lateral shift of 44' visual angle; and one peripheral condition in which the aligned face was placed at 6o to the left of the fixation. The observers were asked to determine whether a test image was happy or sad. The alignment conditions had no effect on happy/sad classification, suggesting that facial expression classification is an analytic process. The model analysis also showed no interaction between upper and lower facial features in foveal expression classification. It was easier to classify a face as sadness in the periphery than in the fovea, suggesting different mechanisms for expression perception in different eccentricities.