以影像為基礎的臉孔辨識：探討線畫臉、色像素與光影以及空間頻率的效應

Image-based Approach to Face Recognition: Effects of Line-drawn Faces, Pigmentation and Shading, and Spatial Frequency

為數眾多的臉孔辨識研究逐漸形成三個主要的研究取向：一個以影像為基礎，一個以結構為基礎，而另一個則是以功能為基礎。以影像為基礎的研究關切臉孔的邊界、空間頻率、色像素與光影等的訊息|研究者為探索這些訊息在臉孔辨識時扮演的角色，以實驗操弄臉孔的空間頻率、正負片、光照、顏色、色像素等因子，從影像的層次了解臉孔辨識的心理歷程。結構為基的研究則關切臉孔中各五官特徵（feature）、臉孔整體（holistic）、組態（configuration）與臉孔3D的訊息，將臉孔視為可拆解、組合的結構性物體，並操弄臉孔五官、局部區塊與空間組態等訊息，從結構描述的層次了解臉孔辨識。另一個取向則從功能性的角度切入，探索臉孔辨識時吾人如何提取臉孔中具有生物及社會性功能的訊息，如臉孔辨識、性別、年齡、種族、情緒等。本文旨在介紹此三大臉孔辨識研究取向的差異並詳細回顧其中屬於初級處理階段的以影像為基礎的研究發現，亦即臉孔辨識中對臉孔的邊界、空間頻率、色像素與光影等訊息的處理。本文之回顧聚焦於臉孔辨識中以影像為基礎的處理歷程，此為臉孔辨識後續以結構為基礎及功能為基礎的歷程提供重要的基礎。

The vast number of studies in face recognition have gradually evolved into three major approaches, namely, the image-based, structure-based, and function-based approach. Studies under the category of image-based approach are concerned with image-level face information, such as edge, spatial frequency, pigmentation, shading and shadow. Researchers manipulate these factors and examine the roles they may play in face recognition. In contrast, those ascribed to the structure-based approach are concerned with the effect of featural, configural, and holistic information on face recognition. Researchers adopting this approach consider faces as decomposable objects and explain face processing in terms of structural descriptions. Finally, studies under the third approach investigate face recognition from a functional perspective, where the main issues concern how humans extract biological as well as social information from faces that are pertinent to social interactions, such as identity, gender, age, race, and emotional expression, among others. In this review, we first discuss the differences between these three approaches, and then focus our review in details on the findings accrued within the image-based approach. Our goal is to provide an integrated overview of the image-based approach and emphasize the importance of image-based processing in understanding face recognition. Specifically, our review highlights how basic processing of image information such as border, spatial frequency, pigmentation, may lay the critical foundation for the later processing of the structural and functional aspects of a face.