Cognitive Templates for Human Face Detection

Faces convey a range of important social information such as a person's identity, emotional state and direction of gaze. However, before such information can be attained from a face, it first has to be localised within the visual environment. Therefore, face detection is a key process that facilitates social navigation in our environment. Detection seems to be most sufficient when faces are presented in frontal, upright views and in colour with all features intact. Yet, the results of such research are mixed, due to the range of methods used. Therefore, this thesis explores the visual processes of face detection, by focusing on how faces are displayed, and the facial features utilised in finding faces in such displays. Firstly, the attentional draw of person stimuli in naturalistic scenes is considered, and the impact of the frequency of person presence on attentional draw (Chapter 2). The findings indicate that people draw attention spontaneously in naturalistic scenes, but at high rates of presentation such stimuli can be actively ignored in favour of other task demands. The drivers of the detection process were then explored, by investigating the influence of display context, whilst noting the facial features that facilitate optimal detection (Chapter 3). The detection of frontal/upright faces and manipulated faces were comparable in blank displays, yet disparities between frontal/upright and manipulated faces emerge in array and scene displays, indicating that detection processes are dependent on scene complexity that requires search. Furthermore, the use of face manipulations revealed that external features, such as overall shape, drive detection. Subsequently, the role of internal and external facial features in peripheral detection was considered using a novel gaze-contingent detection paradigm (Chapter 4). In the periphery, localisation is potentially enhanced for faces without internal features, with faster fixations on initially featureless faces. Yet performance declines when features are removed at fixation, suggesting they are essential for face classification, but not localisation. This was then reinforced, as faces with manipulated internal features (i.e. rotated) were difficult to distinguish from intact faces in the periphery. Therefore, internal features cannot be perceived clearly enough in the periphery to drive detection, but seem to gain importance at fixation when classifying faces. Altogether, this thesis outlines the cognitive templates used in face detection that depend on presentation context. A colour-shape template based on skin tone and oval shapes seems to drive detection when localising faces. Yet a more detailed template with intact featural configuration is utilised closer to fixation once faces are being classified. Separating the processes involved in detection and the features involved in such processes helps to build a more coherent theory of face detection.