Capacity Limits in Face Detection

Our ability to socialise and interact with others is underpinned by our ability to extract social information from faces. However, these faces must first be detected from the complex visual environment before any social information is extracted. Prior face perception work has focused mainly on later stages of face processing. The early stage of face detection has been relatively neglected. As such, some basic questions in face detection are outstanding. Can we know that more than one face is present? How many faces can be detected at once? How do viewing conditions affect performance? This thesis adapted paradigms from face perception and numerical cognition research to address these questions. First, Chapter 2 used a ‘subitizing of faces’ approach to compare multiple target detection for faces and other types of stimuli. A detection advantage was found for faces over non-face stimuli, irrespective of face inversion. Chapter 3 used a novel ‘fixed/mixed’ judgment task to test for capacity limits in face detection. The findings supported an efficient parallel detection mechanism for multiple faces. Chapter 4 adapted search tasks to assess how multiple face detection is affected by different viewing conditions, including aspects of the task and presentations. The findings in this chapter indicate a typical face detection span of four faces, plus or minus one. They also show how the visual complexity and meaningfulness of the surrounding scene affect performance. This thesis establishes that detecting multiple faces in complex visual scenes is an efficient parallel process for up to four faces. It also contributes several methodological innovations that can be adapted to address related research questions.