From parts, whole, to recognition in face and ensemble processing

Abstract

The visual feature binding problem, which asks how multiple visual features are integrated into a unitary percept, is a fundamental question in high-level visual recognition. Feature Integration Theory, initially proposed by Treisman and Gelade (1980), and its descendants such as Wolfe’s Guided Search Model provide an influential psychological framework in addressing the visual feature binding problem and understanding human visual attention. The theory asserts two stages of attentional processing mechanism. The pre-attentive stage focuses on the global properties of the visual scene and enables feature extraction. The focused attention stage then orients to local information and allows integration of multiple features into a global representation. This dissertation presents two separate investigations to explore the interplay between spatial attention, and the two stages of attentional processing in face and ensemble processing.

It is widely accepted that the perception of faces involves holistic face processing, where individual facial features are thought to be automatically integrated into a holistic representation. In Chapter 2, I examine the nature and the function of attention in achieving this representation. Manipulating attention through dual-task, brief exposure, and divided attention, this study explored whether facial features were automatically integrated at sight or were integrated after feature extraction by studying the illusory conjunction error, which is to mistakenly combine features from one stimulus to another when focused attention is unable to be deployed. Across three behavioral experiments, the results demonstrated that when attentional resources were busy with another task, participants were sensitive to facial feature information but not to configural/holistic information. Deploying attention to the faces during study improved processing of configural/holistic information. These findings support the integrative nature of holistic processing and the role of attention to bind facial features into a holistic face representation.

Chapter 3 investigates the relative contribution of selective and distributed attention in perceptual averaging. While the dominant perspective suggested that perceptual averaging resulted from an early integration mechanism facilitated by distributed attention, there remain some arguments in the existence of such an early integration mechanism. I explored these questions using a gaze-contingency paradigm, and computational modelling of the eye-tracking data that I collected. The results showed that participants still achieved similar accuracy in orientation averaging in the free viewing condition and when their foveal vision was blocked but performance significantly deteriorated when their peripheral vision was blocked. Modelling analysis further revealed individual differences in the participants’ distributed attention mechanism, which could be categorized into the spotlight or zoom-lens attention group. The findings confirm the existence of an early integration mechanism that could integrate elements from the periphery in parallel with distributed attention and argue for an interactive mechanism between selective and distributed attention depending on the available processing resources, task demand, and individual differences.

To conclude this dissertation, I will discuss how these studies have informed Feature Integration Theory and propose further studies to explore the temporal dynamics and characteristics of its two attentional processing stages.