Operation of eye-movement control mechanisms during the perception of naturalistic scenes.
Walshe, Ross Calen
MetadataShow full item record
Understanding of visual scenes takes place within very brief episodes known as fixations. To explore the extent of the scene, the eye shifts between fixation locations at intervals of roughly 300 ms. Currently, it is a matter of open inquiry as to what factors influence the timing of these movements. This thesis focuses on understanding the mechanisms that govern the rapid adjustment of fixation and saccade timings when novel stimulus information is encountered during a fixation. In part I, I use an experimental technique known as the fixation-contingent scene quality paradigm to control the quality of incoming visual scene information. This approach is used to assess how fixation timing adapts to moment-by-moment changes in the quality level of the stimulus. I find that quality changes tend to result in an increase in fixation durations and this occurs whether the quality is increased or decreased. Using distributional analytic techniques, I argue that these results reflect the combined influence of a rapid surprise related process and a slower acting encoding related influence. In part II, I study how fixation durations are influenced by the underlying saccade programming mechanisms. An important assumption within the eye-movement control literature is that there exists a threshold called the point-of-no-return. Once this point has been reached, a saccade may no longer be modified or cancelled. I adapt a classic psychophysical technique known as the double-step procedure to study the point-of-no-return within scene viewing tasks. I also provide a measurement of the saccadic dead time, the last point in time that a saccade may be modified. In Part III, a formal model of fixation durations in high-level tasks is presented. I build on recent modelling work and develop a formal account for the early-surprise late-encoding modulation account of fixation durations in scene viewing tasks. The model is tested against data observed in Part I of the thesis. I demonstrate that the model does a very good job of predicting these distributions with relatively few assumptions. In summary, I use experimental techniques in combination with computational modelling to reveal how a composite of low-level (saccade programming) and high-level (information processing) considerations can, and must, be taken into consideration when understanding eye-movement control behaviour in scene viewing tasks.