The starting point for this thesis is a review of Bundesen’s theory of visual attention. This theory has been widely accepted as an appropriate model for describing data from an important class of psychological experiments known as whole and partial report. Analysing data from this class of experiments with the help of the theory of visual attention – have proven to be an effective approach to examine cognitive parameters that are essential for a broad range of different patient groups. The theory of visual attention relies on a psychometric function that describes the ability to identify a stimulus as a function of exposure duration. An important contribution of the thesis is that it investigates whether other psychometric functions than the one originally used with the theory of visual attention could be more appropriate at describing data. The thesis points to two psychometric functions that seem more appropriate. Further the thesis shows that it is possible to incorporate any desired psychometric into the theory of visual attention. Common to the two psychometric functions suggested is that they both have a hazard function that is non-monotonic; a neural argument for this is also presented in the thesis. For the psychometric function it is further investigated how this depends on stimulus contrast. In this respect, we find that the type of psychometric function is independent of contrast, but that the parameters for the psychometric function vary systematically as a function of contrast. An analysis of the psychometric function for the individual letters of the alphabet shows that there are significant differences in the parameters of the psychometric function depending on letter identity. Here we should note that in many cases (also for Bundesen’s theory of visual attention) it has been customary to average performance over the entire set of stimuli, consisting for instance of the 26 alphabetic letters. The fact that each letter is perceived in a different way possibly reflects that each letter is represented differently in our brain. This might have to do with a difference in the set of features representing the individual letters. It is possible that some features are processed faster than others and that overlapping features representing more than one letter in the alphabet play a certain role for the tendency to confuse letters. Hopefully it should be possible, with the dataset that we collected, to directly analyse how confusability develops as a certain letter is exposed for increasingly longer time. An important scientific question is what shapes the psychometric function. It is conceivable that the function reflects both limitations and structure of the physical mechanism underlying perception. For this reason we argue that the alternative psychometric functions that we have suggested are also relevant for models trying to simulate the mechanism leading to perception. The thesis reviews a selection of stochastic models that are well-known candidates when it comes to modelling mechanisms of perception. These candidates include the Ornstein-Uhlenbeck model and the leaky competing accumulator model. A further contribution of the thesis is a demonstration that the leaky competing accumulator model (see Usher & Cohen, 1999) is able to explain a perceptual limit that characterises how many objects can in parallel be perceived. Finally, the thesis suggests five concrete topics for future work. These include as diverse themes as: determination of the visual features representing the individual letters in our brain, neurodynamical modelling of visual perception, investigation of the duration of visual short-term memory as well as psychometric functions and assumptions along with application areas in cognitive diagnostics.
Main Research Area:
Hansen, Lars Kai, Andersen, Tobias, Kyllingsbæk, Søren
Technical University of Denmark, DTU Informatics, Building 321, 2010