Cognitive processes described in terms of Marr's model of vision

Miller, Jane (1986) Cognitive processes described in terms of Marr's model of vision. University of Southampton, Doctoral Thesis.

Abstract

David Marr's A.I. (artificial intelligence) model describes vision as one unified process, from single-cell responses of early processing to higher-level analysis, such as object identification. Thus, it attempts to link the fields of A.I., neurophysiology and psychology. In this research, it was suggested that this model may provide a useful framework for psychological theories of vision and cognition. The first two experiments provide support for some specific aspects of this model. In Experiment 1, subjects required longer to compare unfamiliar objects seen from short-axis views than longer-axis views, as Marr's model would predict. The second experiment supports the hypothesis that vision involves transforming an image into progressively higher-level representaions, by showing that subjects required longer to derive in-depth information (Marr's 2 1/2-D sketch) than two-dimensional (primal sketch) information. Having provided some support for this model, the following experiments used the types of representations proposed by Marr to describe processes beyond vision - memorization and comparison of visual stimuli. In Experiments 3 to 9, subjects compared sequentially-presented unfamiliar objects. The level of representation used for this task seemed to depend on the stimulus. Marr's 2 1/2-D sketch features were compared for mirror-image objects, while 3-D representation features were compared for five-section conical objects. At each level of representation, subjects reported a variey of strategies. These were formulated as flowcharts, which were then used to make predictions for the following experiments in which subjects were given specific strategies. In the final experiment, subjects compared familiar objects, and a tentative processing model was formulated based on the results. The overall findings suggest that various strategies (cognitive processes) can be used to memorize and compare features of each level of representation. There seemed to be a tendency to perform a global comparison first (usually based on primal sketch features), and if this did not produce an unambiguous `same' decision, to then compare higher-level features. In conclusion, Marr's model was found to be useful in formulating cognitive processes involved in memorization and comparison of visual stimuli. However, some modifications were suggested. These included: (1) the peripheral region of an image allowing only low-resolution information, (2) the inclusion of an extra feature in the 3-D representation - the natural `top' of the major axis of an object, and (3) higher-level grouping processes at the level of the 3-D representation.(D72487/87)

This record has no associated files available for download.

Contributors

Download statistics