Year: 2007
Type: dissertation
Abstract: Perceptual similarity is thought to play a fundamental role in human cognition.In this Ph.D. thesis, a set of parametrically-defined, novel 3D objects were used to investigate perceptual similarity from three new angles: (1) studying the effects of seeing versus touching, (2) comparing the effects of touching using different hand movements, and (3) comparing human similarity judgments to similarities computed by state-of-the-art algorithms.In all three studies, similarities were analyzed using multidimensional scaling (MDS) techniques, which provided maps of the objects in "perceptual similarity space," as well as the relative weights of perceptual dimensions.Maps and weights were then compared across conditions.In the first study, differences between judging similarity using vision, touch, or both were demonstrated and modeled by linearly rescaling a single perceptual map.The second study showed that using different hand movement patterns can also lead to changes in perceptual similarity.Furthermore, when a given hand movement allowed for the extraction of more than one object property, relative property weights varied across individuals; these individual biases remained stable over several months.In the third study, good fits were found between perceptual spaces derived from human vision and those derived from algorithms biased towards shape extraction, however none of the algorithms tested provided a good overall fit to haptic data, in which large intersubject variability was observed.Overall, the thesis highlights the role of feature extraction mechanisms in determining similarity and demonstrates how MDS techniques can be used to visualize and quantify these effects.Using MDS, differences in similarities could be accounted for by linear rescaling a single perceptual map, illustrating a potential mechanism for connecting unimodal and multimodal representations.The thesis has also led to the development of a new, similarity-based approach to perceptual validation of virtual environments and feature extraction algorithms.
Source: MPG.PuRe (Max Planck Society)
Author: T Cooke
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W2131269613