Event Details

Shape categorization, by both human and machine, requires the ability to discern qualitatively critical information from a shape for internal representation and future comparative tasks, where both the part structure of objects (Biederman, 1985) and metric information (Basri et al., 1998) play a significant role. In this talk, I will discuss ongoing work that uses the MAP (maximum a priori) skeleton model devised by Feldman and Singh (2006) as a means of investigating human shape similarity. This work explores the idea that in visual shape categorization tasks, humans utilize aspects of natural shapes that correspond to those in a model where shape structure is represented as a mixture of generative and random factors. I�ll show that such a model provides a robust framework for modeling shape similarity judgments for which other (non-generative) shape representation models are inadequate.Any background readings supplied by guest speakers can be found on the perceptual science website, at the following url: http://perceptualscience.rutgers.edu/?title=Calendar