Thermal Identification Through Geon Representations
Abstract
The object of this research was to develop and evaluate a formal measure of shape similarity that could predict human performance in recognizing and discriminating highly similar complex shapes and objects. The similarity measure is the correlation in activation values over a lattice of columns of Gabor filters (termed "Gabor Jets"). Each column is composed of a number of filters at different scales and orientations, all centered on the same position in the visual field, similar to cortical hypercolumns (Lades et al., 1993). For amoeboid blobs, infrared images of tanks, and faces, high correlations were obtained between the similarity measure and psychophysical discrimination/identification performance. To our knowledge, this is the first time that the similarity of complex shapes has been determined directly from a measure of the shapes themselves. A different similarity basis is required, however when distinctive nonaccidental properties are available. We present rigorous evidence for the employment of NAPs in object recognition (Biederman & Bar, 1998) and the preferential tuning of cells in the inferotemporal cortex to NAPs (Vogels, Biederman, bar, and Lorincz. 2000). These findings have provided the bases for an integrated account of basic and subordinate level object classification combining geon theory with the Gabor Jet model (Biederman & Kalocsai; 1987).
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 05, 2000
- Accession Number
- ADA384292
Entities
People
- Irving Biederman
Organizations
- University of Southern California