Panel Summary Foveation, Log-Polar Mapping and Multiscale Approaches to Early Vision
One of the ways by which early human vision is clearly distinguished from current machine vision is the fact that human vision is strongly space variant, and build up a (multiscale) representation of the world from those space variant fixations. In this panel, we will ask how relevant is this principle of human vision to machine vision (Y. Yeshurun), and present the principles of multiscale (C. Guerra) and wavelets (I. DeLotto) approaches that are closely linked to the issue of representation.
Why is foveated vision necessary (if at all.)? or: Should we blindly imitate biological vision?
What is the preferred type of foveated vision? (e. g. complex log, pyramid)
What are the effects of foveated vision on customary vision algorithms? (from edge detection to object recognition)
Effectiveness of multiscale and wavelet techniques.
KeywordsMachine Vision Human Vision Multiscale Approach Early Vision Foveated Vision
Unable to display preview. Download preview PDF.
- 3.A. Rojer and E.L. Schwartz, Design consideration for a space variant visual sensor with complex logarithmic geometry, Proc. 10th IAPR, pp. 278-285 (1990).Google Scholar
- 4.V. Cantoni, L. Cinque, C. Guerra, S. Levialdi, and L. Lombardi, Tree pattern matching for 2D multiresolution objects, Proc. CAMP’93, New Orleans, LA, pp. 43-46 (1993).Google Scholar
- 7.C.K. Chui, Wavelets: a Tutorial in Theory and Applications, Academic Press, NY (1992).Google Scholar
- 8.S.G. Mallat and Z. Zhang, Matching pursuit with time frequency dictionaries, IEEE Trans. SP (in press).Google Scholar