Abstract

If a requirement is to have a minimum range of spatial frequencies that also provide the most general shape sufficient for classification that is independent of the particular elements that make that object, low spatial frequencies of objects are both necessary and sufficient. That finding has been recently questioned by several researchers. They agree that low spatial frequencies of objects can provide sufficient information for general shape but argue that they are not necessary by noting that forms are still classified from high-pass filtered images of objects too. However, that criticism only applies to objects described in the luminance domain. Early nonlinearities in the visual system severely limit the use of high-pass images to question the generalizability of low spatial frequencies being necessary and sufficient for the classification of general shape information. Demonstrations of such early nonlinearities are shown to clearly support that assertion. Finally, in addition to using low spatial frequencies to describe general shape information in objects, Ginsburg has used the concept of low-pass filtering in a more general sense to describe the filtering properties of visual mechanisms at whatever stage of the visual system they occur.

© 1985 Optical Society of America