Expand this Topic clickable element to expand a topic
Skip to content
Optica Publishing Group

Inferring material changes with double-opponent operators

Not Accessible

Your library or personal account may give you access

Abstract

An important goal of early vision is to determine which edges in the image correspond to boundaries between different materials. When there is a “spectral cross-point” at a luminance edge, it can be reliably inferred that, regardless of illuminant color, the edge is due to a material change.1 [If L and S denote two spectral samples of image intensity, there is a cross-point when (LxLy)(SxSy) < 0, where subscripts x and y denote the two regions on either side of the edge.] There will almost never be a cross-point at a nonmaterial edge, such as a shadow, highlight, or surface orientation discontinuity. Another simple condition at an edge is opposite slope sign or (LxSx)(LySy) < 0. When the illuminant is white (or equivalently, after the image has been spectrally normalized), the opposite slope sign condition also reliably indicates material changes2; this condition also does not arise at nonmaterial edges. The cross-point and opposite slope sign conditions are independent. Detecting these two conditions requires two distinct kinds of double-opponent operators. Biological double-opponent cells may detect material changes. It can be tested whether there are two varieties of such cells, one computing the cross-point condition, the other, the opposite slope sign.

© 1985 Optical Society of America

PDF Article
More Like This
Experiments with a spatiochromatic model of early vision

Ron Gershon and John K. Tsotsos
WD2 OSA Annual Meeting (FIO) 1985

Linear opponent-colors model optimized for brightness prediction

Gerald L. Howett
WU3 OSA Annual Meeting (FIO) 1985

Structure into motion: sensitivity to change in early optical flow

Steven W. Zucker
TUB5 OSA Annual Meeting (FIO) 1985

Select as filters


Select Topics Cancel
© Copyright 2024 | Optica Publishing Group. All rights reserved, including rights for text and data mining and training of artificial technologies or similar technologies.