A Vector Signal Processing Approach to Color
Abstract
Color is a useful visual cue for obtaining surface material information in a scene. Unfortunately, the surface color of an object can sometimes be different from its recorded color in an image because of optical effects like highlight and inter-body reflection. These effects often confuse traditional color algorithms that assume all surfaces in a scene to be perfectly lambertian. This thesis adopts a signal processing approach to color vision. It represents surface color as a vector signal in a 3-dimensional color space, from which we extract region and boundary information. We immediately face two problems with this approach. The first, which is the same problem that traditional color algorithms face, is that highlight and inter-body reflection effects make image color different from surface color. We use a simple but effective method based on the theory of polarizing filters and electromagnetic wave reflection to correct for these effects on dielectric material surfaces in the scene. The second problem is to augment traditional scalar signal processing tools for 3-dimensional color vector signals. We linearize color by defining a notion of color similarity and difference, and use these measures in place of their traditional scalar counterparts in our vector signal processing algorithms. The main contribution of this thesis is the systematic approach we propose that allows us to extend scalar signal processing tools and concepts into a multi-dimensional vector domain. We also discuss some ways of integrating surface color information with grey level intensity information.... Color vision, color noise, image segmentation, polarizers, vector signal.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1992
- Accession Number
- ADA259499
Entities
People
- Kah-kay Sung
Organizations
- Massachusetts Institute of Technology