Introduction and Motivations

The goal of this project is to investigate a means of accurately representing the spectral reflectances present in a given scene. The reproduction of the incident spectra is a rather important step in the color balancing of a scene.

Take, for example, a typical digital camera image. The only information encoding the color data of each pixel consists of three numbers representing the R, G, and B intensity values. Obviously, the light striking the device is not composed of three discrete frequencies. In actuality, the light incident on each sensor on each pixel has a continuous spectral distribution. Accurate reconstruction of this spectra is required to color match the image and produce something that resembles true color to the viewer.

For this project, our group has performed a set of experiments and simulations to investigate this area of image processing. The general questions we are attempting to answer with these exercises are:

• Do the spectral estimations (linear models) used in image processing accurately represent the original signal spectra?
• Can a typical camera be used to accurately reconstruct the spectral radiance of a scene?