This report studies the color-balancing capability of the Olympus D-220L digital camera. To gain a better understanding of how consumer-level digital cameras attempt to eliminate objectionable color casts in images taken under non-white illumination, my group developed and performed three experiments on the Olympus D-220L and the Sony DSC-F1. I will discuss the details of two of three experiments, which focus on reverse-engineering the actual color-balancing algorithm used in the camera, below. Finally, I will present Matlab simulations of the algorithms explored in this paper.
It is a well-known fact that the human eye can automatically adjust to color temperature changes within the 2800K (tungsten) to 5500K (daylight) range through a process called color constancy. A white card appears perfectly white to us under tungsten illumination, even though most of the reflected light is yellow. However this visual correction mechanism does not extend to images on a computer display or on photographic print. Simply taking raw RGB data from a camera and dumping them to the screen will not produce an acceptable image. For instance, an image captured in a tungsten-lit room will have an overwhelmingly yellowish tint. The digital camera must figure out a way to color-balance that input data automatically.
What color-balancing achieves
Read about how the image above was color-balanced
Put simply, the goal of color-balancing is to change the
appearance of the scene illuminant to one of constant spectral power density
(i.e. white light). However, with only three spectral samples per pixel,
a perfect reconstruction is impossible. Nevertheless people have developed
empirical methods that solve this problem partially.