All of the image compression techniques described in this book have assumed single-band, i.e., monochrome, images. In many imaging applications, it is necessary to deal with color or multispectral images. Typically, a color image is represented by three bands (or planes), corresponding to red, green, and blue tristimulus values, denoted R(i, j), G(i, j), and B(i, j), respectively, at each pixel location (i, j). In some applications, such as remote sensing via satellites, an image may contain substantially more than three bands in order to provide information over a wide range of wavelengths.
Extending the compression techniques to color images can be done easily by encoding each band independently using the same technique. Unfortunately, this simple approach is generally not optimal in terms of providing the most efficient compression. This is because there is often substantial correlation between the various color planes, and this redundancy is not removed by the independent processing of the planes. The correlation is mainly due to the fact that typical scenes are characterized by smooth spectral reflectances and partly because the spectral shape of the tristimulus color sensitivity functions overlap.
To achieve efficient compression with color images, the problem can be approached in two different ways: one based on the statistical properties of the color planes, and another based on the HVS encoding and perception of color. In the following, we briefly explain each of the two different approaches.
Online access to SPIE eBooks is limited to subscribing institutions.