Remote Sensing: Classification
Multispectral images (images composed of two or more separate image layers acquired in different parts of the spectrum, such as "red", "green", and "blue") can be analyzed and classified by a number of methods. All of the methods share the idea of using the differing spectral profiles of different materials to separate the image elements into a number of meaningful classes.
The upper image compares false-color and (when you roll the mouse pointer over it) natural-color displays. In the natural-color display, the scene's red, green, and blue image layers control the amount of red, green, and blue color. In the false-color display, image layers from the infrared (non-visible) spectrum control the amount of red, green, and blue color, and it's suddenly apparent that features (such as snow and clouds) that look the same in the visible spectrum differ in other parts of the spectrum.
These differences can be quantitatively analyzed by a number of methods to place image pixels into classes. The lower image compares the natural-color display (when you roll the mouse pointer over it) with a supervised maximum-likelihood classification. An analyst identified small regions typical of feature classes such as "snow", "water", "cloud", or "rock", and then the spectral characteristics of those training regions were used to place image pixels into one of the defined classes.