With 24 bit color, there are 3 bands: blue, green, and red. Vivid yellow is (B=0,G=255,R=255). Vivid white is (B=255,G=255,R=255). The pixel value can be the entire vector or one value in a specific band.
With Landsat TM, there are 3 bands of visible, then 3 bands of infrared, then a thermal band as well.
With hyperspectral imagery, there are many bands.
Feature extraction can be done by unsupervised classification. You might look for water by looking for all the pixels that are high in the blue channels and low in the red channels and above. Then you might separate out specific streams or lakes.
It can also be done just by digitizing the imagery -- click with a mouse or other digitizer to get a mountain ridge, road, or other feature.