The attributes used by the classifier to identify the regions can be:

• Spectral: each attribute is a channel pixel value from a color or multi-band image.

• Spatial: each attribute is measurement on a given neighborhood of the target pixel. Generally these measurements reveal the local texture surrounding the pixel.

Pixel classification methods can be split into two categories:

• Unsupervised classification: the input image is automatically partitioned into a predefined number of regions.

• Supervised classification: regions are learned thanks to a training step on representative samples.