Model answer

The decision rule is the formula the computer uses to sort each pixel into a class.

Four common ones:

• Parallelepiped — pixel falls inside a box defined by class min/max. Fast but leaves gaps.
• Minimum Distance — pixel goes to whichever class mean is closest. Fast and never leaves anything unclassified, but ignores class shape.
• Maximum Likelihood — picks the class the pixel is most probably from, using class shape. Most accurate when classes are Gaussian.
• Mahalanobis — like Min Distance, but the distance is scaled by class shape.

The slower rules are smarter; the faster rules are dumber. Pick based on what you have time for and how Gaussian your classes look.

🔬 Show the science / technical version

• The mathematical test a supervised classifier uses to assign each pixel to a class.
• Minimum Distance — nearest class mean.
• Mahalanobis Distance — like Min Distance, scaled by class covariance.
• Maximum Likelihood — highest probability, assumes Gaussian classes.
• Parallelepiped — box-shaped regions by min/max per band. Fast, but has gaps + corner overlaps.