Question

The purpose of thresholding in supervised classification is

• A. to reject homogeneous classes
• B. to correct the geometry of the image
• C. to identify image speckle
• D. to identify and reject pixels not belonging to pre-defined training classes

Hint:

Think “reject option”: apply a decision threshold on class likelihood/distance so that only well-supported pixels are labeled; others are marked as {unknown}.

Answer 1

The Correct Option is D

After training a supervised classifier, each pixel gets a class label along with a confidence/distance measure (e.g., maximum likelihood probability, minimum-distance value). Thresholding sets a limit on this measure so that pixels with low confidence (i.e., not sufficiently similar to any training class) are rejected/left unclassified. It is not used to fix geometry (B), detect speckle (C), or reject “homogeneous classes” (A).