Thresholding & Comparison

The decision is mostly determined by what's on your input side and what your foreground looks like:

One grayscale imageLoading..., foreground brighter than a value: Threshold.

One grayscale image, foreground inside an intensity band: Threshold - LT, GT.

One grayscale image, you don't know the right value: OtsuLoading..., then look at the result and adjust.

Two grayscale images, foreground where one is brighter than the other: Compare.

An RGB image, foreground has a specific color signature: Threshold (RGB) or Compare - inside interval (RGB), depending on whether the test is per-channel intensity or color-band membership.

An RGB image with foreground outside a known color band (e.g., everything not background-blue): Compare - outside interval (RGB).

The engines referenceLoading... has the per-engine parameters; this page is the family map.

A threshold is a single number standing in for a biological boundary — a categorical answer derived from continuous data. The pixels just above the threshold and the pixels just below it are usually similar in intensity; the threshold is the line you draw across that continuum to make a decision the rest of the workflow can act on.

That decision has a sensitivity. If moving the threshold from 100 to 95 changes the foreground area by 30%, the analysis is highly threshold-sensitive — worth reporting the value explicitly, or using automatic methodsLoading... that produce a reproducible choice. If the same change shifts the area by 1%, the foreground is well-separated and the choice is robust across small variations.

The histogramLoading... tells you which regime you're in before you commit. Two clean peaks: robust separation, the threshold lives in the valley between them. One smooth distribution: the threshold is an editorial line on a continuum, and the choice deserves articulation.