Imagine a slider, with the grayscale imageLoading... above and a black-and-white maskLoading... below. Slide right and the mask shrinks toward the brightest pixels. Slide left and it grows toward everything. There's a sweet spot — but it's an editorial choice, not a fact.
The threshold enginesLoading... let you set the slider numerically. Otsu's methodLoading... sets it for you by minimizing intra-class variance — finding the value that best separates the histogram's two modes. The Classifier engineLoading... skips the slider entirely and learns the boundary from your annotations. All three approaches are saying the same thing in different languages: somewhere in this image, foreground stops and background starts.