Global thresholding partitions the image with a single value T: pixels brighter than T become foreground, the rest background. It works best when the histogram is bimodal - two well-separated peaks with a valley between them.
Segmenting an image by intensity thresholdingOtsu's method picks the threshold that maximizes the between-class variance, i.e. it pushes the two class means as far apart as possible. It scans every candidate threshold and scores the split, so it needs no user input.
When global fails
If the object and background have overlapping intensity ranges (uneven lighting), no single threshold separates them - that motivates adaptive and region-based approaches covered later.