Otsu Multi-Threshold¶
The standard Otsu’s method (see Thresholding and Binarization) finds a single threshold that best separates an image into two classes. Multi-threshold Otsu generalizes this to \(N\) classes by finding \(N-1\) thresholds that jointly maximize the between-class variance across all classes at once — directly from the image’s histogram, with no training or seed points needed.
This makes it well suited to images with several genuinely distinct intensity populations (e.g. more than one material phase, plus background), where a single binary threshold would lump some of them together.
Accessible via spice.clustering.otsu_multi_threshold_clustering().
See also
Thresholding and Binarization — the two-class case this generalizes
Histograms — the histogram Otsu’s method operates on
Image Clustering — every other clustering algorithm