Digital signal and image processing
Explainer
How Otsu picks the perfect threshold
Your computer can choose a threshold better than your eyes can
Quick check · tap an answer to continue
Lesson notes
Key points
- Thresholding turns a greyscale image into black and white by labelling each pixel above a chosen grey level as foreground.
- When the histogram has two humps, one for background and one for objects, the best threshold sits in the valley between them.
- Otsu's method tries every threshold and picks the one that maximises the between-class variance, separating the two groups most.
- Guess the catch: uneven lighting ruins one global threshold, so you use adaptive thresholding on local regions instead.
Definitions
- Thresholding
- Pixel becomes 1 if its grey level exceeds threshold T, otherwise 0, splitting object from background.
Takeaway: Otsu picks T that maximises between-class variance.
Quick check: What quantity does Otsu's method maximise when choosing a threshold?
- Within-class variance
- Mean grey level
- Between-class variance
Maximising between-class variance best separates foreground and background.
ThresholdingBimodal histogramOtsu's methodAdaptive thresholding
Lessons like this, for your syllabus.
Short narrated study videos written for your course, level and exam. Free to start.
Get lessons for your own syllabus