Intermediate Course

Learning Objectives

Apply advanced threshold methods (Triangle, Yen, Huang, etc.)
Understand global vs. local thresholding
Use morphological operations (Open, Close, Erode, Dilate) to refine masks
Handle noisy or multi-channel images and overlapping objects
Understand and apply basic watershed segmentation (e.g., common pitfalls)

Key Concepts

Click the arrows to learn more.

Web Version

Desktop Version

Practice Exercise

Try the following steps on a noisy or multi-channel dataset:

Apply Enhance Local Contrast (CLAHE) (Process → Enhance Local Contrast).
Experiment with Triangle, Yen, Huang thresholds (Image → Adjust → Threshold).
Refine with morphological ops (Process → Binary → Open, Close, Erode, Dilate).
Try Watershed Segmentation (Process → Binary → Watershed) separate touching objects.

Quick Quiz

1. What is the main difference between Triangle and Yen threshold methods?

Triangle works only for brightfield images; Yen for fluorescent images Triangle sets threshold at a “valley” in the histogram; Yen uses a criterion maximizing correlation/entropy They are actually identical algorithms

2. What does “Enhance Local Contrast (CLAHE)” accomplish?

It is a morphological operation that erodes away boundaries It globally scales the brightness of the entire image It improves contrast locally, helping with uneven illumination

3. Why might you choose a local thresholding method over a global threshold?

Local thresholding is generally faster than global If illumination or background intensity is uneven across the image Global thresholding is never recommended

4. What is the primary purpose of a morphological Close operation?

Remove small speckles by eroding first, then dilating Fill small holes or gaps by dilating first, then eroding Sharpen the edges of objects

5. Which statement describes watershed segmentation best?

It merges all objects into a single connected component It separates touching objects by treating the image as a topographical surface It is only applicable to time-lapse images

6. What is a common pitfall when using watershed?

It will not work on 8-bit images Over-segmentation in areas of slight intensity variations It deletes the original image data