What is Image Thresholding?
Image thresholding separates pixels into classes by comparing brightness, color, or another measured value with one or more thresholds. A global threshold applies one rule throughout, while adaptive methods vary it by region.
How Image Thresholding works
Thresholding converts a measured image channel or derived score into discrete regions by applying decision boundaries. A single cutoff is adequate when foreground and background distributions remain separated, while local rules estimate cutoffs from neighborhoods under changing illumination. Multiple cutoffs can form more than two classes. The resulting mask often precedes connected-component analysis, OCR, morphology, measurement, or vector tracing in document and inspection pipelines.
Key facts
- Otsu’s method selects a global cutoff by minimizing within-class variance, but overlapping or strongly imbalanced intensity populations can yield a poor separation.
- Adaptive methods depend on neighborhood size: a window that is too small follows texture and noise, while one that is too large fails to compensate for local shading.
- Thresholding an anti-aliased edge chooses a hard contour from partial-coverage pixels; the chosen cutoff therefore changes measured area and apparent stroke width.
When Image Thresholding matters
Use global thresholding for consistently illuminated material and adaptive thresholding for documents or scenes with uneven lighting. A poorly selected threshold can erase faint features or merge foreground noise with the subject.