What to try
- Move the fixed threshold, then let Otsu choose the histogram split.
- Increase the illumination gradient and switch to local mean; vary its neighborhood radius.
What to look for: A global threshold uses one value everywhere. The local curve follows the changing background, and each pixel compares with its own local mean minus 5.
How it works
A fixed threshold compares every pixel with one cutoff. Otsu chooses a global threshold by separating the histogram into classes with low within-class spread. It is useful when the distribution supports such a separation, but is not an object-recognition method.
Adaptive thresholding compares each pixel with a local mean or Gaussian-weighted neighborhood minus a bias. It can handle uneven illumination differently from a single global cutoff. Neighborhood radius determines the scale considered local.
Use it in the Lab
Use Threshold before morphology for hard shape processing, or after a blur to suppress small changes. Set the output colors explicitly when a two-color result is part of the artwork.
Limits & distinctions
These tools use the Lab’s encoded-RGB luminance proxy and preserve source alpha. A two-color RGB result is not necessarily an opaque binary mask.
Sources & references
Guide updated