Auto Local Thresholding does a fine job but only one can be best. Subjective? Sure. It is.
But it’s objectively superior to Auto Thresholding, which I’ll show next.
As you can see, standard Auto Thresholding is inferior but it’s very fast.
When this is used for preprocessing in facial detection for human identification, auto thresholding results in gross errors depending on lighting and skin pigmentation.
So, always choose Auto Local Thresholding for your national identity criminal tracking motion camera multi-layer neural network needs. All the supervised training can’t make up for a bad threshol.d