My thresholding algorithm for semacode is getting pretty good I think. Check out these results on the classic “Barbara” computer graphics test image.
Next up we have my threshold of the image (adaptive, with an algorithm based on Chow and Kaneko, but the bimodal detector isn’t as good as theirs)
Finally, by comparison, a threshold of the image by the person at this page . Mine is better in some places (like the fabric grain and the hand) but not as good in others (like the face).
(If you go there you can also see what you get with global threshold, which isn’t as good.)