Threshold Selection, 4.

TLDR: In this paper, a histogram of gray levels that occur in the picture is used to choose the threshold, where the value of some difference operation is high (e.g., above the p-title, for p =.85 or so).

Abstract: : If a picture contains dark objects on a light background, or vice versa, the objects can be separated from the background by thresholding the picture. A good place to choose the threshold is at the average gray level of those picture points where the value of some difference operation is high (e.g., above the p-title, for p = .85 or so). This idea, suggested over then years ago by Yale Katz, is verified for several classes of pictures (handwriting, chromosomes, cloud cover) and various difference operations. Another standard method of choosing a threshold is to examine the histogram of gray levels that occur in the picture. If this has two peaks, corresponding to the gray level ranges of object points and background points, then a good place to choose the threshold is at the bottom of the valley between these peaks. Mason et al recently describes a method of deepening this valley bottom, to make the choice of threshold easier. This method was tested on the pictures mentioned above; it yielded reasonable thresholds, but the valley deepening effect was not as strong as that obtained using other methods.