Can you clarify your question?

I am tempted to say that without ground truth, you can only compare performance in the sense of computational speed.

If you have a large enough number of techniques, you can make as estimation of the "ground truth" based on the median or average of the threshold regions obtained with each method.

Given this assumption, you can then evaluate which of the individual methods comes closer to the median or average.

Keep in mind however, that being close to the average or median does not necessarily mean it is the best threshold for all applications. This needs to be checked for each case.

@Lambert Zijp.. Actually i implemented 20 different thresholding techniques. But i need to find out, best among them. This can be done using ROC (Receivers Operating Curve ) method, however it needs reference thresholding value, which is not available. So, if any one can suggest how to find the performance of these techniques without any knowledge of "Actual Threshold" vale.

Thanks all of you for your Concern.

In medical imaging, when we want to test some segmentation method, we ask the expert: several physicians will delineate the region of interest. One could compare with the mean or median of these delineations. It will not be the ground truth; just the best we can do at the moment.

If possible, use theoretical images, like April Khademi suggested.

Using the mean or median of your thresholding methods will NOT give you the ground truth.

First, you need to identify what is of interest in your application and what is not.

In radar there is a common saying that "one man's target is another man's chaff".