How to calculate the ZNCC threshold?

I'd like to know how to fix the threshold of ZNCC method and how it is calculated? paper in this field will be desirable

thank you in advance

[Image Analyst]

I've never heard of that acronym that you evidently think is so common that everyone knows about. I doubt I'm the only one who doesn't know what it is. What field of study does ZNCC pertain to?

[Roger Stafford]

@Image Analyst: ZNCC is short for "Zero mean Normalized Cross-Correlation" and is a precisely-defined quantity. For example see

http://vision.deis.unibo.it/fede/papers/iciar04.pdf

[remos remos]

This is an image matching method I want to know how to calculate the threshold to say that points are matched the interval results thank you

[remos remos]

thank you Roger Stafford

Answer 1

[Roger Stafford]

This paper gives a method of computation:

   http://vision.deis.unibo.it/fede/papers/iciar04.pdf

As for choosing a threshold, it is not computed but is something that a user must choose based on comparison with samples that are to be considered alike and those considered different. The ZNCC value lies between -1 and +1. The closer to +1 the greater the similarity.

[remos remos]

I want to know how to justify the choice of a threshold of 0.8 or 0.6 for example is there not a study already done? thank you

[Roger Stafford]

Choosing an appropriate threshold value is something that depends on a number of variables: 1) the nature of the imaging system, 2) the typical variations in the scene to be viewed that are to be disregarded versus those of significance, and 3) the degree of certainly the user wishes to have so as not to miss a significant event versus the expense of responding to non-significant events. I see no way of choosing such threshold values that does not involve a careful consideration and experimentation by the user rather than some automatic computation. Would you trust such an automatic method, Remos?

[remos remos]

so no means for the choice of the threshold. experimentation is the only way thank you my friend for these kind explanations

Answer 2

[Image Analyst]

normxcorr2() may do what you want. It normalizes to zero mean and unit standard deviation. See the definition at https://en.wikipedia.org/wiki/Cross-correlation#Normalized_cross-correlation and see if that will work for you.

What do you need to do, and why do you think you need this specialized version of the normalized cross correlation instead of the regular one? Show us your image and template and show us how normxcorr2() fails on your images so we can help you further.