Greetings. I’m doing a computer vision application with SIFT based feature descriptors matching.
I have for example 500 pictures.
I have a typical pattern how the pictures match.
For example the number of matched keypoints if the object appears in a set of pictures is 10-15 per picture all the others are 0-3.
It is not possible to set a threshold because there may be a situation where good is 6 and bad 1-3.
I am sure that the object appears more or less 5-10% of the time in my 500 pictures or does not appear.
So as I understand there must be a mix of poisson distribution of my data and binary distribution for object found or not.
Would be grateful if someone could help. I think the case is very common for SIFT.
Thanks.