Right. What happens then is that the alignment calculation is likely to get misled by how things change appearance as they go way out of focus, instead of accurately detecting the relative scale of two images that are essentially identical except for scale.mjkzz wrote:I guess shooting 100s of images of some random subject and then take the average (by taking 1/(n-1)th power) would not be a good measure for telecentricity?
As an example of that difficulty, see the graphs shown by mawyatt at http://www.photomacrography.net/forum/v ... 071#243071 just for different tiles of one of his angled chip stack-and-stitch projects. While you're there, revisit http://www.photomacrography.net/forum/v ... 052#243052 for discussion that none of Mike's graphs are really exponential, despite the visual appearance of some of them.
--Rik