Hi list,
I'm strugling with a sampling design problem. It has to do with the
optimal sampling density of an image (regular sampling that is).
I have an image with patches. Every patch has a different value lets say
1 to 10. I want to have an optimal estimate of the whole image based on
N samples. problem is, how to you calculate an optimal N?
Using a regular average would yield very variable results. If by chance
all values are low and some fall in the non patch region (=0) the
average would be very low. The high average value would also exist. So
thinking of the regular design I was thinking of searching some
threshold where the sampling becomes so dense that you actually have
additional information other then the pixel/sample values. Positive
patch samples (x > 0) are related to the total patch coverage of the
image. So I would like to find the value of N for which the patch
coverage non patch samples/patch samples converges to some value.
Is my reasoning faulty because it all sounds iffy to me. To much
interdependence etc...
The problem is actually situated in the fact that you have patches
rather than a continuous image. But this was the only workaround. If
people have their opinion or alternatives, suggestions, feel free to
give comments and feedback of any kind.
Best regards,
Koen
