I am making a preliminary selection of algorithms to use for
recognizing a collection of objects in images. I have a simulation
that creates a set of objects all the combinations of range and size
differences. I want to run a collection of algorithms that calculate
"moments" of an image on each, to determine between what types of
objects (e.g., collections of different shapes) a particular algorithm
can differentiate.

Since I am simulating discrete amounts of range and size and not yet
dealing with random variables, I will get a fixed value for the moment
calculations for each one. There is no random variance in this data
set. (Later I will test with some other, random variables in the
data.)

My question is: Can I use an ANOVA and/or cluster analysis on this
data set (with no random variables), to see between which groups of
shapes --- varying only non-random variables --- the algorithms can
differentiate over the range of factors I am varying?

I think I can validly do this. What is your thought on this?
Thanks, Alan

Alan wrote:
I am making a preliminary selection of algorithms to use for
recognizing a collection of objects in images. I have a simulation
that creates a set of objects all the combinations of range and size
differences. I want to run a collection of algorithms that calculate
"moments" of an image on each, to determine between what types of
objects (e.g., collections of different shapes) a particular algorithm
can differentiate.

Since I am simulating discrete amounts of range and size and not yet
dealing with random variables, I will get a fixed value for the moment
calculations for each one. There is no random variance in this data
set. (Later I will test with some other, random variables in the
data.)

My question is: Can I use an ANOVA and/or cluster analysis on this
data set (with no random variables), to see between which groups of
shapes --- varying only non-random variables --- the algorithms can
differentiate over the range of factors I am varying?

I think I can validly do this. What is your thought on this?
Thanks, Alan

Alan, It has been awhile but I seem to recall the independent variables
can be fixed or random and that can impact how you calculate the F
values. I think one you divide by the Mean Square Error, the other by
another term which I don't recall. The models get interesting when you
have both types of independent variables.