I can understand why definite information is not given for the algorithm (propriatary information!), but I would like to know the general approach taken.
For example, I have processed a set of 5 graphics, where I was comparing them to one of the five. I put the one graphic in a separate folder, and then compared the two groups.
In one "run," the graphics are in color.
In another "run," the graphics are greyscaled.
The "master" image (call it "A," was rated as follows in the two groups (hope the columns show up ok! The numbers indicate percentage of agreement):
It is clear that the last two figures rank about the same. However, the figures in color as comared to greyscale values are "switched." (By the way, the figures involve different images of different objects.
My question: what is stressed more: color, or "shape?"
Does size of the figures have any impact on their similarity ratings?
Thanks for any info!