I am working on retinal identification. I have read alot of papers that have used DRIVE and STARE for evaluation of their proposed system performance. However, in these databases, there is only one image per subject available. The authors have not stated as well that how these data sets have been made suitable for identification.

Atleast two images are needed for identification system, one as input image for training the system and the second as query image. Then how the identification performance is evaluated using such datasets?

I need to comapre my retinal identification system with the exisiting ones. So, how can I use these datasets to make this comaprison? How the authors test their identification with datasets having only sample per subject.