CBIR system - restriction to retrieve top twenty matching images produces same curves of precision, recall and f-measure

Question

I want to know if this restriction is even a valid option?

Because we can't judge the precision recall, f-measure tendency independently. If it's ok then how to justify the same tendency of all precision, recall and f-measure curves?

I have used a dataset of 500 images, 100 images per class. I've used one random image per class to check the algorithms and the idea is to retrieve top 20 matching images (true positive + false positive). Where I am missing the point?

NOTE: I'm using Matlab.

Here are the final graphs: