For training the number of images depend of the process but I think that higher number of images is better. For testing the most popular practice is download the FVC databases and use it, but If you do not have the possibility to do that and have a images to test, you can describe the test process with the number of images, the distribution of the images and the conditions in which the image was taken.
I use the databases for research , but in all the article they only mentioned number of image ans number of user, they dont metione the number of image for test and for training