# Quantifying Results - MULTIMEDIA

Generally speaking, some simple expression of the performance of image search engines is desirable.

In information retrieval theory, precision is the percentage of relevant documents retrieved compared to the number of all the documents retrieved, and recall is the percentage of relevant documents retrieved out of all relevant documents. Recall and precision are widely used for reporting retrieval performance for image retrieval systems as well. However, these measures are affected by the database size and the amount of similar information in the database. Also, they do not consider fuzzy matching or search result ordering.

In equation form, these quantities are defined as

In general, the more we relax threesholds and allow more images to be returned, the smaller the precision, but the larger the recall. The curve of precision versus recall is termed a receiver operator characteristic (ROC) curve. It plots the relationship between sensitivity and specificity over a range of parameters.