Instructions

U.are.U SDK - Developer Guide 15
Developing Applications
Determining an Acceptable Level of Error
When identifying fingerprints, you want to identify strictly enough that you do not let unauthorized people have access
(false positives) but also do not inconvenience legitimate users by rejecting their fingerprints (false negatives). Note
that some people will always experience more false rejections -- the rate of false negatives is a statistical measure, but
individuals may experience higher rejection rates, based on their specific fingerprint characteristics.
There is a trade-off between the frequencies of false positive and false negative errors. Applications have control over
this trade-off by specifying the threshold for the required degree of similarity between two fingerprint images in order
to call it a match.
When choosing the identification threshold, note that increasing the false positive error rate by a factor of 100 will
reduce the false negative error rate only approximately by a factor of 2. There will always be some false negatives. As
the result, every practical fingerprint recognition system should have an alternative means to establish and prove
identity, without using fingerprints.
Setting the Error Threshold when Identifying a Fingerprint in a Collection
When a fingerprint is scanned, the first step is to identify the fingerprint against a set of stored FMDs.
• If you are trying to confirm that a user is allowed access, you will want to identify the fingerprint against all the
valid FMDs that you have stored.
• If you are trying to confirm the identity of a specific person, you must identify against all FMDs for that individual
(typically at least two fingers are stored for each user).
The identification function compares an FMV against a collection of FMDs to produce the candidate list. You can
specify the maximum desired number of candidates: a smaller number can make the execution faster. The candidate
list is sorted by the dissimilarity score, the lower the score the closer the candidate to the beginning of the list. The best
match is the first candidate in the list.
Your threshold determines the trade-off between false positive and false negative error rates where:
• 0 = no false positives
• maxint (#7FFFFFFF or 2147483647) = fingerprints do not match at all
• Values close to 0 allow very few false positives; values closer to maxint allow very poor matches (a lot of false
positives) in the candidate list. The table below shows the relationship between the threshold values and the false
positive identification error rates observed in our test. Note: the actual false positive identification error rates in
your deployment may vary.
For many applications, a good starting point for testing is a threshold of 1 in 100,000. If you want to be conservative,
then you will want to set the threshold lower than the desired error rate (e.g., if you want an error rate that does not
exceed 1 in 100,000, you might set the threshold to 1 in 1,000,000).
Your Threshold Corresponding False
Positive Identification
Rate
Expected number of False
Positive Identifications
Numeric Value of Threshold
.001 * maxint .1% 1 in 1,000
2147483
.0001 * maxint .01% 1 in 10,000
214748
.00001 * maxint .001% 1 in 100,000
21474
1.0e-6 * maxint .0001% 1 in 1,000,000
2147