Fingerprints are always captured from different sensors

and to segment these fingerprints accurately with a united algorithm is very difficult.A universal algorithm based on linear classifier is proposed to sovle the problem.This algorithm uses three new general features extracted from the fingerprint to train a linear classifier with SVM

and then segment the fingerprints with the classifier.These new general features include the dominant frequency band energy ratio

the normalized gray mean value and the gray contrast.The experiments in the fingerprint database FVC2002 show that

this algorithm can segment the fingerprints accurately

with the average error rate of only about 2%

and it is universal in the fingerprint database acquired from different sensors.