This paper extracted characteristic points(end points,
divergent points) and the central point of fingerprints by applying the neural network algorithm. With the back-propagation algorithm of neural network algorithms, ambiguity of the central point was minimized and thus the central point was extracted. Centered on the central point, lengths between the central point and end points(and divergent points) were measured respectively.
During the pre-processing procedure, Butterworth Low-Pass Filter was used to remove noises of fingerprint image. It was learned that the frequency domain was more effective in processing fingerprint images than the space domain. So Butterworth Low-Pass Filter, which processes fingerprint images in the frequency domain, was adopted in this research. Smoothing, binarization, sessionization, and histograms equalization were extracted. Then, the data of orientation were used as input data for the neural network in order to extract the central point of fingerprint.
TI's DSP(TMS320VC5509) was used as the main board of the fingerprint recognition system, and ATMEL's AVR(ATmega16L) was used as the control board. MFC, which was chosen as personal information inquiry system, manages and displays detailed personal data on PC screen by connecting the fingerprint recognition system
through serial communication. In the personal information inquiry system, data sources are made and registered at ODBC Data Sources manager before making programs.
Fifty fingerprints of 10 people(five fingerprints of each person) were used to check the recognition rate. Verification results were retrieved by comparing one fingerprint of a person with forty-five fingerprints of the other persons(FAR(0.1)) and one fingerprint of a person with the rest four images of that person(FRR(4.5)).