Enhancement of fingerprints by adaptive directional filter and development of fingerprint verification system

Abstract

Adaptive directional filter using Fourier short space transform was investigated to enhance fingerprint image. Enhanced image was binarized by adaptive dynamic binarization and weighted averaging. Uni-boundary extraction algorithm and minutia model on $3 \times 3$ mask were applied to extract features of the image and to characterize it. The vergication systeml consisted of a light source with LED array, a prism, a CCD camera, an image frame grabber, image enhancement, feature extraction, and pattern match processor. There were many false minutiae in the image obtained using the total internal reflection property of a prism and no appropriate enhancement process. It was difficults to accomplish a correct match between input and registered patterns with such false minutiae. Thus enhancement process suggested by Tarng was preponderently applied and modified. Major modifications were the window function applied to the directional filter to reduce artifacts, the number of angular orientations, and the number of passband. When the adaptive directional filter was applied, the number of false minutiae was reduced to less than ten, and sometimes it was one or two. Convolution was also applied in spatial domain to reduce noise and to smooth the binarized image. Band pass filter in addition to the directional filter reduced an incorrect filling between ridges compared with the initial directional filter with $45^\circ$ bandwidth and Butterworth type tail. Adaptive binarization with weighting factors -0.2 gave an optimal erosion of binary image. Feature extraction accuracy of at least 82\% was obtained by uni-boundary extraction and this value was higher than that (77\%) obtained without the directional filter.