Wavelet Scalar Quantization
The system the FBI uses is
known as the Wavelet Scalar Quantization (WSQ) method. This algorithm is based a few elementary
steps, where first, the digital source image is broken into a signal, second, the
signal is broken into a sequence of wavelet coefficients, third, a form of
thresholding is applied to the coefficients, fourth, the signal string of
coefficients if quantized, and lastly, entropy coding is applied to compress
the signal.
By scanning in an inked
fingerprint, digitalizing the source image can be done, or a fingerprint can be
obtained from a digital source as mentioned earlier. Clear and distinct ridges can characterize an ideal scanned
fingerprint image. Often images are far
from ideal, and they contain what is known as noise. Noise is any extraneous information in a signal that can be
introduced in the collection and transmission of data through a variety of
means. Some examples of noise in
fingerprint data can be shown in Figure 7.
Figure 7. Fingerprint images with noise. (a) good print (b)
and (c) noised print.
In the case with wavelets,
the image can be enhanced significantly, by assuming a reasonable model for
fingerprint images. Wavelets are used
to filter out such noise found in fingerprints. They use a method of computation averaging and detailing
coefficients.