Compression of fingerprints using Ondicle Transform
Authors: Maribel Borjas, María C. Stefanelli
Number of views: 409
The purpose of this Paper is the compression
of digitalized fingerprints using wavelet transform.
To achieve this, an algorithm that allows the compression
of digitalized fingerprints, using 4 types
of wavelets (Daubechies, Coiflets, Symmlets and
Biorthogonal) was generated. The arrangement of
wavelets and the level of decomposition were variable
in the simulations. The analysis trough wavelets
can be used to divide an image into subimages of
approximation and detail. If the details are sufficiently
small, it can be fixed to zero, without drastically
affecting the image. The value below which the
details are considered irrelevant, is known as threshold.
In order to determine the value of threshold, 2
methods were used: the first one, based strictly on
the decomposition of the original image into wavelets
(Balance sparsity norm and Scarce high), the
second, based on the properties of the data of the
image being studied (entropy and entropy by levels).
The digitalized fingerprints used are those of the FVC
(Fingerprint Verification Competition) database. An
algorithm for the compression of digitalized fingerprints
was designed and programmed in Matlab
7.0. As value figures, the Peak Signal to Noise Ratio
(PSNR) and the compression ratio were used.
Trough evaluation of applied methods, it was decided
to use the one based on the properties of the
data of the image, which guarantees that the quality
of the compressed image is acceptable (higher
than 23 dB) thus and adequate compression rates
are obtained, close to 2 times the original image.
The use of local thresholding entails better results,
since each subband of detail has its own threshold.
The efficiency of each method is intimately related to the type of entry image (original fingerprint), the
wavelet used, the level of decomposition, and the
size of the samples.