Investigations into Extreme Compression of Fingerprint Images for Identification Purpose Using Multiwavelet Transform

Abstract

With increase in number of fingerprints, compression is an inevitable process in the usage newlineof fingerprints. For the purpose of better storage space reduction, extreme compression newlineratios are desired. A good image quality without diminishing fingerprint details is newlinenecessary for a good identification performance. It is possible to increase the compression newlineratio to the level at which relevant attributes of the fingerprint are preserved. Therefore, it newlineis necessary to develop a compression method that preserves useful features, as well as an newlineidentification method that works well with extremely compressed fingerprint images. newlineFor fingerprint images, wavelet-based techniques are typically used for compression. For newlinethese techniques, image quality and identification accuracy decrease at high compression newlineratios. In the wavelet based SPIHT algorithm, the wavelet is replaced by the multiwavelet newlinefor an enhanced performance. The multiwavelet used in this work is SA4 multiwavelet. newlineSPIHT algorithm works by exploiting the spatial correlation of wavelet coefficients and it newlineneeds alterations for applying the multiwavelet transform. Three variants of newlinemultiwavelets are proposed for the SPIHT compression algorithm for a better newlinecompression performance. Unshuffled multiwavelet transform with prefilter (UMT), newlineunshuffled multiwavelet transform with optimum prefilter (UMT-POPT) and unshuffled newlinedecimated multiwavelet transform by a factor 4 (UDMT-4) are the variants considered newlinefor compression. The performance of SPIHT is inversely proportional to the number of newlinenodes used for initialization at high compression ratios. For reducing the number of newlineinitialization nodes, techniques based on UMT and UDMT-4 are proposed for the SPIHT newlinealgorithm. The number of initialization nodes is reduced to 1/4th and 1/16th using the newlinetechniques based on UMT and UDMT-4 respectively. newline newline