Better Fingerprint Image Compression at Lower Bit-Rates: An Approach Using Multiwavelets With Optimised Prefilter Coefficients

Abstract

In this paper, a multiwavelet based fingerprint compression technique using set partitioning in hierarchical trees (SPIHT) algorithm with optimised prefilter coefficients is proposed. While wavelet based progressive compression techniques give a blurred image at lower bit rates due to lack of high frequency information, multiwavelets can be used efficiently to represent high frequency information. SA4 (Symmetric Antisymmetric) multiwavelet when combined with SPIHT reduces the number of nodes during initialization to 1/4 th compared to SPIHT with wavelet. This reduction in nodes leads to improvement in PSNR at lower bit rates. The PSNR can be further improved by optimizing the prefilter coefficients. In this work genetic algorithm (GA) is used for optimizing prefilter coefficients. Using the proposed technique, there is a considerable improvement in PSNR at lower bit rates, compared to existing techniques in literature. An overall average improvement of 4.23dB and 2.52dB for bit rates in between 0.01 to 1 has been achieved for the images in the databases FVC 2000 DB1 and FVC 2002 DB3 respectively. The quality of the reconstructed image is better even at higher compression ratios like 80:1 and 100:1. The level of decomposition required for a multiwavelet is lesser compared to a wavelet.

Improved Fingerprint Compression Technique With Decimated Multi-Wavelet Coefficients for Low Bit Rates

“…The fingerprint images used for validation from various databases were prefiltered by an optimum prefilter [13] whose filter coefficients are given by Eq. (2) followed by the multiwavelet transform.…”

“…(3) is used. Performance of the proposed compression technique is compared with compression achieved using multi-wavelet without decimation [13].…”

“…Compression can be further performed using the conventional wavelet based SPIHT algorithm. In the case of unshuffled multiwavelet transform, compression can be performed with modifications in the initialization of lists in SPIHT algorithm [13]. In the proposed work, an unshuffled multi-wavelet combined with decimation has been used for SPIHT algorithm for further improving its performance.…”

“…If M represents the number of coefficients in the L1L1 band of the multi-wavelet transform, the number of nodes used for initialization of lists is reduced to M/N 2 using this method. The reduction in the number of nodes is directly proportional to the improvement in PSNR [13]. For a typical input image of size 512×512 with an optimum level of decomposition, the L1L1 band contains 64 coefficients and Table.2 shows the corresponding number of nodes used for the initialization of lists with various decimation factors.…”

“…It was reported that the multi-wavelets can reconstruct a better-quality image with lesser number of transform coefficients [11,12] at lower bit rates. For improving the performance of fingerprint compression at lower bit rates, an unshuffled multi-wavelet with optimum prefilter coefficients for SPIHT algorithm was proposed in [13]. An improvement in PSNR was reported at lower bit rates compared to wavelets.…”