A VLSI architecture for lifting-based wavelet packet transform in fingerprint image compression.

摘要

FBI uses a technique called Wavelet Scalar Quantization (WSQ), a wavelet packet transform (WPT) based method, to compress its fingerprint images. Though many VLSI architectures have been proposed for wavelet transform in the literature, it is not the case for the WPT. In this thesis, a VLSI architecture capable of computing the WPT is presented for application of WSQ. In the proposed architecture, Lifting Scheme (LS) is used to generate wavelets instead of the traditional convolution filter-bank (FB) specified in original standard. A comparative study between LS and FB shows that quality of images transformed by LS is completely acceptable (with 30dB∼40dB PSNR at a target bit rate of 0.75dpp) while fewer operations required. In particular, to compare with FB, the hardware consumption, for our WSQ application, is reduced to half due to the LS. Moreover, this architecture can be easily configured to compute any required WPT application.