Improving the fingerprint verification performance of set partitioning coders at low bit rates

Abstract

Wavelet transform combined with the set partitioning coders (SPC) are the most widely used fingerprint image compression approach. Many different SPC coders have been proposed in the literature to encode the wavelet transform coefficients a common feature of which is trying to maximize the global peak-signal-to-noise ratio (PSNR) at a given bit rate. Unfortunately, they have not considered the local variations of SNR within the compressed fingerprint image; therefore, different regions in the compressed image will have different ridge-valley qualities. This problem causes the verification performance to be decreased because minutiae and other useful features cannot be extracted precisely from the low-bit-rate-compressed fingerprint images. Contrast variation within the original image worsens the problem. This paper deals with those applications of fingerprint image compression in which high compression ratios and preserving or improving the verification performance of the compressed images are the main concern. We propose a compression scheme in which the local-SNR (signal-to-noise ratio) variations within the compressed image are minimized (and thus, general quality is maximized everywhere) by means of an iterative procedure. The proposed procedure can be utilized in conjunction with any SPC coder without the need to modify the SPC coder's algorithm. In addition, we used image enhancement to further improve the ridge-valley quality as well as the verification performance of the compressed fingerprint images through alleviating the leakage effect. We evaluated the compression and verification performances of some conventional and modern SPC coders including STW, EZW, SPIHT, WDR, and ASWDR combined with the proposed scheme. This evaluation was performed on the FVC2004 dataset with respect to measures including average PSNR curve versus bit rate, verification accuracy, detection error trade-off (DET) curve, and correlation of matching scores versus the average quality of involved fingerprint images. Simulation results showed considerable improvement on verification performance of all examined SPC coders, especially the SPIHT coder, by using the proposed scheme.