A novel scheme for sweat-pore extraction & performance evaluation on multi-core

Abstract

This paper presents a novel scheme for the detection and extraction of sweat pores (categorized as one of the fingerprint level 3 micro features). The proposed scheme seamlessly fuses two distinct methods into a novel model capable of obtaining good results from fingerprint images captured through optical live-scan devices with resolutions ranging from basic 500 dpi (despite the common belief that images obtained at this resolution are not of high enough quality) to 2000ppi. The performance of the presented model is then evaluated over multiple cores, spanning from single core to quad cores. The proposed scheme is successfully tested on publicly available fingerprint datasets, which were scanned with Cross Match Verifier 300 scanner at 500 dpi (pores are quite visible in these samples), as well as on 144 images at 2000ppi resolution. The experimental results clearly show a significant performance gain of 50.91% (32-bit platform) and 79.11% (64-bit platform) for parallelized approach over sequential approach. The performance gain is surely to increase further with few more code optimizations and with increase in the dataset size.