Improving pore extraction in high resolution fingerprint images using spatial analysis

Abstract

Pore extraction appears to play an important role in high resolution partial fingerprint recognition and in applications involving large population or high security levels. In this paper, we introduce a novel pore extraction approach which takes into account a new relation concerning their spatial and photometric dependence. This relation is given locally by analyzing distinct pores according to their distance and contrast. We evaluate our approach on high resolution pore extraction database and in an application involving partial fingerprint alignment. The proposed segmentation has proved to be quite general, simple and can accurately extract fingerprint pores in real images with different ridge and valley widths. It is also very robust to noise and according to the considered experiments outperforms well-known state-of-art methods.

Pore-based ridge reconstruction for fingerprint recognition

“…We compare DPF's performance with four state-of-the-art approaches, as may be seen in Table 1 (i.e. results for Jain et al [5] and Zhao et al [14,15] were reported in Teixeira et al's work [13]). Table 1.…”

“…DR FDR Jain et al [5] 75.9% 23.0% Zhao et al [15] 80.8% 22.2% Zhao et al [14] 84.8% 17.6% Teixeira et al [13] 86.1% 8.6% DPF [7] 83.5% 9.9% Enhanced DPF 90.8% 11.1%…”

“…sweat pores and incipient ridges) as a complementary information for matching has attracted the attention of many researchers in the literature. Among these features, the use of pores stood out as a promising way to improve the recognition performance [5,7,8,10,[13][14][15].…”

“…Some initial works [5,15] did not adapt themselves to such variations and, as a consequence, faced large numbers of false positives and false negatives. Other works [13,14] addressed adaptability, but at the cost of using computationally expensive techniques, such as ridge reconstruction and frequency estimation. Recently, Lemes et al [7] proposed an accurate method, named Dynamic Pore Filtering (DPF), that efficiently adjusts itself to pore variations without requiring these time-consuming calculations.…”

Pore detection in high-resolution fingerprint images using deep residual network

“…However, it is imperative that the pore coordinates are detected accurately. The existing pore detection techniques can be broadly classified into feature-based techniques [1][2][3][4] and learning-based techniques. 5,6 Since deep convolutional neural networks (CNN) have achieved state-of-the-art performance for various computer vision tasks, researchers have focused on designing learning-based techniques for detecting pores in high-resolution fingerprint images.…”