Výsledky výskumu nášho bývalého študenta Ivana Kovača, ktorý mal inžiniersku promóciu začiatkom júla, a nášho pracovníka Pavla Maráka boli prijaté do prestížneho časopisu Signal, Image and Video Processing. Tento časopis mal v roku 2021 Impact factor na hodnote 1.583.
Ivan Kovač, Pavol Marák: Finger vein recognition: utilization of adaptive gabor filters in the enhancement stage combined with SIFT/SURF-based feature extraction
Abstrakt: Inadequacies of traditional means of human recognition along with one’s possession of unique physiological traits have paved the way for a more reliable mechanism of authentication-biometrics. A myriad of solutions utilizing various biometric traits has been proposed. Notably, in recent years, finger vein recognition has attracted the attention of a wider research community. Due to veins being a part of intrinsic features, it is almost impossible to replicate them. Nonetheless, there is still an inferior portion of conducted research on the discriminative power of finger vein patterns. Hence, we aspire to contribute to this particular branch of biometrics. In this paper, we primarily focus on finger vein pattern enhancement by means of adaptive Gabor filters. First, a region of interest is extracted, followed by adaptive contrast enhancement. Secondly, an orientation map is computed and used as a vein direction estimation. Subsequently, the finger vein pattern is enhanced by a convolution with a Gabor filter that locally adapts its parameters to fit the local orientation and frequency of the underlying vein pattern. Preprocessing algorithm parameters were determined using our interactive GUI tool which allows experimenting with parameter values to see their impact. The extraction phase employs SIFT and SURF features which are passed to a matching stage where we use OpenCV library built-in functions to compute feature distances. Finally, we conclude the paper with a performance evaluation based on FAR/FRR indicators and genuine/impostor distribution graphs. The research is conducted on the SDUMLA-HMT, SCUT-FVD and Vera databases. The best accuracy score for SURF features was obtained with the combination of Gabor filters at 99.94% on the SDUMLA-HMT database. The best result in case of SIFT features with Gabor filter enabled yielded an accuracy of 98.32% on the Vera database. Additionally, the achieved results confirm that utilization of adaptive Gabor filters tends to improve the over-all recognition rate. A comparison table containing our and other finger vein recognition systems from the literature is presented. Our solution has been implemented as a C++ shared library.