[8/28] 제59회 BERC 초청세미나 안내
2009.08.28 16:58
한국과학재단 지정 우수연구센터(ERC)로 지정된 생체인식연구센터의
제59회 초청 세미나를 아래 일정으로 개최합니다.
- 아 래 -
제 목 : Stochastic Resonance for Enhanced Feature Extraction from Low-Quality Fingerprint Images
일 시 : 2009년 9월 10일 (목) PM 4:00 ~
장 소 : 산학협동연구관 310호 생체인식연구센터(BERC)
강 사 : 공성곤 교수 (Temple University, Philadelphia U.S.A.)
Abstract:
This talk addresses enhanced extraction of features from low-quality fingerprint images obtained from challenging conditions. It has been observed that a certain amount of noise helps biological organisms detect the objects from weak environmental signals. This "stochastic resonance" phenomenon reveals that an optimal noise level added to measured signal can enhance the perception of objects, rather than degrade it, in many nonlinear signal processing systems. Inspired by signal sensing mechanisms of biological systems, stochastic resonance enhances the extraction of fingerprint features from low-quality images that many state-of-the-art feature extraction algorithms tend to fail. Experiments with the FVC2004 DB2 database and a fingerprint verification algorithm by NEUROtechnology show that an optimal amount of noise enhances feature extraction and therefore enables the verification of 9 out of 20 fingerprint images in which feature extraction failed. The equal error rate decreases from 7.4% to 5.87% and the receiver operating characteristic curve shows that genuine accept rates are improved for all false accept rates.
참여연구원들의 많은 참여 바랍니다.
제59회 초청 세미나를 아래 일정으로 개최합니다.
- 아 래 -
제 목 : Stochastic Resonance for Enhanced Feature Extraction from Low-Quality Fingerprint Images
일 시 : 2009년 9월 10일 (목) PM 4:00 ~
장 소 : 산학협동연구관 310호 생체인식연구센터(BERC)
강 사 : 공성곤 교수 (Temple University, Philadelphia U.S.A.)
Abstract:
This talk addresses enhanced extraction of features from low-quality fingerprint images obtained from challenging conditions. It has been observed that a certain amount of noise helps biological organisms detect the objects from weak environmental signals. This "stochastic resonance" phenomenon reveals that an optimal noise level added to measured signal can enhance the perception of objects, rather than degrade it, in many nonlinear signal processing systems. Inspired by signal sensing mechanisms of biological systems, stochastic resonance enhances the extraction of fingerprint features from low-quality images that many state-of-the-art feature extraction algorithms tend to fail. Experiments with the FVC2004 DB2 database and a fingerprint verification algorithm by NEUROtechnology show that an optimal amount of noise enhances feature extraction and therefore enables the verification of 9 out of 20 fingerprint images in which feature extraction failed. The equal error rate decreases from 7.4% to 5.87% and the receiver operating characteristic curve shows that genuine accept rates are improved for all false accept rates.
참여연구원들의 많은 참여 바랍니다.
