Multi-modal Presentation Attacks Detection based on Mouth Dynamic and Speech Biometrics

• Main Authors: LEE, CHIEN-PENG, 李建鵬
• Other Authors: CHOU, CHAO-LUNG
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/u68nz3

碩士 === 國防大學 === 網路安全碩士班 === 107 === Biometric technologies have been widely used in daily life due to the advancement of information technology today. However, biometrics still have a high risks of being deceived. For example, an imposter pretends to be a legitimate user illegally accessing the system. This study proposed a countermeasure for the “Video Attack” in face recognition system based on the multi-modal method combined with motion detection and speech recognition. The motion is detected in a continuous time based on the mouth aspect ratio (MAR) while the user is talking. The similarities between the talk and the recognized speech are compared. The score level fusion method is used to fuse these two features, and then the Decision Tree, Random Forest, k-Nearest Neighbor and Naïve Bayes classifiers are used to conduct classifying and testing in the experiments. Experimental results show the accuracy of the proposed method for Video Attack detection reaches as high as 95.17%. It also shows that the proposed multi-modal presentation attacks detection method can effectively improve face recognition system security.