Liveness Detection based Anti-spoofing method in Face Recognition
| dc.contributor.advisor | Menore, Tekeba (Mr.) | |
| dc.contributor.author | Eyob, Haile | |
| dc.date.accessioned | 2021-03-30T08:24:14Z | |
| dc.date.accessioned | 2023-11-04T15:14:43Z | |
| dc.date.available | 2021-03-30T08:24:14Z | |
| dc.date.available | 2023-11-04T15:14:43Z | |
| dc.date.issued | 2020-12 | |
| dc.description.abstract | Today, face liveness detection and recognition are an active research area due to the fact that it’s unique biometric verification, scientific challenges and numerous practical applications. However, the effort of anti-spoofing to biometric identify are limited in accuracy as a result of the fact that its vulnerability to spoofing attack and inappropriate identification techniques. There are several sophisticated attacks in the video stream. Consequently, it becomes difficult to identify the real face from fake face. This research aims to address this challenge and develop an effective framework that can be used to deal with the face liveness detection and recognition. This research consists of two phases. The first phase is face liveness detection. The techniques which are involved in this phase are collecting data set, extracting key eye frames, preprocessing, feature extraction, feature selection and classification. Phase two is face recognition. It is cascading with the result of the face liveness detection. The techniques included in this phase are extracting key face frames, preprocessing, feature selection and classification. Then the face frame is classified according to their respective identification classes with the help of classifiers. Two types of classifiers with three types of descriptors are used to assist us in providing rational and fair comparisons between the state-of-the-art. The proposed approach is evaluated using institute of the automation Chinese academy of sciences (CASIA) anti-spoofing data set. Classification performance reports, 97.8% precision, 98.0% recall and 98.2% f-score were obtained using DBN classifier for liveness detection phase. Compared to the state-of-the-art approaches (DCP, LTP, and SFE method with SVM) an average improvement in f-score of 2.0% was achieved. Face recognition phase also assessed, where, 94.3% precision, 95.1 % recall and f-score 94.7% was achieved using DBN classifier. The overall system provides accuracy of 96.4%. The result showed that the proposed approach has enhanced the security performance of face liveness detection and recognition for a biometric security technology. | en_US |
| dc.identifier.uri | http://etd.aau.edu.et/handle/123456789/25771 | |
| dc.language.iso | en_US | en_US |
| dc.publisher | Addis Ababa University | en_US |
| dc.subject | Liveness detection | en_US |
| dc.subject | Recognition | en_US |
| dc.subject | Deep belief network | en_US |
| dc.subject | Local ternary pattern | en_US |
| dc.subject | Dual cross patterns | en_US |
| dc.title | Liveness Detection based Anti-spoofing method in Face Recognition | en_US |
| dc.type | Thesis | en_US |