A novel texture descriptor using machine learning for face anti‑spoofing detection
| dc.contributor.author | Dr. Mohamed Elrashidy | |
| dc.date.accessioned | 2025-12-16T11:04:56Z | |
| dc.date.issued | 2025-02-20 | |
| dc.description.abstract | Face anti-spoofing has gained increased traction because of its significant contribution to the safety of face recognition systems. In order to gain access to the user resources, the attacker uses a picture or video to execute a face spoofing attack against the authentication system of the legitimate user. This study proposes a novel descriptor to counter spoofing attacks called Comprehensive Correlational Pattern (CCP) based on texture features. The images go through the initial pre-processing stage to eliminate the impact of the background during the feature extraction stage. Using Multi-task Cascaded Convolutional Networks (MTCNN), faces are detected from images. CCP employs 2D filters of 5 × 5 window size to extract spatial information. To cover all the texture characteristics, three new blocks of 3 × 3 window size are proposed, which are computed for each 2D filter, considering all possible acute and obtuse angle directions from the center pixel, and fetching features which are rotational invariance in nature. Support Vector Machine (SVM) classification algorithm is used to compute the hyperplane equation of the extracted features and fit the values of the object characteristics that can categorize a face image as real or spoof. To verify the suggested descriptor, five experiments were conducted on three benchmark datasets Replay-Attack, MSU-MFSD and NUAA datasets. It attained the highest F1-scores of 100%, 98%, and 92%, alongside the lowest Half Total Error Rates (HTER) of 0.071%, 3.496%, and 7.84% on the Replay-Attack, MSU-MFSD, and NUAA datasets, respectively. The experimental results demonstrated that the suggested descriptor CCP achieved a clear improvement rate in the classification accuracy metrics over the texture descriptor approaches and state-of-the-art anti-spoofing methods. | |
| dc.identifier.uri | https://research.arabeast.edu.sa/handle/123456789/485 | |
| dc.language.iso | en | |
| dc.publisher | Springer | |
| dc.title | A novel texture descriptor using machine learning for face anti‑spoofing detection | |
| dc.type | Article |