🎓 Salah Eddine Bekhouche defends his PhD on facial age estimation and drowsiness detection!

🎉 Big congratulations to Salah Eddine Bekhouche on the successful defense of his PhD thesis on July 17, 2025, at the University of the Basque Country (UPV/EHU)! 🎉

Salah’s PhD, supervised by Dr. Fadi Dornaika, tackled two major challenges in computer vision: facial age estimation and driver drowsiness detection. His work is a significant contribution to the development of robust and efficient AI systems for real-world applications.

🧠 Thesis Contributions

📅 Facial Age Estimation

• Comprehensive Study of Feature Types: Salah conducted a thorough comparison between handcrafted texture descriptors (like LBP, LPQ, BSIF) and deep features from pre-trained CNNs (VGG16, ResNet-50, VGGFace, DEX-IMDB-WIKI). He demonstrated the trade-offs between speed and accuracy, highlighting how deep features bring superior robustness, while handcrafted ones are computationally lightweight.

• Multi-Stage Deep Neural Network (MSDNN): He proposed a novel deep architecture that fuses intermediate and final features from CNN backbones (EfficientNet, MobileNetV3), significantly improving age prediction performance while keeping the computational cost low. Evaluations on MORPH2, CACD, and AFAD datasets confirmed state-of-the-art results.

😴 Driver Drowsiness Detection

• Blink Detection with Spatiotemporal CNN: Salah developed a two-stage framework for blink detection using a combination of SVD-based temporal signal extraction and a novel ST-CNN architecture. The integration of Pyramid Bottleneck Blocks (PBBs) allowed the model to effectively detect blinks with high accuracy, even under motion or variable blink durations.

• Hybrid End-to-End Drowsiness System: He introduced FCFS (Features Clustering & Feature Selection) — a powerful combination of K-means++ clustering and three selection techniques (mRMR, NCA, ReliefF) to identify the most discriminative temporal features. This strategy reduced dimensionality drastically (from 4096 to <300 features) while improving detection accuracy, enabling real-time applicability.

🧑‍⚖️ Defense Committee

The PhD evaluation committee was composed of:

• Dr. Abdelmalik Moujahid (Universidad Internacional de La Rioja, Spain)
• Dr. Franck Davoine (Centre National de la Recherche Scientifique, France)
• Dr. Jinan Charafeddine (Pôle Universitaire Léonard de Vinci, France)
• Dr. Karim Hammoudi (University of Haute-Alsace, France)
• Dr. Blanca Cases (UPV/EHU, Spain)

👏 Well done, Salah! Your work opens the door to more intelligent, real-time human behavior understanding systems. We wish you all the best in your next adventure!