Enhancing Facial Recognition Through 3D Depth Map Optimization
Kristina Korkutoska, Ljubinka Sandjakoska, Valentin Cvetanoski, Rasim Salkoski · 2025
Facial recognition systems have traditionally relied on 2D imagery, but they remain vulnerable to challenges such as pose variation and lighting inconsistencies. To address these limitations, this paper presents a novel pipeline that converts 3D facial models into 2D depth maps, thereby enhancing recognition accuracy. Using Open3D for rendering and OpenCV for image preprocessing, we evaluate a range of techniques—including Gaussian blur, bilateral filtering, and Sobel edge detection—to optimize feature extraction. Experiments conducted on the BU-3DFE dataset and a custom dataset reveal that the combination of bilateral filtering and Sobel edge detection yields an 8–12% improvement in recognition accuracy, while maintaining computational efficiency (~1.7 seconds per model). These findings highlight the value of depth map preprocessing in biometric recognition systems and pave the way for future integration with deep learning architectures.