<![CDATA[Face detection plays a vital role in biometric, security, and surveillance systems. Conventional approaches based on the visible light (VIS) spectrum often suffer performance degradation under poor lighting conditions, limiting their reliability. To address this issue, this study employs thermal imagery in the Near-Infrared (NIR) spectrum, which is less affected by ambient light, combined with image morphology operations to enhance segmentation accuracy. Experiments were conducted using the LDHF-DB dataset (300 images at distances of 1 m, 60 m, and 100 m) and a subset of the Tuft dataset (60 images). Face detection was performed using the HOG + SVM method, followed by Otsu thresholding and morphological operations. Performance was evaluated using Peak Signal-to-Noise Ratio (PSNR). Results show that applying morphological operations significantly improves PSNR values, with an average increase of more than 35%. The best performance was achieved on the 1 m subset, while longer distances presented greater challenges. These findings highlight the potential of integrating NIR thermal imagery and morphological processing to improve the robustness and reliability of face detection systems in low-light environments.]]>
Copyrights © 2025
