<p>Segmenting thermal infrared images is essential for identifying physiological and pathological conditions. Accurate detection and monitoring of thermal variations are crucial for timely diagnosis and effective intervention. This study compares advanced and basic denoising techniques for detecting facial thermal dysregulation, specifically hot flushes, in rhesus macaques using infrared thermal images. Using the U-Net architecture for image segmentation, an Intersection over Union (IoU) of 0.896 was achieved without any denoising methods. Advanced denoising techniques, including anisotropic, wavelet, total variation, bilateral, patch-based, and adaptive Wiener denoising, were explored, resulting in IoUs of 0.923, 0.920, 0.917, 0.912, 0.908, and 0.908, respectively. These results were compared with basic denoising filters such as mean, median, and Gaussian. Among these, anisotropic denoising provided the most significant improvement, boosting the IoU by 0.027 (3.0%), with wavelet denoising closely following with a 0.024 (2.7%) increase. Furthermore, combined filters such as wavelet-median, anisotropic-median, wavelet-mean, and anisotropic-mean were tested, resulting in IoUs of 0.927, 0.923, 0.921, and 0.915, respectively. In the current study, these combinations led to a relative improvement of up to 3.5% over non-filtered images, further enhancing the accuracy and effectiveness of thermal image analysis. The research highlights the importance of removing diverse types of noise, such as Gaussian noise and salt-and-pepper noise, while preserving essential image details like edges and textures. This is critical for the early detection and monitoring of facial thermal dysregulation. The insights gained from this study are valuable for both medical and veterinary applications, by enhancing the accuracy and effectiveness of thermal image processing.</p>