Automatic segmentation of regions of interest in thermal images in the facial and hand area

Image processing in the healthcare field has evolved significantly, making it a widely used tool today. Systems for detecting regions of interest are employed in various areas such as medicine and psychology due to their versatility. This work presents an automatic system for detecting and segmenting facial and hand regions of interest in thermographic images. The proposed methodology employs MediaPipe to identify anatomical key points on the face and hands, from which image processing techniques are applied to delineate regions such as the forehead, cheeks, nose, chin, philtrum, palms, and finger phalanges. Unlike other manual or semi-automatic systems, the proposed method enables fully automatic segmentation from a single thermographic image. The system was evaluated on a dataset comprising 10,566 facial and 311 hand thermographic images. Due to the exploratory nature of this study and the limited number of hand images, the dataset was used in full for evaluation without a separate training or validation split. The method achieved a segmentation accuracy of 77% for the face and 69% for the hands, assessed manually. While promising, these results lack validation with standard metrics such as Intersection over Union (IoU) or Dice coefficient. Future work will include dataset partitioning, metric-based evaluation, and thermal-specific model adaptation to improve segmentation robustness in biomedical applications.