In forensic biometrics, archived X-ray images play a crucial role in identification and verification. The rise in homicide cases, driven by population growth, has amplified the demand for cost-effective identification methods, such as forensic odontology. In this field, historical dental X-ray images of victims serve as ante mortem data, allowing forensic experts to compare dental patterns for identification. The success of this post mortem biometric process largely depends on the accuracy of forensic odontologists’ observations. Low-resolution (LR) images can cause perception errors, impacting both disease diagnosis and forensic identification. Therefore, a software solution capable of reconstructing high-resolution (HR) images from LR medical X-rays could be highly advantageous. Traditional approaches, like convolutional neural networks (CNNs), typically require extensive training datasets for effective HR image reconstruction, but existing medical X-ray repositories lack the volume needed for optimal CNN training. This work addresses the issue by introducing a machine-learning-based approach for HR grayscale medical X-ray reconstruction at a granular feature level from LR images. The proposed method generates sufficient training data from a limited number of X-ray images through granular feature extraction. These features capture the influence and orientation of neighboring points, which makes them robust to noise while preserving crucial elements like edges. Finally, a polynomial regression model is applied to reconstruct HR values accurately. By leveraging granular level feature extraction and machine learning, this method effectively reconstructs HR images from LR inputs, ensuring high accuracy and reliability in both disease diagnosis and forensic biometrics.