Neural Network-Based Attention-Guided Residual Interpolation for Division of Focal Plane Polarization Imaging

The increasing demand for high-resolution imaging across diverse applications has catalyzed the development of advanced image interpolation techniques. Traditional interpolation methods often compromise image quality by introducing artifacts and losing critical detail, especially in complex images generated by Division of Focal Plane (DoFP) sensors. These sensors face unique challenges due to their low-resolution outputs and their ability to capture polarized light at different angles, e.g., 0°, 45°, 90°, and 135°. To address these challenges, we propose the Attention-Guided Residual Interpolation Network (AGRIN), a novel deep learning framework designed specifically for the interpolation of DoFP images. AGRIN leverages residual learning and attention mechanisms to enhance interpolation performance by focusing on significant image regions while preserving high-frequency details often neglected by conventional techniques. Our results demonstrate that AGRIN significantly outperforms traditional methods, as measured by Root Mean Square Error (RMSE) and qualitative assessments, especially when interpolating liver cancer images captured at various polarization angles. The results indicate that AGRIN effectively captures essential features with minimal distortion, validating its efficiency and suitability for high-stakes imaging applications that require precise interpolation. This research contributes valuable insights to the field of image processing and lays a foundation for future advancements in DoFP imaging technologies.