We are proposing video quality assessment using green learning principles, with the objective of identifying visual distortions while minimizing computational and energy costs. Instead of relying on global frame analysis or large models, the approach emphasizes efficient, local feature extraction that captures distortion-related characteristics. By analyzing color variations, edges, textures, and structural changes at the patch level, the system is designed to detect degradations caused by compression, processing, or tampering in a scalable and sustainable manner.
To further improve efficiency and discriminative power, our group proposes a method named DFT to identify the most informative features. After spatial filtering, features are transformed into the frequency domain, where DFT is used to analyze their spectral behavior and assign importance scores. This allows the model to focus on frequency components that are most sensitive to distortions while discarding redundant information. The selected features are then used to train a lightweight machine learning model and evaluated on unseen videos, ensuring a balance between accuracy, interpretability, and green learning objectives.
