Q-CIDNet: Perceptual Quality Aware Color and Intensity Decoupling Network for Video Quality Enhancement

Video quality in video conferencing often degrades due to poor lighting conditions and bandwidth limitations, leading to issues such as uneven illumination, noise, and reduced image clarity. Existing solutions struggle to balance foreground and background lighting, resulting in unnatural visuals. To overcome these challenges, we propose a new approach for video quality enhancement that leverages a perceptual quality framework with a color and intensity decoupling network. Our model is initialized with pretrained weights, enabling effective spatial and chromatic feature interaction. Additionally, we introduce a quality loss function based on perceptual and video quality metrics, ensuring that the enhanced outputs prioritize aesthetic appeal while preserving fidelity to the original content. We evaluate our approach on a diverse dataset comprising both real-world and synthetic videos, demonstrating its effectiveness in improving video quality across various lighting conditions.