COMPLEXITY AND SCENE DETECTION FOR VIDEO CODING OPTIMIZATION

The increasing demand for ultra-high-definition video content has significantly elevated the importance of efficient video compression standards. H.266/VVC (Versatile Video Coding) represents the latest advancement in this field, offering substantial improvements in compression performance. However, with enhanced efficiency comes increased encoding complexity, necessitating intelligent strategies to optimize encoding parameters. This thesis explores complexity-driven optimization in H.266/VVC encoding by evaluating two methods of video complexity assessment. The first approach, Scene-Based Complexity Classification, involves detecting scene transitions using PySceneDetect and classifying each scene according to intra and inter prediction complexities based on luminance variance and inter-frame pixel differences. This method allows encoding decisions to be guided by coarse- grained, scene-level complexity insights. To explore a more granular and automated alternative, the Video Complexity Analyzer (VCA) is introduced. VCA provides fine-grained, block-level complexity metrics such as spatial texture energy, temporal gradients, and edge density, extracted from raw YUV video frames. These features offer a detailed, frame-by-frame perspective on content complexity without requiring manual segmentation. A comparative analysis of both methods highlights their respective strengths, the interpretability and scene-awareness of the scene based complexity classification approach versus the precision of VCA. Scene based Complexity Classification stood tall in most of the cases. This Also led to need of more study on different approaches for enhancing the performance of H.266/VVC.