H.264 Video Codec Mode Decision Using SAD and SSD Evaluation

The prediction mode evaluation process is segmented into multiple stages: initial coarse evaluation using sum of absolute differences (SAD), followed by refined evaluation using sum of squared differences (SSD) only for candidate modes, and finally half-pel precision evaluation only for the best candidate. This segmentation allows the system to maintain video quality while reducing overall computational complexity by applying different evaluation rigor at different stages.

Different evaluation methods are applied to different prediction modes based on their likelihood of being optimal. Integer-pel modes are evaluated using SAD for initial screening, while only promising candidates undergo SSD evaluation. Half-pel modes are evaluated only for the single best candidate. This local quality approach ensures high video quality for critical decisions while reducing computational burden for less critical evaluations.

The system performs preliminary evaluation of all prediction modes using the computationally efficient SAD metric before committing to more expensive SSD evaluations. This preliminary action identifies candidate modes that are likely to be optimal, allowing the system to maintain coding accuracy while significantly reducing the number of modes that require full SSD evaluation and half-pel processing.

The system skips detailed SSD evaluation and half-pel precision processing for prediction modes that are unlikely to be optimal, based on initial SAD evaluation results. Only the best candidate mode undergoes half-pel precision evaluation, while other modes are skipped at the refined evaluation stage. This skipping strategy maintains coding accuracy for the selected mode while dramatically reducing overall processing time.