Intra Prediction Efficiency via Gradient-Based Mode Selection

The patent segments the prediction mode selection process into two stages: first determining whether to use directional prediction or average prediction based on gradient calculation, and then selecting the specific directional mode only if needed. This segmentation avoids computing all directional prediction modes when they are not necessary, thereby reducing code amount while maintaining prediction efficiency when edges are present.

The patent applies partial action by selectively applying directional prediction only to blocks where edge detection indicates it is beneficial. Instead of computing all directional prediction modes for every block (excessive action), the method performs gradient calculation first and only proceeds with directional prediction when the gradient threshold is exceeded, thus avoiding unnecessary computation and code overhead.

The patent performs preliminary gradient calculation and edge detection before selecting directional prediction modes. This preliminary action identifies blocks that actually benefit from directional prediction, allowing the system to skip computation of multiple directional modes for blocks where average prediction is sufficient, thereby reducing overall computation amount while maintaining high prediction efficiency where needed.

The method performs partial computation by calculating gradients and only proceeding with directional prediction for blocks exceeding the gradient threshold. This avoids the excessive computation that would result from evaluating all directional prediction modes for every block, thus reducing computation amount while maintaining prediction efficiency for blocks where directional prediction is actually beneficial.

The patent applies local quality by using different prediction strategies for different regions: average prediction for blocks without significant edges and directional prediction for blocks with detectable edges. This localized approach improves encoding efficiency for blocks with edges while keeping the encoding process simple for blocks without edges, thus achieving better overall encoding efficiency without excessive complexity.

The patent introduces dynamic selection between average prediction and directional prediction based on gradient calculation results. This dynamic approach allows the encoding process to adapt to local characteristics of each block, improving encoding efficiency for blocks with edges while maintaining simplicity for blocks without edges, thus achieving a balance between encoding efficiency and process complexity.