Weighted Prediction Synthesis in Image Encoding for Lower Code Amount

The patent changes the parameters of prediction image synthesis by introducing multiple synthesis methods (arithmetic weighting, median filtering, gradient filtering) with different parameters. The arithmetic weighting method uses weight coefficients (e.g., 1:1, 2:1, 3:1) to control the contribution of inter and intra prediction images. The gradient filtering method uses gradient thresholds to selectively combine prediction images. These parameter adjustments allow optimization of both code amount reduction and image quality preservation.

The patent implements dynamic selection of synthesis methods and parameters based on block characteristics. The encoder determines which synthesis method to use (arithmetic weighting, median filtering, or gradient filtering) and what parameters to apply based on the specific block being encoded. This dynamic adaptation allows the system to optimize the balance between code amount and image quality for each individual block rather than using a fixed approach.

The patent segments the prediction process into distinct synthesis methods (arithmetic weighting, median filtering, gradient filtering), each handling specific types of blocks or conditions. This segmentation allows the system to apply the most appropriate and computationally efficient method for each segment, reducing overall processing complexity while maintaining high prediction accuracy where needed.

The patent applies gradient filtering only when gradient conditions are met, rather than applying all synthesis methods to all blocks. This partial action approach reduces processing complexity by selectively applying complex operations only where they provide benefit, while using simpler methods for other blocks.