Image Prediction Method Using Iterative Integer Pixel Search

The patent segments the search process into two distinct phases: first performing multiple iterative searches using integer pixel step size to obtain an intermediate predicted block, then performing a single search using fractional pixel step size to obtain the final target predicted block. This segmentation allows the system to achieve high prediction accuracy through fractional pixel search while limiting the overall computational complexity by restricting fractional pixel searches to only one iteration, unlike conventional methods that alternately perform both types of searches multiple times.

The patent applies preliminary action by first performing multiple iterative searches using integer pixel step size to obtain an intermediate predicted block that serves as a refined starting point. This preliminary integer pixel search establishes a more accurate initial position, which subsequently reduces the search space and data read volume required for the final fractional pixel search, thereby achieving high prediction accuracy with reduced overall data access requirements.

The patent maintains continuity of useful action by performing multiple iterative searches with integer pixel step size to progressively refine the intermediate predicted block, ensuring continuous improvement in prediction accuracy within the integer pixel domain. This continuous refinement through multiple iterations compensates for the limitation of performing only a single fractional pixel search, thereby maintaining high overall prediction accuracy while significantly reducing data read volume compared to conventional alternating search methods.