Decoder-Side Motion Vector Derivation Using Spatial Correlation

The patent combines template matching with spatial correlation analysis by calculating a combined cost function that integrates both template matching cost and spatial correlation cost. This merging of two different approaches allows the system to leverage the motion compensation accuracy of template matching while using spatial correlation to reduce block boundary discontinuities, thereby resolving the technical contradiction between these two competing requirements.

The patent implements a flag-based mechanism that allows spatial correlation consideration to be selectively applied only when beneficial. The encoder can choose to enable or disable spatial correlation processing on a per-block basis, performing the computationally intensive spatial correlation analysis only when it is expected to improve results, rather than applying it universally to all blocks. This partial application reduces overall computational complexity while still achieving the benefit of minimized block boundary discontinuities where needed.

The patent performs motion vector refinement at the decoder side using spatial correlation analysis after receiving the initial motion vectors from the encoder. By conducting this refinement process during decoding rather than during encoding, the system improves video quality through more accurate motion compensation while avoiding the increase in encoding time. The decoder has sufficient computational resources and time to perform this additional refinement without impacting the overall encoding efficiency.