Drift-Free Video Tile Filtering for Parallel Encoding

The picture is divided into multiple independently encodable and decodable tiles, allowing parallel processing while maintaining independent error confinement. Each tile can be processed separately without affecting other tiles, thus improving productivity while managing reliability through proper boundary handling.

Different filtering approaches are applied to different regions: conventional filtering within tiles and drift-free filtering at tile boundaries. This local differentiation ensures high quality at boundaries where error propagation is critical while maintaining efficiency in internal regions.

The harmful effect of error propagation is extracted and eliminated by using drift-free filtering at tile boundaries. This filtering method removes the dependency chain that causes errors to propagate between independently encoded tiles while preserving the boundary smoothness benefits of filtering.

Drift-free filtering is applied as a preliminary measure at tile boundaries to prevent error propagation before it can occur. By proactively using filtering methods that do not create drift or error dependencies, the system prevents the harmful effect from manifesting in the first place.

The contamination issue is converted into a benefit by using the filtered samples for their intended purpose (boundary smoothing) while excluding them from reference block calculations. This separates the beneficial filtering effect from the harmful contamination effect, allowing both to coexist without negative interaction.

The sample set is segmented into two groups: samples affected by loop filtering and samples not affected. Reference blocks are constructed using only the non-filtered samples to avoid contamination, while filtered samples are used separately for boundary filtering operations.