Reference Picture Resampling With Edge-Preserving Interpolation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding standards face issues with visible artefacts and increased bitrate due to the mixup of interpolation filters for motion compensation during reference picture resampling, particularly when motion vectors have different resolutions, leading to poorer motion-compensated inter prediction.
Innovation Solution
Implement a method to selectively choose between two interpolation filters based on the resolution equality of current and reference pictures, using a higher edge-preserving filter for non-equal resolutions to minimize artefacts and adaptively signal motion vector resolution in the bitstream.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If different interpolation filters are applied to different phases during reference picture resampling, then motion compensation can be performed at various resolutions, but visible artefacts occur and prediction quality deteriorates
Solution Approach 1:
The patent applies different interpolation filters to different phases locally. Specifically, a first interpolation filter is applied to first phases while a second interpolation filter is applied to second phases, allowing each phase to receive the most appropriate filtering treatment rather than applying a single filter globally. This resolves the contradiction by enabling resolution flexibility through phase-specific filter selection while minimizing visible artefacts through appropriate local filtering.
Solution Approach 2:
The patent changes the interpolation filter parameters based on the phase of the reference picture samples. By selecting different filters (e.g., smoothing vs. sharpening filters) depending on the phase, the system adapts the filtering characteristics to match the specific requirements of each phase, thereby maintaining motion compensation versatility while reducing artefact formation.
2Adaptability or versatility
If multiple interpolation filters are used for different phases, then various motion vector resolutions can be supported, but the device complexity increases
Solution Approach 1:
The patent implements a phase-based filter selection mechanism where the complexity is localized to phase identification and filter mapping. The decoder determines which phase each reference sample belongs to and applies the corresponding filter, creating a systematic and manageable complexity structure rather than requiring complex global optimization.
Solution Approach 2:
The patent introduces dynamic filter selection based on phase identification. The interpolation filter is not fixed but dynamically chosen according to the phase of the reference picture samples being processed. This dynamic approach enables support for various motion vector resolutions while keeping the selection logic organized and manageable through phase-based categorization.
3Adaptability or versatility
If reference picture resampling is performed with different resolutions, then adaptive resolution change is enabled, but inconsistent filter application increases bitrate
Solution Approach 1:
The patent applies phase-specific interpolation filters during reference picture resampling to ensure consistent and appropriate filtering across all phases. This local quality approach prevents the mixup of filters that would otherwise occur during resolution adaptation, thereby reducing prediction errors and the bitrate required to encode residuals.
Solution Approach 2:
The patent changes the interpolation filter parameters based on the phase of reference samples during resampling. By adapting the filter characteristics to match each phase's requirements, the system enables resolution adaptation while maintaining prediction accuracy, thus avoiding the bitrate increase that would result from inconsistent filter application.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Concepts are described, including encoding of a video into data stream and decoding of a data stream having a video encoded thereinto, using motion compensation prediction between pictures of equal resolution and pictures of different resolution, based on motion vectors at a half-sample resolution and on motion vectors at a different resolution using interpolation filters to obtain sub-sample values within a reference sample array. The interpolation filter is selected from two interpolation filter versions different in a higher edge preserving property, and the selection is depending on whether a current picture is equal in picture resolution to the reference sample array in horizontal and/or vertical dimension, and/or a constraint information in the data stream.