Segment-Level Reference Picture Rescaling for Error-Resilient Inter-Prediction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing video coding technologies face issues with reference picture resampling (RPR) that are not error-resilient, employ inefficient filters, and are limited to whole-picture rescaling, leading to quality degradation and inefficiencies in bandwidth-critical applications.

Innovation Solution

Implementing an in-loop RPR technology that uses efficient filters, is error-resilient, and allows for adaptive rescaling of picture segments, such as tiles, within the coding loop, using syntax elements to control resolution changes between pictures.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If reference picture resampling is applied to the whole picture, then bandwidth requirements are reduced, but video quality degrades and the method is not error-resilient

Engineering Contradiction:

Improvebandwidth requirementsVSAvoiderror resilience

Core Design Contradiction:

Quantity of substanceVSReliability

Solution Approach 1:

The patent divides the picture into multiple segments (tiles or regions) and applies resampling independently to each segment rather than the entire picture. This segmentation allows selective rescaling of specific regions that benefit from it while maintaining full resolution in other areas, thereby reducing overall bandwidth requirements while preserving error resilience through localized processing

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent implements differential resampling where different picture segments maintain different resolution qualities based on their specific requirements. High-motion or important regions maintain full resolution for better quality and error resilience, while static or less critical regions are downsampled to reduce bandwidth, achieving local optimization of both quality and bandwidth efficiency

Inventive Principle:

Principle #3Local quality

2Quantity of substance

If reference picture resampling is applied to the whole picture, then bandwidth requirements are reduced, but video quality degrades

Engineering Contradiction:

Improvebandwidth requirementsVSAvoidvideo quality

Core Design Contradiction:

Quantity of substanceVSManufacturing precision

Solution Approach 1:

The patent segments the picture into multiple regions and applies resampling selectively to only those segments that benefit from resolution reduction. This prevents universal quality degradation while achieving bandwidth reduction in appropriate areas

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Different picture segments maintain different quality levels according to their specific needs. Important regions maintain high quality while less critical regions are downsampled, optimizing the overall quality-bandwidth tradeoff without uniform quality loss

Inventive Principle:

Principle #3Local quality

3Ease of manufacture

If traditional resampling filters are used, then implementation is simpler, but computational efficiency is lower

Engineering Contradiction:

Improveimplementation simplicityVSAvoidcomputational efficiency

Core Design Contradiction:

Ease of manufactureVSProductivity

Solution Approach 1:

The patent employs advanced resampling filters with optimized parameters that balance computational complexity and filtering performance. The filter parameters are adapted based on picture content characteristics, achieving high computational efficiency while maintaining implementation feasibility through parameter optimization rather than algorithmic complexity

Inventive Principle:

Principle #35Parameter changes

4Device complexity

If resolution is changed outside the coding loop, then implementation is simpler, but buffering requirements increase and delay increases

Engineering Contradiction:

Improveimplementation complexityVSAvoidbuffering delay

Core Design Contradiction:

Device complexityVSLoss of time

Solution Approach 1:

The patent performs resampling operations within the coding loop as preliminary actions during the encoding/decoding process itself, rather than as post-processing steps. This allows resolution changes to be integrated into the existing coding workflow without requiring additional buffering stages, thereby reducing delay while maintaining implementation simplicity

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12470730B2Adaptive picture resolution rescaling for inter-prediction and display

Publication Date: 2025.11.11 TENCENT AMERICA LLC

AI summary

A method of decoding a coded picture of a coded video sequence, the method being performed by at least one processor and the method includes decoding, from a first high level syntax structure for a plurality of pictures, a syntax element related to a reference segment resolution, decoding, from a second high level syntax structure that changes from a first coded picture to a second coded picture, a syntax element related to a decoded segment resolution, resampling a sample from a reference picture buffer for use for prediction by a decoder, the decoder decoding a segment at decoding resolution, and the sample from the reference picture buffer is in the reference segment resolution, decoding the segment in the decoded segment resolution into a decoded segment in the decoded segment resolution, and storing the decoded segment into the reference picture buffer.