Linear Model Prediction Mode Sample Access for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video encoding and decoding technologies face inefficiencies in processing and memory bandwidth usage due to the need to fetch unnecessary luma samples for downsampling in linear-model prediction mode, particularly when these samples are not stored in local memory or are outside the range of locally stored samples.
Innovation Solution
The techniques involve configuring video encoders and decoders to fetch only necessary neighboring luma samples for downsampling, excluding those above and left of the top-left luma sample, and using different filters based on the availability of these samples within the line-buffer to reduce processing time and memory bandwidth usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neighboring luma samples are fetched for downsampling in linear-model prediction mode, then chroma prediction accuracy is improved, but memory bandwidth usage and processing time increase due to fetching unnecessary samples
Solution Approach 1:
The patent extracts only the necessary luma samples (above and to the right of the current block) for downsampling in linear-model prediction mode, excluding unnecessary samples (above and left of the top-left sample). This selective extraction maintains chroma prediction accuracy while reducing memory bandwidth usage and processing time.
Solution Approach 2:
The patent applies different sampling strategies based on local requirements: necessary neighboring luma samples are fetched for accurate downsampling, while unnecessary samples are excluded. This localized quality approach ensures prediction accuracy where needed while avoiding redundant processing.
2Measurement precision
If neighboring luma samples are fetched for downsampling, then chroma prediction accuracy is improved, but memory bandwidth usage increases
Solution Approach 1:
The patent extracts only the essential luma samples required for chroma prediction in linear-model mode, specifically samples above and to the right of the current block. By excluding samples above and left of the top-left sample, the patent reduces memory bandwidth consumption while preserving prediction accuracy.
Solution Approach 2:
The patent performs partial downsampling by fetching only the necessary portion of neighboring luma samples rather than all possible samples. This partial action approach maintains sufficient prediction accuracy while significantly reducing memory bandwidth usage.
3Measurement precision
If all neighboring luma samples are fetched for downsampling, then prediction accuracy is maintained, but processing time increases
Solution Approach 1:
The patent extracts only the necessary luma samples (above and to the right) for downsampling in linear-model prediction mode, excluding unnecessary samples (above and left of top-left). This selective extraction reduces the number of samples requiring processing, thereby decreasing processing time while maintaining prediction accuracy.
Solution Approach 2:
The patent segments the neighboring luma samples into necessary and unnecessary portions, processing only the necessary samples for downsampling. This segmentation approach divides the processing task into essential and non-essential parts, reducing overall processing time.
Data Source
AI summary
Techniques are described of linear model (LM) prediction mode. In one or examples, a video encoder or video decoder may limit the number of neighboring luma samples that are fetched for downsampling thereby increasing the speed at which the video encoder or video decoder is able to complete the LM prediction encoding or decoding.