Video Component Prediction Using Filtered Local Reference Samples
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity of video component prediction in existing video coding standards, such as H.265/HEVC, is high due to the need for down-sampling and constructing linear models based on numerous neighboring reference blocks, leading to low chroma prediction efficiency and overall video coding and decoding inefficiency.
Innovation Solution
A method and device for video component prediction that involves selecting first picture component reference values, performing filtering on these values to obtain filtered samples, determining a component linear model based on these filtered samples, and using this model for prediction, thereby reducing the complexity of model construction and improving prediction efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If down-sampling processing is performed using sample values in a luma neighbouring region and maximum and minimum are found from reference samples to construct the linear model, then chroma prediction accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the reference sample region into multiple sub-regions (e.g., top-left, top-right, bottom-left, bottom-right quadrants) and performs down-sampling and model construction separately for each sub-region. This segmentation reduces the number of samples that need to be processed in each sub-region compared to processing the entire reference region as a whole, thereby reducing computational complexity while maintaining prediction accuracy through localized modeling.
Solution Approach 2:
The patent applies partial action by performing down-sampling and model construction only on selected sub-regions of the reference samples rather than processing all reference samples. This selective processing reduces the number of filtering operations and computational steps required, directly addressing the complexity issue while still achieving adequate prediction accuracy for the current block.
2Measurement precision
If a linear model is constructed based on a relatively large number of neighbouring reference blocks, then prediction accuracy is improved, but processing time increases
Solution Approach 1:
The patent divides the large set of neighboring reference blocks into multiple smaller sub-regions and constructs linear models for each sub-region separately. This segmentation reduces the number of reference blocks that need to be processed in each modeling step, thereby reducing processing time while maintaining prediction accuracy through the combined effect of multiple localized models.
Solution Approach 2:
The patent performs model construction on only a subset of the available reference blocks (those in selected sub-regions) rather than using all neighboring reference blocks. This partial processing approach reduces the computational workload and processing time required for model construction while still achieving adequate prediction accuracy for the current block.
3Manufacturing precision
If filtering processing is performed on sample values of multiple reference samples, then model construction accuracy is improved, but computational load increases
Solution Approach 1:
The patent segments the reference sample set into multiple sub-regions and performs filtering processing separately on each sub-region. This segmentation reduces the number of samples that need to be filtered in each processing step, thereby reducing computational load while maintaining model construction accuracy through the aggregation of results from multiple filtered sub-regions.
Solution Approach 2:
The patent applies filtering processing only to reference samples in selected sub-regions rather than to all reference samples. This partial filtering approach reduces the computational load and number of filtering operations required while still achieving adequate model construction accuracy for the prediction task.
Data Source
AI summary
Provided are a picture component prediction method and a video component prediction device. The method includes: determining multiple reference samples from samples in one or more neighboring lines at a top side of the current block according to four preset positions, or determining multiple reference samples from samples in one or more neighboring columns at a left side of the current block according to four preset positions; determining multiple first picture component reference values according to the determined multiple reference samples; performing first filtering processing on sample values of samples corresponding to the multiple first picture component reference values, respectively; determining a parameter of a component linear model; performing mapping processing on a reconstructed value of the first picture component of the current block according to the component linear model; and determining a predicted value of the to-be-predicted picture component of the current block.