Bilateral Filter for Video Intra Prediction Noise
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in efficiently utilizing reference sample filtering to enhance intra prediction, particularly in effectively smoothing noise while preserving edges.
Innovation Solution
The implementation of a bilateral filter for reference sample filtering in video encoding and decoding processes, which is enabled based on specific conditions associated with the current block, such as prediction modes and block sizes, to obtain filtered reference samples for improved intra prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If reference sample filtering is applied to smooth noise in intra prediction, then prediction accuracy is improved, but edge artifacts may be introduced and filtering complexity increases
Solution Approach 1:
The bilateral filter applies different weighting to different reference samples based on their local characteristics. Samples near edges receive different weights than samples in flat regions, allowing noise smoothing in homogeneous areas while preserving edge information. This local differentiation resolves the contradiction by adapting the filtering strength to the local image structure.
Solution Approach 2:
The bilateral filter dynamically adjusts filtering parameters (spatial and range standard deviations) based on local image content. When edges are detected, the range parameter prevents excessive smoothing across intensity boundaries. This parameter adaptation allows the filter to smooth noise where appropriate while maintaining edge sharpness, resolving the contradiction between noise reduction and edge preservation.
2Object-affected harmful factors
If bilateral filter is applied to all reference samples, then noise smoothing is improved, but computational complexity and processing time increase
Solution Approach 1:
The reference samples are segmented into different regions (edge regions and non-edge regions) based on local gradient analysis. The bilateral filter is selectively applied only to non-edge regions where smoothing is beneficial, while edge regions are excluded from filtering. This segmentation approach reduces computational complexity by avoiding unnecessary filtering operations on edge samples.
Solution Approach 2:
Instead of applying the computationally intensive bilateral filter to all reference samples, the method applies partial filtering only to the subset of samples that benefit from smoothing (non-edge samples). This partial action approach achieves sufficient noise reduction without the full computational cost of universal filtering, resolving the contradiction between smoothing quality and processing complexity.
3Reliability
If reference samples are filtered before intra prediction, then prediction quality is improved, but encoding overhead and processing time increase
Solution Approach 1:
The bilateral filtering is performed as a preliminary step before intra prediction, but only on reference samples that meet specific criteria (non-edge samples). By preparing filtered reference samples in advance for only the samples that need filtering, the method improves prediction quality while minimizing the time penalty, as the filtering is done once and reused across multiple prediction operations.
Data Source
AI summary
A method of video decoding includes receiving a bitstream that comprises coded information of one or more pictures. The coded information indicates an intra prediction of a current block. The method also includes: enabling a reference sample filtering with a bilateral filter when the current block satisfies one or more conditions associated with the reference sample filtering; applying the reference sample filtering with the bilateral filter on reference samples of the current block to obtain filtered reference samples when the reference sample filtering with the bilateral filter is enabled; and reconstructing one or more samples of the current block according to the intra prediction using the filtered reference samples.