Matrix-Based Intra Prediction With Restricted Upsampling for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in efficiently handling high-resolution video data, particularly in terms of bandwidth demand and coding efficiency, especially with the increasing number of connected devices and the need for improved compression techniques.
Innovation Solution
The implementation of matrix-based intra prediction (MIP) methods, including boundary downsampling, matrix vector multiplication, and upsampling operations, to enhance video coding efficiency in standards like HEVC and VVC, with specific modes such as affine linear weighted intra prediction (ALWIP) for improved prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional intra prediction methods are used for high-resolution video, then video quality can be maintained, but bandwidth consumption increases and coding efficiency decreases
Solution Approach 1:
The patent segments the prediction process into multiple stages: boundary downsampling to create a reduced-resolution boundary, matrix-based prediction on the reduced boundary, and selective upsampling only where needed. This segmentation allows efficient processing of high-resolution video by working primarily at lower resolutions, reducing bandwidth requirements while maintaining prediction accuracy where it matters most.
Solution Approach 2:
The patent changes the resolution parameter of boundary samples through downsampling before prediction, then selectively restores resolution through upsampling in specific regions. This parameter transformation enables the system to process high-resolution video blocks using lower-resolution boundary data, significantly reducing the computational and bandwidth overhead while preserving video quality through targeted high-resolution reconstruction.
2Measurement precision
If high-resolution video blocks are processed directly, then prediction accuracy is maintained, but computational complexity increases
Solution Approach 1:
The patent divides the video block processing into segmented regions: regions requiring high-resolution prediction and regions that can use low-resolution prediction. By segmenting the processing based on content characteristics, the system maintains high prediction accuracy in critical areas while reducing computational complexity in less critical areas through downsampling and selective upsampling.
Solution Approach 2:
The patent applies different processing qualities to different regions of the video block. High-resolution matrix-based prediction with upsampling is applied locally to regions where it provides the most benefit, while other regions use standard downsampled prediction. This local quality approach maintains overall prediction accuracy while significantly reducing total computational complexity.
3Reliability
If full-resolution boundary samples are used for matrix-based prediction, then prediction quality is improved, but processing time increases
Solution Approach 1:
The patent performs boundary downsampling as a preliminary action before the main prediction process. By pre-reducing the resolution of boundary samples and identifying regions that will benefit from upsampling in advance, the system prepares optimized input data for the matrix-based prediction, reducing the amount of computation required during the main processing phase while maintaining prediction quality in critical regions.
Data Source
AI summary
Devices, systems and methods for digital video coding, which includes matrix-based intra prediction methods for video coding, are described. In one aspect, a method for video processing includes performing a conversion between a current video block of a video and a bitstream representation of the current video block using a matrix based intra prediction (MIP) mode, where the conversion includes performing the upsampling operation in which the final prediction block is determined by using a reduced prediction block of the current video block and by using reconstructed neighboring samples of the current video block according to a rule, and where the reduced prediction block is obtained by performing the matrix vector multiplication operation on reduced boundary samples of the current video block.