Matrix Based Intra Prediction Using Power of Two Coefficients
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression systems face challenges in optimizing memory efficiency and computational complexity for intra prediction, particularly due to the large data requirements for storing weighting parameters and performing matrix operations in video encoding and decoding processes.
Innovation Solution
The method involves using weight matrices with coefficients that are powers of two, allowing for reduced multiplications and storage needs by employing shift operations, and deriving these matrices through a learning stage using Least Mean Square minimization, which reduces the complexity and memory footprint in video encoding and decoding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional intra prediction with down-sampling and matrix products is used, then compression efficiency is improved, but the amount of data needed for storing weighting parameters increases significantly
Solution Approach 1:
The patent changes the parameters of the weight matrices by constraining coefficients to be powers of two. This parameter change allows the same prediction functionality to be achieved with reduced storage requirements, as power-of-two coefficients can be represented more efficiently and enable shift-based operations instead of full multiplications.
Solution Approach 2:
The patent applies different representation methods to different coefficients within the weight matrices. By identifying which coefficients can be represented as powers of two and which require full precision, the system optimizes storage locally for each coefficient while maintaining overall compression efficiency.
2Measurement precision
If conventional intra prediction with full matrix operations is used, then prediction accuracy is improved, but computational complexity increases
Solution Approach 1:
By changing the parameter representation of matrix coefficients to powers of two, the patent enables the use of shift operations instead of multiplication operations. This reduces computational complexity while maintaining prediction accuracy, as shift operations are computationally simpler and can be implemented more efficiently in hardware.
Solution Approach 2:
The patent substitutes multiplication operations with shift operations in the matrix-based intra prediction process. This mechanical substitution reduces the computational burden on the decoder, making the system more suitable for resource-constrained devices while preserving the accuracy benefits of matrix-based prediction.
3Device complexity
If power of two coefficients are used in weight matrices, then computational complexity is reduced through shift operations, but storage requirements for exponents are needed
Solution Approach 1:
The patent applies partial precision storage by only storing the exponent values for power-of-two coefficients rather than full coefficient values. This partial action approach reduces overall storage requirements, as the exponent values require fewer bits than complete floating-point or fixed-point representations, while still enabling accurate reconstruction of the original coefficients.
Data Source
AI summary
Different implementations are described, particularly implementations for video encoding and decoding based on linear weighted intra prediction, also called matrix based intra prediction, are presented. Accordingly, the encoding or decoding comprises obtaining intra predicted samples of a block from a selected weight matrix and associated bias and from a set of neighboring reference samples; wherein any coefficient of the selected weight matrix is a power of two. According to relaxed version, not all but only a part of the coefficients of the selected weight matrix is a power of two. Advantageously, such arrangement allows to reduce the amount of memory for storing data and to reduce the complexity of the intra prediction samples computation.