Rate-Distortion Optimized Quantization for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding standards, such as MPEG and H.26x, do not fully optimize quantization and entropy coding, leading to suboptimal bit allocation and perceptual quality, especially in cases where the distribution of transform coefficients deviates from uniform, and lack flexibility in adjusting quantization schemes to exploit entropy encoding effectively.
Innovation Solution
The proposed methods and systems implement rate-distortion optimized quantization by determining distortion and bit requirements for each level using entropy code tables, allowing for the selection of quantization levels that minimize the rate-distortion cost, and skipping transform blocks based on perceptual quality and rate-distortion tradeoffs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional quantization schemes (Eqs. 1 and 2) are used with fixed entropy code tables, then the encoding process is simple and fast, but the coding performance is suboptimal and does not fully exploit the characteristics of the entropy coding system
Solution Approach 1:
The patent pre-calculates and stores optimal quantization levels in lookup tables during an initialization phase. The entropy encoder is pre-configured with code tables and probability distributions. During actual encoding, the system simply performs table lookups rather than performing complex optimization calculations, thus achieving high coding performance without real-time computational complexity.
Solution Approach 2:
The patent introduces context-adaptive probability distributions that dynamically adjust based on the current encoding state (intra/inter mode, chroma/luma). The system selects different probability distributions and code tables based on the coding context, allowing the quantization scheme to adapt to different scenarios and maximize coding efficiency without requiring complex real-time optimization.
2Loss of information
If the encoder optimizes quantization for each level to match the Laplacian or Exponential distribution of transform coefficients, then the bit allocation becomes more efficient, but the system loses flexibility because entropy code tables are fixed in standards like MPEG and H.26x
Solution Approach 1:
The patent applies different quantization strategies and probability distributions to different local contexts within the video data. Specifically, it uses different probability distributions (Laplacian for intra, Exponential for inter) based on the local coding mode, and different entropy code tables for different coefficient levels and positions. This localized adaptation achieves efficient bit allocation while maintaining compatibility with fixed standard frameworks.
Solution Approach 2:
The patent changes the parameters of the quantization process by introducing context-dependent probability distribution parameters (λ for Laplacian, μ and σ for Exponential). These parameters are selected based on the coding context rather than being fixed, allowing the system to optimize bit allocation for different types of video content while still operating within the constraints of standard entropy coding frameworks.
3Reliability
If all transform coefficients are quantized and transmitted, then the reconstructed video quality is maintained, but the bit rate increases unnecessarily for blocks with small energy
Solution Approach 1:
The patent applies partial quantization by determining a threshold based on the energy of the transform block. If the block's energy is below this threshold, the encoder skips quantization and transmission of that block entirely (or uses a coarser quantization). This partial application of quantization reduces the bit rate for low-energy blocks while maintaining acceptable quality, as these blocks contribute minimally to the overall perceived video quality.
Data Source
AI summary
Provided are methods and systems for rate-distortion optimized quantization and for skipping transform blocks that can take into account the rate-distortion tradeoff as well as the perceptual quality of reconstructed video.