Adaptive Quantization Step Size for Video Block Distortion Control

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current video encoding standards face challenges in balancing distortion and bit rate, particularly due to the uniform quantization step size across all coefficients, which does not account for human visual sensitivity to distortion, leading to inefficiencies in data compression, especially in video coding where distortion in lower frequency components is more noticeable.

Innovation Solution

The method involves determining an activity rank for each block based on its intra-coding mode and size, using a look-up table to calculate a normalization factor that adjusts the quantization step size, allowing for adaptive quantization based on the block's statistical characteristics, thereby optimizing the trade-off between distortion and bit rate.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a uniform quantization step size is used for all coefficients, then the encoding process is simple and fast, but the distortion is not optimized according to human visual sensitivity

Engineering Contradiction:
Improveencoding speedVSAvoiddistortion optimization
Core Design Contradiction:
ProductivityVSManufacturing precision

Solution Approach 1:

The patent applies different quantization step sizes to different frequency components based on their perceptual importance. Specifically, lower frequency coefficients (which are more visible to humans) receive smaller quantization steps, while higher frequency coefficients receive larger steps. This local differentiation of quantization quality resolves the contradiction by optimizing distortion where it matters most while maintaining overall encoding efficiency.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent dynamically adjusts the quantization step size parameter based on the frequency component being processed. By changing this critical parameter according to the coefficient's frequency and perceptual importance, the system achieves both improved distortion optimization and maintained encoding speed, avoiding the need for complex per-coefficient analysis.

Inventive Principle:
Principle #35Parameter changes

2Loss of substance

If larger quantization step sizes are used, then the bit rate is reduced, but the distortion in the reconstructed picture increases

Engineering Contradiction:
Improvebit rateVSAvoidreconstruction quality
Core Design Contradiction:
Loss of substanceVSManufacturing precision

Solution Approach 1:

The patent applies a non-uniform quantization strategy where different step sizes are used for different frequency components. Lower frequency coefficients use smaller step sizes to preserve important visual information, while higher frequency coefficients use larger step sizes to reduce bit rate. This selective approach optimizes the trade-off between bit rate and reconstruction quality according to human visual sensitivity.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The quantization step size parameter is dynamically adjusted based on the frequency component being processed. By changing this parameter to match the perceptual importance of each coefficient, the system achieves efficient bit rate reduction without unnecessarily compromising reconstruction quality in visually critical regions.

Inventive Principle:
Principle #35Parameter changes

3Manufacturing precision

If SSIM-based quantization adaptation is implemented, then perceptual quality is improved, but computational complexity and processing delay increase

Engineering Contradiction:
Improveperceptual qualityVSAvoidcomputational burden
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent extracts only the essential information needed for perceptual optimization - specifically, the frequency component of each coefficient - and uses this extracted information to guide quantization step size selection. This avoids the need for complex full-picture SSIM calculations while still achieving perceptual quality improvement through targeted adaptation to human visual sensitivity.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

Instead of using computationally expensive SSIM metrics that require multiple passes over the entire picture, the patent employs a simpler, faster method based on frequency component analysis. This lighter-weight approach achieves similar perceptual optimization goals with significantly reduced computational burden and processing delay.

Inventive Principle:
Principle #27Cheap short-living objects (Disposable)

Data Source

PatentEP3151560B1Intra-coding mode-dependent quantization tuning
Publication Date: 2019.02.27 BLACKBERRY LTD
  • EP3151560B1 patent drawingFigure 1
  • EP3151560B1 patent drawingFigure 2
  • EP3151560B1 patent drawingFigure 3

AI summary

Methods of encoding and decoding video are described. The methods for encoding and decoding a picture partitioned into blocks include determining an activity rank for a block, based on a block size of the block and an intra-coding mode for the block; calculating a normalization value based on the activity rank divided by an average activity rank determined over a plurality of previously-encoded/decoded blocks; and quantizing/dequantizing a set of transform domain coefficients for the block using a quantization step size adjusted by the normalization value.