Adaptive DCT Coefficient Binarization for Video Entropy Coding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current video compression standards inefficiently binarize discrete cosine transform (DCT) coefficients due to not considering the probability distribution of DCT coefficients, where low-frequency coefficients are concentrated in the upper left portion and high-frequency coefficients in the lower right portion of the DCT coefficient block.

Innovation Solution

Adaptive binarization of DCT coefficients based on whether frequencies are high or low, using different binarization methods such as concatenated unary and exponential Golomb binarization, to efficiently encode video data by grouping coefficients and applying appropriate binarization techniques to each group.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If the same binarization method is applied to all DCT coefficients, then the encoding process is simple, but the compression efficiency is poor due to ignoring the probability distribution of coefficients

Engineering Contradiction:
Improvecompression efficiencyVSAvoidbinarization process complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent segments DCT coefficients into different groups based on their frequency characteristics (low-frequency coefficients with high probability and high-frequency coefficients with low probability). Different binarization methods are applied to each segment: truncated unary binarization for low-frequency coefficients and exponential Golomb binarization for high-frequency coefficients. This segmentation allows the system to optimize compression efficiency for each group while maintaining manageable process complexity through structured classification.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies the local quality principle by using different binarization strategies tailored to specific regions of the DCT coefficient spectrum. Low-frequency coefficients (which occur more frequently) use truncated unary binarization, while high-frequency coefficients (which occur less frequently) use exponential Golomb binarization. This localized optimization ensures that each coefficient group is encoded with the most appropriate method for its statistical characteristics, thereby improving overall compression efficiency without requiring a completely complex unified approach.

Inventive Principle:
Principle #3Local quality

2Quantity of substance

If truncated unary binarization is used for all coefficient values, then the binarization process is simple, but high-level values require many binary strings increasing bitstream length

Engineering Contradiction:
Improvebitstream lengthVSAvoidbinarization method complexity
Core Design Contradiction:
Quantity of substanceVSDevice complexity

Solution Approach 1:

The patent changes the binarization parameter (method type) based on the coefficient value characteristics. For low-frequency coefficients with smaller absolute values, truncated unary binarization is used. For high-frequency coefficients with larger absolute values, exponential Golomb binarization is applied. This parameter change allows the system to adapt the binarization approach to the actual data distribution, reducing the number of binary strings needed for high-level values while maintaining simplicity for low-level values.

Inventive Principle:
Principle #35Parameter changes

3Quantity of substance

If exponential Golomb binarization is used for all coefficient values, then high-level values are efficiently encoded, but low-level values and processing overhead increase

Engineering Contradiction:
Improvebitstream lengthVSAvoidbinarization method complexity
Core Design Contradiction:
Quantity of substanceVSDevice complexity

Solution Approach 1:

The patent applies partial action by using exponential Golomb binarization only where necessary (for high-frequency coefficients with larger absolute values) rather than applying it universally. For low-frequency coefficients with small absolute values, the simpler truncated unary binarization suffices. This partial application of the more complex method avoids unnecessary processing overhead and complexity for cases where it would provide no benefit, while still achieving efficient encoding where needed.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3166316B1Method and apparatus for entropy-coding/entropy-decoding video data
Publication Date: 2019.03.20 SAMSUNG ELECTRONICS CO LTD
  • EP3166316B1 patent drawingFigure 1
  • EP3166316B1 patent drawingFigure 2
  • EP3166316B1 patent drawingFigure 3

AI summary

Provided are a method and apparatus for entropy-encoding/entropy-decoding video data. The method of entropy-encoding video data includes binarizing coefficients of the frequency domain, which are generated by transforming a residual block of a current block into the frequency domain, using different binarization methods and performing binary arithmetic coding on the binarized coefficients. In this way, the coefficients are binarized adaptively according to whether the frequencies of the coefficients are high or low, thereby improving the compression efficiency of the video data.