Image Entropy Coding With Multidimensional Probability Learning

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current video coding techniques, such as those used in MPEG and H.264, suffer from a lack of local adequacy in probability learning due to sequential traversal methods, leading to inefficiencies and limitations in parallel decoding capabilities.

Innovation Solution

A method that employs a multidimensional neighborhood approach for learning probabilities, allowing for parallel decoding by initializing and storing probabilities from adjacent sub-sequences, enabling better local adaptation and synchronization during the coding process.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If sequential raster-scan traversal is used for entropy coding, then the coding process is simple to implement, but the probability learning lacks local adequacy and parallel decoding is prevented

Engineering Contradiction:
Improveease of implementationVSAvoidlocal adequacy of probability learning
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The image is divided into multiple independent sub-sequences that can be processed in parallel. Each sub-sequence is assigned to a different processing thread, allowing simultaneous entropy coding while maintaining local probability learning within each segment. This segmentation resolves the contradiction by enabling parallel processing without sacrificing the simplicity of the coding approach.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a new dimension of processing by organizing blocks into sub-sequences that can be traversed in multiple directions (horizontal, vertical, diagonal) rather than strictly sequential raster-scan. This multi-dimensional approach allows parallel processing while maintaining local context adequacy for probability learning.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Device complexity

If sequential traversal order is used for coding symbols, then probability learning is simple, but parallel decoding of multiple symbols is prevented

Engineering Contradiction:
Improvecoding complexityVSAvoidparallel decoding capability
Core Design Contradiction:
Device complexityVSProductivity

Solution Approach 1:

The symbol sequence is segmented into multiple independent sub-sequences that can be decoded in parallel. Each sub-sequence maintains its own probability context, allowing multiple decoding threads to operate simultaneously without interfering with each other, thus improving productivity while keeping individual thread complexity manageable.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary organization of blocks into sub-sequences during the encoding phase, preparing the data structure in advance to enable parallel decoding. This preliminary action allows the decoding stage to efficiently utilize multiple processing units without increasing the complexity of the decoding algorithm itself.

Inventive Principle:
Principle #10Preliminary action

3Ease of operation

If raster-scan traversal is used, then coding is straightforward, but probability learning at line starts is inadequate due to reliance on end-of-previous-line symbols

Engineering Contradiction:
Improvecoding straightforwardnessVSAvoidprobability learning precision
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent introduces multi-dimensional traversal directions (horizontal, vertical, diagonal) for organizing sub-sequences. This allows probability learning to draw from spatially adjacent blocks in multiple directions rather than only from the end of the previous line, significantly improving probability learning precision while maintaining coding straightforwardness through systematic organization.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent applies local quality by allowing each sub-sequence to have its own optimized traversal direction and probability context. This enables local adaptation of the coding approach to match the local characteristics of different image regions, improving probability learning precision without complicating the overall coding process.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP2486659B1Methods for encoding and decoding images, corresponding encoding and decoding devices and computer programs
Publication Date: 2020.11.25 ORANGE SA
  • EP2486659B1 patent drawingFigure 1~2a
  • EP2486659B1 patent drawingFigure 2b~3
  • EP2486659B1 patent drawingFigure 4a~4b

AI summary

A method is provided for encoding an image represented by a sequence of symbols. The method includes the following steps: defining subsequences within the sequence; and going through the subsequences applying the following sub-steps for a routine subsequence: determining an adjacent routine encoded group including two previously encoded subsequences, one of which belongs to an adjacent encoded group associated with a subsequence that is separate from the routine subsequence; initializing a routine probability of occurrence of a symbol, taking into account two known probabilities of occurrence of the symbol, relating to two previously encoded subsequences of the adjacent routine encoded group; the consecutive entropic encoding of the symbols, and updating the routine probability of occurrence; and storing the routine probability of occurrence after encoding the last symbol.