Partial Sub-Transform Segmentation for Image Decoding Complexity

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current digital image and video coding techniques face complexity and inefficiency in transform operations, particularly with block sizes larger than 4x4, as not all Discrete Cosine/Trigonometric Transforms (DCT/DST) are applicable to all block sizes, and fast algorithms are lacking for larger sizes, leading to increased computational complexity.

Innovation Solution

The method involves decomposing a block into partial sub-vectors and applying separate partial transforms to each, allowing for a reduced number of operations by using orthogonal or quasi-orthogonal transforms, including identity transforms, to achieve a transformed block with reduced computational complexity without compromising compression efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If traditional DCT/DST transforms are applied to blocks larger than 4x4, then transformation coverage is improved, but computational complexity increases significantly

Engineering Contradiction:
Improvetransform applicability to block sizesVSAvoidcomputational complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The transform block is divided into multiple sub-blocks, and the transform operation is segmented into multiple stages: first applying a transform to each sub-block individually, then applying a second transform to the results. This segmentation allows using simpler transforms on smaller sub-blocks rather than applying complex transforms to the entire large block, thereby reducing overall computational complexity while maintaining transform applicability to various block sizes.

Inventive Principle:
Principle #1Segmentation

2Productivity

If fast algorithms are developed for larger block sizes, then transformation efficiency is improved, but algorithm complexity increases

Engineering Contradiction:
Improvetransformation efficiencyVSAvoidalgorithm complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The transformation process is segmented into multiple independent stages operating on sub-blocks. Each stage uses relatively simple transform algorithms that can be efficiently implemented, avoiding the need for complex fast algorithms for large blocks. The segmentation enables parallel processing of sub-blocks, improving overall transformation efficiency without requiring individually complex algorithms.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Instead of applying a single comprehensive transform to the entire block, the method applies partial transforms to sub-blocks and then combines them through a second transform stage. This partial action approach achieves the necessary transformation effect through multiple simpler operations rather than one complex operation, improving efficiency while managing algorithm complexity.

Inventive Principle:
Principle #16Partial or excessive action

3Adaptability or versatility

If multiple transform types are supported for different block sizes, then coding flexibility is improved, but implementation complexity increases

Engineering Contradiction:
Improvecoding flexibilityVSAvoidimplementation complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The implementation complexity is reduced by segmenting the transform process into standardized stages that can be reused across different block sizes. By dividing blocks into sub-blocks and applying the same multi-stage transform process uniformly, the system achieves coding flexibility for various block sizes without implementing entirely different transform algorithms for each size, thereby managing implementation complexity.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3449631B1Method of decoding a digital image, method of coding, devices, and computer programs associated therewith
Publication Date: 2024.06.12 FOND B COM
  • EP3449631B1 patent drawingFigure 1~2B
  • EP3449631B1 patent drawingFigure 3
  • EP3449631B1 patent drawingFigure 4

AI summary

The invention relates to a method for decoding a digital image, on the basis of a binary train comprising coded data representative of said image, comprising the following steps, implemented for a block of said image, the so-called transformed current block: -Decoding of the coefficients of the transformed current block on the basis of coded data read in the binary train; -Transformation of the current block into a decoded block, said step implementing a first sub-step intended to produce an intermediate block, that applies to the column vectors respectively row vectors of the current block, the second intended to produce a block of pixels that applies to the row vectors respectively column vectors of the intermediate block, arising from the first sub-step; -Reconstruction of the image on the basis of the decoded block; Characterized in that: at least one of said first and second transformation sub-steps, comprises, for a said row vector respectively column vector, the said input vector: -the formation of at least one first sub-vector of size K<N respectively N on the basis of adjacent elements of the input vector; -the transformation of the first sub-vector into a first transformed sub-vector by applying a partial sub-transform of size KxK; and -the composition of the transformed vector by inserting the at least one first transformed sub-vector.