Iterative Anchor Pixel Prediction for Image Block Encoding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image and video compression techniques, such as H.264, face challenges in predicting pixels in areas with complex discontinuities and are limited by fixed block sizes, leading to inefficient compression and increased transmission costs.

Innovation Solution

A method that iteratively selects anchor pixels for prediction based on characteristics like prediction origin, direction, and distance, updating the reference region with each iteration to improve prediction quality and adapt to block content.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If fixed block sizes (16x16, 8x8, or 4x4) are used for intra-prediction, then the encoding process is simple and standardized, but prediction quality deteriorates in areas with complex discontinuities

Engineering Contradiction:
Improveencoding simplicityVSAvoidprediction quality
Core Design Contradiction:
Ease of manufactureVSManufacturing precision

Solution Approach 1:

The macroblock is divided into multiple partitions of varying sizes and shapes (rectangular, triangular, L-shaped) rather than fixed square blocks. This segmentation allows the image block to be divided according to the actual content structure, improving prediction quality for complex discontinuities while maintaining encoding efficiency through standardized partition types.

Inventive Principle:
Principle #1Segmentation

2Productivity

If pixels far from previously reconstructed pixels are predicted using classic raster scan order, then the encoding covers the entire block, but prediction accuracy deteriorates and transmission costs increase

Engineering Contradiction:
Improveencoding coverageVSAvoidprediction accuracy
Core Design Contradiction:
ProductivityVSManufacturing precision

Solution Approach 1:

The method performs preliminary prediction of specific anchor pixels (corner and center pixels of partitions) before predicting the remaining pixels. These anchor pixels are predicted first using reference pixels from previously decoded blocks, and their prediction residuals are transmitted. The remaining pixels are then predicted using these anchor pixels as new references, improving overall prediction accuracy while maintaining efficient encoding.

Inventive Principle:
Principle #10Preliminary action

3Manufacturing precision

If linear partitions (rows or columns) are used for intra-1D prediction, then the quantity of previously reconstructed pixels increases, but prediction quality still deteriorates when blocks have complex discontinuities

Engineering Contradiction:
Improveprediction qualityVSAvoidpartition adaptability
Core Design Contradiction:
Manufacturing precisionVSAdaptability or versatility

Solution Approach 1:

The method divides the macroblock into multiple partitions with different shapes (rectangular, triangular, L-shaped) rather than using only linear row or column partitions. This multi-shaped segmentation allows the partitioning to adapt to complex discontinuities in the image content, improving prediction quality by aligning partition boundaries with actual content structures.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The partitioning structure is made dynamic by allowing different partition shapes and sizes to be selected based on the content characteristics of each macroblock. The encoder can choose from multiple partition types (rectangular, triangular, L-shaped) and configure them according to the actual image content, making the prediction structure adaptable rather than fixed.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP2652952B1Verfahren und vorrichtungen zur codierung und decodierung mindestens eines bilds durch selektierung zu predizierender pixels, und entsprechendes computerprogramm
Publication Date: 2019.05.29 ORANGE SA
  • EP2652952B1 patent drawingFigure 1~2
  • EP2652952B1 patent drawingFigure 3~5B
  • EP2652952B1 patent drawingFigure 6~8

AI summary

The invention relates to a method for decoding a signal that represents at least one image. According to the invention, such a method implements at least one iteration of the following steps, for at least one block of said image to be decoded: determining (61) at least one selection parameter from at least one characteristic associated with at least one reference pixel of a reference region of said image, said reference region including at least one previously encoded/decoded block and/or at least one pixel predicted in a preceding iteration; selecting (62) at least one pixel to be predicted in said block to be decoded from said at least one selection parameter; for at least one selected pixel to be predicted, referred to as an anchor pixel, predicting (63) said anchor pixel from at least one reference pixel, thereby outputting a predicted pixel.