Classification-Based Loop Filter for Video Artifact Removal

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current video coding standards, such as the MPEG-4 AVC Standard, face challenges in effectively removing compression artifacts due to their block-based methods, which fail to capture spatial variations in video signals, particularly edge sharpness, leading to degraded visual quality.

Innovation Solution

A classification-based loop filter approach is introduced, where pixels in a video frame are categorized based on local geometric characteristics, such as edge orientation, and filtered using corresponding filter coefficients, either adaptively trained frame-by-frame or pre-trained offline, to improve the removal of compression artifacts.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If block-based filtering methods are used, then filtering can be applied to remove compression artifacts, but the method fails to capture spatial variations and edge sharpness, leading to degraded visual quality

Engineering Contradiction:
Improveartifact removal effectivenessVSAvoidedge sharpness preservation
Core Design Contradiction:
ReliabilityVSManufacturing precision

Solution Approach 1:

The image is segmented into multiple blocks, and each block is further divided into sub-blocks for independent filtering decisions. This hierarchical segmentation allows the filter to adapt to local variations while maintaining computational efficiency, resolving the contradiction between artifact removal and edge preservation by treating different regions differently

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements local quality by classifying pixels within each sub-block based on their specific characteristics (edge presence, orientation, variance) and applying appropriate filtering strength to each pixel. This ensures that edges are preserved where present while artifacts are removed in flat regions, directly addressing the contradiction between artifact removal and edge sharpness

Inventive Principle:
Principle #3Local quality

2Reliability

If Wiener filters are estimated for each picture with spatial adaptation, then filtering performance improves, but overhead for transmitting filter coefficients and region partition information increases

Engineering Contradiction:
Improvefiltering performanceVSAvoidoverhead transmission
Core Design Contradiction:
ReliabilityVSLoss of information

Solution Approach 1:

The patent divides the picture into blocks and sub-blocks, allowing filter coefficients to be estimated and transmitted only for necessary regions rather than the entire picture. This segmentation reduces the overall overhead while maintaining filtering performance in critical areas

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Instead of applying full spatial adaptation with region partitioning to the entire picture, the patent applies filtering selectively at the block and sub-block level where needed. This partial action approach maintains essential filtering performance while significantly reducing the overhead associated with comprehensive region partitioning and coefficient transmission

Inventive Principle:
Principle #16Partial or excessive action

3Ease of manufacture

If equal-size blocks are used for region partitioning, then implementation is simple, but the method does not effectively characterize spatial variations within video

Engineering Contradiction:
Improveimplementation simplicityVSAvoidspatial variation characterization
Core Design Contradiction:
Ease of manufactureVSManufacturing precision

Solution Approach 1:

The patent uses equal-size blocks as a base structure for simplicity, but introduces sub-block segmentation within each block to capture finer spatial variations. This hierarchical approach maintains the implementation simplicity of regular blocks while adding the precision needed to characterize local spatial variations through sub-block classification

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Within each equal-size block, the patent applies local quality analysis by classifying pixels and sub-blocks based on their specific characteristics. This allows the simple block structure to be enhanced with local adaptability, achieving both ease of implementation and effective spatial variation characterization

Inventive Principle:
Principle #3Local quality

4Manufacturing precision

If variable-size blocks are used for region partitioning, then spatial adaptation improves, but overhead for coding block size and location increases

Engineering Contradiction:
Improvespatial adaptationVSAvoidoverhead for block size and location
Core Design Contradiction:
Manufacturing precisionVSLoss of information

Solution Approach 1:

The patent adopts a hierarchical segmentation structure with equal-size blocks containing variable sub-blocks. This approach captures spatial adaptation benefits at the sub-block level while avoiding the overhead of encoding variable block sizes and positions, as the hierarchical structure is more compact to represent

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Instead of implementing full variable-size block partitioning throughout the picture, the patent applies variable-size sub-blocks only within equal-size blocks where needed. This partial implementation achieves spatial adaptation where necessary while minimizing the overhead associated with variable block coding

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP2545711B1Methods and apparatus for a classification-based loop filter
Publication Date: 2018.05.09 THOMSON LICENSING SA
  • EP2545711B1 patent drawingFigure 1
  • EP2545711B1 patent drawingFigure 2
  • EP2545711B1 patent drawingFigure 3

AI summary

Methods and apparatus for a classification-based loop filter are provided. An encoding method encodes an input picture by transforming and quantizing a residue of the input picture to obtain quantized transform coefficients. Then inverse quantizing and inverse transforming the quantized transform coefficients to obtain a reconstructed version of the residue. A reconstructed version of the input picture is obtained by combining at least one reference picture with the reconstructed version of the residue (315). Pixels in the reconstructed version of the input picture are classified (330) within a respective one of a plurality of categories responsive to local geometric characteristics. Pixels are selected for filtering responsive to a corresponding classification of the pixels with respect to the plurality of categories (350).