Directional Filter Adaptation for Video Coding Edge Artifacts

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Hybrid video coding schemes face challenges in coding efficiency and visual quality due to lossy nature, with existing adaptive loop filtering methods being complex and unsuitable for reducing artifacts around sharp edges, as they primarily rely on linear Wiener filters that are isotropic and low-pass.

Innovation Solution

The method involves classifying video frame blocks based on local directional features, designing directional filters with symmetric constraints, and optionally combining linear spatial filtering with directional filtering using a similarity mapping function to adapt to edges, thereby improving coding efficiency and visual quality.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If linear Wiener filters are used for adaptive loop filtering, then coding efficiency is improved, but artifacts around sharp edges cannot be effectively reduced due to isotropic and low-pass characteristics

Engineering Contradiction:
Improvecoding efficiencyVSAvoidartifacts around sharp edges
Core Design Contradiction:
ProductivityVSObject-affected harmful factors

Solution Approach 1:

The patent applies asymmetry by transitioning from isotropic Wiener filters to directional filters that exploit local edge orientations. Blocks are classified based on dominant edge directions (horizontal, vertical, diagonal), and filters are designed with asymmetric coefficient patterns matched to these directions. This allows the filter to adapt to the specific orientation of edges in each block, effectively reducing artifacts around sharp edges while maintaining coding efficiency.

Inventive Principle:
Principle #4Asymmetry

Solution Approach 2:

The patent implements local quality by classifying blocks into different directional categories based on local edge characteristics and designing separate filters for each class. Instead of using a single isotropic filter for the entire image, the system adapts the filter characteristics to match the local edge orientation and texture properties of each block, thereby optimizing artifact reduction at each location while preserving overall coding efficiency.

Inventive Principle:
Principle #3Local quality

2Object-affected harmful factors

If pixel-based filter adaptation is used to reduce artifacts, then visual quality is improved, but device complexity increases significantly

Engineering Contradiction:
ImproveartifactsVSAvoidfilter adaptation complexity
Core Design Contradiction:
Object-affected harmful factorsVSDevice complexity

Solution Approach 1:

The patent applies segmentation by dividing the image into blocks (e.g., 8x8 or 16x16 pixels) and classifying each block based on its edge characteristics. This block-level segmentation reduces complexity compared to pixel-based adaptation, as fewer classification decisions need to be made. The blocks are then grouped into directional classes and assigned appropriate filters, achieving a balance between artifact reduction and computational complexity.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent utilizes parameter changes by modifying filter characteristics based on block classification results. Different filter coefficient sets are applied to different directional classes, and the filter strength is adjusted based on block variance and edge strength parameters. This parameter adaptation allows the system to optimize artifact reduction for each block type while maintaining manageable complexity through structured parameter variation rather than complete redesign.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP2477403B1Filter adaptation with directional features for video/image coding
Publication Date: 2019.07.03 SAMSUNG ELECTRONICS CO LTD
  • EP2477403B1 patent drawingFigure 1
  • EP2477403B1 patent drawingFigure 2
  • EP2477403B1 patent drawingFigure 3

AI summary

A method for processing video information by a video encoder includes classifying video information based on at least one local directional feature of the video information to design multiple filters. The encoder encodes filter coefficients of the multiple filters. The multiple filters are designed with symmetric constraints on the filter coefficients configured according to the classification of the video information based on at least one local directional feature of the video information. The method also includes applying, by the encoder, the multiple directional filters to the video information. A decoder is configured to decode filter coefficients of multiple filters and apply the multiple filters to the video information. The decoder constructs, from decoded filter coefficients, multiple filters with symmetric constraints on the filter coefficients configured according to the classification of the video information based on at least one local directional feature of the video information.