Neural Network Image Filtering for QP-Adaptive Video Coding

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing video coding technologies, such as HEVC and VVC, face limitations in achieving superior coding efficiency despite advancements, necessitating improved image filtering methods to enhance compression while maintaining video quality.

Innovation Solution

Implementing neural network-based image filtering techniques, including fully connected neural networks (FC-NN), convolutional neural networks (CNN), and residual networks (ResNet), to adaptively adjust and filter video frames, utilizing QP maps and adjustment factors for enhanced coding efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional video coding standards (HEVC, VVC) are used, then video compression is achieved, but coding efficiency is limited and video quality degradation occurs

Engineering Contradiction:

Improvecoding efficiencyVSAvoidvideo quality

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The patent replaces traditional mechanical video filtering systems with neural network-based intelligent filtering. The neural network model learns optimal filtering operations from training data and applies adaptive filtering to video frames, substituting conventional signal processing algorithms with machine learning-based approaches that achieve superior coding efficiency and quality preservation

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent dynamically adjusts filtering parameters based on local video content characteristics. The neural network model processes input video data and generates adaptive filtering parameters that vary across different regions and frames, enabling optimal filtering strength and type selection for each local area, thereby maintaining video quality while improving compression efficiency

Inventive Principle:

Principle #35Parameter changes

2Productivity

If neural network based image filtering is implemented, then coding efficiency is improved, but computational complexity increases

Engineering Contradiction:

Improvecoding efficiencyVSAvoidcomputational complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent performs preliminary training of the neural network model offline using large datasets. The model learns optimal filtering strategies during the training phase, and during actual video coding, the pre-trained model is deployed to perform inference. This preliminary action transfers complex learning computations from the real-time coding process to the offline training phase, reducing runtime computational complexity

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent implements dynamic adjustment of neural network filtering operations based on input video characteristics. The system adapts filtering strength, kernel size, and network depth according to local content complexity, motion levels, and quality requirements. This dynamic approach allows the system to use full neural network power when needed while reducing computational load for simpler content, balancing efficiency and complexity

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS12587646B2Network based image filtering for video coding

Publication Date: 2026.03.24 BEIJING DAJIA INTERNET INFORMATION TECH CO LTD

AI summary

A method and an apparatus for image filtering in video coding using a neural network are provided. The method includes: loading, a plurality of quantization parameter (QP) map (QpMap) values at a plurality of QpMap channels into the neural network; and adjusting, according to a QP scaling factor, the plurality of QpMap values for the neural network to learn and filter a current input frame to the neural network.