Neural Network Image Filtering With QP Scaling for Video Coding

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing video coding techniques, such as HEVC and VVC, face limitations in achieving superior coding efficiency despite advancements, necessitating improved methods to compress video data with lower bit rates while maintaining video quality.

Innovation Solution

Employing neural network-based model filtering, specifically using QP scaling factors to adjust quantization parameter maps (QpMap) values, and applying neural networks like FC-NN, CNN, and ResNet for image filtering in video coding to enhance compression efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If neural network-based model filtering is applied to enhance video coding efficiency, then compression efficiency and video quality are improved, but computational complexity increases

Engineering Contradiction:

Improvevideo coding efficiencyVSAvoidcomputational complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent applies preliminary action by pre-training neural network models (such as FC-NN, CNN, ResNet) offline using training data, then deploying the trained models for inference during video coding. This separates the complex training phase from the coding phase, allowing efficient real-time filtering without requiring complex computations during actual video processing.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent replaces traditional mechanical filtering systems (such as deblocking filters, sample adaptive offset, adaptive loop filter) with neural network-based filtering. The neural networks learn optimal filtering parameters from training data and apply them automatically, substituting complex manual filter design with data-driven automated filtering that achieves superior compression efficiency.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Manufacturing precision

If adaptive filtering techniques are used to maintain video quality, then filtering performance improves, but processing time increases

Engineering Contradiction:

Improvevideo qualityVSAvoidprocessing time

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent applies dynamics by making the filtering process adaptive to different video content characteristics. The neural network models dynamically adjust filtering strength and parameters based on input video data, allowing optimal quality maintenance while reducing processing time for simple content and applying stronger filtering only when needed for complex content.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent changes filtering parameters dynamically during video processing. The neural networks adjust convolution kernel sizes, filtering strengths, and other parameters based on input video characteristics, enabling quality optimization without fixed processing time constraints.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20250386015A1Network based image filtering for video coding

Publication Date: 2025.12.18 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; obtaining a QP scaling factor by adjusting a plurality of input QP values related to an input frame; and adjusting, according to a QP scaling factor, the plurality of QpMap values for the neural network to learn and filter the input frame to the neural network.