Unified Deep-Learning Picture Filtering for Intra and Inter Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing loop filters based on deep learning in picture processing have performance optimization defects and require further improvement, with manually designed filter coefficients leading to suboptimal performance.
Innovation Solution
A picture filtering method using a general filtering model trained with deep learning that classifies encoding modes (intra or inter) to reduce model parameter storage and enhance filtering efficiency across different encoding modes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If separate filtering models are built for different encoding modes (intra and inter), then filtering performance for each mode is optimized, but model parameter storage space increases significantly
Solution Approach 1:
The patent applies universality by designing a single filtering model that can handle both intra-frame and inter-frame encoding modes. The model uses classification information to identify the encoding mode and applies appropriate filtering strategies accordingly, eliminating the need for separate models for each mode while maintaining optimal filtering performance across different scenarios
Solution Approach 2:
The patent merges the functionality of multiple filtering models into one unified model. By combining intra-mode and inter-mode filtering capabilities within a single model architecture, the patent reduces parameter storage requirements while preserving the filtering effectiveness that would otherwise require separate specialized models
2Device complexity
If manually designed filter coefficients are used, then device complexity is reduced, but filtering performance is suboptimal
Solution Approach 1:
The patent replaces manual mechanical design of filter coefficients with an automated deep learning system. The filtering model is trained using deep learning algorithms that automatically learn optimal filtering parameters from data, eliminating the need for manual coefficient design while achieving superior filtering performance
Solution Approach 2:
The patent transitions from fixed manually-designed filter coefficients to dynamically learned parameters. The deep learning model adapts filter coefficients based on input characteristics and encoding modes, allowing optimal parameter selection without manual intervention and achieving better filtering performance
Data Source
AI summary
A picture filtering method includes determining encoding information of a picture, the encoding information comprising classification information indicating one of an intra encoding mode or an inter encoding mode of the picture. The method further includes inputting the picture and the classification information indicating one of the intra encoding mode or the inter encoding mode into a general filtering model trained using deep learning, and obtaining a filtered picture based on the filtering model performing filtering on the picture based on the encoding information.