Deep Learning Intra Prediction Refinement for Lower Residual Video Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression techniques struggle with high data volume due to increasing image sizes, resolutions, and frame rates, necessitating improved coding efficiency and image enhancement.
Innovation Solution
A video coding method and apparatus using a deep learning-based refinement model to generate refined prediction signals from intra prediction, employing variable and fixed coefficient networks to reduce residual data and enhance coding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image size, resolution, and frame rate are increased, then image quality and visual performance are improved, but data volume and hardware resource requirements increase
Solution Approach 1:
The patent applies deep learning-based prediction models in advance to generate prediction blocks before actual encoding. By pre-computing prediction signals using neural networks trained on image characteristics, the system prepares optimized prediction data that reduces the magnitude of residuals needing transmission, thereby reducing overall data volume while maintaining high image quality
Solution Approach 2:
The patent transforms the encoding approach by changing from traditional pixel-difference encoding to deep learning-based prediction encoding. It introduces learned parameters including prediction filter coefficients, residual scaling factors, and block partitioning strategies that adapt to local image characteristics, enabling more efficient representation of high-resolution content
2Productivity
If traditional compression techniques are used, then implementation simplicity is maintained, but coding efficiency is insufficient for high-resolution video
Solution Approach 1:
The patent segments the encoding process into distinct deep learning-based modules: prediction block generation using convolutional networks, residual computation, residual encoding, and reconstruction. Each module is independently optimized and can be selectively applied to different block types and resolutions, enabling high coding efficiency while managing complexity through modular architecture
Solution Approach 2:
The patent replaces traditional mechanical/mathematical prediction algorithms with deep learning-based neural networks. Instead of using fixed mathematical models for prediction, the system employs trained neural networks that learn optimal prediction strategies from data, achieving superior coding efficiency for complex high-resolution video content
Data Source
AI summary
A video coding method and an apparatus for refining predicted signals in intra prediction are disclosed. The video coding method and apparatus generate refined prediction signals approximating original video signals from predicted signals of intra prediction using a variable and fixed coefficient-based deep learning model to reduce the amount of data for residual signals, which are to be encoded.