Regression Model Intra Prediction for Video Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in accurately predicting and encoding video data, particularly in utilizing neighboring samples for intra prediction, which can lead to inefficiencies in compressing video data.
Innovation Solution
The proposed solution involves a regression model-based intra prediction method, where a video bitstream is processed to generate template predictions for a current block based on various prediction modes. These template predictions are then used to derive filters, which are applied to determine the final prediction of the current block, thereby improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional intra prediction methods are used, then the encoding process is simple, but the prediction accuracy is insufficient
Solution Approach 1:
The patent performs preliminary actions by generating multiple template predictions using different prediction modes before final decoding. The decoder derives filters based on template predictions and template reconstructions in advance, then applies these filters to predictions. This preliminary processing improves prediction accuracy without significantly increasing decoder complexity, as the computationally intensive filter derivation is performed using already-available template data.
Solution Approach 2:
The patent introduces filters as an intermediary between template predictions and final predictions. These filters are derived based on the relationship between template predictions and template reconstructions, and then applied to refine the final predictions. This intermediary filtering mechanism bridges the gap between simple template-based prediction and accurate reconstruction, improving prediction accuracy while maintaining a relatively simple encoding process.
2Measurement precision
If more complex prediction methods are applied, then prediction accuracy improves, but computational complexity increases
Solution Approach 1:
The patent changes parameters by deriving filters based on template predictions and template reconstructions. Instead of using a single fixed prediction method, the system dynamically adjusts prediction by applying derived filters to predictions. This parameter-based approach improves prediction accuracy by adapting to local characteristics while controlling computational complexity through efficient filter derivation and application mechanisms.
3Reliability
If template-based prediction is used, then spatial redundancy is exploited, but prediction smoothness and naturalness are insufficient
Solution Approach 1:
The patent implements feedback by deriving filters based on the comparison between template predictions and template reconstructions. The filter derivation process uses the difference between predicted and actual template values as feedback to create correction filters. These filters are then applied to improve the smoothness and naturalness of final predictions, ensuring that spatial redundancy is effectively exploited while maintaining prediction quality and stability.
Data Source
AI summary
One or more template predictions are generated for a template of a current block based on one or more prediction modes. The template includes neighboring samples of the current block. Each of the one or more template predictions is generated based on a respective one of the one or more prediction modes. One or more filters are derived for the current block. Each of the one or more filters is derived based on (i) filter index information or (ii) a respective one of the one or more template predictions and a template reconstruction of the template. One or more predictions of the current block are determined. Each of the one or more predictions is determined based on a respective one of the one or more prediction modes. A final prediction of the current block is determined by applying the one or more filters to the one or more predictions.