Inter Prediction LIC Timing for Faster Video Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional encoders and decoders require excessive processing time when applying both bi-directional optical flow (BIO) and local illumination compensation (LIC) modes for inter prediction, particularly in low-performance systems.
Innovation Solution
The encoder and decoder implement a local illumination compensation (LIC) process using a finally-derived motion vector in a stage before correction, without applying additional corrections, and derive LIC parameters near the end of the inter prediction process, utilizing a decoder motion vector refinement (DMVR) when in merge mode.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If both bi-directional optical flow (BIO) and local illumination compensation (LIC) modes are applied for inter prediction, then coding accuracy is improved, but processing time increases excessively
Solution Approach 1:
The patent derives LIC parameters at a later stage in the encoding process, after motion vectors are finalized through DMVR or other refinement processes. This preliminary action of deferring parameter derivation allows the system to use more accurate motion information, reducing the need for extensive feedback iterations and thereby reducing processing time while maintaining coding accuracy
Solution Approach 2:
The patent extracts and applies only the necessary LIC correction using the finally-derived motion vector at a specific stage, rather than applying multiple correction passes. This extraction approach focuses computational resources on the most critical correction step, reducing overall processing time while preserving the essential illumination compensation effect
2Measurement precision
If LIC parameters are derived using feedback process, then coding accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs LIC parameter derivation at a later stage after motion vectors are finalized, rather than performing multiple feedback iterations earlier in the process. This timing strategy reduces the number of feedback loops needed while still achieving accurate parameter values, thereby improving processing speed without sacrificing coding accuracy
3Measurement precision
If multiple correction processes are applied to prediction image, then coding accuracy is improved, but device complexity increases
Solution Approach 1:
The patent extracts and applies only the essential LIC correction using the finally-derived motion vector, rather than implementing multiple separate correction processes. This extraction approach simplifies the overall process complexity while maintaining the critical illumination compensation function needed for coding accuracy
Data Source
AI summary
An encoder that encodes a moving picture using an inter prediction process, and includes circuitry and memory coupled to the circuitry. In the inter prediction process, when performing a correction process which is a local illumination compensation (LIC) process for a prediction image, the circuitry, in operation; performs the correction process on a prediction image generated using a finally-derived motion vector that is finally derived in a stage before the correction process; and after the correction process, determines, as a final prediction image, the prediction image subjected to the correction process, without applying other correction process on the prediction image.