Image Decoding with GPM Intra-Inter Blending in LMCS Space
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing geometric partitioning mode (GPM) in image decoding is limited to inter prediction, leading to suboptimal coding performance.
Innovation Solution
An image decoding device that blends motion compensation (MC), overlapped block motion compensation (OBMC), and intra prediction samples after luma mapping, aligning signal spaces to generate a prediction sample, especially when GPM, OBMC, and luma mapping with chroma scaling (LMCS) are valid and include inter and intra prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If geometric partitioning mode (GPM) is limited to inter prediction only, then the implementation complexity is reduced, but the coding performance is insufficient
Solution Approach 1:
The patent extends the geometric partitioning mode (GPM) to support both inter prediction and intra prediction functions within a unified framework. The partitioning structure and blending mechanism are designed to be versatile, accommodating different prediction types without requiring separate implementation paths, thus improving coding performance while maintaining implementation efficiency.
Solution Approach 2:
The patent introduces a parameter (prediction type flag) that controls whether GPM applies inter prediction or intra prediction. By changing this parameter, the system can adapt the GPM behavior to different prediction scenarios, enabling improved coding performance for both inter and intra predicted blocks without increasing structural complexity.
2Productivity
If motion compensation sample, OBMC sample, and intra prediction sample are blended without luma mapping, then the processing speed is faster, but the prediction accuracy is reduced
Solution Approach 1:
The patent applies luma mapping as a preliminary transformation to the motion compensation sample and OBMC sample before blending. This pre-processing step aligns the signal spaces of different prediction samples, ensuring that the subsequent blending operation produces accurate predictions. The luma mapping is performed once before blending rather than repeatedly, maintaining processing efficiency.
Solution Approach 2:
Luma mapping serves as an intermediary transformation that converts samples from different signal spaces into a common reference space. This intermediary step enables accurate blending of motion compensation, OBMC, and intra prediction samples by ensuring they all operate in the same signal space, thereby improving prediction accuracy without significantly impacting processing speed.
3Measurement precision
If luma mapping is applied to align signal spaces of MC, OBMC, and intra prediction samples, then the prediction accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent applies luma mapping selectively only to the luma components of the motion compensation sample and OBMC sample, leaving chroma components and intra prediction samples unchanged. This localized application of luma mapping reduces unnecessary computational operations while maintaining the signal space alignment needed for accurate prediction in the luma domain.
Solution Approach 2:
The patent performs luma mapping only when necessary - specifically when blending motion compensation samples or OBMC samples with intra prediction samples. In cases where only inter prediction modes are used, luma mapping is skipped. This partial application of luma mapping reduces overall computational complexity while maintaining prediction accuracy when the transformation is actually needed.
Data Source
AI summary
In an image decoding device (200) according to the present invention, a blending unit (243) blends, in a case where all of a geometric partitioning mode, overlapped block motion compensation, and luma mapping with chroma scaling are valid for a current block, and the geometric partitioning mode includes inter prediction and intra prediction, luma-mapped luma components of the motion compensation sample, the overlapped block motion compensation sample, and the intra prediction sample to generate a prediction sample for a luma component of the current block.