Intra Prediction Mode Derivation for High-Resolution Image Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image compression technologies struggle to efficiently handle high-resolution images by lacking diverse prediction methods, which affects prediction accuracy and compression performance.
Innovation Solution
The method and device derive an intra prediction mode for a current block, construct reference samples, and generate prediction samples based on predefined planar modes, including normal, horizontal, and vertical modes, using decoder-side intra mode derivation (DIMD) and template matching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional intra-prediction modes are used, then the prediction process is simple, but prediction accuracy is insufficient for high-resolution images
Solution Approach 1:
The patent divides the prediction process into multiple independent planar modes (first planar mode using first reference samples, second planar mode using second reference samples). Each mode can be selected or combined based on image characteristics, allowing the system to achieve high prediction accuracy by choosing the most appropriate segmented prediction approach without processing the entire image as a single complex unit.
Solution Approach 2:
The patent introduces a new dimension of prediction by allowing combination of multiple planar modes with different reference sample configurations. Instead of using a single planar mode, the system can combine first planar mode prediction with second planar mode prediction, effectively adding a dimensional layer of complexity that enables more accurate representation of complex image structures.
2Productivity
If multiple planar modes are introduced to improve prediction accuracy, then compression performance improves, but signaling overhead increases
Solution Approach 1:
The patent allows the encoder to selectively signal only the necessary planar mode information rather than always signaling complete mode data. When a single planar mode is sufficient, only that mode is signaled; when multiple modes are needed, only the combination information is signaled. This partial action approach reduces signaling overhead while maintaining compression performance.
Solution Approach 2:
The patent changes the parameter representation by using flags and indicators to denote planar mode usage rather than explicitly signaling all mode parameters. The system can indicate whether first planar mode, second planar mode, or both are used through compact flag structures, reducing the amount of information that needs to be transmitted while preserving the ability to achieve high compression performance.
3Device complexity
If decoder-side intra mode derivation is used, then encoding complexity is reduced, but derivation accuracy may be insufficient
Solution Approach 1:
The patent incorporates feedback mechanisms where the decoder uses previously derived planar mode information and reconstruction results to refine subsequent mode derivation. The system can adjust derivation parameters based on feedback from prediction error analysis, allowing the decoder-side derivation to progressively improve accuracy while maintaining reduced encoding complexity.
Solution Approach 2:
The patent performs preliminary planar mode derivation at the encoder side during the encoding process, and the decoder uses these pre-derived modes as starting points for its own derivation process. This preliminary action reduces the computational burden on the decoder while maintaining high derivation accuracy, as the decoder only needs to refine rather than compute modes from scratch.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image decoding/encoding method and device according to the present disclosure may derive an intra prediction mode of a current block, construct reference samples of the current block, generate a prediction sample of the current block based on the intra prediction mode and the reference samples, and decode/encode the current block based on the prediction sample of the current block.