Decoder Side Intra Prediction Mode Selection Using No Reference Quality Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding technologies face challenges in efficiently signaling intra prediction modes, leading to increased overhead and reduced compression gain due to the need for fixed-length binarization of mode indices, especially with a large number of modes like in VVC.
Innovation Solution
Implementing a no reference image quality assessment method that allows for decoder-side prediction, using visual parameter measurement indices (VPMI) or deep learning-based Blind Image Quality Assessment (DeepBIQ) to predict the intra prediction mode, reducing the need for explicit signaling of intra prediction modes and minimizing overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If fixed-length binarization is used for signaling intra prediction modes, then the signaling process is simple and deterministic, but the overhead increases and compression gain decreases
Solution Approach 1:
The decoder autonomously determines the intra prediction mode by performing quality assessment on candidate modes without requiring explicit signaling from the encoder. This self-service mechanism eliminates the need for overhead bits while maintaining decoding accuracy, directly resolving the contradiction between signaling simplicity and compression efficiency.
Solution Approach 2:
The patent extracts the intra prediction mode determination function from the encoder and relocates it to the decoder. By taking out the mode selection task from the encoding process and performing it independently at the decoder side through quality assessment, the signaling overhead is eliminated while preserving the ability to select optimal modes.
2Measurement precision
If the number of intra prediction modes is increased (e.g., in VVC), then the prediction accuracy improves, but the signaling overhead increases due to fixed-length binarization
Solution Approach 1:
The decoder independently evaluates multiple candidate intra prediction modes using quality assessment metrics and autonomously selects the best mode without requiring the encoder to signal the chosen mode. This approach enables support for a large number of prediction modes (improving accuracy) while eliminating the increased signaling overhead that would result from encoding more mode options.
Solution Approach 2:
The decoder performs quality assessment on a subset of candidate modes (e.g., most probable modes) rather than all possible modes. This partial action approach maintains high prediction accuracy by evaluating sufficient candidates while avoiding the need to signal all mode possibilities, thus reducing overhead even as the total number of available modes increases.
3Reliability
If explicit signaling of intra prediction modes is performed, then the encoder and decoder maintain synchronization, but the bitstream overhead increases
Solution Approach 1:
Both encoder and decoder independently determine the same intra prediction mode through identical quality assessment procedures applied to the same candidate modes. This self-service approach ensures synchronization between encoder and decoder without requiring explicit mode signaling in the bitstream, thereby eliminating overhead while maintaining reliability.
Solution Approach 2:
The patent merges the mode determination function into the quality assessment process itself, which is already performed at the decoder for other purposes (e.g., rate-distortion optimization). By combining mode selection with existing decoder operations, synchronization is achieved without additional signaling overhead.
Data Source
AI summary
A decoder may receive, for a block, an indication that an intra prediction mode is to be determined at the decoder to decode the block. The decoder may generate, based on the indication, a plurality of reconstructed blocks for a plurality of intra prediction modes. Each of the plurality of reconstructed blocks is generated based on a prediction block generated using a respective intra prediction mode of the plurality of intra prediction modes. The decoder may further select, based on a visual quality of each of the plurality of reconstructed blocks, a prediction mode, from the plurality of intra prediction modes, as the intra prediction mode of the block. The decoder may decode the block based on the intra prediction mode.