Pixel-based intra prediction for HEVC coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current intra prediction schemes in video coding face challenges in achieving accurate prediction, especially as block size increases, leading to degraded prediction accuracy and increased residual data, which in turn increases the data to be encoded, stored, and transmitted.
Innovation Solution
The proposed solution involves a pixel-based intra prediction scheme that uses both external and internal reference pixels within a current block to generate prediction pixels sequentially, rather than in parallel, allowing for improved prediction accuracy by maintaining a consistent prediction distance regardless of block size. This scheme can be implemented in various coding schemes, including transform bypass, transform without quantization, and quantization without transform, using reconstructed pixels to avoid decoding drift.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional intra prediction schemes are used, then the coding process is simple, but prediction accuracy degrades as block size increases
Solution Approach 1:
The current block is divided into multiple sub-blocks, and intra prediction is performed separately for each sub-block using reference pixels from neighboring blocks. This segmentation allows each sub-block to use locally appropriate reference pixels, maintaining prediction accuracy even for large block sizes while keeping the overall scheme manageable through systematic processing
Solution Approach 2:
The patent introduces a new dimension to intra prediction by utilizing reference pixels from spatially adjacent neighboring blocks in addition to traditional reference pixels. This extends the prediction space beyond the current block boundaries, providing more accurate reference information for large blocks without significantly increasing computational complexity
2Productivity
If larger block sizes are used, then coding efficiency improves, but prediction accuracy degrades leading to increased residual data
Solution Approach 1:
Large blocks are segmented into smaller sub-blocks for independent prediction processing. This allows the system to maintain the efficiency benefits of large block processing while achieving prediction accuracy comparable to smaller blocks by applying the segmentation principle to preserve local prediction quality
Solution Approach 2:
Different prediction modes and reference pixel selections are applied to different sub-blocks based on local characteristics. This local quality approach ensures that each region is predicted using the most appropriate method, maintaining high prediction accuracy across the entire large block and reducing residual data
3Measurement precision
If more reference pixels are used, then prediction accuracy improves, but computational complexity increases
Solution Approach 1:
The patent uses a selective and partial approach to reference pixel utilization, employing only the necessary reference pixels from neighboring blocks for each sub-block prediction. This avoids the computational overhead of using all available reference pixels while maintaining sufficient prediction accuracy through targeted reference pixel selection
Data Source
AI summary
A decoding method including receiving a bitstream corresponding to a residual block, decoding the residual block having a plurality of residual pixels represented as transform coefficients, and computing a reconstructed block based on the residual pixels. The reconstructed block includes reconstructed pixels and uses an intra prediction mode to generate prediction pixels in sequence vertically or horizontally based on reconstructed pixels in the reconstructed block. The reconstructed block includes initial reconstructed pixels based on initial prediction pixels. The intra prediction mode is used to generate the initial prediction pixels based on external reference pixels located in neighboring blocks decoded before the reconstructed block. Computing the reconstructed block includes combining prediction pixels with residual pixels to generate additional reconstructed pixels used to generate additional prediction pixels. The additional prediction pixels are located in one position before each the additional prediction pixels according to a prediction order of the intra prediction mode.