Intra Prediction Direction Selection for Image Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The H.265/HEVC image coding standard requires examining all 35 intra prediction directions, leading to significant processing overhead, especially in real-time or low-delay systems, and prioritizing directions like DC and planar prediction reduces coding efficiency.
Innovation Solution
An image coding method that selects a first prediction direction from a set of directions based on cost calculation, such as distortion amount, and optionally performs intra prediction in that direction, then selects a second prediction direction from an adjacent set to optimize processing while maintaining coding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If all 35 intra prediction directions are examined, then coding efficiency is improved, but processing complexity increases significantly
Solution Approach 1:
The patent divides the 35 prediction directions into multiple groups (e.g., vertical group, horizontal group, diagonal group) and processes them separately. This segmentation allows the system to evaluate multiple directions without examining all 35 in a single exhaustive pass, reducing processing complexity while maintaining coding efficiency through selective group evaluation.
Solution Approach 2:
The patent implements a two-stage prediction direction selection process where only a subset of prediction directions is evaluated in the first stage, and only if certain conditions are met does the system proceed to evaluate additional directions in a second stage. This partial action approach prevents the need to always process all 35 directions, reducing average processing complexity while preserving coding efficiency when needed.
2Manufacturing precision
If all 35 intra prediction directions are examined, then coding efficiency is improved, but processing time increases
Solution Approach 1:
The patent implements a periodic evaluation mechanism where prediction directions are examined in alternating stages or passes rather than all at once. The system periodically switches between evaluating different groups of directions based on intermediate results, which reduces the time required for any single evaluation cycle while maintaining overall coding efficiency through multiple passes.
Solution Approach 2:
The patent performs preliminary evaluation of certain prediction directions (such as vertical and horizontal directions) before evaluating more complex diagonal directions. This preliminary action allows the system to quickly eliminate obviously suboptimal directions and focus processing time on more promising candidates, reducing total processing time while maintaining coding efficiency.
3Device complexity
If prediction directions are prioritized (e.g., DC and planar), then processing complexity is reduced, but coding efficiency deteriorates
Solution Approach 1:
The patent implements a dynamic prediction direction selection mechanism that adapts the evaluation strategy based on the characteristics of the current block being encoded. Rather than statically prioritizing only DC and planar directions, the system dynamically adjusts which directions to evaluate and in what order based on block size, block position, and preliminary distortion measurements, thereby maintaining coding efficiency while reducing processing complexity through adaptive prioritization.
Solution Approach 2:
The patent changes the evaluation parameters (such as the set of directions to evaluate, the order of evaluation, and the threshold for proceeding to second-stage evaluation) based on block characteristics. This parameter adaptation allows the system to reduce processing complexity for certain block types while maintaining coding efficiency for others, resolving the contradiction between simplified processing and optimal coding performance.
Data Source
AI summary
Prediction directions for cost calculation are reduced by preferentially selecting a direction in which a prediction pixel is generated by direct copy without filtering from among intra prediction directions to perform intra prediction. In addition to the roughly selected prediction direction, an amount of processing is reduced while deterioration of coding efficiency is prevented through comparison with preferentially selected prediction, such as DC prediction and planar prediction, and selection of second-stage prediction direction focused on a vicinity of roughly selected prediction direction.