Flexible Tree Partitioning for Non-Power-of-Two Video Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing video coding technologies, such as HEVC and its extensions, face inefficiencies and computational challenges due to unbalanced tree structures like the two-level MTT, which can degrade coding performance and introduce significant computational expense, especially in cases where block sizes are not powers of two, and HEVC's implicit quad-tree splitting is not flexible enough for advanced block partitioning.
Innovation Solution
Implementing a split-to-square partition pattern, horizontal binary tree partition pattern, and horizontal ternary tree partition pattern with equal priority, allowing these patterns to be used sequentially, interleaved, or before/after each other, to generate sub-blocks, and ensuring balanced tree depth for efficient block partitioning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If HEVC's implicit quad-tree splitting is used, then the block partitioning is simple to implement, but it lacks flexibility for advanced block partitioning and degrades coding performance
Solution Approach 1:
The patent segments the block partitioning process into multiple independent tree structures (first tree structure for spatial partitioning, second tree structure for frequency partitioning) that can be applied separately and independently. This allows flexible combination of different partitioning strategies without requiring a single complex unified structure, thereby improving adaptability while managing complexity.
Solution Approach 2:
The patent introduces a second tree structure that operates in a different dimension (frequency domain partitioning) alongside the first tree structure (spatial domain partitioning). This multi-dimensional approach enables more versatile block partitioning capabilities by adding another layer of segmentation without significantly increasing the complexity of individual tree structures.
2Adaptability or versatility
If unbalanced tree structures like two-level MTT are used, then advanced block partitioning is achieved, but computational expense increases significantly
Solution Approach 1:
The patent divides the partitioning task into two separate balanced tree structures rather than using one unbalanced two-level MTT structure. Each tree handles a specific aspect (spatial and frequency partitioning), which distributes the computational load more evenly and avoids the exponential complexity growth associated with unbalanced deep trees.
Solution Approach 2:
The patent applies partitioning in two sequential stages rather than attempting to achieve all partitioning in a single unbalanced structure. The first tree structure performs initial partitioning, and the second tree structure performs additional partitioning only where needed, avoiding the excessive computational action that would result from forcing a single unbalanced structure to handle all cases.
3Adaptability or versatility
If block sizes are not powers of two, then flexible partitioning is achieved, but transform requirements become more complex
Solution Approach 1:
The patent segments the transform processing into two independent stages corresponding to the two tree structures. Each stage handles transforms for blocks of various sizes independently, allowing flexible block size support without requiring a single complex transform algorithm to handle all size variations, thereby reducing overall transform processing complexity.
Data Source
AI summary
A method and apparatus for partitioning a block to permit the encoding or decoding of a video sequence includes partitioning the block using a split-to-square partition pattern, a horizontal binary tree partition pattern, or a horizontal ternary tree partition pattern to generate a set of sub-blocks, wherein a same priority is allocated to the split-to-square partition pattern, the horizontal binary tree partition pattern, and the horizontal ternary tree partition pattern such that the split-to-square partition pattern may be used before, after, or interleaved with the horizontal binary tree partition pattern and the horizontal ternary tree partition pattern. Encoding or decoding the video sequence is performed based on partitioning the block using the split-to-square partition pattern, the horizontal binary tree partition pattern, or the horizontal ternary tree partition pattern.