Asymmetric Video Block Transform via Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression technologies, such as the HEVC standard and the Quad-Tree plus Binary-Tree (QTBT) coding tool, face limitations in achieving further improved compression efficiency and reduced complexity.
Innovation Solution
The proposed method involves dividing a block of video data into sub-blocks, where at least one sub-block has a size that is not a power of two in height or width, and encoding these sub-blocks using processing that avoids the need for transforms that are not powers of two.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If asymmetric coding units (non-power-of-two sizes) are used in video compression, then compression efficiency is improved, but transform complexity increases due to the need for non-power-of-two transforms
Solution Approach 1:
The asymmetric coding unit is segmented into multiple sub-blocks, where at least one sub-block has a power-of-two size. This allows the transform operation to be applied only to the power-of-two sized sub-blocks, avoiding the need for complex non-power-of-two transforms while still supporting asymmetric coding unit structures for improved compression efficiency
Solution Approach 2:
Different parts of the coding unit are treated differently: power-of-two sized sub-blocks undergo transform operations while other regions may use alternative processing. This local differentiation allows the system to maintain compression efficiency benefits of asymmetric units without incurring the full complexity of non-power-of-two transforms across the entire block
2Adaptability or versatility
If non-power-of-two transform sizes are introduced to support asymmetric coding units, then coding flexibility is improved, but memory usage increases
Solution Approach 1:
By segmenting the asymmetric coding unit into power-of-two sized sub-blocks, the system can reuse existing power-of-two transform kernels that are already stored in memory. This avoids the need to allocate additional memory for new non-power-of-two transform sizes, maintaining coding flexibility while controlling memory usage
Solution Approach 2:
The power-of-two sized sub-blocks can utilize the same transform kernels for multiple different asymmetric coding unit configurations. This universal approach allows a single set of transform kernels to serve multiple coding scenarios, reducing the overall memory footprint required for transform operations
Data Source
AI summary
A coding unit having a size multiple of three in horizontal or vertical direction is coded through one of several embodiments. In one embodiment, for some block sizes, the coding unit is coded and decoded systematically through SKIP mode. In another embodiment, the coding units can be coded in SKIP mode or with a DC coefficient. In another embodiment, an asymmetric division of a common coding unit parent is performed and transform coefficients are factorized among at least two sub-blocks to encode a coding unit. In another embodiment, a separable two dimensional transform can be applied by applying a transform over the block in one direction, and using two one-dimensional transforms on sub-blocks in the other direction to code. Methods, apparatus, and signal embodiments are provided for encoding and decoding.