Sub-pixel Motion Compensation Using Direct Polyphase Interpolation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current motion compensation approaches in video standards, such as H.264 and proposed H.265, incur increased bus-bandwidth and cost due to hierarchical interpolation methods that require multiple levels of interpolation, leading to higher complexity and bandwidth consumption.
Innovation Solution
Implementing direct polyphase interpolation using programmable multi-tap filters to derive sub-pixel positions directly from integer-pixel data, reducing the need for intermediate interpolation levels and minimizing bus-bandwidth, while maintaining performance for future video coding standards.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If hierarchical interpolation techniques are used to achieve sub-pixel motion compensation, then motion compensation performance is improved, but bus-bandwidth consumption and device complexity increase significantly
Solution Approach 1:
The patent segments the interpolation process by using separate filter structures for different interpolation directions (horizontal and vertical). Instead of applying multiple levels of hierarchical filtering, it uses segmented filter banks that can independently process different directional components, reducing the overall computational complexity and bus-bandwidth requirements while maintaining sub-pixel accuracy.
Solution Approach 2:
The patent transitions from the traditional hierarchical approach (which operates sequentially through multiple refinement levels) to a direct polynomial phase modulation approach. This dimensional change allows all sub-pixel positions to be generated simultaneously from integer positions using closed-form polynomial expressions, eliminating the need for intermediate hierarchical processing stages and significantly reducing bus-bandwidth consumption.
2Measurement precision
If programmable multi-tap filters are used for direct sub-pixel interpolation, then filtering sharpness and performance are improved, but implementation cost increases
Solution Approach 1:
The patent changes the fundamental parameters of the filtering approach by using polynomial phase modulation instead of traditional multi-tap convolutional filtering. This parameter change allows the system to achieve sharp filtering through closed-form mathematical expressions that are computationally more efficient and less costly to implement than programmable multi-tap filters, while maintaining sub-pixel precision.
Solution Approach 2:
The patent substitutes the mechanical/iterative hierarchical filtering process with a direct mathematical computation approach using polynomial phase modulation. This replacement eliminates the need for iterative refinement stages and complex filter tap programming, reducing implementation cost while achieving the same or better filtering sharpness through efficient closed-form calculations.
3Measurement precision
If multiple levels of hierarchical interpolation are implemented, then sub-pixel accuracy is improved, but the number of required bus transactions and memory accesses increases
Solution Approach 1:
The patent performs preliminary action by pre-computing polynomial phase coefficients and organizing filter bank structures in advance. This allows the system to directly generate sub-pixel values from integer positions using pre-prepared mathematical expressions, eliminating the need for multiple sequential bus transactions that would otherwise be required to fetch intermediate results at each hierarchical level.
Solution Approach 2:
The patent uses polynomial phase modulation to directly copy and transform integer-position pixel values into sub-pixel positions through mathematical transformation. This copying mechanism eliminates the need for multiple hierarchical interpolation stages, each requiring separate bus transactions, by directly generating all required sub-pixel values in a single computational pass.
Data Source
AI summary
A method of motion compensation for an input block is disclosed. The method generally includes the steps of (A) generating a plurality of tap values in response to a motion vector for the input block, (B) generating an interpolated block by programmable filtering a reference block using the tap values, the interpolated block being spatially offset from the reference block by less than a half-pel separation and (C) generating an output block in response to the input block and the interpolated block.