Motion Vector Refinement With Padded Access Windows
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current motion vector refinement methods in video encoding and decoding require increased on-chip memory or external memory access, which can slow down processing and are inefficient.
Innovation Solution
Implement motion vector prediction by padding sample values outside a predefined memory access window with replacement values based on samples within the window, using interpolation filtering to limit memory access and maintain accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion vector refinement is performed using traditional methods, then prediction accuracy is improved, but memory access requirements increase and processing speed decreases
Solution Approach 1:
The patent segments the reference picture into a current block and a padding block, where the padding block contains synthesized reference samples generated through interpolation. This segmentation allows the motion vector refinement process to access only the necessary current block data from memory, while the padding block is generated locally, thereby reducing memory access requirements and improving processing speed without compromising prediction accuracy.
Solution Approach 2:
The patent performs preliminary interpolation to generate reference samples for the padding block before the motion vector refinement process. By pre-computing these synthetic reference samples and storing them in the padding block, the system eliminates the need for repeated memory accesses during refinement operations, thus maintaining high prediction accuracy while significantly reducing memory bandwidth consumption and improving processing throughput.
2Measurement precision
If on-chip memory size is increased to store more reference samples, then motion vector refinement accuracy is improved, but device complexity and cost increase
Solution Approach 1:
The patent extracts only the essential reference samples needed for motion vector refinement from the reference picture and places them in a compact current block within on-chip memory. The remaining reference samples that would traditionally be stored in large on-chip memory are instead synthesized on-demand in a padding block using interpolation filters, thereby reducing on-chip memory requirements while maintaining refinement accuracy.
Solution Approach 2:
The patent employs a disposable padding block that contains synthesized reference samples generated through interpolation. Instead of permanently allocating large on-chip memory for all possible reference samples, the system creates temporary synthetic samples only when needed for motion vector refinement. These synthetic samples are computed locally and discarded after use, significantly reducing hardware complexity and cost while maintaining the necessary refinement accuracy.
3Measurement precision
If external memory access is increased to provide more reference data, then prediction quality is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary interpolation to generate the padding block containing synthetic reference samples before the motion vector refinement process begins. By pre-computing these samples and making them locally available, the system eliminates the need for time-consuming external memory accesses during refinement operations, thus maintaining high prediction quality while significantly reducing processing time.
Solution Approach 2:
The patent introduces a padding block as an intermediary structure that contains synthesized reference samples. This padding block acts as a local cache that mediates between the limited on-chip memory and the external memory, providing quick access to necessary reference data without requiring frequent external memory accesses, thereby improving prediction quality while reducing processing time.
Data Source
AI summary
The present disclosure relates to motion vector determination using template or bilateral matching and predictor generation based on the motion vector. The template or bilateral matching and/or the predictor generation use interpolation filtering. The interpolation filtering operation accesses integer sample positions within a window, and further uses padded sample values for integer sample positions outside the window, which are based on at least one sample within said window, and uses the accessed integer sample position values as well as the padded sample values to perform the template or bilateral matching and/or predictor generation.