Motion Estimation Using Compression Metadata for Adaptive Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional motion estimation techniques in graphics processing require significant computational resources, energy, and latency due to the need to classify regions as edges or shades before determining an appropriate search space, and they often necessitate additional hardware for pre-processing.
Innovation Solution
Utilize compression meta data to determine the smoothness of regions, allowing for the setting of reduced or extended search spaces based on smoothness information without additional processing, thereby reducing computational requirements and latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a larger search space is used to improve matching accuracy, then the likelihood of finding a better matching region increases, but more processing resources, energy, and time are required
Solution Approach 1:
The patent applies local quality by differentiating search space sizes based on region characteristics. Edge regions use a larger search space (e.g., 32x32 pixels) to capture motion across boundaries, while shade regions use a smaller search space (e.g., 16x16 pixels) since they are likely to move with adjacent regions. This selective approach improves matching accuracy where needed while reducing overall processing load.
Solution Approach 2:
The patent implements dynamics by adaptively adjusting search space size based on detected region properties. The system dynamically determines whether each region is an edge or shade region and adjusts the search space accordingly, rather than using a fixed search space for all regions. This dynamic adaptation optimizes the balance between matching accuracy and processing efficiency.
2Productivity
If edge detection and region classification are performed to reduce search space, then processing energy and time are reduced, but additional hardware and pre-processing steps are required
Solution Approach 1:
The patent merges the motion estimation process with region classification by performing both operations in an integrated manner. The region classification (edge vs. shade) is performed as part of the motion estimation pipeline rather than as a separate pre-processing step, reducing hardware overhead and simplifying the overall system architecture.
Solution Approach 2:
The motion estimation algorithm itself performs the classification function as a byproduct of its operation. By analyzing pixel variations during the matching process, the algorithm automatically identifies edge and shade regions without requiring separate dedicated hardware for classification, thus reducing device complexity.
3Measurement precision
If reference frame data is retrieved from memory for each region matching operation, then accurate motion estimation can be performed, but significant memory bandwidth and energy are consumed
Solution Approach 1:
The patent segments the reference frame access pattern by region type. For shade regions, it accesses only the necessary smaller search space portions from memory, while for edge regions it accesses the full search space. This segmentation of memory access patterns reduces overall memory bandwidth consumption and energy usage compared to uniformly accessing the entire search space for all regions.
Data Source
AI summary
A method of performing motion estimation for spatially arranged data in a data processor, the spatially arranged data having been previously compressed using a compression algorithm that generates compression meta data representing the spatially arranged data includes for a first region of the spatially arranged data: obtaining compression meta data for the first region, the compression metadata including smoothness information indicative of a smoothness of the first region; determining a search space size based on the obtained smoothness information, the search space size being a size of a search space for applying a region-matching motion estimation algorithm to the first region to match the first region with a corresponding region in reference spatially arranged data within the search space.