Motion Search with Scaled Reference Pictures
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The processing of video information for encoding, particularly in standards like H.264, requires generating motion vectors by comparing pictures, which strains memory bandwidth and reduces precision due to the need for downscaling images, leading to reduced accuracy in motion vector generation.
Innovation Solution
The technique involves downsampling reference pictures while maintaining current pictures at their original resolution, dividing them into blocks and sub-blocks, and comparing these to the downscaled reference images to generate candidate motion vectors, thereby reducing memory bandwidth usage while enhancing precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If pictures are downscaled prior to generating motion vectors, then memory bandwidth consumption is reduced, but motion vector precision is reduced
Solution Approach 1:
The patent divides the motion search process into two stages: coarse motion vector generation using downscaled pictures, and refinement using full-resolution pictures. This segmentation allows the system to benefit from both reduced memory bandwidth (in the coarse stage) and maintained precision (in the refinement stage), resolving the contradiction between memory efficiency and motion vector accuracy.
Solution Approach 2:
The patent introduces an intermediate refinement stage that takes the coarse motion vectors from downscaled pictures and improves them using full-resolution pictures. This intermediary process acts as a bridge, allowing the system to use downscaled pictures for initial search (reducing memory bandwidth) while still achieving high precision through the refinement step.
2Measurement precision
If full-resolution pictures are used for motion search, then motion vector accuracy is improved, but memory bandwidth consumption increases
Solution Approach 1:
The patent segments the motion search into a coarse search phase using downscaled pictures and a refinement phase using full-resolution pictures. This allows the system to minimize memory bandwidth consumption during the coarse phase while reserving full-resolution picture access only for the refinement phase where high accuracy is critical, thus balancing memory efficiency with motion vector accuracy.
Solution Approach 2:
The patent applies partial action by using downscaled pictures for the initial coarse motion search, which provides sufficient accuracy for many applications without requiring full-resolution pictures. This partial use of full-resolution detail reduces memory bandwidth consumption while maintaining acceptable motion vector accuracy for the majority of cases.
Data Source
AI summary
Reference pictures received via a video signal are downscaled to a specified resolution by a video encoder/decoder. For each current picture being processed by the video encoder/decoder, the current picture is maintained at its original received resolution, but is divided into blocks. Each block is further divided into sub-blocks, and each sub-block is compared, for a set of specified positions, to a corresponding block of the downscaled reference image to generate a set of candidate motion vectors. The candidate motion vectors are scored according to how closely their corresponding sub-block matches the corresponding block of the reference picture at the corresponding position, and a motion vector for each block of the current image is selected based on the scores. The selected motion vectors are used to process (e.g. encode) the video signal.