Image Encoding Sub-Area Motion Vector Derivation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image encoding devices face low prediction accuracy and decreased encoding efficiency when using merge or Affine modes for inter prediction, particularly at borders between objects, due to reliance on single or fixed motion vectors.
Innovation Solution
An image encoding device that divides blocks into sub-areas, determines reference directions, derives motion vectors by extrapolating from neighboring blocks, and encodes reference direction information, allowing for improved inter prediction accuracy without increasing information overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If merge mode is used to reduce motion vector information, then encoding efficiency improves, but prediction accuracy decreases for complicated motion and border portions
Solution Approach 1:
The patent divides the encoding-target CU into multiple sub-areas (e.g., 4×4 pixel regions within a larger CU) and derives separate motion vectors for each sub-area based on neighboring CUs. This segmentation allows the system to capture spatial variations in motion across different regions of the same CU, thereby improving prediction accuracy for complicated motion patterns and border portions while maintaining the efficiency benefits of merge mode.
Solution Approach 2:
The patent applies different motion vector derivation strategies to different sub-areas within a CU. By determining reference directions and deriving motion vectors locally for each sub-area rather than using a single motion vector for the entire CU, the system adapts the prediction to local motion characteristics. This local quality approach enables accurate prediction at object borders and in regions with non-uniform motion while keeping the overall encoding efficient.
2Measurement precision
If Affine mode is used to improve prediction for rotation motion, then prediction accuracy improves, but encoding efficiency decreases due to increased information requirements
Solution Approach 1:
Instead of using a full Affine model that requires multiple parameters and significant information for each CU, the patent segments the CU into sub-areas and derives motion vectors for each sub-area independently from neighboring CUs. This approach achieves variable motion prediction (similar to Affine mode's capability to handle rotation and non-uniform motion) without the heavy information overhead, as each sub-area's motion vector is derived efficiently from available neighbor data.
Solution Approach 2:
The patent changes the approach from using a fixed set of Affine parameters (which require significant bitrates) to using a variable number of motion vectors derived from neighboring CUs. By adjusting which neighboring CUs are referenced and how motion vectors are interpolated for each sub-area, the system achieves flexible motion modeling with lower information cost, effectively changing the parameter representation to balance accuracy and efficiency.
3Measurement precision
If encoding is performed by finely dividing CUs to enhance prediction accuracy at object borders, then prediction accuracy improves, but information amount increases and encoding efficiency decreases
Solution Approach 1:
The patent applies local quality by deriving motion vectors specifically for sub-areas where they are most needed (such as border portions and regions with complicated motion) while using a coarser approach for uniform regions. By determining reference directions and deriving motion vectors locally rather than uniformly across the entire image, the system improves prediction accuracy at critical locations without unnecessarily increasing the overall information amount. The motion vector derivation is adapted to local characteristics rather than applying a fixed high-resolution approach everywhere.
Data Source
AI summary
An image encoding device that divides an image into blocks and encodes on each of the blocks, the image encoding device includes: a sub-area divider configured to divide a target block to be encoded into a plurality of sub-areas; a reference direction determiner configured to determine a reference direction that indicates in which direction an encoded block to be referenced is located among encoded blocks around the target block; a motion vector deriver configured to derive a motion vector for each of the sub-areas by referencing a motion vector applied in inter prediction of an encoded block located in the determined reference direction; and a prediction image generator configured to generate a prediction image on the target block by performing inter prediction on each of the sub-areas by using the derived motion vector. The image encoding device configured to encode reference direction information that indicates the determined reference direction, and output a stream.