Rounding Non-Integer Motion Vectors for Video Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The high complexity in selecting an optimal motion vector from a candidate motion vector set in video compression, particularly due to the need for numerous sub-pixel interpolation operations in non-integer pixel accuracy motion vectors, hampers coding efficiency.
Innovation Solution
A method and apparatus that round the coordinate value of non-integer pixel position motion vectors by determining the nearest integer pixel position based on equal distances to adjacent integer positions, using the sign of the coordinate value to decide between two possible integer positions, thereby reducing the complexity of motion vector selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If non-integer pixel accuracy motion vectors are used to improve inter-frame prediction accuracy, then prediction precision is improved, but the complexity of motion vector selection increases due to numerous sub-pixel interpolation operations
Solution Approach 1:
The patent segments the motion vector selection process into two stages: first selecting candidate motion vectors from a reduced set with lower precision (integer or half-pixel), then performing detailed evaluation only on these candidates. This segmentation reduces the total number of sub-pixel interpolation operations while maintaining prediction accuracy for the final selected motion vector.
Solution Approach 2:
The patent applies different precision levels to different parts of the motion vector selection process. Candidate motion vectors are generated with lower precision to reduce computational complexity, while the final optimal motion vector is determined with high precision through sub-pixel interpolation. This local quality differentiation optimizes the balance between accuracy and complexity.
2Reliability
If non-integer pixel position motion vectors are used, then coding performance is improved, but the number of bits required to represent motion vectors increases
Solution Approach 1:
The patent changes the precision parameter of motion vectors dynamically based on the coding context. Instead of using high-precision non-integer pixel motion vectors throughout, the system adjusts the precision level (integer, half-pixel, or full sub-pixel) according to motion complexity and prediction requirements, thereby reducing the average bit rate while maintaining coding performance.
3Measurement precision
If sub-pixel interpolation operations are performed for each candidate motion vector, then motion vector accuracy is improved, but processing time and computational load increase
Solution Approach 1:
The patent performs preliminary action by pre-selecting candidate motion vectors from a reduced set before performing computationally intensive sub-pixel interpolation operations. This preliminary filtering based on lower-precision vectors eliminates many candidates that would not be optimal, thereby reducing the total number of interpolation operations required and improving coding efficiency.
Data Source
AI summary
The present disclosure provides a method and an apparatus for rounding a coordinate value of a non-integer pixel position motion vector. The method includes: rounding a coordinate value of a non-integer pixel position motion vector, which includes: for each dimension of coordinates of the non-integer pixel position motion vector, when a coordinate in the dimension is a non-integer pixel position, rounding the coordinate value of the non-integer pixel position motion vector in the dimension to one of integer pixel coordinate positions A and B if distances from the coordinate of the non-integer pixel position motion vector in the dimension to the integer pixel coordinate positions A and B adjacent to the non-integer pixel position in the dimension are the same, where an integer pixel coordinate position is determined by a sign of the coordinate value of the non-integer pixel position motion vector in the dimension.