Sub-pixel Motion Compensation via Shifted DCT Basis Functions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding devices with motion estimation modules struggle with accurate motion compensation at sub-pixel accuracy due to limitations in interpolation filters, which restrict precise motion estimation and compensation, especially when motion vectors have components less than 1/8 of a pixel.
Innovation Solution
A method for coding pixel blocks involves determining an intermediate prediction block larger than the original block by adding pixel lines and columns in the direction of motion, transforming it using a discrete cosine transform, and applying an inverse transform with shifted basis functions to match the motion vector components, allowing for more accurate prediction and residue calculation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional interpolation filters are used for motion compensation, then the device complexity is reduced, but the measurement precision of motion vectors is limited to 1/8 of a pixel
Solution Approach 1:
The patent segments the motion compensation process into multiple stages: first performing motion estimation at coarse precision using simple interpolation, then refining the motion vector components separately. The integer parts and fractional parts of motion vector components are processed through different transformation stages, allowing high precision without requiring a single complex interpolation filter.
Solution Approach 2:
The patent transforms the motion compensation problem from spatial domain interpolation to frequency domain transformation. By using discrete cosine transform (DCT) and its inverse with shifted basis functions, the method achieves sub-pixel precision through parameter transformation rather than spatial interpolation, effectively moving the solution to another dimensional space.
2Measurement precision
If larger interpolation filters are used to improve accuracy, then the measurement precision of motion vectors improves, but the processing time increases
Solution Approach 1:
The method divides the transformation process into separate stages handling integer and fractional parts of motion vectors independently. This segmentation allows each stage to use optimized, computationally efficient transformations rather than requiring a single large, time-consuming interpolation filter.
Solution Approach 2:
The patent changes the parameters of the transformation basis functions (shifting the DCT basis functions by fractional amounts) rather than changing the spatial extent of interpolation filters. This parameter transformation achieves higher precision without increasing the computational complexity proportional to filter size.
3Manufacturing precision
If precise motion compensation at sub-1/8 pixel accuracy is achieved, then the manufacturing precision of image coding improves, but the device complexity increases
Solution Approach 1:
The patent uses a universal discrete cosine transform framework that can handle both integer-pixel and sub-pixel motion compensation through parameter adjustment. The same DCT basis functions are used across different precision requirements, with only the shift parameters changing, making the device multi-functional without requiring separate complex modules for each precision level.
Solution Approach 2:
The method performs preliminary motion estimation at integer pixel accuracy, then applies refined transformations only to the fractional parts of the motion vector components. This preliminary action separates the coarse and fine adjustment stages, allowing the complex high-precision transformation to be applied only where necessary rather than to the entire image processing pipeline.
4Reliability
If traditional interpolation methods are used, then the ease of operation is maintained, but the reliability of motion compensation decreases at high sub-pixel accuracy
Solution Approach 1:
The patent uses inverse discrete cosine transform to reconstruct the prediction block from transformed coefficients, creating a precise copy of the motion-compensated data. This copying mechanism through mathematical transformation ensures reliability by maintaining exact relationships between original and transformed data, avoiding the approximations inherent in traditional interpolation methods.
Data Source
AI summary
A method for coding a block of pixels is described. The coding method comprises:determining a prediction block for the pixel block from a motion vector;calculating a residue between the pixel block and the prediction block; andcode the residue.The determination of the prediction block comprises the following steps:determining an intermediate prediction block of size strictly greater than the size of the pixel block;transforming the intermediate prediction block into a first block transformed with a first transform; andtransforming the first transformed block into a second block transformed with a second transform of inverse the first transform whose basis functions are shifted by at least a part of each component of the motion vector, the prediction block being extracted from the second transformed block.