Unified Convolution Engine for Video Intra Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding technologies face challenges in achieving improved compression ratios with minimal sacrifice in picture quality, particularly in streaming and storage due to the substantial amount of video data and limited bandwidth or memory resources.
Innovation Solution
The method involves a unified directional intra prediction with affine linear weighted intra prediction (ALWIP) that aligns the accuracy of multiplication operations, allowing for a unified convolution step and eliminating hardware redundancy by using adaptive coefficient sets and matrix-based operations, enabling efficient processing of both directional and ALWIP methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If separate hardware is used for directional intra prediction and ALWIP, then each method can be processed independently, but hardware redundancy increases and processing efficiency decreases
Solution Approach 1:
The patent merges the separate hardware processing paths for directional intra prediction and ALWIP into a unified convolution operation. By aligning the coefficient precision (both using 6-bit coefficients) and using a single convolution engine that can handle both prediction types, the system eliminates redundant hardware components while maintaining the ability to process both methods independently when needed.
Solution Approach 2:
The convolution engine is designed with universal functionality to handle both directional intra prediction and ALWIP operations. The same hardware block can be configured to perform different prediction types by loading appropriate coefficient sets, making the hardware multi-functional and eliminating the need for separate dedicated hardware for each prediction method.
2Measurement precision
If different precision coefficients are used for directional intra prediction and ALWIP, then each method can maintain its optimal accuracy, but hardware complexity increases due to multiple coefficient sets
Solution Approach 1:
The patent changes the parameter of coefficient precision to be uniform (6-bit) for both directional intra prediction and ALWIP. This parameter change allows both methods to operate with the same precision level while using a single coefficient storage structure, thereby reducing hardware complexity without significantly compromising the accuracy of either prediction method.
3Quantity of substance
If video data is compressed more aggressively to reduce data size, then bandwidth and storage requirements decrease, but picture quality deteriorates
Solution Approach 1:
The patent employs parameter changes in the form of adaptive coefficient selection and precision alignment to improve compression efficiency. By optimizing the convolution operation and eliminating hardware redundancy, the system reduces the overall bitrate required for encoding while maintaining prediction accuracy, thereby achieving better compression ratios without significant picture quality loss.
Data Source
AI summary
A method of intra prediction of a block, comprising obtaining two lines of reconstructed neighboring samples; deriving a set of reference samples based on the two lines of reconstructed neighboring samples; obtaining a set of MIP coefficients based on a intra prediction mode obtained from a bitstream, wherein a MIP coefficient CMIP of the set of MIP coefficients is obtained based on CMIP=vsgn·(q<<s), where q is a magnitude of the MIP coefficient, wherein s is a left shift value and vsgn is a sign value of the MIP coefficient; and obtaining a prediction block based on the set of reference samples and the set of MIP coefficients, wherein a reconstruct picture is obtained based on the prediction block.


