Lapped Biorthogonal Transform Vectorization for Codec Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Block transform-based coding methods, such as those used in image and video compression, often introduce perceptible artifacts like blockiness due to quantization of high-frequency components, which are not tolerable in many applications.
Innovation Solution
The technique reformulates lapped biorthogonal transforms into SIMD-friendly operations by remapping input and output sampling grids, allowing for vectorization of data and parallel processing, thereby reducing computational complexity and minimizing artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of substance
If block transform-based coding is used for compression, then data compression efficiency is improved, but perceptible artifacts like blockiness are introduced
Solution Approach 1:
The patent divides the transform process into multiple stages: overlapping block transforms are applied to segmented portions of the input signal, followed by a synthesis stage that combines the results. This segmentation approach allows compression while reducing blockiness artifacts through the overlapping and synthesis operations.
Solution Approach 2:
The patent introduces an intermediary synthesis stage that processes the outputs from multiple overlapping block transforms. This intermediary stage combines the transformed blocks using weighted averaging or other synthesis techniques, acting as a mediator that reduces the harmful blockiness artifacts while preserving the compression benefits.
2Object-affected harmful factors
If traditional lapped transform is used to reduce artifacts, then perceptual quality is improved, but computational complexity increases
Solution Approach 1:
The patent segments the computational process into independent parallel block transform operations that can be executed simultaneously. Each block transform operates on a localized portion of the signal, allowing for parallel processing that reduces overall computational complexity compared to traditional sequential lapped transform implementations.
Solution Approach 2:
The patent processes multiple overlapping blocks simultaneously in a parallel dimension, rather than sequentially processing a single lapped transform. This dimensional change from sequential to parallel processing reduces computational complexity while maintaining artifact reduction through the overlapping block approach.
3Measurement precision
If reversible block transform is used for lossless compression, then data fidelity is improved, but transform choice is constrained
Solution Approach 1:
The patent creates a universal transform framework that can accommodate multiple different block transform types (DCT, DST, Hadamard, etc.) while maintaining reversibility. The overlapping block structure with synthesis stage provides a multi-functional platform that supports various transform choices, enabling both lossless and lossy compression modes with flexible transform selection.
Solution Approach 2:
The patent allows dynamic adjustment of transform parameters including block size, overlap extent, and transform type selection. These parameter changes enable the system to adapt between different compression modes (lossless/lossy) and optimize for different content types while maintaining the reversible transform property when needed.
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
A block transform-based digital media codec achieves faster performance by re-mapping components of the digital media data into vectors or parallel units on which many operations of the transforms can be performed on a parallel or single-instruction, multiple data (SIMD) basis. In the case of a one-dimensional lapped biorthogonal transform, the digital media data components are re-mapped into vectors on which butterfly stages of both overlap pre-/post-filter and block transform portions of the lapped transform can be performed on a SIMD basis. In the case of a two-dimensional lapped biorthogonal transform, the digital media data components are re-mapped into vectors on which a Hadamard operator of both overlap pre-/post-filter and block transform can be performed on a SIMD basis.