Length-Based Transform Kernels for High-Resolution Video Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image compression technologies face challenges in efficiently configuring transform sets and determining transform kernel candidates for current blocks, particularly in high-resolution and high-quality image encoding/decoding processes.
Innovation Solution
The method and apparatus utilize length-based transform kernels, including non-trigonometric functions, to derive and apply transform coefficients for current blocks, with specific rules for horizontal and vertical kernels based on block dimensions and thresholds, and assign these kernels to groups of allowable widths or heights.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional transform kernels are used for image compression, then the encoding process is simple, but the compression efficiency and image quality are insufficient for high-resolution images
Solution Approach 1:
The patent changes the functional form of transform kernels from traditional trigonometric functions to non-trigonometric functions (such as polynomial-based kernels). This parameter change in the mathematical representation enables better compression efficiency for high-resolution images while maintaining manageable computational complexity through optimized kernel designs.
Solution Approach 2:
The patent employs composite transform kernel designs that combine multiple mathematical components (e.g., polynomial terms, scaling factors, and adaptive parameters) to create transform kernels with superior compression performance. These composite kernels integrate different functional elements to achieve both high compression efficiency and controlled complexity.
2Productivity
If transform set configuration is simplified, then the encoding process is faster, but the adaptability to different block sizes and types is reduced
Solution Approach 1:
The patent implements dynamic transform set configuration where the selected transform kernel automatically adapts to different block sizes, shapes, and prediction modes. The system dynamically chooses appropriate non-trigonometric transform kernels based on block characteristics, maintaining high encoding speed through pre-defined adaptation rules while achieving versatile adaptability across different image regions.
Solution Approach 2:
The patent performs preliminary classification of transform blocks based on size and type, then pre-selects appropriate non-trigonometric transform kernels for each category. This preliminary action enables fast encoding by avoiding real-time complex selection processes while maintaining adaptability through category-specific optimized kernels.
3Manufacturing precision
If more transform kernel candidates are considered, then the transform performance is improved, but the complexity of determining the optimal kernel increases
Solution Approach 1:
The patent segments the transform kernel selection process into distinct stages: block classification, candidate kernel identification, and optimal kernel selection. By dividing the overall process into manageable segments with clear decision criteria at each stage, the system achieves high transform precision through multiple candidates while controlling complexity through structured segmentation.
Solution Approach 2:
The patent considers a limited set of carefully selected non-trigonometric transform kernel candidates rather than exhaustively evaluating all possible kernels. This partial action approach focuses computational resources on the most promising kernel types for high-resolution imaging, achieving sufficient precision without the complexity of exhaustive search.
Data Source
AI summary
A video decoding method and device disclosed herein may: derive transform coefficients of the current block from a bitstream; perform at least one of inverse quantization or inverse transform on the transform coefficients of the current block and derive residual samples of the current block; and recover the current block on the basis of the residual samples of the current block. Here, the inverse transform is performed on the basis of a length-based transform kernel, and the length-based transform kernel may include at least one of the horizontal transform kernel having the same length as the width of the current block or the vertical transform kernel having the same length as the height of the current block.