Visual Data Compression Transform Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing visual data compression methods often fail to optimize transform selection for efficient compression, leading to suboptimal storage and transmission efficiency, particularly in cases with locally anisotropic features and varying prediction residuals.
Innovation Solution
A method that computes multiple sets of coefficients using different one-dimensional and two-dimensional transforms over various portions of the data array, choosing the set that achieves maximum compression by selecting the most suitable transform based on the data's characteristics, and encoding these coefficients along with relevant parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single transform is used for all data blocks, then device complexity is reduced, but compression efficiency deteriorates due to inability to adapt to locally anisotropic features
Solution Approach 1:
The data array is divided into multiple sub-arrays or blocks, allowing different transforms to be applied to different segments. This segmentation enables localized optimization where each block can use the most appropriate transform for its specific characteristics, resolving the contradiction between adaptability and complexity by applying complexity only where needed.
Solution Approach 2:
Different transforms are applied to different portions of the data based on local characteristics such as anisotropy. This local quality approach ensures that each region undergoes the transform most suited to its specific features, improving overall compression efficiency without requiring complex global transform selection.
2Productivity
If multiple transforms are computed for each data block, then compression efficiency is improved through optimal transform selection, but loss of time increases due to additional computation
Solution Approach 1:
A set of candidate transforms is pre-computed and prepared before the actual compression process. This preliminary action allows the encoder to quickly select from pre-prepared transforms rather than computing them on-the-fly, reducing the time penalty associated with transform selection while maintaining compression efficiency.
Solution Approach 2:
Instead of computing all possible transforms for every block, the method computes a selective subset of candidate transforms that are most likely to be optimal. This partial action approach reduces computation time while still achieving good compression by focusing resources on the most promising transform options.
3Manufacturing precision
If transforms are applied to cover entire data arrays, then compression is simplified, but manufacturing precision deteriorates due to inability to capture local anisotropic features
Solution Approach 1:
The global transform application is segmented into multiple local transform applications on sub-arrays. Each local transform operates on a smaller portion of the data, allowing it to capture local anisotropic features more effectively while maintaining manageable complexity through modular processing.
Solution Approach 2:
The approach introduces a new dimension of granularity by applying transforms at the block level rather than only at the global array level. This dimensional change in the scope of transform application enables precise local optimization while preserving the overall structure through hierarchical processing.
Data Source
AI summary
Encoding data includes: computing a first set of coefficients based on a plurality of transforms each computed over a different portion of an array of data, and a second set of coefficients based on a transform computed over the array of data; choosing a set of coefficients to represent the array of data from a group of multiple sets of coefficients, the group including the first set of coefficients and the second set of coefficients; and encoding the chosen coefficients and one or more parameters related to the chosen coefficients.