Content-Style Data Compression for High-Ratio Reversible Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image and document compression systems are limited in achieving high compression ratios while maintaining decompressibility without significant loss, and they are not suitable for tasks beyond storage and transmission.
Innovation Solution
A data compression apparatus that uses an encoder trained with grouped training data to separate content and style factors, allowing for highly compressed representations that can be manipulated and decompressed to produce similar outputs, leveraging machine learning technology with neural networks and stochastic variational inference.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional compression systems are used to compress images or documents, then the data can be stored and transmitted efficiently, but the compressed data is not suitable for tasks other than storage and transmission and cannot be manipulated to generate new variations
Solution Approach 1:
The patent segments the compressed representation into distinct components: content factors (describing what the data depicts) and style factors (describing how the data appears). This segmentation allows independent manipulation of content and style, enabling versatile applications such as generating variations with different styles while preserving content, or modifying content while maintaining style consistency.
2Quantity of substance
If conventional compression systems compress data to a certain extent, then storage and transmission efficiency is improved, but the amount of compression that can be achieved is limited while still enabling decompression without significant loss
Solution Approach 1:
The patent changes the parameter representation from conventional pixel-by-pixel or block-by-block encoding to a semantic parameter space with content and style factors. This allows for higher compression ratios because the compressed representation captures essential semantic information rather than all visual details, while still enabling accurate decompression by reconstructing data from these semantic parameters.
3Adaptability or versatility
If machine learning technology with neural networks is used to separate content and style factors, then highly compressed representations can be achieved with manipulation capability, but the device complexity increases
Solution Approach 1:
The patent applies preliminary action by training the neural network encoder offline using grouped training data before deployment. During actual compression operations, the pre-trained encoder directly separates content and style factors without requiring complex real-time computations. This preliminary training phase handles the complexity, while the operational phase remains relatively simple and efficient.
Data Source
AI summary
A data compression apparatus is described which has an encoder configured to receive an input data item and to compress the data item into an encoding comprising a plurality of numerical values. The numerical values are grouped at least according to whether they relate to content of the input data item or style of the input data item. The encoder has been trained using a plurality of groups of training data items grouped according to the content and where training data items within individual ones of the groups vary with respect to the style. The encoder has been trained using a training objective which takes into account the groups.