Image Encoding via Invertible Neural Network Latent Transformation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding technologies face challenges in efficiently compressing image data due to limited network bandwidth, necessitating the development of high-level video coding techniques.
Innovation Solution
The proposed solution involves an image encoding method that utilizes an invertible neural network to transform an image block into a first latent representation, followed by transformation into a second latent representation using a non-invertible neural network. This method estimates a probability distribution of the first latent representation based on the second latent representation and performs entropy encoding accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional image compression methods are used, then network bandwidth is consumed, but compression efficiency is insufficient
Solution Approach 1:
The patent transforms image data through neural networks to change its representation parameters, converting the image into latent representations with different statistical properties that enable more efficient compression while maintaining quality
Solution Approach 2:
The patent replaces traditional mechanical compression algorithms with neural network-based transformation, using learned representations to achieve superior compression efficiency and reduce bandwidth requirements
2Manufacturing precision
If lossless compression is achieved, then data quality is maintained, but compression ratio is limited
Solution Approach 1:
The patent segments the compression process into multiple stages: first transforming the image through an invertible neural network to latent representations, then applying additional transformation through a non-invertible neural network, finally performing entropy encoding. This multi-stage segmentation enables both lossless compression and high compression ratio
Solution Approach 2:
The patent transforms the image from its original pixel dimension to a latent representation dimension through neural network transformations, creating a new dimensional space where compression can be achieved more efficiently while maintaining data quality through the invertible transformation
3Productivity
If neural network transformation is applied, then compression ratio increases, but computational complexity increases
Solution Approach 1:
The patent performs preliminary transformation of the image through neural networks before the actual compression encoding. By pre-transforming the image into latent representations with favorable statistical properties, the subsequent compression steps become more efficient and less computationally intensive
Data Source
AI summary
There is provide an image encoding method including transforming an image block into a first latent representation based on an invertible neural network, transforming the first latent representation into a second latent representation based on a non-invertible neural network, estimating a probability distribution of the first latent representation, and performing entropy encoding on the first latent representation based on the probability distribution by using an entropy encoder.