Block-Based Image Compression Latent Space Intra Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing end-to-end image and video compression methods process entire images, limiting flexibility in controlling the quality of reconstructed regions and failing to adaptively minimize spatial redundancies at a block level.
Innovation Solution
A block-based end-to-end image and video compression method that partitions images into non-overlapping or overlapping blocks, using latent space intra prediction to reduce spatial redundancy by predicting current blocks from neighboring blocks in the latent space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If end-to-end compression methods process entire images, then compression is simplified, but flexibility in controlling quality of reconstructed regions is limited
Solution Approach 1:
The image is divided into multiple blocks that can be processed independently. Each block can be reconstructed with different quality levels by selectively applying pixel reconstruction only to certain blocks, enabling region-specific quality control while maintaining the simplicity of end-to-end compression.
2Adaptability or versatility
If traditional methods use pre-defined prediction modes, then processing is faster, but adaptability to different content is reduced
Solution Approach 1:
The method uses learned parameters from neural network training to adaptively determine prediction modes and reconstruction strategies for different blocks. The system learns optimal parameters during training that balance adaptability to various content types with processing efficiency, eliminating the need for complex runtime decisions while maintaining high adaptability.
3Productivity
If spatial redundancy is not reduced, then decoding is simpler, but compression efficiency decreases
Solution Approach 1:
The encoder performs latent space prediction and identifies spatial redundancy before encoding. By pre-computing predicted blocks from neighboring regions and encoding only the residuals, the system achieves high compression efficiency while keeping the decoder relatively simple, as it only needs to perform the inverse transformation and selective pixel reconstruction.
Data Source
AI summary
In one implementation, we propose a block-based end-to-end image and video compression method that takes non-overlapping or overlapping split blocks of input images or frames of videos as input. Then, the proposed decoder network reconstructs non-overlapped split blocks of the input. We also introduce an intra prediction method to reduce spatial redundancy in the latent space, i.e., one or more previously decoded latent tensors from neighboring blocks are used as references to predict the current block's latent tensor. Additionally, the decoder can selectively complete the pixel reconstruction process for decoded latent blocks without causing any error drift to neighboring blocks since the prediction is made in the latent space. Enabling and disabling the pixel reconstruction can be signaled by the encoder as metadata in the bitstream or decided at the decoding stage using a computer vision task.