Graph Weight Prediction for Lower-Bitrate Image Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing graph-based image encoding and decoding methods face challenges in efficiently compressing graph weights, leading to high bitrate and reconstruction artifacts, especially when dealing with natural images, as they require transmitting or storing large weights matrices and coefficients vectors, which is not feasible in bandwidth-constrained channels or limited storage environments.
Innovation Solution
The method predicts graph weights for each block from reconstructed pixel samples of adjacent blocks, using directional graph weight prediction modes, allowing for effective encoding without transmitting graph weights as side information, and combines this with existing graph-based intra-prediction techniques to achieve coding gains.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If graph-based transforms are used for image compression, then coding efficiency is improved, but the overhead of graph transmission increases bitrate
Solution Approach 1:
The patent extracts only the essential graph structure information (adjacency relationships) from the full graph representation, transmitting only the connectivity pattern rather than complete graph data including weights and all structural properties. This selective extraction maintains the transform's ability to exploit image correlations while dramatically reducing the amount of graph data that must be transmitted to the decoder.
Solution Approach 2:
Instead of transmitting graph weights and properties and having the decoder reconstruct them, the patent inverts the approach by having the decoder generate the graph structure locally using the same deterministic algorithm that the encoder uses. This eliminates the need to transmit graph definition data, as both encoder and decoder independently create identical graph representations from the image data itself.
2Measurement precision
If graph weights are transmitted as side information, then transform accuracy is improved, but bitrate increases
Solution Approach 1:
The patent makes the graph structure self-generating at the decoder side. Rather than receiving pre-computed graph weights as side information, the decoder autonomously reconstructs the exact same graph structure and weight values that the encoder used, by applying the identical deterministic algorithm to the decoded image data. This self-service approach ensures transform accuracy without requiring additional bitrate for weight transmission.
Solution Approach 2:
The patent changes the fundamental parameter being transmitted from graph weights (continuous values requiring high precision) to a simple adjacency structure indicator (discrete topology information). By transforming what needs to be communicated from precise numerical weights to topological relationships, the system maintains transform accuracy while dramatically reducing the information that must be transmitted.
3Reliability
If complete graph representation is used, then transform performance is improved, but storage requirements increase
Solution Approach 1:
The patent performs preliminary graph structure generation at the encoder side before compression, using the image data itself to create the graph representation. This pre-computed structure is then used for the transform, ensuring optimal performance. The key insight is that the same preliminary action is performed at the decoder, eliminating the need to store or transmit the pre-computed graph structure, as it is regenerated when needed.
4Productivity
If adaptive graph transforms are used, then compression gain is improved, but complexity of graph description increases
Solution Approach 1:
The patent segments the graph representation into essential topological components (adjacency relationships) that can be efficiently encoded and transmitted. By dividing the complex graph description into fundamental connectivity elements rather than attempting to transmit the complete graph object with all its properties, the system achieves adaptive transform performance with manageable description complexity.
Data Source
AI summary
A method for encoding digital images or video streams, includes a receiving phase, wherein a portion of an image is received; a graph weights prediction phase, wherein the elements of a weights matrix associated to the graph related to the blocks of the image (predicted blocks) are predicted on the basis of reconstructed, de-quantized and inverse-transformed pixel values of at least one previously coded block (predictor block) of the image, the weights matrix being a matrix comprising elements denoting the level of similarity between a pair of pixels composing said image, a graph transform computation phase, wherein the graph Fourier transform of the blocks of the image is performed, obtaining for the blocks a set of coefficients determined on the basis of the predicted weights; a coefficients quantization phase, wherein the coefficients are quantized an output phase wherein a bitstream comprising the transformed and quantized coefficients is transmitted and/or stored.