Graph Transform Image Encoding Using Edge Map Transmission
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing graph-based image encoding and decoding methods face challenges in efficiently compressing the weights matrix, leading to high bitrate requirements and reconstruction artifacts, as they struggle to effectively compress the weights graph, which is crucial for real-world applications with bandwidth constraints.
Innovation Solution
The method involves quantizing the weights matrix elements using a non-linear function and transmitting an edge map instead of the weights matrix, allowing for reduced bit representation and further compression using entropy coding techniques, while exploiting inter-pixel prediction to enhance compression efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If graph-based transforms are used for image compression, then compression efficiency is improved, but the overhead of graph transmission increases bitrate
Solution Approach 1:
The patent extracts only the essential information from the full weights matrix by identifying and transmitting only the edge map (binary representation of significant edges) rather than the complete weights matrix. This extraction principle reduces the transmitted data from O(N^2) elements to O(N) elements, directly resolving the contradiction between compression efficiency and transmission overhead
Solution Approach 2:
Instead of transmitting the weights matrix and deriving the edge map at the decoder, the patent inverts the approach by directly transmitting the edge map and reconstructing the weights matrix at the decoder. This inversion eliminates the need to transmit redundant weight information while preserving the essential graph structure needed for transformation
2Measurement precision
If the weights matrix is transmitted in full, then transformation accuracy is maintained, but bitrate requirements increase significantly
Solution Approach 1:
The patent uses a simplified binary edge map representation instead of the full-precision weights matrix. The edge map serves as a disposable, low-bitrate approximation that contains sufficient information for accurate transformation, eliminating the need to transmit high-precision weight values while maintaining transformation accuracy
Solution Approach 2:
The patent changes the parameter representation from continuous weight values (requiring high bitrate) to discrete binary edge indicators (requiring low bitrate). This parameter transformation from O(N^2) real-valued weights to O(N) binary edge markers maintains the essential graph topology information while dramatically reducing bitrate requirements
3Quantity of substance
If binary unweighted graphs are used for compression, then bitrate is reduced, but encoding noise increases
Solution Approach 1:
The patent performs preliminary action by computing the edge map from the weights matrix at the encoder before transmission. This pre-computation of the binary edge representation captures the essential graph structure in advance, allowing the decoder to reconstruct an accurate weights matrix without introducing encoding noise, thus maintaining quality while reducing bitrate
Data Source
Figure 1
Figure 2(a)~2(b)
Figure 3(a)~3(b)
AI summary
The present invention relates to methods and apparatuses for encoding and/or decoding digital images or video streams, wherein the encoding apparatus (1100) comprises processing means (1110) configured for reading a portion of said image (f), computing difference values between pixel values of the image, quantizing such pixel difference values for obtaining a quantized weight map (W), computing an edge map (f') composed by elements (f'i) indicating whether a corresponding pixel of said portion of the image is an edge or not on the basis of the quantized weight map, determining a reconstructed weight map (W') on the basis of the edge map (f'), determining a graph transform matrix (U) on the basis of the reconstructed weight map (W'), computing transform coefficients (f^) on the basis of the graph transform matrix (U) and said portion of the image (f), transmitting the computed transform coefficients (f^) and the edge map (f').