Diffusion based scalable semantic communication framework for image compression and transmission

Tools

Gamage, Namesh Pannigala and Fernando, Thanuj and Fernando, Anil; (2026) Diffusion based scalable semantic communication framework for image compression and transmission. In: 2025 IEEE International Conference on Image Processing Workshops (ICIPW). IEEE, USA, pp. 657-662. ISBN 979-8-3315-7799-5 (https://doi.org/10.1109/icipw68931.2025.11385841)

[thumbnail of Gamage-etal-2025-Diffusion-based-scalable-semantic-communication-framework]
Preview
 Text. Filename: Gamage-etal-2025-Diffusion-based-scalable-semantic-communication-framework.pdf
Accepted Author Manuscript
License: Creative Commons Attribution 4.0 logo
Download (1MB)| Preview

Abstract

Semantic communication for image transmission enhances efficiency by focusing on meaning rather than pixel accuracy. However, existing methods struggle to preserve semantic integrity due to challenges in identifying key content and contextual relationships. To address this, we propose a scalable semantic communication model that prioritizes object-based semantic extraction and dynamic weighting to enhance image representation. The framework detects and extracts objects within an image, assigning significance through a weight map that emphasizes key elements while de-emphasizing irrelevant details. The system dynamically adjusts weights in a weight map for scalable image transmission, adapting to network infrastructure constraints like bandwidth and latency. This ensures only the most relevant data is sent. The weighted semantic representation feeds into a diffusion-based reconstruction model, regenerating high-fidelity images at low bit rates. Experimental results show this approach significantly improves image representation while minimizing transmission overhead, scaling effectively with network conditions. The ability to adaptively control semantic granularity makes it ideal for real-world applications, paving the way for efficient, context-aware image transmission with improved resource utilization and semantic preservation.

ORCID iDs

Gamage, Namesh Pannigala, Fernando, Thanuj and Fernando, Anil ORCID logoORCID: https://orcid.org/0000-0002-2158-2367;
CORE (COnnecting REpositories)