Semantic communication and residual deep learning-driven bitplane coding for scalable image compression and communications

Tools

Samarathunga, Prabhath and Fernando, Thanuj and Ganearachchi, Yasith and Pollwaththage, Nimesh and Fernando, Anil; (2026) Semantic communication and residual deep learning-driven bitplane coding for scalable image compression and communications. In: 2025 IEEE International Conference on Image Processing Workshops (ICIPW). IEEE, USA, pp. 568-573. ISBN 979-8-3315-7799-5 (https://doi.org/10.1109/icipw68931.2025.11385999)

[thumbnail of Samarathunga-etal-2025-Semantic-communication-and-residual-deep-learning-driven-bitplane-coding]
Preview
 Text. Filename: Samarathunga-etal-2025-Semantic-communication-and-residual-deep-learning-driven-bitplane-coding.pdf
Accepted Author Manuscript
License: Creative Commons Attribution 4.0 logo
Download (842kB)| Preview

Abstract

Digital imaging applications which require both reconstruction and semantic accuracy pose challenges to both conventional codecs, which fail to preserve semantics in very noisy conditions, and learned codecs, which fail to achieve reconstruction accuracy even though they preserve semantics. To address this challenge, we propose a semantic autoencoder that preserves high-level image structures in a compact latent representation combined with a progressive bitplane refinement mechanism employing Enhancement Networks and Bitplane Correction Functions to minimize reconstruction errors, which also effectively prevents error propagation across bitplanes while enabling flexible rate adaptation. Experimental results demonstrate significant improvements over conventional image compression codecs, achieving up to 10 dB higher PSNR and 0.12 higher SSIM at low bitrates. In particular, the proposed system reaches quality saturation at just 3.2 bpp, while conventional codecs require 8 bpp to achieve comparable performance. These advances establish a new paradigm for scalable compression, where latent semantics guide iterative refinement.

ORCID iDs

Samarathunga, Prabhath, Fernando, Thanuj, Ganearachchi, Yasith ORCID logoORCID: https://orcid.org/0000-0002-8337-3739, Pollwaththage, Nimesh ORCID logoORCID: https://orcid.org/0009-0002-4822-978X and Fernando, Anil ORCID logoORCID: https://orcid.org/0000-0002-2158-2367;
CORE (COnnecting REpositories)