Ordered Hierarchical Encoding for Robust Image Semantic Communication

doi:10.1109/LCOMM.2025.3642888

Ordered Hierarchical Encoding for Robust Image Semantic Communication

2025-12-26

0

PRE

AI

DOI：10.1109/LCOMM.2025.3642888

Original

Original request for help

Save

Abstract

En 中文

Current semantic communication methods encode all features uniformly, however, downstream models are far more sensitive to distortions in high-frequency semantic features than in low-frequency ones. To address this issue, we propose an Ordered Hierarchical Encoding (OHE) framework for semantic image transmission. OHE employs a hierarchical encoding scheme that combine cascaded pooling with cross-attention to construct multi-scale semantic representations, effectively decoupling and extracting low-frequency and high-frequency features. Furthermore, an ordered representation mechanism with random prefix masking enforces a natural prioritization of semantic information, enabling progressive reconstruction and supporting simple, flexible rate control. Extensive experiments conducted on various datasets demonstrate that our method outperforms the baselines.

Keywords:

Semantic communication

ordered hierarchical encoding

multi-scale

ordered representation

Ordered Hierarchical Encoding for Robust Image Semantic Communication

Abstract

Journal

Organization