Feature-Driven Semantic Communication for Efficient Image Transmission

Area of Science:

• Computer Science
• Electrical Engineering
• Information Theory

Background:

• Semantic communication enhances transmission efficiency by conveying information's core meaning.
• Current image semantic communication systems often use uniform feature compression, neglecting feature importance for recovery.
• Bandwidth limitations are a critical challenge in real-world image transmission.

Purpose of the Study:

• To propose a novel semantic communication system for image transmission that addresses bandwidth constraints.
• To introduce non-uniform quantization techniques for differential feature processing.
• To improve image recovery quality under limited bandwidth.

Main Methods:

• Developed a semantic communication system incorporating non-uniform quantization.
• Implemented a dynamic bit allocation algorithm for features based on their contribution to receiver-end tasks.
• Varied quantization levels according to feature importance for image reconstruction.

Main Results:

• The proposed system significantly reduces bandwidth requirements compared to uniform processing.
• Image reconstruction quality is maintained or improved under bandwidth constraints.
• Non-uniform feature processing demonstrates superior performance over uniform methods.

Conclusions:

• Non-uniform quantization is an effective strategy for semantic image communication under bandwidth limitations.
• Dynamic bit allocation based on feature importance optimizes data transmission.
• The developed system offers a practical solution for efficient and high-quality image transmission in constrained environments.