Bandwidth-Adaptive Semantic Communication System with Non-ROI Image Compression
Chao Wu, Celimuge Wu, Jiale Wu, Yangfei Lin, Zhaoyang Du, Ha Si, Wugedele Bao · 2024
Semantic communication systems aim to reduce data transmission by extracting semantic information. However, under bandwidth-constrained conditions, users may still experience significant delays, which negatively impacts the Quality of Service (QoS) and Quality of Experience (QoE). To address this issue, we propose a bandwidth-adaptive semantic communication system that leverages the varying importance of different regions within an image. Specifically, an image is divided into regions of interest (ROI) and non-regions of interest (Non-ROI), where Non-ROI typically consists of less critical background elements, such as the sky, ground, or walls. By compressing Non-ROI more efficiently, we can alleviate bandwidth limitations without compromising the quality of the crucial content. The proposed system employs an instance segmentation model to accurately identify ROI and Non-ROI, followed by coarse-grained bandwidth adaptation and fine-grained bandwidth adaptation methods to dynamically adjust the compression level of Non-ROI based on available bandwidth. Experimental results demonstrate that the proposed system improves image compression rates, achieving up to 16fold compression under low bandwidth conditions (11–15 Mbps), while maintaining high image quality with a peak signal-to-noise ratio (PSNR) of up to 29.89.