Efficient Image Transmission using Semantic Communication for Static Environment Surveillance

Semantic Communication is a novel approach that focuses on transmitting only essential information, leading to significant reductions in the number of bits required and conserving energy. This paper introduces a novel transmission technique using the semantic communication paradigm. Utilizing the shared knowledge of the static background, the technique involves the extraction of the Region of Interest (RoI), from an image frame, using deep learning frameworks at the transmitter, followed by transmission of the RoI and blending of the RoI onto the background at the receiver. The simulation was executed using the YOLOv3 object detection model and the FastSeg semantic segmentation model for the extraction of RoI. Using a custom dataset, our system demonstrates superior performance compared to existing semantic image transmission systems. Reconstruction Metrics such as Normalized Mean Squared Error (NMSE), Peak Signal-to-Noise Ratio (PSNR), and Mean Structural Similarity Index Measure (MSSIM) show significant improvements. Specifically, using object detection module enhances PSNR by 4.40 dB, while the semantic segmentation module improves it by 10.47 dB. Additionally, we achieve average bit savings of $\mathbf{8 6. 7 2 \%}$ with object detection and $\mathbf{5 4. 4 2 \%}$ with semantic segmentation. Although object detection provides comparable bit savings to the baseline with $\mathbf{9 1. 5 7 \%}$ savings, it avoids the pitfalls of Generative Adversarial Networks (GANs), which often produce poor or unrelated reconstructions due to dataset biases. Considering the balance between bit savings and reconstruction quality, our method proves to be the most effective.