ROI-Based Image Transmission With Channel Feedback for Semantic Communications

We propose an image transmission scheme for semantic communications, by exploiting multi-round channel feedback and region of interest (ROI) in images. First, a multi-round channel feedback regime is proposed. Specifically, each image is encoded and transmitted in multiple rounds. At each round, received symbols are transmitted back to the transmitter for generating coded symbols at the subsequential rounds. Then, to reduce communication overheads, ROI is utilized to allocate more coded symbols to interested regions by determining the forward and backward of the corresponding symbols. Finally, extensive experiments demonstrate the advantages of our proposed scheme over other traditional schemes.