A Robustness and Low Bit-Rate Image Compression Network for Underwater Acoustic Communication

Image compression algorithm is an important technology in the process of image transmission. Algorithm faces more difficult challenges in underwater acoustic communication. Images are required to be transmitted at a low bit-rate due to the limited underwater bandwidth and the noisy underwater acoustic environment will cause errors like random bit flip or packet loss. Therefore, the performance of common compression algorithms (JPEG, BPG, etc.) will be greatly reduced. Based on deep neural network (DNN), we propose an image compression algorithm that compresses the image texture and color separately for reducing the bit-rate. Moreover, we simulate the underwater acoustic environment and add different types of errors to compressed bit codes in our training process. Extensive experiments show that this training method improves the robustness of decoder and reconstruction performance. Besides, the algorithm is better than common compression algorithms and DNN based algorithms for underwater acoustic communication.