An Efficient and Robust Underwater Image Compression Scheme Based on Autoencoder

Jiahui Liu,Fei Yuan,Chang Xue,Zhenyu Jia,En Cheng
DOI: https://doi.org/10.1109/joe.2023.3249243
IF: 3.883
2023-01-01
IEEE Journal of Oceanic Engineering
Abstract:Images occupy a prominent place in data because they are more visually appealing than sounds or texts. Acoustic waves are the only feasible alternative for long-distance underwater transmission since seawater has a high absorption impact on lighting and electromagnetic signals. However, underwater acoustic (UWA) communication technology can only provide relatively limited bandwidth (low effectiveness) and insufficiently stable links (low reliability). As a result, it is challenging to send high-resolution underwater images via the UWA channel. This article proposes an effective and robust underwater image compression scheme. First, an autoencoder is used for underwater image extreme bit rate compression. Then, a multistep training strategy is proposed to improve the robustness of the decoder by gradually learning channel degradation features. Finally, the autoencoder encodes images in two paths to achieve efficient compression and higher image quality when reconstructing images. The main path completes the compression task with a low bit rate and high robustness, while the branching path implements the image block retransmission compensation through the feedback signal. The experimental results demonstrate that the content of the reconstructed image can still be recognized under the conditions of a compression ratio of up to 1/768 and an average bit error rate of up to $ \text{10}^{-1}$ . The joint multistep training strategy and multidescription coding achieve a low bit rate and high robustness for underwater image communication, which has good application prospects.
What problem does this paper attempt to address?