An efficient adaptive compressive sensing technique for underwater image compression in IoUT

In recent years, internet of underwater things (IoUT) has become a hot topic of research and attracted many researchers to use them for applications such as underwater environment monitoring and exploration, monitoring geological and biological changes, defense, disaster monitoring and prevention. Smart underwater objects in IoUT are generally equipped with high definition cameras for capturing underwater images. Storage and transmission of high-quality images captured by these smart objects is a challenging task due to huge volumes of data generated. Compressive sensing (CS) based data sampling and acquisition strategy provides promising ways to compress image at low computational cost and power. However, the samples are chosen randomly compromising reconstructed image's visual quality. To select signal samples with more importance, Adaptive Block Compressive Sensing (ABCS) can be utilized. In this research paper, we implement two-modified fast haar wavelet transform (MFHWT) coefficient thresholding based ABCS strategies for compression of underwater surveillance images in IoUT. The novelty lies in the fact that the proposed strategies choose only 5-20% of samples/measurements to achieve a rise in PSNR of about 3-5dB with respect to conventional techniques. The results show that the proposed strategies have NCC and SSIM values closer to 1 and NAE values closer to 0. Also, around 70% memory space can be saved.

Automated summary

In recent years, Internet of Underwater Things (IoUT) has received significant attention for applications such as underwater environment monitoring and exploration, defense, disaster monitoring, etc. This research paper proposes two-modified fast Haar wavelet transform (MFHWT) coefficient thresholding based Adaptive Block Compressive Sensing (ABCS) strategies for compressing underwater surveillance images in IoUT. The proposed strategies achieve a rise in PSNR of about 3-5dB with respect to conventional techniques, while also saving around 70% memory space.